149
Cost assessment proposal Chapter 7: Supplementary document Document Reference: S6002 In this report, we provide detailed evidence supporting our approach to assessing efficient costs for the wholesale and retail price controls for AMP7. This includes proposed cost models, and triangulation between models for base costs, and other assessment methods for enhancement costs and other policy items. It also explains our proposals for setting an efficient benchmark cost, and for predictions of forward looking dynamic efficiency. United Utilities Water Limited

Chapter 7: Supplementary Document - S6002 · Reckon, and Equifax) to better understand the drivers of cost, and developed alternative variables. • We have developed a robust approach

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Cost assessment proposal Chapter 7: Supplementary document

Document Reference: S6002

In this report, we provide detailed evidence supporting our approach to assessing efficient costs for the wholesale and retail price controls for AMP7. This includes proposed cost models, and triangulation between models for base costs, and other assessment methods for enhancement costs and other policy items. It also explains our proposals for setting an efficient benchmark cost, and for predictions of forward looking dynamic efficiency.

United Utilities Water Limited

Chapter 7: Supplementary Document - S6002 unitedutilities.com

Contents Executive summary ................................................................................................................................................. 4

Wholesale cost assessment ................................................................................................................................ 4 Residential Retail cost assessment ..................................................................................................................... 5

1 Wholesale Cost Assessment .......................................................................................................................... 6 1.1 Introduction .......................................................................................................................................... 6

1.1.1 Objective and structure of this document ........................................................................................ 7 1.2 Econometric benchmarking within cost assessment ............................................................................ 8

1.2.1 Data and modelled costs .................................................................................................................. 9 1.2.2 Aggregated or disaggregated botex modelling? ............................................................................. 11 1.2.3 Model assessment framework and selection criteria ..................................................................... 15 1.2.4 Modelling costs with small datasets ............................................................................................... 18

1.3 Water botex modelling ....................................................................................................................... 19 1.3.1 Cost drivers and explanatory factors .............................................................................................. 19 1.3.2 Summary of Water botex modelling results ................................................................................... 26

1.4 Wastewater botex modelling .............................................................................................................. 31 1.4.1 Cost drivers and explanatory factors .............................................................................................. 31 1.4.2 Summary of Wastewater modelling results ................................................................................... 38

1.5 Assessing enhancement expenditure ................................................................................................. 41 1.5.1 Modelling enhancement expenditure ............................................................................................ 41 1.5.2 Botex+ modelling ............................................................................................................................ 48 1.5.3 Alternative approach ...................................................................................................................... 54

1.6 Setting an effective baseline ............................................................................................................... 56 1.6.1 Triangulation ................................................................................................................................... 56 1.6.2 Efficiency adjustment ..................................................................................................................... 60 1.6.3 Adjustments to the baseline ........................................................................................................... 72 1.6.4 Water (Network plus & Resources) totex baselines ....................................................................... 84 1.6.5 Wastewater (Network plus & Bioresources) totex baselines ......................................................... 85 1.6.6 Alternative approach to Bioresources cost assessment ................................................................. 86

2 Residential retail cost assessment ............................................................................................................... 87 2.1 Introduction ........................................................................................................................................ 87 2.2 A Background to retail cost assessment at PR19 ................................................................................ 87 2.3 Prior expectation of retail cost drivers ................................................................................................ 87

2.3.1 Measures of deprivation: average vs extreme ............................................................................... 89 2.3.2 Transiency ....................................................................................................................................... 91 2.3.3 Regional wages ............................................................................................................................... 92

2.4 United Utilities’ approach to econometric modelling of retail costs .................................................. 92 2.4.1 Bad debt and debt management model choice ............................................................................. 93 2.4.2 Remaining retail cost model choice ................................................................................................ 94 2.4.3 Total cost model choice .................................................................................................................. 95

Copyright © United Utilities Water Limited 2018 2

Chapter 7: Supplementary Document - S6002 unitedutilities.com

2.4.4 United Utilities’ Preferred Model Suite .......................................................................................... 96 2.4.5 Model results .................................................................................................................................. 97 2.4.6 Validation of model results ............................................................................................................. 99 2.4.7 Ofwat’s draft retail cost models ................................................................................................... 100 2.4.8 Differences in approach between wholesale and retail ............................................................... 101

2.5 Building the Cost Threshold .............................................................................................................. 102 2.5.1 Incorporating time fixed effects into the cost threshold .............................................................. 102 2.5.2 Efficiency Adjustment ................................................................................................................... 103 2.5.3 Real price effects and dynamic efficiency ..................................................................................... 104 2.5.4 Triangulation ................................................................................................................................. 105

2.6 Additional Methodological Considerations ....................................................................................... 106 2.6.1 Forecast method ........................................................................................................................... 106 2.6.2 Separated cost to serve for metered and unmetered customers ................................................ 107

2.7 Retail Totex Baselines ....................................................................................................................... 107 References .......................................................................................................................................................... 108 Appendix ............................................................................................................................................................. 110

Model assessment framework – consultation results/application ............................................................ 110

Copyright © United Utilities Water Limited 2018 3

Chapter 7: Supplementary Document - S6002 unitedutilities.com

Executive summary

Cost assessment is a key element of any Price Review not only for the impact on current and future bills but it also sets the (wholesale) incentive rates against which companies will operate for the AMP7 period. This document sets out the approach that we have taken in developing totex baselines for each of the Wholesale price controls and the Residential Retail control for AMP7, which we can then compare to the business plans. Wholesale cost assessment In developing each Wholesale totex baseline, we have undertaken a thorough review of each of the models contained within the “Cost Assessment for PR19 – a consultation on econometric cost modelling”, rationalising the large number of proposed models by way of a robust Model assessment framework and selection criteria. Our selections have resulted in a diverse range of models for the industry that properly represent the cost drivers within each of the value chains and complement one another when used as a suite. We then combine these models using an innovative approach to Triangulation that we have developed to reflect the characteristics of each individual company. Our assessment of the Efficiency adjustment comes in two forms, a static and dynamic component. As at PR14, the static adjustment is derived using the model results and we have found no evidence to support that adopting a position beyond the upper quartile percentile is appropriate given the uncertainties around measurement error and the potential for omitted variables. We have evidenced that the PR14 efficiency adjustment was artificially high (+6% in Wastewater) due to the poor predicative capability of the enhancement models and as such, combined with our preferred approach to assessing enhancement requirements, we do not include enhancement within the calculation of the static adjustment. The dynamic adjustment builds on the evidence proposed by a number of protagonists over recent years in deriving a stretching adjustment given the measure of inflation (CPIH) and the real price effects against it. Lastly, we make an assessment on the required Adjustments to the baseline, adjusting for both the items excluded from the models as well as the impact that changes to accounting guidelines has had on the comparability of historic expenditure to that which is being proposed by companies in AMP7. We assess the efficient costs to be £5,889m versus our business plan costs of £5,434m, split between the price controls as follows.

• We have proposed a reasonable approach to cost assessment for the four Wholesale controls and the Residential Retail control.

• We have assessed an efficient cost for wholesale of £5,889m (compared with our wholesale plan totex of £5,434m), and assessed an efficient retail cost of £546m (compared with our retail plan costs of £490m). Therefore, we believe that our cost plans are both efficient and stretching.

• We have undertaken significant work with third parties (Vivid Economics, Arup, Reckon, and Equifax) to better understand the drivers of cost, and developed alternative variables.

• We have developed a robust approach to model selection, placing greater weight on operational experience and economic validity.

• We have proposed an innovative approach to triangulation between models and model suites for wholesale.

• We have proposed options for assessing enhancement expenditure within Water and Wastewater network, and assessed policy items required for those items not covered by econometric models.

Copyright © United Utilities Water Limited 2018 4

Chapter 7: Supplementary Document - S6002 unitedutilities.com

£m 2017/18 CPIH Water Resources

Water Network

plus

Bioresources Wastewater Network

Plus

Total

Business plan 374 2,074 372 2,613 5,434 Totex baseline 400 2,254 387 2,847 5,889 Out/(under)performance 26 180 15 234 455

Residential Retail cost assessment We have developed a dataset that contains new information on measures of deprivation and arrears risk, and shared this with the industry well in advance of business plan submission. We are encouraged by stakeholders’ responses so far, and are confident that this previously unavailable dataset represents a substantial expansion of the evidence base available at PR19. We worked with Reckon to develop these new measures of arrears risk into company-level econometric models of cost assessment. We have based our proposals on sound a priori economic and operational expectations of retail cost drivers and we consider them to fit well with Ofwat’s stated criteria for good models. We have sought to incorporate efficiency into our assessment of retail expenditure in a way that strikes the right balance between risk and reward. Triangulation ensures that any biases within one particular model do not penalise or reward any one company unfairly. We present additional evidence, based on internal and external data, which validates our preferred approach. We assess efficient costs in the retail price control to be £545.7m vs our business plan of £489.6m.

£m outturn 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 PR19 Business Plan 98.0 98.1 97.5 98.0 97.9 489.6 Totex gross of cost pressure 103.3 105.3 107.2 109.3 111.3 545.7

Out/(under)performance 5.3 7.2 9.7 11.3 13.4 56.1

Copyright © United Utilities Water Limited 2018 5

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1 Wholesale Cost Assessment 1.1 Introduction Cost assessment is the process within a Price Review by which an effective total expenditure (totex) baseline is determined for each company for the subsequent regulatory period. Ofwat will use assumptions on required totex in the calculation of the allowed revenues for each service/company, which results in impacts on customer bills both now and in the future. This means that an effective and appropriate approach to cost assessment is fundamental to ensuring that customers only pay for an efficient service, as well as enabling companies to recover sufficient revenues to finance the costs required to deliver their performance commitments. The benchmarking of expenditure across an industry through econometric analysis is an approach that has been adopted within many regulated sectors; and many countries. The benefits of an independent benchmarking exercise is that it creates a bridge for the information asymmetries that exist between the company and the customer (and the regulator). It also provides a way to use information on historical costs (albeit of other firms) to set a price/revenue control for a company while leaving that company with financial incentives to control and reduce its own costs, not just in the short term but over the long term. Whilst many regulators have adopted this process, the technical approach of calculating the baseline varies significantly, highlighting the fact that there is no one-size-fits-all approach. Therefore, an important characteristic of cost assessment is that it must be flexible. Often, one of the principal limitations of any econometric benchmarking exercise is that it inherently assumes homogeneity between companies for areas not covered by the explanatory factors that may not always be appropriate, particularly given the potential for regional variations in both the operating environment and levels of performance provided to their customers. It is therefore important to supplement econometric benchmarking with the ability to adjust the baselines in order to better account for company specific circumstances. This is particularly true for the water sector, given the differences in cost and quality requirements between regions. Figure 1 High-level approach to cost assessment as at PR14

Whilst the intricacies of benchmarking approaches by different (regulatory) bodies may vary Figure 1 illustrates the key components that an approach to cost assessment will contain when constructing an effective totex baseline for companies. We have retained the same naming conventions from PR14 for the basic cost threshold, the cost threshold and the baseline for ease. Over the last few years, we have been actively engaged in supporting the development of the knowledge base that underpins the econometric analysis that comprises a large element of the benchmarking assessment. We have commissioned reports with Arup and Vivid Economics to investigate the exogenous factors which drive

TOTEX BASELINE

COST THRESHOLD

Efficiency adjustment Real price effects 'Policy' items Cost adjustment claim

BASIC COST THRESHOLD (BCT)Econometric assessment Unit cost assessment Triangulation

Copyright © United Utilities Water Limited 2018 6

Chapter 7: Supplementary Document - S6002

unitedutilities.com

differences in the cost of delivering wholesale wastewater services (T6001 - Arup & Vivid Economics, 2017) and, using the knowledge gained and the evidence gathered, to present recommendations on the use of econometric models for PR19 (T6005 - Arup & Vivid Economics, 2018). Whilst these reports focused on the Wastewater service, many of the principles that underpin the recommendations can be extrapolated to all areas of the Wholesale value chain. Implementing the recommendations from these reports, alongside consideration of publications by other protagonists, we have developed econometric models for both the Water and Wastewater services. This document sets out our approach to assessing an efficient totex baselines required to deliver the prescribed levels of performance for each Wholesale price control in the regulatory period covering 2020-2025 (AMP7). Often, when deliberating any comparative cost assessment, it can be tempting to rely solely on statistical models. Whilst econometrics is typically a key component of cost assessment, it is not (and should not be viewed as) the only method, as there are many occasions when it is not the best method. Accordingly, this document accounts for more than merely the most appropriate econometric modelling to undertake; it addresses each component of cost assessment as part of an integrated process where key components supplement one another. It builds upon the information contained within final methodology for the 2019 price review (Ofwat, 2017), which have been factored into our cost assessment proposals. Where appropriate, it uses evidence gathered through Ofwat’s Cost Assessment working Group (CAWG) in recent years as well as the lessons learned from PR14 from both our own experiences and the recommendations of the Competition and Markets Authority (CMA) in the Bristol Water referral (CMA, 2015). We have combined econometric models with, where appropriate, other approaches to cost assessment and used the resulting baseline to validate that our PR19 business plans is efficient and stretching when compared to the leading benchmarked companies. 1.1.1 Objective and structure of this document The objective of this document is to detail the steps that we believe should be taken in developing effective totex baselines for each of the four Wholesale price controls in AMP7. The approaches to developing totex baselines can then be used to assess the business plans for all companies in a similar vein to that at PR14. This document does not discuss or reference the development of the totex requirements within the PR19 business plans for the Wholesale price controls for United Utilities. We structure the remainder of the document as follows; • Section 1.2 Econometric benchmarking within cost assessment presents and combines the evidence

gathered to date into proposals on appropriate econometric models and best how to, to use them in formulating a basic cost threshold for all Wholesale price controls

• Section 1.3 Water botex modelling summarises the main cost drivers for botex within the Water value chains and the results of the models that we have selected to derive the AMP7 baseline for the Water Resources and Water Network plus price controls

• Section 1.4 Wastewater botex modelling summarises the main cost drivers for botex within the Wastewater value chains and the results of the models that we have selected to derive the AMP7 baseline for the Bioresources and Wastewater Network Plus price controls

• Section 1.5 Assessing enhancement expenditure investigates three potential approaches to assessing enhancement expenditure at PR19

• Section 1.6 Setting an effective baseline presents evidence on how the basic cost thresholds should subsequently be adjusted in order to generate totex baselines for companies, accounting not only for efficiency adjustments; both static and dynamic, but also items excluded from botex models which are termed ‘policy items’

Copyright © United Utilities Water Limited 2018 7

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.2 Econometric benchmarking within cost assessment A simple, yet fundamental question that needs answering prior to any econometric analysis is undertaken is what is the ultimate goal? Do we want to seek a predictive solution that best fits historic observations (models that have internal validity) or to understand the causal relationship between expenditure and the core drivers which can be used to generate appropriate predictions of future requirements (models that have external validity)? Understanding the ultimate goal has a large influence on not only the technical decisions that are made when developing the models but also the application of the models within the overall cost assessment. Both methods are equally valid under the right conditions, but concerns over data quality combined with the limited number of observations restricts the range of technical approaches that are available to the modeller which can limit the effectiveness of “predictive modelling” approaches. Predictive modelling is primarily backward looking and therefore the underpinning assumption is that relationships between factors that were present in the past will continue to hold in the future. Under this approach, explanatory factors do not necessarily need to be associated to the dependent variable (i.e. cost), as the primary aim is to maximise the internal validity (R2) of the regression, and therefore no weight is placed on the legitimacy and correctness of the factors used. For some areas of expenditure within a value chain, historically this may be a logical assumption to make. However, changes to regulatory environment (particularly those driven by the switch to totex and outcomes, combined with the separation of the resource controls) means that historic relationships between disassociated factors may no longer hold. Furthermore, predictive modelling approaches are typically less transparent than utilising causal factors within a model and can be more reliant on having a large, comparable dataset with which to construct their relationships.

“Everything should be made as simple as possible, but no simpler”

Albert Einstein

As these benchmarking models are used for setting future totex assumptions for companies then it is important that they are able to predict future expenditure requirements for the majority of the industry, more so than it is that they fit the historic data perfectly. Our starting point for developing appropriate econometric models focused on understanding the causal relationship between expenditure and the drivers of costs. If engineering logic underpins the selection and validation of key drivers of cost, this means that model results should be less susceptible to changes to the operating environment and are less reliant on large datasets, providing a reliable relationship is established. Consequently, whilst we still need to predict the expenditure requirements for the industry within these models, seeking a high R2 is of less importance1 as the focus within model development and selection emphasises the correctness of the key explanatory factors and their coefficients (see section 1.2.3 Model assessment framework and selection criteria). Adopting a causal modelling approach should be more amenable to cost assessment, and be more complementary to claims for company specific adjustments, as where models are unable to capture company specific circumstances, claims should be more transparent and relevant to the models used. Homogenous attributes between companies in certain aspects of the value chain means that a ‘one size fits all’ approach may be appropriate but as the level of comparability reduces then the ability of a single model to predict the requirements of all companies diminishes. This can be addressed in several ways, such as through company specific cost adjustments (discussed in section 1.6.3.13 Cost adjustment claims), clustering companies together and modelling the similar groups separately or adopting a suite of models and allowing for differing triangulations dependent upon characteristics (discussed in section 1.6.1 Triangulation). Whilst clustering companies that have similar characteristics would increase the comparability of the observations, water companies are diverse and small in number, meaning that appropriate triangulation through a variety of

1 However, it is still relevant as the static efficiency challenge calculated using the residual spread of companies against the model results (see 1.6.2.4 Static efficiency adjustment)

Copyright © United Utilities Water Limited 2018 8

Chapter 7: Supplementary Document - S6002

unitedutilities.com

models combined with a cost adjustment process is likely to be the most appropriate method through which we can address a lack of comparability within cost assessment. This section sets out how we have constructed the basic cost threshold (BCT) through the combination of a suite of econometric models, which triangulate to form a ‘pre-efficiency’ view of the required expenditure for each price control given specific characteristics. 1.2.1 Data and modelled costs

Key findings and decisions:

We have excluded a number of costs from our botex model development in line with the approach taken by Ofwat for the consultation

We have used the dataset as published by Ofwat on 7th February 2018 and have not undertaken any smoothing or extension of the dataset

We have developed models using the adjusted information provided by companies as part of the 2017 July Cost Assessment Information Request (Ofwat, 2017 July Cost Assmt Information Request, 2018). In some instances we have supplemented this information with data constructed by Arup and Vivid Economics in their investigation into the drivers of Wastewater expenditure (T6001 - Arup & Vivid Economics, 2017) or with that provided by Ofwat through their development of additional explanatory factors (Ofwat, Constructed data, 2017). This results in having 6 years of information across 10 Water and Sewerage (WaSC) and eight Water only companies (WoCs) with which to construct benchmarking models2. We have excluded certain areas of expenditure from benchmarking models on the basis that the available explanatory factors cannot reliably predict the correct costs, that they are sufficiently outside of management control or that proposed incentive mechanisms in AMP7 preclude their inclusion. These ‘policy items’ are in line with those specified by Ofwat in their draft benchmarking models published alongside the consultation in early 2018 (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018) and are summarised in Table 1 below.

2 We have adopted the approach proposed by Ofwat (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018) in adjusting for the merger between South West Water and Bournemouth Water. This involves calculating explanatory factors that are not additive e.g. measures of density, based on the ratio of wholesale energy consumption between the two companies for the newly formed entity in 2016-17. The result is a maximum of 107 observations for Water benchmarking models rather than other approaches, which merge the companies for all years resulting in fewer observations.

Copyright © United Utilities Water Limited 2018 9

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Table 1: Items excluded from dependent variable within botex modelling

Items excluded from models

Wat

er

Net

wor

k pl

us

Wat

er

Reso

urce

s

Was

tew

ater

N

etw

ork

Plus

Bior

esou

rces

Local authority and Cumulo rates Grants and contributions (price control) Total cost of abstraction charges by the environment agency or canal & river trust.

Costs associated with industrial emissions directive (IED) Third party services (capital and operating expenditure) Costs of permits associated with the Traffic Management Act. Costs associated with statutory requirements for the softening of water

Cash Expenditure (excluding Atypical expenditure) Atypical expenditure

These costs cannot simply be excluded from cost assessment and so section 1.6.3 Adjustments to the baseline assesses how those costs that have been excluded from the BCT should be accounted for within the overall cost assessment. Whilst we acknowledge that the often ‘lumpy’ annual profile of investment risks potentially spurious relationships occurring (or a worse ‘fit’), we have not applied any smoothing to the data as we believe that the loss of additional observations outweighs the benefits gained through smoothing out lumpy investments. Another option would have been to extend the dataset using other sources such as the June Return submissions or the Water and Wastewater data inputs derived from the PR14 August Datashare. Whilst the additional observations could potentially result in more statistically robust models which otherwise would be discounted, using a longer time series not only incorporates expenditure that was less efficiently incurred (assuming year on year efficiencies have been made by companies) but changes to the Regulatory Accounting Guidelines (RAGs) makes historic allocations of expenditure less comparable. Additionally, care should be taken when smoothing so as not to lessen any of the legitimate variations that correlate with changes in explanatory factors, which is of particular importance should activity based explanatory factors be used in a model. We have tested the impact of extending the dataset to incorporate the use of additional years gathered through the PR14 data collection process (Ofwat, Setting price controls for 2015-20 – wholesale cost assessment, 2014) and found there to be little benefit in extending the series to incorporate these additional years. All benchmarking models have been created using Stata/IC 14.13 with basic cost thresholds constructed by use of four excel files (two for Water price controls and two for Wastewater price controls), all of which have been provided alongside this document4 (S6002a).

3 Some functionality utilised with the do-files may not be compatible with earlier versions of Stata. 4 S6002a_Cost_assessment_proposal_modelling

Copyright © United Utilities Water Limited 2018 10

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.2.2 Aggregated or disaggregated botex modelling?

Key findings and decisions:

Utilising granular allocations of cost allows more explanatory factors to be included within the models

Utilising aggregated allocations of cost allows substitution effects to be netted off within the models

For both Water and Wastewater modelling, we have developed a suite of aggregate and granular models covering all elements of the value chains

We apportion aggregate models to each of the AMP7 price controls using the company historic weighting between the two controls to reflect the relative asset configurations

The movement to four wholesale price controls for PR19 poses some interesting possibilities for cost assessment, namely, what level of granularity (the value chains) is the most appropriate to model? With limited available degrees of freedom due to a small number of observations, modellers face a trade-off between the ability to capture all of the appropriate and exogenous drivers of expenditure and the need to develop robust models with significant and relevant coefficients. Whilst granular modelling e.g. at a value chain level, has the obvious benefits of being able to capture a more explicit set of variables within the available degrees of freedom, it ignores the potential for substitution effects across the different elements of the value chain. This is particularly relevant given that Wholesale companies will still be vertically integrated entities with an asset configuration developed over time to be the most optimal for their circumstances that cannot adjusted in the short run.

(T6005 - Arup & Vivid Economics, 2018) “Costs can be substituted between wastewater treatment and bioresources to a significant degree. For example, company choices over the type of thickening and dewatering process at each sewage treatment works affect the volume of sludge amount produced and its consistency conveyed between sludge treatment centres, which in turn affects bioresources costs.”

Whilst it may be sensible to expect that comparative ‘efficiencies’ in one value chain due to substitution effects would be offset by similar ‘inefficiencies’ in others5 it has clear implications for how (static) efficiency targets should be set (discussed in section 1.6.2.4 Static efficiency ) and how best to triangulate the final suite of models. It is important to draw the distinction between granular modelling in order to support the development of a suite of models to capture the relative trade-offs faced by the industry and company specific factors within certain aspects of the value chain (which are best addressed ex post). In the PR19 methodology consultation (Ofwat, 2017), the notion of generating a cost assumption for both of the resource controls by developing a suite of models covering various parts of the value chain was presented within Appendix 12 and shown below in Figure 2.

5 Assuming that each model were able to appropriately predict the required expenditure and that neither form were not biased due to omitted variables, measurement error etc.

Copyright © United Utilities Water Limited 2018 11

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 2 Options for assessing value chain costs

Providing that we can develop suitable models for each of the assessment options, we agree that modelling different combinations of value chains is the most appropriate method to capture any substitution effects given the available data gathered as part of the information request6. A diverse suite of models covering different aggregations of the value chain, offers better opportunities to include more and relevant explanatory factors at the chosen levels of aggregation. The result of this increased should enable us to better distinguish between actual efficiency (at a company level) and noise as well as capturing company level substitution effects between the value chains by use of a wider selection of explanatory factors. Adopting an approach consistent with that in Figure 2 is desirable for not only the Resource controls but also for the Network plus controls. We can apportion expenditure generated by aggregate models to the AMP7 price controls by use of the company specific historical expenditure ratios within the value chains as per Equation 1 below. Equation 1 Calculation method for apportionment of expenditure to Resources controls

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝐵𝐵𝐵𝐵𝐵𝐵𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 = � ∑𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑎𝑎𝑅𝑅𝑎𝑎𝑅𝑅𝑎𝑎𝑎𝑎 𝑅𝑅𝑠𝑠𝑅𝑅𝑠𝑠𝑠𝑠𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐20112017 �

� ∑𝐴𝐴𝐴𝐴𝐴𝐴𝑅𝑅𝑅𝑅𝐴𝐴𝑎𝑎𝑎𝑎𝑅𝑅 𝑎𝑎𝑅𝑅𝑎𝑎𝑅𝑅𝑎𝑎𝑎𝑎 𝑅𝑅𝑠𝑠𝑅𝑅𝑠𝑠𝑠𝑠𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐20112017 �

× 𝐴𝐴𝐴𝐴𝐴𝐴𝑅𝑅𝑅𝑅𝐴𝐴𝑎𝑎𝑎𝑎𝑅𝑅 𝑚𝑚𝑅𝑅𝑠𝑠𝑅𝑅𝑎𝑎 𝑠𝑠𝑅𝑅𝑅𝑅𝑠𝑠𝑝𝑝𝑅𝑅𝑎𝑎𝑝𝑝𝑅𝑅𝑠𝑠

Using company specific weightings rather than an industry average or a weighting based on modelled proportions is important as it ensures that predictions reflect the relative asset configurations in each of the controls and that should provide a more appropriate BCT for each individual company. Under these circumstances, the remaining expenditure, once an apportionment for Resources is removed (1 - 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝐵𝐵𝐵𝐵𝐵𝐵𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐), can then be used as part of the derivation of the Network plus control BCT (providing any missing value chains are modelled separately). For the Water Resources and Network plus price controls, Ofwat will calculate a single totex sharing rate using the combined totals, meaning that this allocation does not influence the incentive mechanism ratio. For Wastewater, there will be separate incentive mechanisms for Bioresources and Wastewater Network plus, this means that getting the correct allocation between controls is more significant. There may be justification for companies to represent on these weightings given the nature of their historic investment over and above what has been called out as ‘atypical’ spend by e.g. weighting over a smaller period of time.

6 The confirmation that a single (combined) totex sharing rate will apply to the Water price controls will negate some of the issues caused by substitution effects between the value chains but the issue will remain for wastewater.

Copyright © United Utilities Water Limited 2018 12

Chapter 7: Supplementary Document - S6002

unitedutilities.com

In developing an appropriate and diverse suite of models that has the best possibility of capturing efficient differences between companies, we have adopted models utilising different levels of granularity across the value chains. Figure 3 and Figure 5 illustrate the splits of the value chains that we have utilised in deriving the BCTs for Water and Wastewater price controls respectively. Figure 3 Modelling approach for AMP7 Water price controls which incorporates a mixture of granular and top down modelling approaches

For Water, in addition to developing separate suites of models to predict each of the AMP7 price controls, we supplement this with two further aggregations of the value chains. A Wholesale Water suite combining all value chains and what is termed as ‘Water Resources plus’ which combines the Water Resources, Raw Water Distribution and Water Treatment value chains which enables us to reflect more explanatory factors within the specific value chains. The purpose of these aggregations is to remove any bias in the predictions that comes about due to substitution effects or opposing forces not captured by the explanatory factors, which can be prominent between the different parts of the value chains. Substitution effects arise legitimately because of a company’s response to the trade-offs that exist between the different parts of the value chain that result in companies having different configuration of assets. Opposing forces result where a specific attribute might make one part of the value chain cheaper to operate but involves greater costs further down the value chain (e.g. surface water sources may require less pumping, but require more treatment than groundwater sources). The effect of these differences is that a company can appear as opposing outliers across the two value chains in which the substitution occurs if the factor is not included as an explanatory variable. Within Water, these effects are most evident between Water Resources and Water Treatment shown below using basic unit cost assessments of each value chain.

Copyright © United Utilities Water Limited 2018 13

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 4 Average annual modelled expenditure per number of properties served (2011/12 - 2016/17)

Without accounting for any differences between companies other than the number of properties, a simple unit cost model would overestimate the requirements within Water Resources (we are below the line), but would conversely underestimate the requirements within Water Treatment (we are above the line). If the only distinguishing factor between companies was the amount of properties served i.e. the ‘goods’ were homogenous, then this approach would be sufficient and any variances in residuals would solely be due to differences in efficiency. However, we know that companies are not homogenous and employ different source types, have different levels of urbanisation, different topography, and different water quality to name but a few. For this reason, we introduce additional explanatory factors into models, albeit there are limits to the number of factors that we can control. Therefore modelling aggregations of the value chains can, to some degree, better enables these cross subsidisations to be ‘netted off’ from one another.

NWT

NWT

Copyright © United Utilities Water Limited 2018 14

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 5 Modelling approach for AMP7 Wastewater price controls which incorporates a mixture of granular and top down modelling approaches

For wastewater, we have adopted the same approach as within Water and developed suites of models to capture the substitution effects that are most evident between Wastewater treatment and Bioresources (sludge) in addition to the price control models7. Companies make decisions on the locations of assets and what part of the value chain to undertake work in response to conditions within their operating environment. These choices affect how much cost is borne within each value chain and so it is important to capture these differences through the apportionment of a combined model using the asset split for that company. This has resulted in developing suites at both an aggregate Wastewater and a treatment and sludge suite plus sewage collection as was modelled for PR14. 1.2.3 Model assessment framework and selection criteria

Key findings and decisions:

We adopted a robust model assessment framework to select the most appropriate models for inclusion within the final suite.

We sourced our models are sourced from the consultation and prioritise engineering and economic justification as well as transparency/relevance over statistical validity.

An important part of cost assessment is the use of comparative benchmarking as this enables customers and the regulator to identify relative performance and efficiency across the industry. To support this, we have developed a suite of benchmarking models and calculated the resulting baselines against which we have been able to validate that our business plans are amply stretching when assessed against our peers. In March 2018, Ofwat held a consultation on econometric cost modelling (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018) with models proposed by both companies and Ofwat, covering all aspects of the Wholesale and Retail value chains, and feedback on their appropriateness sought. With more than 320+ different wholesale models shared, the coverage and variation of possible models was large, but our reviews and analysis have highlighted that some are more suitable for use in an industry benchmarking exercise. In order to rationalise down the 320+ models into a final suite of models, we adopted

7 There are substitution effects between Sewage treatment and Sewage collection but the Network plus model suite and the top down approach will capture these.

Copyright © United Utilities Water Limited 2018 15

Chapter 7: Supplementary Document - S6002

unitedutilities.com

a model assessment framework similar to those proposed by CEPA (CEPA, 2018) and Ofwat (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018).

The model selection criteria consists of a series of tests falling under three broad categories and applied sequentially. These criteria are similar to those previously been adopted by others (CEPA, 2018) however a noticeable difference between our criteria and that of others is the ordering at which each of tests occurs. Similar to the Ofwat assessment criteria, we are primarily concerned with engineering/economic justification and practicality, only after which we then look at the statistical performance. Given that we are assessing models from a pool which has already been rationalised down by companies (companies were asked to put forward their ‘best’ models), it stands to reason that each proposal was the most promising of that form, and so prioritising engineering priors should help distinguish between proposals at the first stage. The benefit of this sequencing over prioritising statistical performance is that is allows for a more balanced approach and prevents the exclusion of models that whilst independently may not be the most proficient at fitting historic expenditure, may add greater value when utilised as part of a diverse suite of models to predict future costs. 1.2.3.1 Engineering and economic justification The reason for assessing models against this criterion first is that if strong engineering and economic narratives underpin selected models, then it increases the likelihood that models will have superior external validity. This is of particular importance given the small sample size increases the risk of outliers affecting the coefficients. External validity is the ability of the models to forecast accurately outside of the dataset (e.g. into the future) as opposed to internal validity which is the ability of the models to predict within the dataset. Given that the primary function of benchmarking models within cost assessment is to provide an initial assumption as to the required future expenditure, then having external validity is a key requirement and should be prioritised above all else. Statistical tests and criteria will by their very nature focus on the internal validity of the models rather than the external validity and if models are compared on statistical properties only (or first), then this will tend to place greater emphasis on overfitted models in isolation rather than facilitating the best possible suite of models for any given area. 1.2.3.2 Transparency and relevance One of the key requirements of cost assessment is that it reduces the information asymmetry between companies and customers (and the regulator) and provides a comparable independent estimate of a

Engineering and/or Economic justification

What are the key drivers of cost within the proposed area of modelling?

Is there strong engineering/economic evidence for the factor(s) utilised in the model being signfiicant drivers of cost

for the industry?

Are the signs and magnitdues of the coefficients consistent with engineering

and economic theory?

Do companies perform in line with prior expectations?

Transparency and relevance

Is the model replicable? Does it use any information that is not readily

obtainable?

Is the model intuitive and are the results transparent and easily interpretable?

Can the level of granularity be used to develop a prediction for PR19?

Does the model support the wider cost assessment process? e.g. cost

adjustments.

Statistical validity

Does the model pass the key statistical tests?

Is the model stable to changes in data?

Does the model possess appropriate predictive power

Can a suite of models perform better than a single model?

Figure 6 Model assessment framework for benchmarking model selection

Copyright © United Utilities Water Limited 2018 16

Chapter 7: Supplementary Document - S6002

unitedutilities.com

company’s efficient costs. In order to achieve this for all parties, the approach and modelling within cost assessment should be as transparent as possible. This is of significant importance in the event that there are material differences between the independent assessment and the bottom up business plan of a company. Should gaps occur then it is in the interest of all parties concerned to understand what is driving the differences and then for appropriate corrective measures to be taken whether that be a well specified and targeted cost adjustment claim or a further challenge to efficiency delivered by the company. Whilst this does not necessarily preclude the use of more advanced econometric approaches, it does require there to be a sufficient advantage in adopting such approaches over more simple and transparent approaches. It is also important that suites of models compliment alternative approaches to developing an effective baseline e.g. policy items, enhancement assessments. We should prioritise models that are supportive of the wider cost assessment process over those that can cause further issues, e.g. the inclusion or exclusion of grants and contributions within the benchmarking models. 1.2.3.3 Statistical validity Once the range of credible models has been rationalised, we then set about testing the comparative performance of these models. Statistical tests help to differentiate between equally credible models and narrow down the potential range of models available for use in cost assessment. However, it is important not to simply reject models on an individual basis as performance needs to be assessed at the level at which the predictions will ultimately be made. If it is unlikely that a single model will be capable of capturing all efficient variations between companies (if for example the service is highly complex and differentiated across companies), then it is more likely that a suite of models needs to be adopted. If this is the case, individual model performance against a set of statistical test may be less informative - instead, attention should be centred on the performance across the suite of models and whether or not the inclusion of another model within the suite offers additional explanatory power, for example by accounting for an additional explanatory factor. The principal statistical tests that we utilised for model selection were: • Ramsey’s RESET test • Variance inflation factor (VIF) [max ≤ 10] • Statistical significance of individual coefficients (T-test) • Residual analysis and adjusted R2 • Robustness testing through the exclusion of years, companies and explanatory factors The first three tests are easily quantifiable metrics that facilitate an unbiased comparative assessment of model performance. Whilst the final two tests have elements of subjectivity, the economic and engineering priors assessed within the first element of the framework inform them. In selecting the proposed models for inclusion within each suite, we have used robustness testing of variables extensively within model selection, particularly when deciding between two (or more) equally credible models, by assessing the distribution of the resulting coefficients under different exclusion scenarios. Companies and years are (iteratively) dropped from models and the regressions run so that the resulting distribution of the coefficients for each variable within that model can be analysed in order to assess the relative stability of the coefficients as well as the amount of leverage present within the model from the excluded company/year. This is a key benefit to assessing significance in isolation as it permits for variables that might not pass statistical tests to still be considered for inclusion if the distributions for coefficients under robustness testing is within an acceptable range and the coefficients align with the underpinning engineering and economic priors. The results shown in Figure 7 show the distributions for the coefficients in each of the charts below with each stack showing the percentage of scenarios where the coefficient lies within a specified range (the width of the bars). The orange vertical line within each graph represents the average coefficient generated through each of the scenarios.

Copyright © United Utilities Water Limited 2018 17

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 7 Example of the results received from a robustness test.

Analysing the data in this way enables us to see whether any one company or year is having an undue impact on the model coefficients. If we determine that there are outliers within the data, causing a skew in the coefficient of one or more variables then it may be more appropriate to exclude this company/year from the models in order to prevent inaccurate coefficients. 1.2.4 Modelling costs with small datasets In its initial report, Vivid Economics (T6001 - Arup & Vivid Economics, 2017) looked into the options for modelling in the face of a small number of observations, noting that:

(T6001 - Arup & Vivid Economics, 2017) “The small sample of data points available at the company level represents a formidable barrier to including many drivers in models, with only ten companies and modest variation in summary variables over time. This lack of degrees of freedom means that models that use of a long list of variables are unlikely to produce stable results with statistically significant coefficients. However, modelling practices can ameliorate the problem of a small sample size. Such practices include adopting functional forms that allow for the inclusion of more variables and avoiding methods whose performance is most adversely affected by small samples. The PR14 models employ a translog functional form, in which the squares and cross products of some variables are included as additional explanatory variables. Though this makes the translog a flexible functional form, there is no strong engineering or statistical case that these additional variables have a significant effect on costs. The flexibility it brings is thus of little value relative to adding further variables for which there is such a case. In addition, given the limits imposed by the small sample size, the translog’s use of additional explanatory variables makes it less feasible to include other factors for which there is a solid engineering narrative. This study therefore recommends the adoption of

Copyright © United Utilities Water Limited 2018 18

Chapter 7: Supplementary Document - S6002

unitedutilities.com

the more parsimonious Cobb-Douglas functional form in order to accommodate more engineering factors.”

An additional finding was that “panel data methods cannot be implemented with the small number of data points available”. Panel methods have the benefit of being able observe companies repeatedly over time, which has useful implications for assessing efficiency between companies when setting the static efficiency challenge. Whilst our starting point for model selection and development is to utilise ordinary least squares (OLS), unlike decisions regarding the functional form, we do not automatically disregard the use of random effects (RE). Models were selected for inclusion based on our Model assessment framework and selection criteria and tested with both OLS and RE formulations, with the RE version of the model being adopted if statistically superior to the OLS (passing RESET etc.).

1.3 Water botex modelling This section details the primary cost drivers that affect the Water service that we expect to lead to differences in efficient expenditure between companies. For each cost driver, we state the explanatory factor that we have used in explaining that element and the parts of the value chain we have used it. 1.3.1 Cost drivers and explanatory factors

Key findings and decisions:

We have adopted a diverse suite of models that capture all of the primary cost drivers across the Water value chains

Accepting that there is no one perfect measure for e.g. density, we have utilised a variety of explanatory factors to capture the differences between companies for the same cost drivers.

Having assessed all proposed models (and in the absence of acceptable models from the consultation, proposed our own), we have arrived at a suite of models for the Water price controls that best reflects the majority of companies within the industry, whilst supporting the wider cost assessment process. It is impractical to suppose that any one model or any suite of models could (or should) cover every single permutation and cost driver across the industry with the data available therefore we have selected diverse suite of models covering the primary drivers presented within the consultation by the majority of companies. The proposed models cover a wide range of explanatory factors and, importantly complement one another when used as part of the overall suite in generating a prediction for the period. Part of assessing the overall engineering and economic justification for each model is that they use only appropriate explanatory factors to explain the dependent variable. As a result, we rejected any model that included independent variables that were not associated with the dependent. The following tables set out the key drivers of cost within the water values chains and the explanatory factors that we have utilised within each of the model suites to explain these drivers. As our primary concern is representing the cost driver in the model, the choice of which independent variable to explain that cost is often a result of which variable performs best against the dataset (engineering priors, statistical testing etc.).

Copyright © United Utilities Water Limited 2018 19

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.3.1.1 Primary scale driver Explanatory factor Wholesal

e Water Water

Resources Water

network plus

Water resources

(plus)

Treated Water

distribution

Total number of household and non-household connected properties at year end

The total length of potable and all non-potable and partially treated water mains on 31 March of report year

Scale drivers enable econometric models to account for the varying sizes of companies and the customers that they serve. The choice of explanatory factor for scale is important as it drives a significant amount of expenditure within the predictions. All suites that we have selected use the total number of connections as the factor for scale but with the treated water distribution element of the Water Resources plus suite utilising the total length of mains instead. We discounted the use of distribution input (DI) as a primary scale driver within any botex models during the first stage of the Model assessment framework and selection criteria as it creates perverse incentives due to the relationship that it has with leakage. Within the aggregate models, (i.e. where the scale driver is an independent variable) the coefficient of the scale driver is typically (as expected) around unity8, indicating that economies of scale within the primary scale driver are not significant. Whilst some protagonists may find this result unexpected, the assets that are utilised rather than the physical number of connections drive economies of scale. Therefore, where we believe that economies of scale can occur, we have attempted to capture this impact through a separate independent variable. Because of observed constant elasticities, we have employed a selection of unit cost models that utilise the number of connected properties as the denominator. This has the effect of capping the elasticity at one thereby preventing any diseconomies of scale entering into the resulting predictions as well as adding further diversity to our proposed suites of models. 1.3.1.2 Source type

One of the key drivers of cost and variability within Water price controls is the type of source(s) that a company employs to supply water to its customers. We have broken out the impact of source types into two sections, one covering the upstream cost implications followed by the downstream impacts on Water Treatment and so there are certain trade-offs to be made, particularly within aggregated modelling. Whilst in our proposed suites we have only utilised an explanatory factor for the proportion of DI from impounding reservoirs, we model this alongside average pumping head thereby accounting for groundwater sources equitably.

8 Models used are log-log models and therefore a coefficient of one represents constant elasticity i.e. a 1% change in the independent results in a 1% change in the dependent.

Explanatory factor Wholesale Water

Water Resources

Water network

plus

Water resources

(plus)

Treated Water

distribution

Proportion of distribution input derived from impounding reservoirs

Copyright © United Utilities Water Limited 2018 20

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.3.1.3 Water Treatment complexity and size

The cost of treating water (Water Treatment value chain) accounts for approximately 31% of industry botex9 during the 2011-17 period (Ofwat, 2017 July Cost Assmt Information Request, 2018) across the Water value chains, with significant variation between companies. The amount of Water that needs to be treated and the level of treatment that is required primarily drive expenditure within Water Treatment. The main aspect of treatment and one that is dependent upon the source type employed concerns complexity and the amount of treatment undertaken. The underlying engineering priors that supports treatment level as an independent variable is that water from surface water sources requires more intensive (and therefore more costly) treatment in comparison the water sourced from below ground. This effect can be captured within models in one of two ways, either through the percentage of water treated at complex (or simple for the inverse relationship) treatment works or by differentiating between the amount of water sourced from surface water and groundwater sources. We have utilised several measures of treatment complexity that all result in the expected sign on the coefficient and are significant in most cases. In aggregated Water modelling, we have favoured using treatment complexity variables rather than source types to explain these additional costs as the linkage to expenditure is more direct and explicit but both types of variable should capture the same effects in replacement for one another. As stated above, the economies of scale that matter concern opportunities for economies of scale at the level of individual assets or sites (e.g. treatment works) rather than the overall scale of the company. This has little to do with the total number of customers served by a company, or the total scale of the system, and instead reflects more local conditions. For instance, treatment works in more rural areas treating smaller volumes of water may have relatively high unit costs (e.g. on basis of costs per cubic metre treated) as there are not the same opportunities for economies of scale as at larger treatment works. Therefore, it is appropriate to consider the impact here within this cost driver. The proportion of DI within bands 1 to 4 is intended to capture the effects of economies of scale at an asset level in a similar manner as the % load treated at bands 1-3 in Wastewater has done in the past. Whilst it does not perform as well as we would have hoped (lacking significance) it does have the expected sign and is constantly positive during robustness testing10 and therefore has been included within one of the suites. We have looked at alternative formulations in an attempt to improve the performance without success but still believe that it is valid to include this within models, particularly when the other proposed measures for economies of scale tended to focus on the number of assets that we consider inferior as they ignore where water is treat.

9 Modelled botex as per the definition within Data and modelled costs 10 Robustness testing of the variable involves separately dropping each of the years and companies within the dataset from the regressions and assessing the impact on the significance and sign of the coefficient.

Explanatory factor Wholesale Water

Water Resource

s

Water network

plus

Water resources (plus)

Treated Water

distribution

Percentage of water treated at treatment works categories simple to 2

Percentage of water treated at treatment works categories 3 to 6

Percentage of water treated at treatment works categories 4 to 6

Percentage of water treated at treatment works categories 5 to 6

Percentage of water treated at treatment works using surface water sources

Ratio of groundwater to surfacewater sourced treatment works

Proportion of Total DI in bands 1 to 4

Copyright © United Utilities Water Limited 2018 21

Chapter 7: Supplementary Document - S6002

unitedutilities.com

As the driver of costs are associated with Treatment, we have only included these variables within those suites that have Treatment costs within the dependent variable. 1.3.1.4 Measures of density/sparsity

Explanatory factor Wholesale Water

Water Resource

s

Water network

plus

Water resources (plus)

Treated Water

distribution

Proportion of company population in highly dense area : >=4000 people per km2

Proportion of company population in highly sparse area : <=600 people per km2

Percentage of population living in urban areas Length of mains per connected property Total number of household and non-household connected properties at year end per km of potable and non-potable mains (demeaned)

The level of density can be a significant cost driver within Water chains with the greatest impact felt at the extremes of the scale, that where there is high density or high sparsity; a ‘U’-shaped relationship. There are various methods proposed to capture the differing densities between companies but they can be broken down into two general groups, asset density and population density. Detailed population density measures such as those created by Ofwat as well as Arup & Vivid Economics are more highly specified measures of density that have the ability to capture differences within a region. Asset based measures of density are less precise as they use aggregated company data but do capture a reasonable amount of variation between companies and can form a reasonable proxy. Additionally, the different groupings of density will better associate with different splits of the value chains where population density measures are more appropriate for splits that include treated water distribution and asset-based measures are better associated with upstream value chain splits. The relationship between density and Water Resources is less explicit and we believe that any additional costs associated with high density or sparsity are better reflected by the type of source and therefore not essential for inclusion within Water Resources models. Whilst we believe that there is potentially a benefit in the use of a credible density metric within benchmarking models of downstream services we do have some concerns around the validity of some formulations of the measure. Simple aggregated measures of density for a company, particularly those that capture asset density, will often fail to recognise the different characteristics that occur within a company. A company may have areas of both high density and high sparsity within their operating region (as is the case for UU with, for example Greater Manchester is densely populated and Cumbria is sparsely populated) both of which drive additional costs over and above the average. However, when density is measured in aggregate, these differences would inappropriately ‘net off’ from one another, giving the impression that no such cost pressures exist for that company. Given that it is primarily the extremes (high density or sparsity) where costs are materially different from average, this limits the amount of companies to which we expect to see differential costs. It may therefore be more appropriate to adjust for these impacts ex post (e.g. through cost adjustment claims) in some model formulations, nonetheless, we have included some measures of density within our proposed models for downstream services.

Copyright © United Utilities Water Limited 2018 22

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.3.1.5 Network characteristics and activity Explanatory factor Wholesa

le Water Water

Resources

Water network

plus

Water resources (plus)

Treated Water

distribution

Percentage of mains laid or structurally refurbished prior to 1941

Total length of mains renewed or relined in report year

Number of booster pumping stations within the distribution system (potable only) per km of potable mains

Treated water distribution contributes a significant amount toward the total expenditure requirements of the Water service with an average of 58% across the industry during the period and greater than 70% for some companies11 highlighting not only significant expenditure but also variation between companies (when normalised by properties). The drivers of expenditure within the network can be attributed to the age of the system and the complexity of the system that transports water to customers. As such, it is pertinent to include an explanatory factor for these drivers within those value chains that account for treated water distribution. The age profile of a company’s mains can be used to predict the expenditure requirements due to failure (reactive) or for preventative actions (proactive). Using actual data on failure rates across our network, our analysis indicates that peak failure rates on different age groups of mains act in ‘waves’ over time, increasing to a point once the asset reaches a certain age as evidenced below. Whilst this is based only on our own information, the underlying engineering logic will be consistent across all firms albeit with some potential differences across neighbouring periods. Figure 8 Nominal distributions of failure rates for mains laid in each decade

11 Affinity Water and South Staffordshire Water.

Copyright © United Utilities Water Limited 2018 23

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 9 Cumulative distributions of failure rates for mains laid in each decade

Figure 8 above shows nominal distributions of failure rates for mains laid in each decade. Each cohort follows the same distribution with a 5% improvement in failure rates every 10 years. Figure 9 shows the same trends as cumulative totals through the period. The proportion of failures from each period of installation varies depending on the snapshot taken; in the 1950s, the majority of failures will be from pipe laid up until 1930 with small contributions from later mains. By the 1990s, the earliest mains have passed their peak failure rate and make a smaller contribution; the majority of failing pipes will have been laid between 1920 and 1940. The bar chart to the right of Figure 9 shows that a snapshot taken between 2010 and 2020 would represent that mains laid around the mid-part of the 20th century are currently experiencing the highest failure rate. We have tested this hypothesis within the datashare using different variables from this period and find that the data supports the engineering logic, particularly for the ‘percentage of mains laid or structurally refurbished prior to 1941’ which we find to be the most appropriate measure to use when differentiating between network failure rates and therefore the expenditure requirements for companies. Other drivers of cost differentials between companies can be due to the complexity of the network or the cost of interventions on the network. If more assets are required to transport water to customers, either due to topography or due to geological conditions, then this will increase the maintenance requirements of the company. We have also used measures that account for renewal or relining activity that a company undertakes as an explanatory factor. We have some reservations about the use of this in that it has the ability to create perverse incentives as it may compensate companies for undertaking inefficient levels of activity. Renewing or relining a main is an intervention undertaken to solve a leak or to deal with ingress (taste, odour and colour issues). Whilst the actual additional cost of the intervention is not under dispute, only accounting for these types of interventions risks penalising those companies which may have taken more innovative approaches to dealing with the underlying need for example through pressure management. The inclusion of these types of (endogenous) activity based factors will likely improve the internal validity of the models but do not provide any external validity – we therefore question the appropriateness of their use within cost assessment, particularly if the coefficient is large. 1.3.1.6 Pumping head

Explanatory factor Wholesale Water

Water Resource

s

Water network

plus

Water resources (plus)

Treated Water

distribution

Average pumping head - resources

Average pumping head - water network plus

Average pumping head - water resources (plus)

Whilst the engineering and economic narrative for the inclusion of pumping head within a benchmarking assessment is clear, it can perform poorly in models, sometimes with the wrong sign and often being

Copyright © United Utilities Water Limited 2018 24

Chapter 7: Supplementary Document - S6002

unitedutilities.com

statistically insignificant. Rather than not being a genuine driver of cost differentials between companies, we believe that the quality of the supporting data hinders the performance of these factors. We have performed some analysis using internal unit costs for power to check whether numbers in the datashare seem sensible. Using our unit cost of pumping head, we are able to estimate how much other companies should spend on power costs, given pumping costs make up the majority of spend on power. Table 2 Analysis of industry average datashare pumping head and power costs 2012-17 assuming 60% pump efficiency.

Company Total Estimated Cost (£m)

Reported power cost (£m)

Difference (£m)

Difference (%)

ANH 35.57 28.22 7.34 26% NES 23.04 17.46 5.58 32% NWT 24.73 25.03 -0.30 -1% SRN 16.49 12.65 3.85 30% SVT 55.40 42.70 12.70 30% SWT 11.70 10.65 1.06 10% TMS 65.49 50.06 15.42 31% WSH 25.52 19.48 6.03 31% WSX 8.64 7.16 1.48 21% YKY 30.24 25.00 5.24 21% AFW 22.36 20.06 2.31 11% BRL 10.02 7.66 2.36 31% DVW 2.79 1.66 1.13 68% PRT 2.33 1.97 0.35 18% SBW 3.61 2.58 1.03 40% SES 5.91 4.86 1.05 22% SEW 16.64 14.64 2.00 14% SSC 12.71 9.15 3.56 39%

If we assume that unit cost of power is comparable across the industry (and it seems reasonable to assume that they should at least be very similar), then the variances above may indicate widespread: • Over-reporting distribution input • Over-reporting pumping head; and/or • Under-reporting power costs • Significant differences in pump efficiencies across the industry It is less likely that there would be material differences in data quality for power costs or distribution input given the ability to measure this more accurately and the scrutiny within the WRMP. Furthermore, the relative efficiencies of pumps employed is likely to be comparable across the industry and so it is far more plausible that there are inconsistencies in the estimation of pumping head between companies which is causing the factor to perform poorly within the models. The data quality issue is further exacerbated when conducting the analysis at a more granular level (on each value chain) which further precludes its use in many model forms. We have attempted to include pumping head to maintain balance across the models and suites as we also account for the additional costs associated with operating and maintaining sources that require lower levels of pumping but its performance is not as robust as we would prefer. One important point to note is that the value chains within the aggregated pumping head factors must correspond to the value chains included within the dependent variable. The use of non-corresponding value chains e.g. only including treatment pumping head within a Water network plus model, will not result in a credible association between explanatory factor and cost.

Copyright © United Utilities Water Limited 2018 25

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.3.1.7 Additional explanatory factors Explanatory factor Wholesa

le Water Water

Resources

Water network

plus

Water resources (plus)

Treated Water

distribution

Number of sources per distribution input

Distribution input per connected property

All potable water supplied as part of the appointed business per connected property

We find that the drivers and explanatory factors covered in the preceding tables cover a significant proportion of the variation in expenditure between companies. There are some instances where additional explanatory factors further enhance the predictive power of the models proposed. The first of these relates to capturing the impact of the relative sizes of sources employed by companies by normalising the number of sources by distribution input (DI). The percentage of DI from various source types used within the source type and treatment drivers cannot capture the relative differences between companies caused by needing to maintain fewer or more sources. A company that has fewer sources per DI will need lower expenditure due to reduced maintenance requirements as sources are typically long life assets that need lower levels of cyclical maintenance. There are potential issues in the definition of the number of sources but we have included this factor as it still provides a reasonable differentiation between companies and the underlying engineering priors remain it is just that some companies will have an understated ratio. An additional variable selected is usage, captured through either DI per connected property or potable water per connected property. Usage acts as a proxy for the relative amount of ‘activity’ that a company must undertake in order to meet the requirements of its customers (their total demand) and therefore allows per capita consumption to vary between companies and the scale drivers used in models. The impact of increased usage covers a wide range of potential activities, particularly on the downstream value chains whereby if there is greater demand from customers, there will be increased costs associated. We have included these factors as a measure of balance and whilst the inclusion of these factors will improve the internal validity of the models and their predictions it is questionable whether a benchmarking model should incorporate such differences as they are potentially within management control and might provide the wrong incentives to tackling leakage and/or customer demand-side reductions. 1.3.2 Summary of Water botex modelling results

Key findings and decisions:

We have selected 9 models to predict Water Resources and Water Network plus botex by applying our Model Assessment Framework to the models proposed within the consultation alongside models we have developed internally.

We have utilised a mixture of total cost and unit cost models Models complement one another when included as part of an overall suite rather than

attempting to be individually perfect predictors.

This section presents the statistical model outputs of each of the models proposed within the Water value chains and the modelled aggregations. The format is the same as that used for the consolation and for each model chosen; the unique reference (the proposing company) is included within the top of each table and the key statistical diagnostics (R2, VIF(max), RESET) are included at the bottom of each.

Copyright © United Utilities Water Limited 2018 26

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.3.2.1 Summary of botex model statistical outputs We have presented all of the results using the same template as used in the consultation. Ofwat provided Table 3 within the consultation to help explain each of the statistical diagnostic test results within the template. Table 3 A simple glossary of statistical diagnostics in the templates

P-value of an estimated coefficient

The p-value gives the probability of observing the estimated coefficient (or one more extreme) if the true value was in fact zero. A lower value indicates a lower probability of observing the estimated coefficient if the true value was zero, and can thus be interpreted as giving a higher degree of confidence that the true value is not zero – i.e. that there is a relationship between the dependent and explanatory variables. In practice, the p-value indicates our confidence in the estimated coefficient. The lower the p-value, the more confident we are in the value of the estimated coefficient. Technical comment: due to the panel nature of the data, p-values in this appendix are based on cluster robust standard errors.

Asterisks for estimated coefficients

Next to the estimated coefficients, we use a common asterisks notation to indicate their statistical significance. *** indicates 1% significance level ** indicates 5% significance level * indicates 10% significance level The more stars, the more confident we are in the value of the estimated coefficient. No star indicates a lower level of statistical significance (ie there is less confidence in the value of the estimated coefficient). However, there is a wide range of confidence levels in this category. As we say in section 2.1 of the consultation, statistical significance of 80% and even 70% are may deemed valid in practical work.

R2 adjusted The adjusted R-squared measures how accurately the model fits the data. It measures the proportion of variation in the dependent variables (in our case, variation in costs) that can be explained by the model. The statistic ranges from 0 to 1. The higher the value the better the model fits. Importantly, R2 measures should only be used to compare models with the same dependent variable.

Variance Inflation Factor (VIF)

Used to detect multicollinearity. High collinearity means that we cannot estimate the coefficients with confidence – their variance is high and statistical significance low. As a consequence the individual coefficient estimates are not precise and unstable. As a rule of thumb, a VIF>4 indicates medium risk and VIF>10 indicates harmful collinearity. An exception to this rule is when the model includes a variable and its quadratic term. In such cases, the VIF becomes high due to the correlation between these two related terms. But while the high collinearity may impair our ability to accurately estimate the impact of the individual terms on the dependent variable, it should not impair our ability to accurately estimate their collective impact. Since these two terms always move together, the collective impact is what is important.

Reset test Regression specification error test. Used to detect an inadequate functional form. Particularly powerful for detecting if the model is missing non-linear terms. The higher the p-value the more confident we are that the functional form is adequate.

Copyright © United Utilities Water Limited 2018 27

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.3.2.2 Wholesale water models WSHWW1 SEWWW4

Dependent variable ln (Water botex) ln (Water botex)

ln (connected properties) (000’s) 0.997*** 1.080*** {0.000} {0.000}

ln (property density [demeaned]) (000’s/km) 0.009

{0.954}

ln (property density [demeaned])2 (000’s/km) 1.107** {0.028}

Number of Sources/DI (nr/Ml/d) 0.643*** 0.560*** {0.004} {0.003}

% water treated at complexity band 2 and below -0.666***

{0.000}

% water treated at complexity band 5 and above 0.176

{0.277}

% water treated at complexity band 3 and above 0.486*** {0.000}

% of area with more than 4000 people per km2 0.597*** {0.000}

% of area with less than 600 people per km2 0.471*** {0.002}

Constant -2.253*** -3.548***

{0.000} {0.000} R2 adjusted 0.974 0.974 VIF (max) 1.98 3.051 Reset test 0.346 0.205 Estimation method (e.g. OLS or RE) OLS OLS N (sample size) 107 107

1.3.2.3 Water network plus models

BRLNPW3 SVTNPW* UUNPW

Dependent variable

ln (Water Network+ botex per property)

ln (Water Network+

botex)

ln (Water Network+

botex)

ln (length of main) (km) 0.962*** 1.015*** {0.000} {0.000}

ln(DI/connected property) (Ml/d/000’s) 0.679

{0.139}

ln (Number of Sources/DI) (nr/Ml/d) 0.152

{0.239}

ln (property density) (000’s/km) 1.009*** 0.673** {0.004} {0.028}

ln(length of mains/connected property) (km/000’s) 0.369

{0.233}

% of mains laid pre-1940 1.004*** {0.005}

ln (length of mains relined and renewed) (km) 0.105*** {0.005}

Proportion of total DI at bands 1-4 0.022 {0.927}

% water treated at complexity band 2 and below -0.192

Copyright © United Utilities Water Limited 2018 28

Chapter 7: Supplementary Document - S6002

unitedutilities.com

{0.232}

% water treated at complexity band 4 and above 0.041 {0.866}

% water treated at complexity band 5 and above 0.085 {0.592}

% of total water treated at all SW works 0.379** {0.031}

Ratio of GW works to SW works (nr) -0.015** {0.041}

ln (average pumping head – Water network plus) (m.hd) 0.163 {0.226}

Constant -3.028*** -2.955*** -3.273***

{0.004} {0.000} {0.000} R2 adjusted 0.394 0.966 0.957 VIF (max) 2.078 Reset test 0.36 0.234 0.134 Estimation method (e.g. OLS or RE) Random

Effects OLS Random Effects

N (sample size) 107 107 107 1.3.2.4 Water resources models

OWR1 UUWR

Dependent variable ln (Water Resources botex)

ln (Water Resources botex)

ln (connected properties) (000’s) 1.022*** 1.016*** {0.000} {0.000}

% of distribution input sourced from impounding reservoirs

0.838** {0.047}

ln (average pumping head – Water Resources) (m.hd) 0.199** {0.023}

Constant -4.751*** -5.494***

{0.000} {0.000} R2 adjusted 0.888 0.902 VIF (max) 1 1.384 Reset test 0.677 0.374 Estimation method (e.g. OLS or RE) OLS OLS N (sample size) 107 107

Copyright © United Utilities Water Limited 2018 29

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.3.2.5 Water resources (plus) models UUWRP

Dependent variable ln (Water Resources+ botex per property)

% water treated at complexity band 2 and below -0.356* {0.094}

% of total water treated at all SW works 0.241 {0.165}

ln (property density) (000’s/km) -0.421** {0.036}

ln (average pumping head – Water Resources (plus)) (m.hd)

0.210* {0.061}

Constant -5.097***

{0.000} R2 adjusted 0.503 VIF (max) Reset test 0.617 Estimation method (e.g. OLS or RE) Random Effects N (sample size) 107

1.3.2.6 Treated Water Distribution models

UUTWD

Dependent variable ln (Treated Water Distribution botex)

% of mains laid pre-1940 0.925** {0.039}

ln (length of main) (km) 1.041*** {0.000}

ln (Water delivered (potable) per property) (Ml/d/000’s) 0.946** {0.036}

% of population living in urban areas 2.167*** {0.000}

ln (Number of booster pumping stations per km) (nr/km)

0.514*** {0.001}

Constant -4.770***

{0.000} R2 adjusted 0.96 VIF (max) Reset test 0.127 Estimation method (e.g. OLS or RE) Random Effects N (sample size) 107

Copyright © United Utilities Water Limited 2018 30

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.4 Wastewater botex modelling This section details the primary cost drivers that affect the Wastewater service that we expect to lead to differences in efficient expenditure between companies. For each cost driver, we state the explanatory factor that we have used in explaining that element and the parts of the value chain we have used it. 1.4.1 Cost drivers and explanatory factors

Key findings and decisions:

We have adopted models that capture all of the primary cost drivers across the Wastewater value chains

Accepting that there is no one perfect measure for e.g. density, we have utilised a variety of explanatory factors to capture the differences between companies for the same cost drivers.

Having reviewed all models proposed within the consultation against the above framework we have arrived at a suite of models for the Wastewater price controls that best reflects the industry whilst supporting the wider cost assessment process. It is impractical to suppose that any one model or any suite of models could (or should) cover every single permutation and cost driver across the industry with the data available therefore we have selected diverse suite of models covering the primary drivers presented within the consultation by the majority of companies. The proposed models cover a wide range of explanatory factors and importantly, complement one another when used as part of the overall suite in generating a prediction for the period. Part of assessing the overall engineering and economic justification for each model is that only appropriate explanatory factors are used to explain the dependent variable. As a result, any model that included independent variables that were not associated with the dependent would not be accepted without changes. The following tables set out the key drivers of cost within the wastewater values chains and the explanatory factors that we have utilised within each of the model suits to explain these drivers. 1.4.1.1 Primary scale drivers

Explanatory factor Wholesale Wastewat

er

Bioresources

Wastewater

network Plus

Treatment &

Sludge

Sewage collectio

n

Total number of household and non-household properties billed including voids

Total length of "legacy" public sewers as at 31 March

Total load received by all STW

Total sewage sludge produced

As within Water value chains, the selection of an appropriate scale driver is an important decision within Wastewater benchmarking models. However, within Wastewater, there are fewer perverse incentives associated with using volume-based measures, enabling a more diverse suite of models capturing a wider range of scale drivers to be developed. Furthermore, explanatory factors for the wastewater service need to account for more than waste produced by households or customers (e.g. surface water runoff) which means that volume based measures are more appropriate in some value chains. For Bioresources, the price control for AMP7 will be an average revenue control using the tonnes of dry solids produced (as the denominator) therefore, it is only appropriate to use the total sewage sludge produced as the

Copyright © United Utilities Water Limited 2018 31

Chapter 7: Supplementary Document - S6002

unitedutilities.com

scale driver within Bioresources models in order to maintain symmetry between the revenue control and the underpinning cost estimation. 1.4.1.2 Economies of scale

Explanatory factor Wholesale Wastewat

er

Bioresources

Wastewater

network Plus

Treatment &

Sludge

Sewage collectio

n

Percentage of load received by STW in size bands 1 to 3

Through their investigation into cost drivers for Wastewater (T6001 - Arup & Vivid Economics, 2017), Arup & Vivid Economics confirmed the engineering priori that economies of scale are present at an asset level (particularly at treatment works), driven by the size of the asset rather than the aggregated characteristics of a company or the customers that they serve. The unit costs of larger treatment works are lower than smaller treatment works due to efficiency savings, in particular energy, which accounts for a significant proportion of operating expenditure. Treatment works with population equivalent greater than 2,000 (bands 4–6) have much lower unit costs than bands 1-3 and there is significant variation in the proportion of load treated at these sites across the industry which enhances its reasonableness as an explanatory factor. They also found that there are significant links between economies of scale and the levels of urbanisation whereby the former increases in the presence of the latter.

(T6001 - Arup & Vivid Economics, 2017): “Companies with a higher-than-average percentage of treatment capacity in rural areas generally show a corresponding higher-than-average proportion of band 1-3 assets, both in terms of number of assets and load treated. In other words, the more rural the company area, the higher the reliance on smaller treatment works. This indicates that the profile of population settlements has an impact on a water company’s ability to utilise efficiencies offered by economies of scale.”

We have selected models that utilise asset level variables to account for the economies of scale within wastewater value chains. Given that evidence highlights how economies of scale occurs at treatment works, we do not include an additional variable to account for economies of scale within sewage collection. 1.4.1.3 Measures of density and urbanisation

Explanatory factor Wholesale

Wastewater

Bioresources

Wastewater

network Plus

Treatment & Sludge

Sewage collection

Percentage of population living in urban areas

Percentage of WWTW assets in sparse areas

Number of properties per km of public sewer

As mentioned above, there are significant links between economies of scale and the levels of urbanisation whereby the former increases in the presence of the latter. There are typically two proposed methods by which the levels of urbanisation (or density) faced by a company is estimated, by using a generalised/aggregated measure such as the number of properties per km, or more granular metrics that account for the specific location of assets or customers within a company’s boundary. We have used both types of measures within our proposed suite when attempting to capture the effects of urbanisation although we do use the more granular measure more extensively as we feel that it is a better approximation of the cost pressures that a company actually faces compared to that of an aggregated measure. Although we have not

Copyright © United Utilities Water Limited 2018 32

Chapter 7: Supplementary Document - S6002

unitedutilities.com

selected a final model that utilises it as a factor, the measures of a company’s population in highly dense areas developed by Ofwat performs well within Wastewater modelling as a measure of density and is also considered viable12. As Figure 10 shows, there is significant variation between companies in their levels of urbanisation that should improve its validity within models. Figure 10 Variation in (average) urbanisation for all WaSCs within granular measures

Similar to what occurs within Water, the impact of urbanisation/density on expenditure will have varying effects on different parts of the value chain. Evidence highlights that we should expect higher levels of urbanisation to cause increased expenditure requirements within Sewage treatment and collection as evidenced by the Arup & Vivid investigation (T6001 - Arup & Vivid Economics, 2017) but to result in reduced expenditure requirements within Bioresources due to lower intersiting requirements because of increased colocation between Network plus and Bioresource assets and a more centralised system13.

(T6001 - Arup & Vivid Economics, 2017): “The location of populations and settlements in wastewater company service areas determines where wastewater assets need to be located. Populations living in sparse areas must have the same standard of wastewater services as large, centralised populations in urban areas. This results in capital and operating costs differing substantially between the two demographics. Overall, the weight of evidence points to the total costs of networks

12 Final selection of the most appropriate variable to use will likely be dependent upon statistical performance rather than there being any engineering or economics preference for one particular measure. 13 Correlation between urbanisation/density and percentage colocation variables ranges from 0.40 to 0.70 depending upon the measure of urbanisation used.

Copyright © United Utilities Water Limited 2018 33

Chapter 7: Supplementary Document - S6002

unitedutilities.com

and wastewater treatment being higher in urban areas; for sparse areas the picture is more mixed and would benefit from further evidence.”

1.4.1.4 Wastewater Treatment complexity

Explanatory factor Wholesale

Wastewater

Bioresources

Wastewater

network Plus

Treatment &

Sludge

Sewage collectio

n

Percentage of load received by STW with Ammonia consents of less than 1mg/l

Percentage of load received by STW with Ammonia consents of less than 1mg/l and BOD consent of less than 10 mg/l

A significant proportion of Wastewater botex is a result of Sewage Treatment, comprising 47% of overall industry Wholesale Wastewater and 57% of Wastewater Network Plus botex14 during the 2011-17 period (Ofwat, 2017 July Cost Assmt Information Request, 2018). Because of the significance of this and the variation of consents and the resulting technologies adopted by companies, we believe that it is important to try to capture the relative differences within benchmarking models.

(T6001 - Arup & Vivid Economics, 2017): “The level of discharge permits drive the choice of treatment technologies, which in turn can have an appreciable effect on unit costs of treatment. For a large works, moving from a ‘basic’ permit of 20 mg/l BOD5 to a more ‘stringent’ permit that limits discharges to 3 mg/l NH3 requires the installation of a nitrifying activated sludge process, a shift that causes unit costs of treatment to increase by around 47 per cent.”

We have analysed the ability of a variety of explanatory factors to capture all of the different operating environments and cost pressures faced by companies in treating wastewater each of which account for differing variances between companies as shown within Figure 11. Other factors were considered to help explain treatment quality within the model suites; such as those that capture variations in the amount of tertiary treatment, Phosphorus or BOD5 consents, but those listed within the table above consistently perform better within models and so we have used in our final proposed models. Whilst there are underlying engineering priors which preclude the use of some treatment variables, ultimately, it is the driver that we are interested in representing. Once we know the thresholds at which cost differentials are expected then the most appropriate variable will be the one that performs best within the model. In adopting models that were proposed by Arup & Vivid Economics (T6005 - Arup & Vivid Economics, 2018), we have substituted the % of load received by STW with Ammonia consents of less than 1mg/l for % of total load received that undergoes tertiary treatment to improve statistical performance. This change is merely attempting to explain the same variation in a cost driver but utilising a different variable and so does not detract from the original findings or engineering prior that supported the construction of the models in the report.

14 Modelled botex as per the definition within Data and modelled costs

Copyright © United Utilities Water Limited 2018 34

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 11 Variation in treatment technologies and consents for each company (2016-17)

As well as receiving load from households, Wastewater treatment works also receive load from industrial processes (known as trade effluent (TE)) which can be more costly to treat than household waste. There is a signifcant amount of variation across the industry in the amount of load received from trade effluent customers with between 2.0-15.2% (Ofwat, 2017 July Cost Assmt Information Request, 2018) of total load (including TE) coming from trade effluent customers. In order to try and capture the variance between companies in the amount of trade effluent that companies must treat, we have also included these factors within some models where appropriate. 1.4.1.5 Network characteristics and activity

Explanatory factor Wholesale

Wastewater

Bioresources

Wastewater

network Plus

Treatment &

Sludge

Sewage collectio

n

Annual urban runoff Total length of sewer (including rising mains) laid or structurally refurbished post 2001

Percentage of total sewer length that are gravity combined sewers

Total installed pumping capacity of all sewage pumping stations (including standby pumps) per km of legacy sewers

Sewage networks can vary significantly between regions due to topography, environmental impacts or growth rates and the age or type of the network. These variances will drive differences in expenditure requirements meaning that the inclusion of network characteristics can improve the internal validity of a model. We have included models that attempt to capture differences in the volume of surface water that the network has to

Copyright © United Utilities Water Limited 2018 35

Chapter 7: Supplementary Document - S6002

unitedutilities.com

accommodate (drainage), the relative complexities required to transport sewage across the network (topography) and information on network type which distinguish between age and service (surface water, foul or combined).

(T6001 - Arup & Vivid Economics, 2017): “The relationship between drainage and costs is generally supported by hydraulic modelling evidence. This indicates base costs vary by 13 to 17 per cent when dry weather flow (DWF) volumes pumped are varied by up to 25 per cent. The analysis assumed a baseline amount of DWF pumping (75 per cent) as a starting point, which was varied to simulate the additional flows due to drainage in the system. The costs of maintaining additional storage assets to accommodate drainage flows were not included in the modelling exercise but, from a separate simulation, would be expected to be substantial given the level of capex required to build them.”

For drainage, most scale variables focus on customer numbers and associated volumes meaning that they do not appropriately account for the impact of surface water run-off which the sewage network needs to be able to accommodate. A network that has installed additional capacity (either through larger sewers or installed capacity such as storm tanks) to accommodate higher surface water flows will require additional maintenance to maintain services levels. Similarly, combined sewers need to be large enough to accommodate the flows from both households as well as surface water runoff. This then increases the amount of flow into a sewage treatment works as the surface water cannot simply be released back in to the environment (at no cost) as within surface water only sewers, it must go through the treatment process incurring additional costs for the company in the process.

(T6001 - Arup & Vivid Economics, 2017): “For the existing combined drainage systems to perform to the required levels of service, companies in areas of high runoff will have already invested in larger infrastructure - such as storage tanks and conveyance capacity – in order to meet equivalent service levels to companies in areas with low urban runoff.”

Unlike Water where there are ranges, there is only one available measure of network age for use in modelling: sewers constructed post 2001. Given the short nature of time that this covers, it implies all other parts of the network are the same age/quality, which is an oversimplification, but there is justification for inclusion within botex models. A younger network should be less susceptible to structural failures and will have less sediment and waste (fats/oils/greases, wet wipes etc.) build up thereby reducing the chances of blockages. Additionally, new sewers are constructed to serve the projections for current local environment and therefore will have sufficient capacity in place to deal with anticipated household waste and surface water run-off. An issue faced with many old sewers is that the surrounding areas have ‘outgrown’ their original function (urban creep) which leads to increased costs to maintain or storage tanks being required. The amount of installed pumping capacity per km gives a measure of network complexity in terms of both the amount of pumping required to deal with additional flows (sewage is pumped out of storage tanks) as well as topographical attributes for regions that require pumping in order to transport sewage through the network because of hilly terrain.

Copyright © United Utilities Water Limited 2018 36

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.4.1.6 Bioresources Explanatory factor Wholesal

e Wastewa

ter

Bioresources

Wastewater

network Plus

Treatment &

Sludge

Sewage collectio

n

Average distance travelled in disposal

Percentage of intersiting 'work' done by truck and tanker

Percentage of sludge produced and treated at a site of STW and STC co-location

Whilst on average, 63% of expenditure15 during the 2011-17 period (Ofwat, 2017 July Cost Assmt Information Request, 2018) within the Bioresources price control is associated with sludge treatment, we have not selected any model that attempts to include variables that capture differences in treatment technologies as this is an endogenous response to deal with an exogenous driver.

(T6005 - Arup & Vivid Economics, 2018): “There is no exogenous driver that corresponds exclusively to treatment quality over the long run. In contrast to wastewater treatment, where permit tightening can drive more advanced treatment quality which in turn affects efficient costs (see the June 2017 Report), enhanced treatment quality is for the most part either an endogenous response to land availability or an unforced management decision justified by energy recovery or reduced water content.”

We would have preferred not to utilise the measures of ‘work’ done within models as they too are endogenous and the investigation by Arup & Vivid Economics (T6005 - Arup & Vivid Economics, 2018) highlighted significant variances between the actual distances that companies have travelled and the optimal distances calculated through spatial analysis of available landbank. This poses significant risks within cost assessment of remunerating inefficient behaviours if we derive a forecast of future activity based on historic activities. However, we have included the factor within the model as we believe that there are suitable adjustments that can be made to account for these variances by adjusting forecast ‘work’ done to remove unexplained differences to the optimal distance or by intervening ex-post to correct for the endogeneity through a 2 sided adjustment to all companies.

(T6005 - Arup & Vivid Economics, 2018): “Short-run optimised distance measures do not perform as expected, with insignificant coefficients in models 1 and 2. This lack of correlation with costs does not refute the engineering narrative on which it is based and could simply reflect the small sample size with limited variation in most drivers over time….it may also reflect issues within cost data, limitations of the driver, or inefficiency in company sludge disposal activity.”

15 Modelled botex as per the definition within Data and modelled costs

Copyright © United Utilities Water Limited 2018 37

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.4.2 Summary of Wastewater modelling results

Key findings and decisions:

We have selected 11 models to predict Bioresources and Wastewater Network Plus botex by applying our Model Assessment Framework to the models proposed within the consultation alongside models we have developed internally.

We have only utilised total cost models, as unit cost models were not viable. Models complement one another when included as part of an overall suite rather than

attempting to be individually perfect predictors.

This section presents the statistical model outputs of each of the models proposed within the Wastewater value chains and the modelled aggregations. The format is the same as that used for the consolation and for each model chosen; the unique reference (the proposing company) is included within the top of each table and the key statistical diagnostics (R2, VIF(max), RESET) are included at the bottom of each. 1.4.2.1 Wholesale Wastewater models

YKYWW2 UUWWW1

Dependent variable ln (Wastewater botex) ln (Wastewater botex)

ln (Total number of properties) (000’s) 0.831*** {0.000}

ln (Total load received) (kg BOD5/day) 0.837*** {0.000}

% Population living in urban areas 1.815* {0.081}

% of load received by STWs in size bands 1-3 9.071*** {0.002}

% of load with Ammonia <=1mg/l 1.811*** {0.002}

% of load with BOD<=10mg/l and Ammonia <=1mg/l 0.805*** {0.000}

ln (Pumping station capacity per km sewer) (kW/km)) 0.279*** {0.000}

% of combined sewers 0.620*** {0.000}

Constant -1.306*** -6.830***

{0.001} {0.000} R2 adjusted 0.971 0.955 VIF (max) 7.852 Reset test 0.108 0 Estimation method (e.g. OLS or RE) Random Effects OLS N (sample size) 60 60

Copyright © United Utilities Water Limited 2018 38

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.4.2.2 Wastewater network plus models ONPWW9 UUNPWW1

Dependent variable ln (Wastewater Network Plus botex)

ln (Wastewater Network Plus botex)

ln (Total number of properties) (000’s) 0.760*** {0.000}

ln (Total load received) (kg BOD5/day) 0.806*** {0.000}

% Population living in urban areas 2.496** {0.025}

ln (Properties per km public sewer) (000’s/km) 0.838* {0.055}

ln (lengths of sewer laid post 2001) (km) -0.725 {0.112}

% of load received by STWs in size bands 1-3 11.091*** {0.001}

% of load with Ammonia <=1mg/l 1.014*** 2.264*** {0.000} {0.000}

Constant 1.826 -7.258*** {0.213} {0.000}

R2 adjusted 0.918 0.943 VIF (max) 7.852 Reset test 0.658 0 Estimation method (e.g. OLS or RE) Random Effects OLS N (sample size) 60 60

1.4.2.3 Bioresources models

WSHBR1 UUBR2

Dependent variable ln (Bioresources botex)

ln (Bioresources botex)

ln (Total sewage sludge produced) (ttds) 1.110*** 1.072*** {0.000} {0.000}

% of load received by STWs in size bands 1-3 6.426*** 6.839** {0.009} {0.013}

% of WwTW assets in sparse areas 0.089 {0.777}

% of sludge produced and treated at a site of STW and STC co-location

-0.207 {0.560}

ln (Average measure of 'work' done in sludge disposal operations (all forms of transportation) / Total sewage sludge disposed) (km)

0.205**

{0.013}

Constant -1.641*** -2.360***

{0.006} {0.001} R2 adjusted 0.817 0.821 VIF (max) 2.751 2.845 Reset test 0.003 0.005 Estimation method (e.g. OLS or RE) OLS OLS N (sample size) 60 60

Copyright © United Utilities Water Limited 2018 39

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.4.2.4 Treatment & sludge models OBP5 OBP7 UUBP1

Dependent variable ln (Treatment

& Sludge botex)

ln (Treatment & Sludge

botex)

ln (Treatment & Sludge

botex)

ln (Total number of properties) (000’s) 0.961*** 0.761*** {0.000} {0.000}

ln (Total load received) (kg BOD5/day) 0.927*** {0.000}

% of load received by STWs in size bands 1-3 5.473*** 10.207*** {0.007} {0.003}

% of intersiting ‘work’ done by truck or tanker -0.779*** -0.388** {0.003} {0.017}

% of load with Ammonia <=1mg/l 1.378** 1.678** {0.012} {0.011}

% Population living in urban areas 1.875 {0.132}

% of WwTW assets in sparse areas 0.329 {0.196}

Constant -1.672** -0.319 -8.558*** {0.025} {0.682} {0.000}

R2 adjusted 0.93 0.897 0.94 VIF (max) 2.411 1.814 8.065 Reset test 0 0 0 Estimation method (e.g. OLS or RE) OLS OLS OLS N (sample size) 60 60 60

1.4.2.5 Sewage collection models

UUSWC1 OSWC2

Dependent variable ln (Sewage collection botex)

ln (Sewage collection botex)

ln (Total length of "legacy" public sewers as at 31 March) (km)

0.456*** {0.000}

ln (Annual urban runoff) (million m3) 0.197*** {0.000}

% Population living in urban areas 1.235* {0.093}

ln (Total number of properties) (000’s) 0.785*** {0.000}

ln (Properties per km public sewer) (000’s/km) 0.981** {0.017}

ln (lengths of sewer laid post 2001) (km) -0.043** {0.024}

Constant -2.381** 1.448 {0.035} {0.224}

R2 adjusted 0.811 0.91 VIF (max) Reset test 0.586 0.369 Estimation method (e.g. OLS or RE) Random Effects Random Effects N (sample size) 60 60

Copyright © United Utilities Water Limited 2018 40

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.5 Assessing enhancement expenditure Enhancement expenditure can form a significant proportion of a company’s expenditure requirements with an average of 20% of Water and 25% of Wastewater modelled expenditure across the industry invested on improving service levels for customers in the 2012-17 period (Ofwat, 2017 July Cost Assmt Information Request, 2018). Within the industry, there is significant variability between companies in both the amount of enhancement expenditure and the driving force behind the investment. It is therefore important to develop a robust approach to assessing all enhancement expenditure, not only for its primary use in accurate forecasting of a company’s requirements but also for the impact that it can have if it is included within the static efficiency adjustment, which can be significant. We cover the impact of enhancement modelling on the efficiency adjustment and the appropriate method by which to address this within section 1.6.2.4 Static efficiency . As noted within the final methodology (Ofwat, 2017), “enhancement expenditure can be quite company-specific, irregular and difficult to predict” which makes developing credible and appropriate enhancement models a challenge. With this in mind, we have looked into three options available to assess the future requirements for enhancement expenditure; Modelling enhancement expenditure, including some enhancement within botex (termed Botex+ modelling) and finally, assessing all enhancement expenditure on an individual company basis based on evidence provided within the business plans (what we have termed an Alternative approach). 1.5.1 Modelling enhancement expenditure

Key findings and decisions:

We have replicated the approach adopted at PR14 for developing enhancement models for Water and Wastewater.

We have encountered the same problems as PR14 resulting in some unsuitable model formulations for certain enhancement areas.

Whilst some enhancement models are potentially viable, we do not propose that Ofwat should use modelling approaches to assess any enhancement requirements for each company at PR19.

1.5.1.1 The PR14 approach One type of analysis of enhancement expenditure that we have explored is the econometric (unit cost) modelling for specific categories of enhancement expenditure that Ofwat used at PR14. We have built on and refined the PR14 approach. We recap briefly below on the approach to enhancement modelling/benchmarking that Ofwat used for PR14 and then describe the analysis we have done. • For assessing Water enhancement requirements at PR14, Ofwat developed econometric assessments for

enhancements to the supply/demand balance, new developments (gross and net) and meeting lead standards.

• For assessing Wastewater enhancement requirements at PR14, Ofwat developed econometric assessments for 13 areas of enhancement as well as a separate company specific assessment of the Chemicals Investigation Programme.

For each area of enhancement, the approach involved four calculation methods, which were then averaged to result in a pre-efficiency prediction being made based on a forecast of future activity. The four calculation methods were; • Unweighted unit cost • Weighted unit cost • Linear regression

Copyright © United Utilities Water Limited 2018 41

Chapter 7: Supplementary Document - S6002

unitedutilities.com

• Log-linear16 regression The unmodelled uplift then formed the basis of the initial assessment for all remaining enhancement expenditure. This uplift was later rationalised against the business plan of each company in order to prevent any over remuneration as a result of the simplistic approach. 1.5.1.2 Issues caused by the approach/dataset A difficulty when developing many of the assessments for enhancement is that it is often the case that the recording of the expenditure and the explanatory variable (the output) are not aligned in time. This is primarily due to the nature of major capital interventions, which can take several years to deliver, but where the ‘output’ is claimed in a single year. Given the short duration of time covered by the datashare, the possibility of outliers due to timing differences increases significantly as the period may not capture all relevant information for either cost or output. An example of this evident when assessing Wastewater enhancement for ‘NEP - Reduction of sanitary parameters’ where South West has a unit cost more than five times greater than the average and forty times higher than the average when the outliers are removed as shown in Figure 12. In some instances (as you would expect), cumulative expenditure tables have been useful in negating this issue but in general this information did not add sufficiently to the predictive capability of the enhancement models. In developing the unit costs models, we have removed those observations that we consider are clear and unjustifiable outliers within the dataset in order to prevent any incredulous results due to anomalous data. We have annotated all excluded observations within the supporting Stata do-files17 for enhancements in Water and Wastewater. Figure 12 Unit cost for NEP - Reduction of sanitary parameters (£m/000s population equivalent)

16 Sometimes referred to as a log-log model whereby both the dependent and independent are regressed in their logged form. 17 5. Water enhancement modelling.do and 7. Wastewater enhancement modelling.do

Copyright © United Utilities Water Limited 2018 42

Chapter 7: Supplementary Document - S6002

unitedutilities.com

For PR14, all four approaches were included within the final BCT derivation for each area of enhancement, even if predictions were significantly different from one another. Rather than including all approaches within the BCT derivation, we believe that it is only appropriate to include approaches that have a reasonable predictive power. In order to discern which approaches are appropriate, we assessed the ability to predict historic expenditure requirements at both a company and industry level18 so that approaches can be included/excluded based on their fitness. As Figure 13 illustrates19, some model approaches have been more successful in predicting historic expenditure both across the industry and for each company. Figure 13 Approach to assessing quantitative methods for predicting: ‘Meeting lead standards’ expenditure requirements for the period 2011-12 to 2016-17 (2017/18 CPI(H) FYA)

Within both the Water and Wastewater information, some companies have extensively used freeform lines, accounting for 11% and 23% of total enhancement expenditure respectively. This can pose issues for trying to develop credible assessments, as it may be appropriate to include these lines within some models but without further information as to what drives each of these investments, the only option available to us is to assume that it is reoccurring spend and include it within an unmodelled uplift percentage. Other than lead, supply/demand enhancements and new development expenditure represent significant areas of investment within the industry and are likely to continue to be of similar importance for AMP7 and beyond. Within Water value chains, historic enhancement investment has varied across companies, with four of the predefined drivers/lines seeing more than £500m of investment during the 2011/12-2016/17 period.

18 It is important to assess not just whether or not an approach can predict expenditure requirements at an industry level but also that the allocation between companies is reflective. 19 This enhancement model suffered the same issues at PR14.

Copyright © United Utilities Water Limited 2018 43

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 14 Total industry water enhancement expenditure by company (2011/12 - 2016/17)

Within Wastewater Network Plus value chains, historic enhancement investment has varied across companies but is dominated by investment into environmental improvements driven by the National Environmental Programme (NEP), with one company (United Utilities) investing more than £1,000m on NEP driven activities. Figure 15 Total industry wastewater enhancement expenditure by company (2011/12 - 2016/17)

Copyright © United Utilities Water Limited 2018 44

Chapter 7: Supplementary Document - S6002

unitedutilities.com

An issue for many enhancement areas is that the historic expenditure is dominated by a few companies e.g. Resilience within Water, which calls into question the validity of developing industry models that are only representative of a small number of companies. 1.5.1.3 Results For Water, we have developed enhancement models for the same three areas as PR14 and also created a new model for total metering enhancement activities (for optants, company introduced and non-households) using all years covered by the datshare (2011/12-2016/17). There are potential endogeneity and forecasting issues for a metering model but if there are adequate protective measures in place (e.g. through an ODI) then company forecasts can be used to derive a prediction without exposing customers. For new development expenditure, we have replicated the same model as PR14. However, as we cover in section 1.6.3.1 Grants and contributions, the new connections charging rules which come into force from April 2020 (Ofwat, 2017); specifically how changes to the income offset impacts asset payments and incomes, mean that whilst the change is neutral from a net totex perspective, the historic reporting of gross expenditure and grants and contributions are no longer comparable. This means that whilst the unit cost model may be replicatable, it is not appropriate to use within the BCT development as it would be inconsistent with future charging rules regarding asset payments. Table 4 Potential PR19 Water enhancement models and explanatory factors Water enhancement area As

PR14? Volume/scale measure used for explanatory variable

Enhancements to the supply/demand balance

Y Total supply side enhancements to the supply demand balance (dry year critical / peak conditions) + Total demand side enhancements to the supply demand balance (dry year critical / peak conditions)20

Lead pipes Y Number of lead communication pipes replaced for water quality

New development (gross) Y Total number of new non-household connections + Total number of new household connections

Metering N Number of meter optants + Number of selective meters installed The performance of the approaches varied across the areas of enhancement as set out in Table 5 for Water enhancements highlighting where model forms fail to offer a credible prediction for companies (as indicated by a ) and where they provide a sufficiently credible prediction that we feel it could be used within cost assessment (as indicated by a ).

20 As companies often duplicate the benefits of an intervention between the dry year critical/peak conditions and dry year annual average conditions, we only use one line for each supply and demand. If critical/peak equals zero but there are additions to dry year annual average, we use these instead. This is consistent with the approach adopted at PR14.

Copyright © United Utilities Water Limited 2018 45

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Table 5 Results to the four approaches for assessing Water enhancement expenditure. Water enhancement area Unweighted

UC Weighted UC Linear

regression Log-linear regression

Enhancements to the supply/demand balance 21 22

Lead pipes New development (gross) Metering

For Wastewater, we have developed enhancement models for the many of the enhancement areas as covered at PR14 using all years covered by the datshare (2011/12-2016/17). As we have excluded grants and contributions from the BCT, we also needed to create an assessment for gross new development expenditure which was assumed to be 100% developer funded at PR14, and therefore net totex was zero and so no model was required. The majority of historic NEP related investment has been centred around storage, phosphorus removal and sanitary determinands each of which had a model at PR14. We have updated assessments for each of these areas (as well as event duration monitoring and CIP trials) but other lesser used lines were either deemed not material enough (e.g. flow monitoring), there were too few significant observations to develop a credible industry model (e.g. groundwater schemes), or a clear explanatory factor was not present within the data (e.g. investigations). Table 6 Potential PR19 Wastewater enhancement models and explanatory factors

Wastewater enhancement area

As PR14? Volume/scale measure used for explanatory variable

First time sewerage UC: Y Reg: N23

Unit cost: Connectable properties served by s101A schemes completed in the report year Regression (multiple): Connectable properties served by s101A schemes completed in the report year and Number of s101A schemes completed in the report year

NEP: Event Duration Monitoring at Intermittent Discharges

Y Number of intermittent discharge sites with event duration monitoring

NEP: storage schemes for intermittent discharges

Y Volume of storage provided at CSOs, storm tanks, etc to meet spill frequency objectives

Sewer flooding Y Household properties billed for sewage + Non-household properties billed for sewage

Private sewers - pipes N24 Length of formerly private sewers and lateral drains (s105A sewers) WINEP / NEP ~ Reduction of sanitary parameters

Y Current population equivalent served by STWs with tightened/new sanitary parameter consents

21 Weighted unit costs predict appropriate expenditure at an industry level but not when assessing predictions on a company basis. 22 Linear regression models predict appropriate expenditure at an industry level but not when assessing predictions on a company basis. 23 Ofwat proposed model form from (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018) has been used instead of the single factor regression at PR14 24 PR14 model accounted for both botex and enhancement expenditure

Copyright © United Utilities Water Limited 2018 46

Chapter 7: Supplementary Document - S6002

unitedutilities.com

WINEP / NEP ~ Nutrients (P removal at all STWs)

N25 Current population equivalent served by filter bed STWs with tightened/new P consents + Current population equivalent served by activated sludge STWs with tightened/new P consents

Growth at sewage treatment works (excluding sludge treatment)

N26 Population equivalent treatment capacity enhancement

New development and growth

N27 Household properties connected during the year + Non-household properties connected during the year

Table 7 Results to the four approaches for assessing Wastewater enhancement expenditure.

Wastewater enhancement area Unweighted UC

Weighted UC

Linear regression

Log-linear regression

First time sewerage NEP: Event Duration Monitoring at Intermittent Discharges

NEP: storage schemes for intermittent discharges

Sewer flooding Private sewers - pipes WINEP / NEP ~ Reduction of sanitary parameters

WINEP / NEP ~ Nutrients (P removal at all STWs)

Growth at sewage treatment works (excluding sludge treatment)

New development and growth Chemicals investigations programme28 n/a n/a n/a n/a

All reoccuring enhancement expenditure which does not have a specific model or approach is added together and used to calculate an ‘unmodelled uplift’ in line with the approach at PR14 which is then applied as an upward adjustment to the BCT. Equation 2 Calculating the unmodelled uplift percentage

𝑈𝑈𝑠𝑠𝑚𝑚𝑅𝑅𝑠𝑠𝑅𝑅𝑎𝑎𝑎𝑎𝑅𝑅𝑠𝑠 𝑅𝑅𝑠𝑠𝑎𝑎𝑝𝑝𝑢𝑢𝑎𝑎 % = ∑ 𝐼𝐼𝑠𝑠𝑠𝑠𝑅𝑅𝑅𝑅𝑎𝑎𝑅𝑅𝐼𝐼 𝑅𝑅𝑠𝑠𝑚𝑚𝑅𝑅𝑠𝑠𝑅𝑅𝑎𝑎𝑎𝑎𝑅𝑅𝑠𝑠 𝑅𝑅𝑠𝑠ℎ𝑎𝑎𝑠𝑠𝑅𝑅𝑅𝑅𝑚𝑚𝑅𝑅𝑠𝑠𝑎𝑎6𝑡𝑡=1

∑ 𝐼𝐼𝑠𝑠𝑠𝑠𝑅𝑅𝑅𝑅𝑎𝑎𝑅𝑅𝐼𝐼 𝑚𝑚𝑅𝑅𝑠𝑠𝑅𝑅𝑎𝑎𝑎𝑎𝑅𝑅𝑠𝑠 𝑏𝑏𝑅𝑅𝑎𝑎𝑅𝑅𝑏𝑏6𝑡𝑡=1 + ∑ 𝐼𝐼𝑠𝑠𝑠𝑠𝑅𝑅𝑅𝑅𝑎𝑎𝑅𝑅𝐼𝐼 𝑚𝑚𝑅𝑅𝑠𝑠𝑅𝑅𝑎𝑎𝑎𝑎𝑅𝑅𝑠𝑠 𝑅𝑅𝑠𝑠ℎ𝑎𝑎𝑠𝑠𝑅𝑅𝑅𝑅𝑚𝑚𝑅𝑅𝑠𝑠𝑎𝑎6

𝑡𝑡=1

Given the large amounts of enhancement expenditure that companies allocated to freeform lines, this results in a large unmodelled uplift % within both Water (10.3%) and Wastewater (11.1%)29 BCTs using all of the enhancement models. High percentages for unmodelled uplifts aren’t desirable as they do not capture any of the relative differences between companies and therefore simply assume that scale of all other investment is a credible predictor of future requirements, but no appropriate model form could be found for these areas of enhancement and these costs cannot be ignored in setting a baseline.

25 We have combined both P removal enhancement lines (activated sludge SWT and filter bed STW) together to form a single enhancement model for all phosphorous removal 26 PR14 model used total population growth rather than the more asset based metric 27 No model at PR14 –Assumed that all new development and growth activities were fully recovered from developers and not customers. 28 Chemicals investigations programmes have been assessed bottom up based on company business plan as per PR14. 29 PR14 equivalent uplifts were 8.4% within Water and 3.77% within Wastewater.

Copyright © United Utilities Water Limited 2018 47

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.5.2 Botex+ modelling

Key findings and decisions:

We do not believe it is appropriate to include enhancement expenditures within botex models for the purposes of developing effective baselines for PR19.

One option explored by some protagonists is to include some areas of enhancement within botex models. This is not to say that full totex models are appropriate, rather it is on the assumption that some areas of enhancement can equally likely be predicted using the same variables as what are utilised within botex models as they can be by having separate models of their own. Some of the unit cost models we have developed do utilise the same variables that we have adopted within our proposed botex models (e.g. annual property numbers). This may facilitate their inclusion within a botex model from a theoretical point of view, but we do not believe that it is appropriate to do so in the context of the wider cost assessment for PR19 given the predictive ability of botex+ models and other mechanisms that are in place. In its consultation document (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018), Ofwat set out nine enhancement areas across water and wastewater that it will investigate to determine whether it may be more amenable to inclusion with botex+ models as shown in Table 8. We have assessed the validity (on a RAG basis) of each of these areas of enhancement and do not believe that any of these should automatically be included within botex as evidenced below. Table 8 Enhancement costs proposed to be included within base costs for modelling.

Wholesale Water RAG Wholesale Wastewater RAG Expenditure in local network assets associated with new development and growth in water services

Expenditure in local network assets associated with new development and growth in sewerage services

Expenditure to enhance the balance of supply and demand

Expenditure to address growth at sewage treatment works (excluding sludge treatment)

Expenditure associated with metering (excluding metering to new connections)

Expenditure related to transferred private sewers and pumping stations

Expenditure to improve resilience Expenditure to improve resilience Expenditure to reduce flooding risk for

properties

Generally, we do not believe it is practical to include enhancement expenditures within botex models for the purposes of developing effective baselines for PR19. The correlation between what explains the enhancement (the underlying drivers rather than an explanatory factor in a unit cost model) and the factors used within botex models (typically the scale driver) is not sufficient to predict future requirements for each individual company. Given this, it will be more likely that botex+ models will lead to a misallocation between companies and for companies to appear more or less ‘efficient’ than they otherwise would. Whether or not this poses an issue in the final PR19 models should be tested by running botex and botex+ models simultaneously and comparing the results not by assessing the statistical performance of the model but its ability to predict individual company requirements.

Copyright © United Utilities Water Limited 2018 48

Chapter 7: Supplementary Document - S6002

unitedutilities.com

The example in Table 9 below uses results from one of our proposed botex models (ONPWW9) including and excluding STW growth enhancement expenditure. Table 9 Average Wastewater Network Plus expenditure for ONPWW9 botex and botex+ modelling £m (2011/12 - 2016/17)

STW growth

Actual cost

botex+

Predicted cost

botex+

Ratio Botex+

Actual cost

botex

Predicted cost

botex

Ratio Botex

Variance

ANH 16.2 293.1 243.7 1.20 276.9 232.3 1.19 1.09% NES 3.0 124.4 153.0 0.81 121.4 150.0 0.81 0.41% NWT 14.5 422.8 407.2 1.04 408.3 388.3 1.05 -1.32% SRN 2.4 243.1 241.8 1.01 240.7 239.4 1.01 0.00% SVT 6.3 349.6 348.8 1.00 343.3 338.9 1.01 -1.08% SWT 4.6 120.2 102.3 1.18 115.6 100.7 1.15 2.72% TMS 27.8 511.6 539.2 0.95 483.7 525.2 0.92 2.77% WSH 4.1 201.0 170.6 1.18 196.9 166.9 1.18 -0.11% WSX 0.0 114.8 133.2 0.86 114.8 130.1 0.88 -2.04% YKY 1.0 209.2 237.3 0.88 208.2 231.1 0.90 -1.94% Total 79.9 2,589.7 2,576.9 2,509.8 2,502.8

The inclusion of STW growth enhancement expenditure within the botex model does increase the aggregate industry expenditure prediction roughly in line with the expenditure incurred which is positive but the allocation of the additional expenditure between companies is not appropriate. This generally leads to those companies with higher than average expenditure as shown in Figure 16 (ANH, SWT, TMS) being under-remunerated and becoming less ‘efficient’ and those with lower expenditure (SVT, YKY, WSX, SRN) being over-remunerated and becoming more ‘efficient’. The clearest example would be that of Wessex, whom despite not investing in STW growth during the period, would see an improvement of 2% in their efficiency score because of the botex+ model predicting them a higher expenditure requirement. As such, if we were to assess a botex+ model through our Model assessment framework and selection criteria, we would fail these models due to a lack of transparency in the dependent and from an economic justification standpoint given the misallocation that occurs.

Copyright © United Utilities Water Limited 2018 49

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 16 Average expenditure to address growth at sewage treatment works (excluding sludge treatment) per property (2011/12 - 2016/17)

This misallocation of expenditure between companies occurs even when using an enhancement that is fairly highly correlated with the scale driver, the lower the correlation, the even poorer the predictions will be. As stated, this is a general rule that we have followed but we have addressed each of the proposed enhancements to include within botex separately below. Whilst botex models might technically be capable of predicting the required expenditure if the scale driver is property numbers, we do not believe that policy makers can include new development and growth (water and wastewater) enhancement expenditure within botex models for three reasons. Firstly, as we exclude them from the BCT development, we need to adjust for grants and contributions as a policy item, making an explicit valuation of the gross predicted expenditure of greater importance in generating an appropriate adjustment. Secondly, and more importantly, the new connections charging rules which come into force from April 2020 (Ofwat, 2017); specifically how changes to the income offset impacts asset payments and incomes, mean that whilst the changes are neutral from a net totex perspective, the historic reporting of gross expenditure and grants and contributions are no longer comparable. This means that whilst the development of a botex+ model may be possible, it is not appropriate to use within the BCT development as it would be inconsistent with future charging rules regarding asset payments. Finally, the links to the developer services incentive mechanism will be strengthened if an explicit prediction is made for gross new development expenditure to serve as a baseline in which the actual forecast values of developer activity are used. If an independent assessment of this activity is developed then it will create a discrepancy between the expenditure assumption within baseline and the proposed activities by companies against which they will be assessed for the developer services forecasting incentive mechanism. We do not believe that it is appropriate to predict the expenditure to enhance the balance of supply and demand at a company level by explanatory factors that are included within botex models. Not only do companies have very different long-term views on future supply/demand (im)balances as evidenced by the draft Water Resource Management Plans, a company can make a wide variety of interventions in order to address their particular supply/demand imbalance with the cost of these interventions differing significantly depending on the nature of the investment. Such variety is evident within the historic data whereby the per connection intervention costs during the period vary from no expenditure up to £22 per connection (Southern

Copyright © United Utilities Water Limited 2018 50

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Water) and is highlighted within Figure 17. For these reasons, it makes it inappropriate to include this within a botex assessment, which would then remunerate them the average industry unit cost scaled up by whatever scale variable is utilised within the botex model.

Figure 17 Average expenditure to enhance the balance of supply and demand per connected property (2011/12-2016/17)

There are endogeneity issues faced by using activity based measures (the number of meters installed) as proposed in the Modelling enhancement expenditure to predict expenditure requirements for metering. However, we do not believe that a suitable alternative is to include expenditure associated with metering (excluding metering to new connections) within botex models. Including this expenditure within botex makes two untenable assumptions. Firstly, it assumes that meter enhancement activity is dependent only on aggregate scale and not as a response to addressing other issues such as affordability, leakage, or a supply demand imbalance, neither of which will be captured by the scale variable. Secondly, it will implicitly assume that companies have the same levels of current meter penetration and therefore remunerate companies based on the scale of historic investment strategies. Implicitly then, a company could end up being predicted expenditure to achieve greater than 100% meter penetration which is obviously unfeasible. Table 10 2016-17 Percentage of Households and Non-households billed for measured water

Company Meter Penetration Company Meter

Penetration Company Meter Penetration

SRN 84% DVW 62% NES 44% SWT 77% SES 51% SVT 44% ANH 77% AFW 51% WSH 42% SEW 71% YKY 50% NWT 40% SBW 70% BRL 50% TMS 38% WSX 62% SSC 44% PRT 31%

Copyright © United Utilities Water Limited 2018 51

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 18 Average expenditure assocaited with metering per connected property (2011/12 - 2016/17)

Table 10 and Figure 18 both highlight the high levels of variety across the industry that should preclude the inclusion of this expenditure within the botex assessment. We do not believe that it is appropriate to include the expenditure to improve resilience within the expenditure for botex modelling in water or wastewater for two reasons. Firstly, within Water, a single company (Severn Trent) has dominated historic expenditure for this line within the industry, presumably due to the Elan Valley Aqueduct Resilience scheme. This will affect cost assessment by both incorrectly assessing that company as being more inefficient than would otherwise be the case or by remunerating all companies in the future a proportion of this historic investment that is clearly atypical in nature. Secondly, as this line is has been sparingly used in the past (other than for Severn Trent) it stands to reason that it is associated more with atypical and large investments which are better off addressed separately and not included within a botex assessment. Given that companies are likely to be placing additional focus on resilience in future periods, it may be more valuable to maintain this separation from botex in the event of any cost adjustment claims raised by companies and any resulting implicit allowance discussions. The underlying assumption that growth in the size of treatment works will correlate to growths in some scale variables (properties or population) does warrant consideration to include expenditure to address growth at sewage treatment works (excluding sludge treatment) within botex models. However, we have not proposed to include this within the expenditure for botex modelling due to the variability across the industry in the historic expenditure incurred in dealing with growth as illustrated at the beginning of this section in Figure 16. As a basic scale driver or botex+ model is unlikely to capture these requirements (as investment is not due to changes in aggregate scale but localised supply/demand imbalances) then this will likely manifest as efficiency variances between companies rather than appropriate predictions. Furthermore, the likelihood that future interventions across the industry to address growth at treatment works can require significant investments that could not be predicted by a botex+ model and so may require cost adjustment claims, which are more amenable to separate assessments of enhancement. Whilst the argument can quite easily be made for its inclusion (they are now operationally companies’ sewers just like any other) we do not believe that it is appropriate to include the expenditure related to transferred private sewers and pumping stations within the expenditure for botex modelling for two reasons. Firstly, whilst the length of formerly private sewers adopted by companies does correlate fairly reasonably with other

Copyright © United Utilities Water Limited 2018 52

Chapter 7: Supplementary Document - S6002

unitedutilities.com

scale drivers, namely the total number of properties (R2 = 0.86), company investment activities in enhancing those assets since adoption have been significantly different from one another indicating different strategies and/or asset conditions upon adoption. If the investment strategies are caused by timing differences then it were logical to deduce that those companies who have not historically invested will need to invest more in future periods and therefore a simple scale derived assumption would not be sufficient. If the investment strategies are caused by differences in asset conditions then this is company specific and so is better captured through separate enhancement models, as this will prevent over remuneration for those companies that adopted a higher quality sewer. Indeed, as Figure 19 shows, some companies have not incurred any expenditure on enhancing transferred private sewers Figure 19 Average transferred private sewers and pumping stations per total number of properties (2011/12 - 2016/17)

Lastly, we do not believe that it is appropriate to include the expenditure to reduce flooding risk for properties within the expenditure for botex modelling. The need for increased investment into reducing the flood risk for properties is dependent upon the amounts of rainfall and the levels of urban run-off and not the size of a company. Given the move to more comparable reporting of flooding performance, there may be increased scrutiny placed on flooding expenditure and so separate reporting of the expenditure assumptions may be more beneficial. Whilst some factors can be included within a botex model that might appropriately capture these differences (T6001 - Arup & Vivid Economics, 2017), these investments can often be large and atypical schemes (particularly for hydraulic investments). Therefore, it may not be that historical information can be an accurate prediction of future requirements as each scheme can be unique for the topographical requirements. This will increase the likelihood of cost adjustment claims being required for significant investments that are easier to address if assessing expenditure separately.

Copyright © United Utilities Water Limited 2018 53

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 20 Average expenditure to reduce flooding risk per total number of properties (2011/12 - 2016/17)

1.5.3 Alternative approach

Key findings and decisions:

We believe that the most appropriate method by which Ofwat should assess all enhancement expenditure is using the evidence provided by the company for each individual enhancement area in line with IN 18/11.

In light of guidance within IN 18/11 (Ofwat, 2018), the final approach, which is also our proposed approach, would be to assess all enhancement expenditure planned by each company separately using the evidence contained within the respective business plans and data tables. This approach gives the best opportunity to setting an effective baseline as the burden of generating the information is passed from Ofwat onto the companies, whilst simultaneously preventing any windfall gains due to lower than forecast activity on the part of the company being made to the detriment of their customers. Ofwat can still use a variety of information such as the unit cost models, independent engineering intelligence or even comparative company assessments of future unit costs as a basis for setting an initial assumption but the burden of information is on the company to evidence that their proposed expenditure is what their customers want and incurred efficiently. The benefit of this approach is that any material variations to the business plan can be assessed during the initial assessment of plans (IAP) using the evidence provided by companies rather than subsequent submissions of cost adjustment claims based on the inability of a historic industry model to capture the atypical and unique requirements of a particular future enhancement. Some of the most significant areas of enhancement are likely to be those associated with the WINEP and given the push to achieving even lower consents than historically has been the case. In a large proportion of cases, this will result in changes in technology or higher than previous unit costs to achieve and so assessments based on historic costs will not be reflective of future requirements. Importantly, it also negates the need to make an assumption for unmodelled

Copyright © United Utilities Water Limited 2018 54

Chapter 7: Supplementary Document - S6002

unitedutilities.com

expenditure, which, even with a correction mechanism as was in place at PR1430, results in an ineffective initial baseline prediction. Whilst it is probable that the majority of expenditure within this assessment will be driven by capital intensive schemes, the opex impact of AMP7 investments may need to be accounted for within this assessment. Depending upon the explanatory factors included within the botex models it may be that some of these growths are capable of being predicted through the changes in botex over time e.g., whether increases in scale will predict the increased expenditure requirements. One disadvantage of solely utilising this approach to assessing enhancement expenditure is that is removes any potential for companies to be seen to be submitting stretching plans through outperforming the BCT and obtaining a more favourable cost sharing rate as a consequence31. This is primarily a disadvantage to the companies and not to the customer, nevertheless companies should take greater comfort from the increased certainty and transparency around the outcome and the ability for their entire programme to be justifiable and not just those elements that pass a materiality threshold.

30 The correction mechanism for the unmodelled uplift at PR14 assigned the unmodelled total based on industry historic % weights of unmodelled expenditure and then recast it based on the AMP6 business plan where over-predicted lines were equated to the business plan and the residual prediction reallocated to “over-bid” lines which were then assessed separately. Deep dives were conducted for over-bid lines that passed the materiality threshold (1% of business plan totex) to see if further adjustments to the baseline were required. 31 As even if all enhancement expenditure is accepted the highest value that can be added to the baseline is the value within the business plan.

Copyright © United Utilities Water Limited 2018 55

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.6 Setting an effective baseline Having reviewed the explanatory factors that we consider most appropriate to include within base totex and enhancement totex models, we now move to translating those models into an assessment of efficient overall totex for each price control. Before we discuss the approach to setting an effective baseline, it is important to distinguish what an effective baseline is and what it is not. It is not a constraint, nor is it a target or an allowance, it is an independent, initial assumption of expenditure required for a company with certain characteristics32 to operate, maintain and enhance the service offered to customers for the five-year period. As such, deviations from the baseline should not be attributed solely to differences in efficiency without proper consideration of the ability to predict costs due to regional differences and relative service quality. An effective baseline is an important component of the overall regulatory contract as not only does it help prevent customers paying for inefficiently incurred expenditure but also provides reasonable remuneration for companies to deliver an efficient service. It strikes the right balance of risk between customer and company, working in conjunction with the various incentive mechanisms that are in place. If the balance of risk is not properly appropriated between the customer and the company, it is ultimately the customer that will be worse off in the long run either as a result of higher bills than would have otherwise been, or a lower quality service offering as companies accept inappropriate levels of risk. The previous section focused solely on the assessment of econometric models from which we can derive the ‘basic cost threshold’ (BCT) which is simply one component of the wider cost assessment undertaken. Whilst this does enable a comparative position against a significant proportion of the expenditure within a company business plan, it alone does not constitute the setting of an effective baseline. This section focuses on how the econometric assessment is supplemented in order to develop an effective baseline covering: • How to setting of an efficient frontier using both historical and future information; and • Assessment of a company’s business plan that are not covered by the totex models, referred to as ‘policy

items’ Finally, we look at how an effective baseline provides the right signals in light of the various incentive mechanisms such as those for totex and performance commitments/outcome delivery incentives (ODIs). 1.6.1 Triangulation

Key findings and decisions:

We have developed a unique and innovative approach to triangulation that weights each suite(s) on their predictive ability for companies individually.

Having adopted a diverse suite of benchmarking models (as set out in sections 1.3 and 1.4), an important part of developing an effective baseline is the manner in which we combine the results from each of these models in order to form a single prediction for each price control, for each company. Triangulation is not only important from the point of view of the final prediction but it also affects the efficiency adjustment that is derived from the models as it influences the spread of residuals across companies and time. Triangulation can occur at a variety of levels, at an individual model level within a suite or at an aggregate level across potential suites, or both.

32 The explanatory factors of any given company used in the econometric models.

Copyright © United Utilities Water Limited 2018 56

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 21 Options for applying triangulation to proposed suites within the Water price controls

Because companies are heterogeneous, likely requiring different explanatory factors to one another to explain their costs within their company specific circumstances, it is practical to postulate that models that utilise more of a single company’s specific explanatory factors will be more accurate predictors of their efficient cost than those that do not. This is not to say that they will automatically predict higher expenditure requirements (indeed they may predict lower), rather they will predict a more accurate view of the requirements for a company with those characteristics. The impact, and ultimately the benefits of appropriate triangulation are shown in Figure 22 below. Each of the individual models contains different explanatory factors to attempt to predict the same expenditure. This results in different residuals for each company depending on the how well each model can predict the expenditure for that company. Taking a suite of diverse models and then triangulating the results therefore narrows the spread of residuals significantly indicating a more appropriate balance between companies.

Copyright © United Utilities Water Limited 2018 57

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 22 Triangulating within a suite between Water Network plus models to reduce the spread of residuals

For PR14, Ofwat implemented a simple approach to triangulation whereby it averaged the results of each of the component models within a suite and then averaged across suites in order to construct the final cost prediction. Simple averaging therefore implicitly assumes that all models and all suites are equally suited to predicting expenditure and that no one approach can generate a more realistic baseline for a company. The greater the degree of homogeneity between the companies then the less important targeted triangulation becomes as the resulting model predictions should account for all companies circumstances equally. However, if company circumstances are more heterogeneous, or if there are threats to the internal validity of the models caused by e.g. omitted variable bias or errors (see section 1.6.2.3 Data quality and measurement error/attenuation bias), then triangulation on an individual company basis can be a useful tool to use to minimise the risks of extreme outliers as a result of model deficiencies. There are a number of instances where regulators have historically used a more targeted approach to triangulation as a method of correcting for modelling deficiencies. The German energy regulator, Bundesnetzagentur, utilises both SFA and DEA33 approaches when developing their benchmarks for each distribution system operator (DSO). Accepting that neither statistical method is perfect and that the different approaches will have varying success in predicting required expenditure for individual companies, they determine the actual efficiency score (ES) using a ‘best off approach’ whereby the maximum efficiency score is applied, with a minimum of 60%.

𝐸𝐸𝐸𝐸 = max(𝐷𝐷𝐸𝐸𝐴𝐴1,𝐷𝐷𝐸𝐸𝐴𝐴2, 𝐸𝐸𝑆𝑆𝐴𝐴1, 𝐸𝐸𝑆𝑆𝐴𝐴2, 0.6)

33 Stochastic frontier analysis (SFA) and data envelopment analysis (DEA) are alternative statistical approaches for the estimation of comparative benchmarks.

Copyright © United Utilities Water Limited 2018 58

Chapter 7: Supplementary Document - S6002

unitedutilities.com

The same principal occurs within Austrian gas distribution regulation but in this case, the weighting assigned to the ‘best off’ estimation method (DEA and MOLS34) is pre-determined using a 60/40 ratio with the larger weighting applied to the models that give the highest ES.

𝐸𝐸𝐸𝐸 = 0.6 × max(𝐷𝐷𝐸𝐸𝐴𝐴,𝑀𝑀𝑀𝑀𝑀𝑀𝐸𝐸) + 0.4 × min(𝐷𝐷𝐸𝐸𝐴𝐴,𝑀𝑀𝑀𝑀𝑀𝑀𝐸𝐸) “The weighting of the different approaches is a compromise between industry and the regulator. Originally, the regulator argued in favour of an equal weighting of DEA(average) and MOLS(average), whereas the industry preferred a best-off calculation, which means that the highest score of all four models should be used to determine the cost reduction requirements” (WIK-Consult, 2011). Whilst we recognise that such weightings are somewhat arbitrary, we agree that it is unattainable to expect that any one model can act as a perfect prediction for every company within the industry. As a result, an appropriately triangulated diverse suite of models is the most suitable method to pursue in order to ensure that the resulting baselines are a fair reflection of each company’s required expenditure. Rather than arbitrarily assigning a greater weighting to the suite/model that result in the largest prediction, we believe that minimising the residuals (and therefore equally addressing both over and under predictions) is the most appropriate approach to take as it prevents baselines being set too high because of a model/suite overestimating the requirements. In implementing this decision, we have developed the following matrix whereby weightings (Figure 23 showing a 50% gap between the ‘best’ and ‘worst model/suite and Figure 24 showing 100% gap), whether they are applied at a model level or a suite level, scale linearly based on the performance and minimising the resulting residuals for each company. The matrix determines the ‘gap’ between the weightings of the ‘best’ and ‘worst’ model or suite. The resulting weights (in the blue box) can then be applied to the suites or models based on the rankings of the residuals for each company (the ratio between actual cost and modelled cost). Figure 23 UUW proposed triangulation matrix using 50% gap between high and low weightings

34 MOLS = modified ordinary least squares

Copyright © United Utilities Water Limited 2018 59

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 24 UUW proposed triangulation matrix using 100% gap between high and low weightings

We have illustrated the results of 50% and 100% weightings but the scaling can be set to any level which best reflects the suitability of the models/suites to predict costs for all companies. A higher weighting places a larger differential between the ‘best’ model/suite and the ‘worst’ model/suite that will minimise the sum of the residuals further. Economic priors and the spread and magnitude of the residuals will largely inform the most appropriate final weighting requirements and using our proposed models. We have set a 50% differential between the highest and lowest weighting and applied the resulting triangulation weightings to the suites set out in section 1.2.2 ‘Aggregated or disaggregated botex modelling?’ rather than to the individual models themselves (although that would be another equally valid approach). This gap between the highest and lowest apportionment is consistent with the triangulation allowed by the Austrian energy regulator between SFA and DEA (a 60/40 split). The benefit of adopting this approach over simple averaging or arbitrarily weighting based on resulting predictions is that averaging implies that each model/suite is equally adept at predicting requirements for all companies (which we know not to be true) and an arbitrary weighting will simply result in predictions increasing for all companies which is to the detriment of customers. Weighting models/suites by seeking to minimise the residuals on the other hand will prevent models or suites that are too generous for a company obtaining as high of a weighting whilst minimising undue challenge due to a particularly poor performing model/suite. 1.6.2 Efficiency adjustment

Key findings and decisions:

The perceived ‘outperformance’ by companies during the current period is largely due to outperformance of enhancement expenditure rather than botex that is caused, in part, by the non-delivery of NEP schemes rather than efficiency.

The poor performance of the PR14 Ww enhancement models drove more than half of the efficiency adjustment at PR14 (6%) despite comprising only a quarter of historic expenditure.

We see no justification in considering a position more towards the frontier than the Upper-Quartile is appropriate given the likelihood of measurement error.

A key aspect of setting an effective baseline is an appropriate and well-specified efficiency adjustment. As a benchmarking exercise, using econometric models is solely backwards looking, then the resulting predictions too are backward looking. This means that the expenditure predicted by these benchmarking models are that of a company with average historical efficiency. As such, adjustments are required in order to ensure that we only include efficient expenditure within the baseline both now and in the future. In this section, we subdivide the efficiency adjustment into two components, one that ensures that companies ‘catch up’ to a current

Copyright © United Utilities Water Limited 2018 60

Chapter 7: Supplementary Document - S6002

unitedutilities.com

efficient benchmark, based on econometric analysis of historic performance (static efficiency) and one that ensures companies continue to deliver further efficiencies in future periods (dynamic efficiency) thereby ensuring that the baseline remains effective throughout the future period. 1.6.2.1 Perception of AMP6 performance When setting static and dynamic efficiency adjustments, it is natural to consider the potential for companies to deliver future efficiencies by reference to historic performance against cost assumptions set at previous price controls. Ofwat has previously stated that “In the first two years of this AMP period more than half of companies outperformed their innovative TOTEX allowances which were set at PR14” (Ofwat, 2018), which may lead to the perception that the efficiency adjustment was too lax and that it should be tougher at PR19. However, whilst companies have invariably sought to deliver greater efficiencies and innovation, the observation of lower costs in AMP6 risks confusing efficiency savings with differences in the levels of activity (or non-delivery), particularly within enhancement. Analysis of company performance across the industry for the first 3 years of the current period does not support this suggestion of widespread outperformance. Whilst admittedly, some companies have seen some reduced expenditure against their final menu expenditure assumption, analysing the variances in the type of expenditure implies that whilst savings may have occurred on totex, there has been little or no overall outperformance of PR14 botex assumptions to date, but widespread outperformance of expected enhancement programmes. Whether these cost reductions are due to under-delivery of enhancement programmes, timing or increased efficiency against the PR14 models is unknown to us without further details of each company’s actual deliveries. Figure 25 Actual Water (out)/under-performance against PR14 allowed menu expenditure split by investment type (excl Thames)35

35 We have excluded Thames from both the Water and Wastewater analysis, as there have been significant movements in expenditure from enhancement to maintenance that is assumed to be due to accounting differences between PR14 and now rather than due to legitimate changes in investment.

Copyright © United Utilities Water Limited 2018 61

Chapter 7: Supplementary Document - S6002

unitedutilities.com

The assumption is even more apparent within Wastewater where botex performance has been largely in line with PR14 assumptions but the outperformance on enhancement expenditure is significant as shown in Figure 26. Figure 26 Actual Wastewater (out)/under-performance against PR14 allowed menu expenditure split by investment type (excl Thames)

Broadly speaking, when a baseline is determined it assumes two things: • most often discussed, the assumed level of efficiency that should be delivered by the company over the

period; and • most often forgotten, the assumed level of activity that underpins the assessment of efficient cost. It would be inappropriate to conclude that apparent cost outperformance resulted from a too relaxed approach to the efficiency adjustment at PR14, rather than to investigate whether the baseline incorrectly represented unnecessary levels of activity for a specific or group of companies, leading to an ineffective baseline. As the review periods of the environmental regulators are not align with that of the Price Review, some of the more costly areas of a company’s business plan are benchmarked by reference to unconfirmed requirements. Without an appropriate correction mechanism in place, this has meant that if a company subsequently received fewer requirements than previously assumed, then this results in totex outperformance, with the perception that this has been achieved through greater efficiency, rather than a different level of activity. The key conclusions of this are twofold. • First, the upper quartile targets set by Ofwat were challenging in respect of Botex costs with the industry

not showing any significant outperformance against the PR14 assumptions in Water or Wastewater. • Secondly, there is a need to take a different approach to assessing required activity levels when assessing

the efficient cost of enhancement activities. 1.6.2.2 Use of forecast activity levels For PR19 (and AMP7), Ofwat has sought for companies to propose cost adjustment mechanisms, to account for ex post changes in relation to unconfirmed requirements – this should remove the potential for future

Copyright © United Utilities Water Limited 2018 62

Chapter 7: Supplementary Document - S6002

unitedutilities.com

windfall gains. Forecasting the level of activity for a given company is therefore an extremely important part of the effective baseline derivation and particular care should be taken in setting baselines independent of company business plans and subsequently making judgements on efficiency.

“Like driving a car by looking in the rear view mirror.”

W. Edwards Deming

For PR14, the exercise of forecasting future activity levels for botex was largely a desktop exercise (Jacobs, 2014) which simply projected forwards using historic data (either a trend, and average or a single point estimate). This approach may be acceptable if the assumption is that what has occurred in the past will continue to occur in the future but leads to an ineffective baseline if circumstances of historic trends change. (Quality) Enhancement expenditure was forecast using the most up to date version of the National Environmental Programme (NEP) at the time. Whilst it is likely to give the most comparable view of future activity requirements in relation the business plan expenditure, the use of company forecasts for all explanatory factors within the baseline is not something that we would necessarily recommend as being a practical approach as there is not always protection for customers from companies arbitrarily inflating future forecasts. Where there is some form of customer protection or an appropriate incentive mechanism designed to promote accurate forecasting of explanatory factors, we believe derivation of the baseline must use company forecast values in order for it to be effective. This currently applies to all factors that are associated with the developer services incentive mechanism, the WINEP cost adjustment mechanism and the Bioresources volume forecasting incentive mechanism. However, many other ODIs are based on forecast levels of activity, such that any arbitrary changes to forecasts for the purpose of cost assessment risks misalignment with performance incentives, which could harm the overall coherence of a company’s price determination, to the detriment of customers or the company. If we use any value other than the company central estimate for these factors in the derivation of the baseline then it will not produce comparable results or support the incentive mechanism thereby making the resulting baseline ineffective36. We do appreciate there may be risk in using a company’s forecast of explanatory factors given the potential to overinflate forecasts in search of a more favourable cost assessment (where there are not mechanisms to protect from this e.g. developer services forecasts). For this reason, a desktop exercise may be more pragmatic from a policy maker’s point of view in deriving the initial baseline. This does not mean that we should disregard company forecasts completely, even if significant gaps do not occur between the baseline and the business plan, a check of proposed activity levels to ensure that the baseline is effective and comparable is required as comparative performance may be over/understated due to activity assumptions. Once we are comfortable that a baseline appropriately reflects the appropriate level of activity for a company, we can make assessments over cost (and therefore efficiency) more confidently, as there can be greater certainty that the baseline is effective. 1.6.2.3 Data quality and measurement error/attenuation bias It is important to recognise the strengths and weaknesses of any quantitative analysis, particularly when using the results to predict future requirements. One of the key risk areas, not just for regression analysis, is the quality of the data used and the relative certainty of the reported value. With a small number of observations available, the quality and accuracy of the data becomes even more important as the impact of inconsistencies are emphasised which can lead to biased model results.

36 Where forecasts are substantially different to historic trends we would expect companies to have provided sufficient explanation within their table commentary and/or business plan but further challenge may still be appropriate.

Copyright © United Utilities Water Limited 2018 63

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Uncertainty and issues around data quality within the industry datashare means that, even balanced suites of high quality models will contain error and, without due care, measurement error may be introduced into the efficiency adjustment. Errors can be introduced into the prediction by one of two routes (or both); • Errors or inconsistencies in companies' reporting of data on costs drivers (for example, as indicated by

companies reporting wide confidence intervals for certain data items) • Cost data is also potentially unreliable, with interpretations of the reporting guidelines and quality of

measurement differing between companies. o For example, one company reallocation assigned 26% of base costs to enhancement37.

Our models all draw on the same dataset, with the same expenditure and attributes for companies in each. If equally credible model specifications, drawing on the same data, produce different efficiency adjustments then it is impossible for these different efficiency scores to reflect real variations in efficiency across companies, and must therefore be due to the inability of individual models to predict cost accurately. This further reiterates why we cannot presume that any individual model (regardless of how well it passes statistical tests) could be a perfect predictor of efficient cost. Therefore, the appropriate level of efficiency adjustment should be informed by other sources, rather than simplistically using a “frontier” position derived from model residuals. This has two implications for both the models and the resulting efficiency adjustment: • Measurement error may be driving a large proportion of unexplained variation between companies

depending on the variables utilised; • It is likely impractical to correct for the effect of measurement error for in the predicted expenditure for

each company – however, we can (and should) consider it in selecting the most appropriate level of efficiency for the static adjustment.

Adopting a suite of models that predict different levels of aggregation can reduce the risk of measurement error within the dependent variable but reducing the errors within the explanatory variables is more difficult. We could adopt an explicit rule whereby we only include variables that have a minimum confidence rating, but this might then risk excluding some key cost drivers from the model development causing omitted variable bias within the results. Additionally, the likelihood is that models that incorporate variables with wide confidence intervals will produce insignificant results and so would be discounted from model selection. The alternative approach then would be to seek to understand the level of error within model results and allow for this in the selection of the static efficiency adjustment. In attempting to do so, Vivid (T6005 - Arup & Vivid Economics, 2018) undertook Monte Carlo simulations of differing assessments of explanatory variables based on their confidence intervals to highlight the potential error within a model result. The impact of these measurement errors on the resulting predictions can be significant and affect companies differently. They estimated that the impact of measurement error on the PR14 wastewater models could be as much as £520m across the industry per AMP for historic Wastewater expenditure, which is further accentuated through the impact on the efficiency adjustment that added a further £590m of potential variation. Determining the level and impact of measurement error within the AMP7 proposed models will help to understand the overall level of confidence that can be placed on the ability of the models to predict expenditure requirements appropriately for all companies. Ofwat can then use this information to inform the most suitable level of adjustment to place on the resulting BCTs.

37 Whilst queries from companies and Ofwat helped eliminate a significant amount of these issues, there are likely still inconsistencies within the observations.

(T6005 - Arup & Vivid Economics, 2018) “Measurement error produces ranges in assessed company costs worth many hundreds of millions of pounds, which will be compounded by further error in the efficiency challenge.”

Copyright © United Utilities Water Limited 2018 64

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.6.2.4 Static efficiency adjustment The first part of setting an efficiency adjustment within an effective baseline relates to static efficiency, also referred to as the ‘catch-up’ efficiency. It typically involves applying a percentage reduction based on a percentile ranked company historic performance against the models used38 as illustrated by Figure 27 below. At PR14, the upper-quartile ranked company for both Water and Wastewater BCT was used to set the efficiency adjustment, but this does not mean that the PR19 adjustment has to be the set on the same basis, the upper-third or even the median might be more appropriate. Figure 27 illustrating the potential choices for the efficiency frontier using the residuals within the proposed Water BCT.

In deriving the efficiency adjustment, it is important that we calculate the efficiency adjustment at the aggregate service level (Water and Wastewater) rather than at an individual model or value chain split in order to prevent issues that arise due to the substitutability of inputs between the services that lead to comparative differences within a business unit. As highlighted in section 1.2.2 Aggregated or disaggregated botex modelling?, companies can have perceived relative efficiencies to others within a price control simply by the configuration of their assets with respect to the boundary definition. This means that parts of the value chain can appear comparatively efficient or inefficient simply due to differing asset configurations between network plus and resources, or different operational strategies. It would be inappropriate to set separate efficiency adjustment for e.g. Wastewater Network Plus and Bioresources, as this would lead to an artificially high percentile being applied that no company would be able to achieve.

38 It is therefore an assessment of the residuals of the predicted values of the company using the benchmarking models relative to the actual expenditure incurred by that company for the period.

Copyright © United Utilities Water Limited 2018 65

Chapter 7: Supplementary Document - S6002

unitedutilities.com

As there are asymmetric totex sharing incentive rates and opportunities to make efficiency savings between the Bioresources and Network Plus control, it is not an equal trade off and therefore may unduly benefit or penalise companies. As a result, it may be more appropriate to allow for disaggregation of an aggregate efficiency assumption, particularly if that assumption is excessively stretching. An important point to note is that static efficiency percentages for companies are based on the residuals of the models, not an independent assessment of actual company relative efficiencies (the internal validity of the models). Whilst it would be entirely possible to conduct a comparative assessment of actual efficiency between all companies and then to apply an individual adjustment based on the results, this would disassociate it from the results of the benchmarking exercise and it is pivotal to maintain this association if an effective baseline is to be constructed. This is because, when the static efficiency adjustment remains a function of company historic performance using these models, it captures the same effects in both the efficiency derivation as well as the forecast baseline position. Whilst this can lead to issues of circularity (especially in the case of omitted variable bias), it is consistent and therefore leads to the most transparent outcome. One issue with this approach is that, without any corrections, it infers that all unexplained differences between the prediction and the actual expenditure incurred (the residual) is solely down to efficiency – which is unlikely given the aforementioned reasons or measurement error, industry complexity leading to omitted variables, and so on. In choosing the appropriate level of efficiency adjustment to be applied, it is important that we understand the residuals for each company and therefore the relative certainty we can apply in setting an efficiency adjustment. Rather than viewing the residual as being solely driven by company inefficiency, it important to remember that the residual for any observation consists of noise, omitted variables and then ‘true’ inefficiency as illustrated by Figure 28. This means that adopting a position that is on the frontier (COLS) will inherently bake noise, omitted variable and measurement error into the adjustment. Figure 28 Understanding the composition of the residual for all predictions

Practitioners may attempt to use statistical methods such as stochastic frontier analysis (SFA) as a means to better explain variations between the baseline and actual costs. These methods enable the residual to be decomposed into ‘noise’, firm effects and true underlying differences in costs (efficiency), but are often performed on large data samples (thousands) rather than the 107 that we have within Water (or 60 within Wastewater) and so is less appropriate for such a small sample size. In their assessment of potential benchmarking within cost assessment, Vivid Economics investigated the potential to utilise stochastic frontier

Copyright © United Utilities Water Limited 2018 66

Chapter 7: Supplementary Document - S6002

unitedutilities.com

analysis in decomposing the residuals and better inform the selection of the efficient frontier but found that it was not sufficiently robust to enable a credible decomposition of the residual.

(T6005 - Arup & Vivid Economics, 2018) “Stochastic Frontier Analysis (SFA)…lacks robustness in this dataset due to serial correlation between explanatory variables…. the procedure is not robust to omitted variable biases or one-off forms of measurement error. Such persistent effects would be reflected in company efficiency terms, rather than the noise term. The strength of the evidence shown suggests that efficiency approaches which continue to conflate these effects do not represent an improvement over the UQ efficiency score approach.”

Without these types of analysis available to use, it is important not to lose sight of the fact that predicted residuals are not solely variations in efficiency levels between companies. This means that when determining the most appropriate percentile of residual adjustment, we must first consider both the accuracy with which the benchmarking models make their predictions (the overall spread of residuals) as well as the accuracy of the underlying information used to develop the benchmarking assessment. As mentioned in section 1.5 Assessing enhancement expenditure, the inability of enhancement models to capture the company specific and atypical nature of many enhancements can cause a bias within the resulting static efficiency adjustment. For PR14, this was most evident within the Wastewater static efficiency adjustment39 (Ofwat, 2013) as illustrated in the tables below. The variance between companies ‘efficiency’ scores on enhancement models was unrealistically large with almost 100% variation between the 1st and 10th companies. These wide ranging results point more towards the inability of the models to predict the activities undertaken by each of the companies within the period more than they do to actual variations in relative efficiency achieved. Table 11 PR14 Wastewater enhancement model suite and resulting efficiency rankings

Rank (most efficient

first)

Company Actual cost / predicted cost

Enhancement model implied ‘efficiency’

1 Yorkshire 0.5214 48% 2 Southern 0.6615 34% 3 Wessex 0.6798 32% 4 Severn Trent 0.7171 28% 5 South West 0.8407 16% 6 Welsh 0.9197 8% 7 United Utilities 0.9784 2% 8 Anglian 0.9961 0% 9 Thames 1.1043 -10%

10 Northumbrian 1.4638 -46% The upper quartile ‘efficiency’ adjustment on enhancement alone was over 30%, highlighting the lack of predictive ability within the suite. This impact flows through to the overall efficiency rankings albeit with a lower weighting as enhancement expenditure forms a smaller percentage of totex. By removing the enhancement expenditure completely from the efficiency calculation, we can see the Wastewater enhancement models that were utilised at PR14 drove almost 6% (i.e. more than half of the total 10% wastewater efficiency target) of the static efficiency adjustment applied to totex, despite only reflecting 27% of totex within the efficiency calculation. The problem with this is that the same efficiency adjustment applied equally to all aspects of totex, irrespective of the size of the enhancement programme for each company. This

39 Two thirds of the PR14 Water models were totex models and therefore did not include separate enhancement models.

Copyright © United Utilities Water Limited 2018 67

Chapter 7: Supplementary Document - S6002

unitedutilities.com

equated to an additional £1,148m efficiency adjustment on Wastewater botex across the industry at PR14 than would have been the case under a more appropriate efficiency derivation. Table 12 Impact of removing enhancement from the PR14 Wastewater efficiency rankings

Interpretation PR14 Efficiency factor Efficiency factor PR14 botex only

Variance due to enhancement

Frontier -25.19% -8.38% -16.81% Upper third -7.20% -3.74% -3.46% Upper quartile -10.40% -4.41% -5.99% Median -0.45% -0.19% -0.26% Lower third 1.30% 1.96% -0.65% Lower quartile 3.72% 2.56% 1.17% Least efficient 7.85% 16.08% -8.23%

Our proposed approach to assessing enhancement expenditure within cost assessment is to utilise an Alternative approach whereby Ofwat assess each area of enhancement on a case-by-case basis, with companies being responsible for providing sufficient evidence to justify why expenditure is required and incurred efficiently. Adopting this approach automatically removes the enhancement expenditure from the efficiency calculation, thereby avoiding similar issues as faced at PR14. Even if Ofwat do not adopt this approach to assessing enhancement requirements, given the clear difficulties faced in modelling enhancement expenditure, we believe that it is only correct to calculate the efficiency adjustments separately for botex and enhancement in order to prevent the same issues occurring at PR19. The percentiles that result from our proposed models (as set out in sections 1.3.2 and 1.4.2) for Water and Wastewater are within Table 13 and Table 14 below. Table 13 Water efficiency rankings for proposed models and suites

Percentile Top down: Minimal split

Bottom up: price control

Bottom up: Resources (plus)

Triangulated40

Frontier -15.0% -13.7% -20.1% -15.8% Upper Quartile -4.0% -9.3% -4.4% -6.2% Upper Third -2.6% -7.7% -1.1% -4.3% Median 0.7% -2.0% 4.0% 0.5% Lower Third 3.6% 4.2% 6.3% 4.5% Lower Quartile 5.5% 15.2% 6.9% 9.8% Least efficient 27.6% 26.0% 27.0% 26.8%

Table 14 Wastewater efficiency rankings for proposed models and suites

Percentile Top down: Minimal Split

Bottom up: price control

Bottom up: PR14 Split

Triangulated41

Frontier -9.1% -9.7% -15.0% -11.6% Upper Quartile -1.8% -6.7% -5.7% -4.7% Upper Third -1.7% -6.2% 0.6% -2.0% Median 0.3% 2.5% 2.1% 1.6% Lower Third 3.6% 4.7% 7.2% 5.3% Lower Quartile 3.8% 7.4% 9.0% 6.8% Least efficient 9.9% 14.8% 11.0% 11.7%

Whilst the first and last companies within these suites can be large outliers, the residual spread across the rest of the industry is reasonably narrow, indicating that these suites are well suited to estimating efficient costs

40 Using United Utilities’ triangulation weightings with a 50% gap between the highest and lowest suite as per section 1.6.1. This applies weights of 33.3%, 40.0% and 26.7% respectively across the three suites. 41 Using United Utilities’ triangulation weightings with a 50% gap between the highest and lowest suite as per section 1.6.1. This applies weights of 33.3%, 26.7% and 40.0% respectively across the three suites.

Copyright © United Utilities Water Limited 2018 68

Chapter 7: Supplementary Document - S6002

unitedutilities.com

for the majority of the industry. It is logical to presume that those companies that are towards the bottom of the rankings are likely to have submitted cost adjustment claims for those areas where they believe that industry benchmarking models cannot capture their circumstances sufficiently which may lead to 2-sided adjustments causing a further narrowing of the results (albeit not impacting the efficiency rankings). As we can see from the tables above, for both Water and wastewater, equally reasonable modelling suites can produce different levels of unexplained variation, including at the upper quartile. As the Model assessment framework and selection criteria explains, the models and the suites are deliberately chosen to minimise errors and biases and the suites themselves are considered equally valid, but they nonetheless produce markedly different distributions of unexplained variation. The rankings of the companies do change slightly between the suites that reflects the appropriateness of each suite for that company, supports the basis for appropriate triangulation rather than pointing towards inappropriate models. We observe this variation despite there being no underlying difference in relative efficiency in the data. There was a single efficiency adjustment for PR14, based on the upper quartile ranked company to all expenditure generated by the BCT. We are proposing to assess all enhancement expenditure separately which means that our proposed static efficiency adjustment apply to botex predictions only. Given that the suite of models utilised at PR14 were over-fitted (T6001 - Arup & Vivid Economics, 2017), their spread of residuals was narrower than would have been the case under the approaches that we have adopted for generating the PR19 baselines. As a result, we believe that applying the same upper quartile adjustment against our proposed suite of models for PR19 is a more stretching adjustment than the efficiency adjustment applied at PR14. The impact of Data quality and measurement error/attenuation bias on the results means that any percentile beyond the upper quartile, risks baking into the efficiency adjustment some of these discrepancies. If Ofwat decides to assess enhancement expenditure through a unit cost assessment rather than by an alternative approach, we believe that the relative predictive abilities of botex and enhancement modelling should facilitate applying a separate efficiency adjustment to enhancement and base expenditure in deriving the baseline for each company. Furthermore, given the likely differences in scale between company enhancement programmes, it is actually necessary to apply separate distinct static efficiency adjustments to botex and enhancement costs, rather than a single combined efficiency percentage. 1.6.2.5 Dynamic efficiency adjustment Comparative econometric cost models support assessment of “static” differences in efficiency – i.e. differences in efficiency at the point in time that the measurement is made. They do not contain any inherent forecast of future changes in cost relative to the inflation assumption applied (CPIH in the case of PR19). In general, this may consist of two main components, for which we must make an assumption to inform the assessment of efficient costs: • Change in input prices beyond those reflected in the measure of inflation – this may include inflation of

specific items that do not form part of the CPIH basket. It can also include sector specific differences in the balance of inputs, which may indicate that one should expect input prices to increase (or decrease) relative to the main inflation measure. For example, it may be that energy prices are expected to rise above forecast CPIH, and that energy costs for a greater part of water company costs than is assumed in the CPIH basket.

• Differences in sector specific productivity trends relative to economy-wide productivity gains reflected in CPIH – it is important to remember that measures of inflation reflect economy-wide productivity gains. Therefore if Ofwat assumed zero (relative to CPIH) for its dynamic efficiency assumption, it would be incorrect to conclude that no dynamic efficiency is assumed – it is just that no additional dynamic efficiency is assumed over and above that being delivered by the economy as a whole, as embodied within the inflation measure.

On the latter point, it is also important to note that Ofwat changing the main inflation measure applied to costs (from RPI to CPIH) implicitly applies an additional dynamic efficiency expectation of c.1% per annum due

Copyright © United Utilities Water Limited 2018 69

Chapter 7: Supplementary Document - S6002

unitedutilities.com

to the differences between RPI and CPIH42. At PR14, Ofwat made no explicit assumption for dynamic efficiency beyond application of RPI. If Ofwat were to make the same assumptions again (i.e. no dynamic efficiency relative to inflation applied) then that would, in effect, assume an additional 1% per annum dynamic efficiency adjustment onto the industry relative to PR14. It is common for regulators and companies to make use of analyses of Total Factor Productivity (TFP) to inform expectations for dynamic efficiency. Such studies (generally using the ubiquitous EU KLEMS dataset) commonly observe historic trends in productivity gains of between 0% and 1% relative to RPI. KPMG (KPMG, 2018) observed similar results in its recently work for Ofwat, which indicated the following results (which KPMG confirmed were evaluated relative to RPI, not relative to CPIH): • Wholesale: 0.4% to 1.2% per annum (relative to RPI) • Retail: 0.8% to 1.8% per annum (relative to RPI)

In order to translate these values into appropriate PR19 assumptions, one must make two adjustments (a) to deduct 1% pa to reflect the differences between RPI and CPIH, and (b) to deduct inflation altogether from the assessment of Retail, as Ofwat assumes that Retail businesses can absorb all inflationary pressures (i.e. to additionally deduct 2% pa from Retail to reflect non-application of CPI(H) to the Retail cost assessment). Adjusting for those two factors results in the following expectations for dynamic efficiency, based on comparable TFP analysis: • Wholesale: -0.6% to 0.2% per annum (relative to Ofwat’s PR19 inflation assumption for wholesale costs of

CPIH) • Retail: -2.2% to -1.2% per annum (relative to Ofwat’s PR19 inflation assumption for retail costs of zero) NB. negative values imply cost growth relative to the measure of inflation. It is also helpful to note that these findings are not too dissimilar (albeit somewhat more challenging) to the findings of previous TFP studies performed during PR09, by both First Economics (First Economics, 2008) for Water UK and by Reckon for Ofwat: • First Economics concluded c.0% pa relative to RPI for operating costs, and -1% to -2% pa dynamic

efficiency relative to RPI for capital costs (i.e. costs growth relative to RPI) • Reckon also concluded 0% pa relative to RPI for operating costs, and 0.5% pa dynamic efficiency relative to

RPI for capital costs

It is also worth noting some of the further analysis undertaken in these more comprehensive studies, which is not evident within the published work recently undertaken by KPMG for Ofwat. In particular, First Economics sought to decompose TPF between companies which had the scope to obtain productivity gains from substantial global outsourcing, and those (such as the water sector) where such opportunities were limited. First Economics concluded:

(First Economics, 2008) “…it would be a bold move to expect businesses from anywhere but the goods sector to reduce costs in real terms. Substantial outsourcing of work to developing economies appears to be a pre-requisite for entry into the group of firms whose costs are falling in real terms. Having a UK based workforce would appear to place a firm among companies that see their costs move up in real terms.”

42 Differences in the basket of goods/services used to calculate each index as well as the impact of the ‘formula effect’

Copyright © United Utilities Water Limited 2018 70

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Whilst such studies are clearly dated, the themes remain valid, and the underlying values remain in the historic productivity datasets used to estimate current estimates of TFP. In conclusion, we believe it is entirely reasonable that Ofwat should conclude that a reasonable expectation for the net impact of dynamic efficiency and real price effects for the water sector is to assume some cost growth (not cost reductions) relative to CPIH. There is certainly little evidence to support the application of any adjustment in addition to CPIH. This is supported both the KPMG’s analysis of TFP (rebased to the PR19 context) and relevant information from historic studies. We recommend that Ofwat assume: • Wholesale: Net Dynamic Efficiency of -0.2% pa relative to CPIH (i.e. 0.2% pa cost growth relative to CPIH).

This is the midpoint of KPMG’s assumptions, rebased to CPIH. We believe that the midpoint is sufficiently stretching given the lack of opportunities for global outsourcing as noted by First Economics in its report from 2008.

• Retail: Net Dynamic efficiency of -1.5% pa (i.e. cost growth of 1.5% per annum relative to Ofwat’s assumption of zero inflation for retail businesses). This towards most stretching end of KPMG’s assessment.

We have therefore reflected these values within our estimates of cost assessment, which we have used to demonstrate that our PR19 cost proposals are both efficient and stretching. In applying this adjustment, rather than making an annual adjustment, which risks perverse incentives for the company to alter the profile of expenditure, we calculated the average percentage for the AMP and applied this value to the total AMP7 baseline for each price control which can then be pro-rated back. It is further worth noting that Ofwat should not feel that such assumptions are lenient towards companies as they imply cost growth relative to the applicable inflation measure. They are actually (in conjunction with the applicable inflation measure) significantly more stretching that those applied at PR14. We understand the attraction for Ofwat in representing that water companies should be expected to deliver productivity gains relative to inflation (CPIH for wholesale, zero for retail) – however such an assumption would be indefensible in light of the available evidence, including Ofwat own evidence produced by KPMG. Such an assumptions would also fail to enable companies to fairly represent the level of stretch being assumed within company plans. UU has proposed to deliver efficiencies in AMP7 in excess of these assumptions – however, those proposals should be judged by reference to a reasonably assessed benchmark (as we have set out above), not an arbitrary assumption. Consequently, it is also helpful to compare these values to those we have reported as part of table App24a – Real price effects (RPEs) and efficiency gains, which sets out the increasing levels of efficiency which we plan to deliver over AMP7. • Wholesale – annual efficiency net of RPE from App24a = 2.82% pa • Retail – annual efficiency net of RPE from App24a = 1.10% pa This clearly shows that our plan is delivering significantly higher levels of expected efficiency delivery that our reasonable estimates of dynamic efficiency, demonstrating the level of ambition and stretch that our plan is delivering to the benefit of customers. Finally, it is important that Ofwat fully recognises the ambition and stretch being assumed within our plan – i.e. that it is recognised that the gap between our efficiency proposals (within table App24a) and our reasonable assessment of dynamic efficiency represents genuine stretch beyond reasonable expectations – not that these values should inform an unreasonable expectation of sector-wide dynamic efficiency.

Copyright © United Utilities Water Limited 2018 71

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.6.3 Adjustments to the baseline

Key findings and decisions:

We have made adjustments for ‘Policy items’ in line with PR14 and the costs excluded from modelling.

The changes to future charging rules has complicated the assessment of new development expenditure and the associated grants and contributions.

Adjustments are required for differences between AMP7 and the datashare accounting principles to correct for liquor treatment recharges, IFRS16 and principal use opex recharges.

A regional wage adjustment is not required as it is immaterial for all companies and the uncertainties around the true input price pressures faced by companies.

As highlighted in section 1.2.1 Data and modelled costs, we exclude some types of expenditure from the econometric model assessment on the basis that the available explanatory factors cannot predict the correct costs, that they are sufficiently outside of management control or that proposed incentive mechanisms in AMP7 preclude their inclusion. In order to derive the totex baseline for each price control, we add those excluded costs that will continue in AMP7 to the BCT to form the. For PR14, such adjustments were often referred to as ‘policy items’ and a separate assessment of the required expenditure to be included within the baseline was conducted. The table below summarises the net adjustments to account for these policy items that are required to the basic cost thresholds in order to derive the baseline for each price control.

Table 15 Summary of all AMP7 adjustments to the baseline for each Wholesale price control £m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus 55.496 53.730 53.833 51.794 50.748 265.599 Water Resources 36.075 35.825 35.722 35.529 35.481 178.634 Wastewater Network Plus (19.501) (17.073) (16.761) (14.236) (12.452) (80.023) Bioresources 20.807 20.688 20.865 20.818 20.918 104.097 Total 92.877 93.170 93.658 93.906 94.695 468.307

This rest of this section assess each of these policy items in detail, including the potential impact of revaluations occurring during the AMP7 period for cumulo rates and abstraction charges. 1.6.3.1 Grants and contributions Table 16 Summary totex adjustments to the baseline because of Grants and contributions.

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus (10.061) (9.972) (9.882) (10.259) (10.681) (50.855) Water Resources 0.000 0.000 0.000 0.000 0.000 0.000 Wastewater Network Plus (15.240) (15.401) (15.564) (15.328) (15.101) (76.634) Bioresources 0.000 0.000 0.000 0.000 0.000 0.000 Total (25.301) (25.374) (25.445) (25.586) (25.782) (127.489)

For PR14, enhancement and botex model predictions within the BCT were net of capital grants and contributions using the historic assumptions, but we have constructed our suite of models gross of grants and contributions (i.e. not including) and so we adjust the BCT ex post to correct for this. Companies will specify the amount and types of grants and contributions that they expect to receive for the activities undertaken

Copyright © United Utilities Water Limited 2018 72

Chapter 7: Supplementary Document - S6002

unitedutilities.com

within their business plans43. At a high level, we need to make two adjustments for grants and contributions, one to botex (diversions income) and one to enhancement activities (all other income). The correct method by which to calculate the enhancement adjustment depends heavily on how new development and growth expenditure is assessed within Water and Wastewater enhancement (as grants and contributions are a function of gross expenditure). If, as we have proposed, we assess gross enhancement totex through an Alternative approach then we can just subtract the associated business plan grants and contributions (or a percentage of if the full value of gross totex is not accepted) to form the correct net totex within the baseline. Furthermore, the new connections charging rules that come into force from April 2020 (Ofwat, 2017); specifically how changes to the “income offset” impacts asset payments and incomes, mean that whilst the change is neutral from a net totex perspective, the historic reporting of gross expenditure and grants and contributions are no longer comparable to future predictions. Our proposed method is simple, transparent and more importantly in line with companies’ future charging rules which is particularly important given the proposal to exclude grants and contributions from the AMP7 totex incentive mechanism. For all of these reasons, we believe a bottom up assessment and equitable adjustment using the business plan grants and contributions is the most appropriate for this net enhancement expenditure to be calculated. It would not appropriate to deduct the business plan grants and contributions from the BCT if making an independent assessment of gross capex where the underlying activity levels e.g. the volumes and/or types of connection, and therefore the gross totex prediction are not equivalent. Table 17 AMP7 enhancement grants and contributions received - wholesale water service (table App28)

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Connection charges (s45) (6.156) (6.496) (6.842) (7.195) (7.553) (34.243) Infrastructure charge receipts (s146) 15.723 16.814 17.925 18.585 19.216 88.263

Requisitioned mains (s43, s55 & s56) (9.462) (9.951) (10.449) (10.955) (11.470) (52.287)

Other contributions (price control) (3.027) (3.131) (3.236) (3.343) (3.451) (16.189)

Total (2.922) (2.764) (2.603) (2.908) (3.259) (14.456) Table 18 AMP7 enhancement grants and contributions received - wholesale wastewater service (App28)

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Infrastructure charge receipts (s146) (6.559) (6.624) (6.689) (6.354) (6.027) (32.253)

Requisitioned sewers (s100) (0.361) (0.365) (0.368) (0.372) (0.375) (1.841)

Other contributions (price control) (2.971) (3.011) (3.051) (3.092) (3.133) (15.258)

Total (9.892) (9.999) (10.108) (9.817) (9.535) (49.352) For diversions income, the gross expenditure is not readily identifiable within the business plan tables and so a simple approach is likely the only option available to policy makers. As diversion activities can often correlate with scale, we propose to adjust the baseline by applying an adjustment to the BCT (once correcting for adjustments due to RAG changes) based on the percentage of modelled botex within a company’s business plan that is associated with diversions income.

43 PR19 Business plan table App28 – Developer services (Wholesale)

Copyright © United Utilities Water Limited 2018 73

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Table 19 Derivation of the required adjustment to the baseline to account for Water diversions income £m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Diversions (s185) (6.290) (6.352) (6.414) (6.477) (6.540) (32.074) Business plan modelled botex 277.477 305.974 325.224 294.274 266.842 1,469.790

Percentage of business plan -2.18% Modelled botex 320.156 321.117 321.981 323.229 324.531 1,611.014 RAG adjustments - IFRS16 (0.104) (0.128) (0.106) (0.104) (0.128) (0.569) RAG adjustments - Principal use 13.841 11.982 11.931 10.221 9.575 57.549

Modelled botex (consistent RAG) 333.893 332.971 333.806 333.346 333.978 1,667.994

Diversions adjustment (7.286) (7.266) (7.284) (7.274) (7.288) (36.399) Table 20 Derivation of the required adjustment to the baseline to account for Wastewater diversions income

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Diversions (s185) (4.712) (4.759) (4.806) (4.854) (4.903) (24.033) Business plan modelled botex 316.263 332.358 329.340 357.910 334.654 1,670.526

Percentage of business plan -1.44%

Modelled botex 400.020 402.693 404.361 406.424 413.777 2,027.275 RAG adjustments - IFRS16 (1.428) (1.426) (1.476) (1.289) (1.426) (7.044) RAG adjustments - Principal use (18.310) (15.888) (15.807) (13.755) (12.965) (76.724)

RAG adjustments - Liquor recharges (9.234) (9.301) (9.387) (9.506) (9.663) (47.091)

Modelled botex (consistent RAG) 371.048 376.078 377.691 381.874 389.723 1,896.416

Diversions adjustment (5.338) (5.410) (5.434) (5.494) (5.607) (27.283) For both Water and Wastewater given the proposed botex models, this results in applying a larger downward adjustment than if we were simply to just deduct the business plan diversions income from the BCT but we feel that this better reflects the relative predictive capabilities of the models when compared to the business plan. It is important to make the RAG adjustments prior to the derivation to ensure that the percentage derived from the business plan is comparable. 1.6.3.2 Water Cumulo rates and Wastewater business rates Table 21 Summary totex adjustments because of Water Cumulo rates & Wastewater business rates (tables WS7/WWS7)

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus 46.926 46.926 46.926 46.926 46.926 234.632 Water Resources 17.903 17.903 17.903 17.903 17.903 89.514 Wastewater Network Plus 23.624 23.850 24.371 24.532 25.583 121.960 Bioresources 7.147 7.147 7.147 7.147 7.147 35.733 Total 95.600 95.826 96.347 96.507 97.559 481.839

The rates for our operational water assets are assessed on a cumulo basis, which results in United Utilities being allocated an RV (Rateable Value) for all water assets. This RV is multiplied by a Business Rates Multiplier (BRM) to result in a rates liability for the year. The Valuation Office Agency (VOA) is responsible for setting RV’s

Copyright © United Utilities Water Limited 2018 74

Chapter 7: Supplementary Document - S6002

unitedutilities.com

and BRM’s44. The RV is reassessed periodically with the most recent revaluation on 1st April 2017, the next revaluation will be on 1st April 2021, and then every 3 years after that, meaning that there will be two revaluations in AMP7. Due to the high level of uncertainty surrounding the forthcoming revaluations, we forecast the total rates for AMP7 by taking the 2017-18 operational water business rates expenditure, deducting the one-off rates refund that artificially lowered the 2017-18 rates for operational sites (as per business plan table WS7) and apply this to all years in AMP7. This approach is consistent with that taken by Ofwat at PR14 in deriving the baseline for Water and so we add these values to the baseline rather than applying any other adjustment. The rates for our operational wastewater assets are assessed on a site-by-site basis, which results in United Utilities getting an individual RV (Rateable Value) for each of their 426 Wastewater sites. This RV is then multiplied by a Business Rates Multiplier (BRM) to result in a rates liability for the year. The Valuation Office Agency (VOA) is responsible for setting RV’s and BRM’s. The RV is reassessed periodically with the most recent revaluation on 1st April 2017, the next revaluation will be on 1st April 2021, and then every 3 years after that, meaning that there will be two revaluations in AMP7. The RV’s for each site can be reassessed on an ad hoc basis and if the assets on site change significantly, the VOA may want to adjust the RV between rating periods. For sites where we are aware that additional civil building works have taken place since the last valuation, we accrue for the potential additional business rates liability that may be charged if the VOA decide to reassess that site, this is called the growth accrual. There is also a transitional scheme in place for Wastewater RV changes, where if the RV of the Wastewater sites comes down45 there is a limit to the level of annual reduction. The assumption for the baseline mirrors that of the business plan, which accounts for changes in the asset stock as new assets are constructed within Wastewater Network Plus and a constant transitional relief. We assume that business rates within Bioresources are flat throughout the period as there are no new assets expected to be constructed. We believe that this is a credible and transparent approach to assessing future requirements and have added these values to each of the baselines. 1.6.3.3 Abstraction charges & EA fees Table 22 Summary totex adjustments because of Abstraction charges & EA fees (table WS1)

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus 0.303 0.303 0.303 0.303 0.303 1.514 Water Resources 16.564 16.564 16.564 16.564 16.564 82.818 Total 16.866 16.866 16.866 16.866 16.866 84.332

The Environment Agency (EA) sets abstraction charges nationally, and each company has limited management control available to influence the level of charge applied other than to control the amount of licenses (permits) that it requires. Abstraction charges are set to recover the regulator’s costs of water resource management and apportioned between the total amounts of licenses (permits) that are in operation at that time. The Water abstraction management reform in England could potentially lead to changes to the structure of the charging mechanism and therefore there is a large amount of uncertainty surrounding the future abstraction charges. These costs form a substantial proportion of Water Resources total expenditure requirements and any change to the charging mechanism could have significant impacts if not taken into account within the baseline. The recent review of discharge consents within Wastewater will result in a substantial increase46 to the amount paid and a comparable increase to abstraction charges would be difficult

44 The BRM is subject to inflation each year and while historically this has been RPI, this changed to CPI from 1st April 2018 meaning that it is broadly in line with the inflation measure used to index expenditure. 45 Which it tends to do at each reassessment date, because generally the age of the wastewater assets increases over time, which means a reduced RV. 46 PR19 Business plan table WWS5 - Other wholesale wastewater expenditure, Block B. Approximately a 25% increase in costs within Wastewater discharge consents.

Copyright © United Utilities Water Limited 2018 75

Chapter 7: Supplementary Document - S6002

unitedutilities.com

to absorb. “These changes are not intended to come into force until the early 2020s. The move to a reformed system will be gradual and more detailed guidance and information will follow as we make progress towards reform” (Department for Environment Food & Rural Affairs, 2016). Given this uncertainty, we have provisionally maintained the most recent levels of charging (2017-18) within our Water Resources business plan for the AMP7 period and included this value within the baseline. Should more information regarding future charges across the industry become available, we would expect that Ofwat updated this policy item (and business plan equivalent) prior to the final determination. The expenditures paid to the Canals & Rivers Trust (British Waterways Board) for the utilisation of Llangollen Canal and subsequent abstraction at Huntington as well as the service charge paid to Severn Trent for the utilisation of Vyrnwy are included within this. 1.6.3.4 Industrial emissions directive (IED) Table 23 Summary totex adjustments because of Industrial emissions directive (IED) (table WWS5)

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Wastewater Network Plus 0.180 0.177 0.177 0.178 0.179 0.892 Bioresources 1.638 1.735 1.758 1.767 1.775 8.673 Total 1.818 1.912 1.935 1.945 1.954 9.564

As set out within section 1.2.1 Data and modelled costs we have excluded the reported expenditure associated with the Industrial Emissions Directive from our benchmarking models. These costs are borne by a small number of companies (those that have incineration assets) and are therefore best assessed outside of the botex models. We will continue to incur these costs in AMP7 to a similar level as experienced in the current period and therefore an adjustment to the baselines is needed to account for this. Historically, all sites with IED permits were classed as co-located with the IED costs being reported in Sludge Treatment. With revised site operations and application of RAG guidance, some of these sites are now classed as Wastewater Network Plus and so adjustments are required to both controls and not just the Bioresources baseline. The 2017-18 reported costs47 of £1.443m are for the permits and the manpower costs associated with inspecting and maintaining assets to the required permit standard. The forward forecast assumes that there will be additional costs for maintenance, power for assets specifically required for the permits and the costs of new permits. Within Bioresources, these costs are included within the cost adjustment claim for “Distance to landbank”. Should IED costs be assessed as a policy item as we have proposed then we would need to remove the IED component from the cost adjustment claim in order to prevent any double counting of requirements. 1.6.3.5 Traffic Management Act Table 24 Summary totex adjustments because of Traffic Management Act (tables WS5/WWS5)

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus 3.312 3.345 3.377 3.411 3.444 16.889 Water Resources 0.000 0.000 0.000 0.000 0.000 0.000 Wastewater Network Plus 0.781 0.788 0.796 0.804 0.812 3.980 Bioresources 0.000 0.000 0.000 0.000 0.000 0.000 Total 4.093 4.133 4.173 4.214 4.256 20.869

Costs associated with Traffic Management Act relate to permit costs that are segregated based upon job type. The costs do not include fines or the costs of employees administering permits, as our view is that these employees existed to administer notification schemes prior to the commencement of any permit schemes. In 2018-19, 3 Highways Authorities (HA) plan to come online (Liverpool, Cumbria and Blackpool) which would see step change increase in costs meaning all HA’s in the United Utilities area would be operating a permitting scheme. As such, the numbers reflect half-year cost impact in 2018-19 and full year effect from 2019-20 onwards. From 2019-20 onwards, the costs remain broadly similar as we expect the number of jobs to remain

47 PR19 Business plan table WWS5 – Other wholesale wastewater expenditure

Copyright © United Utilities Water Limited 2018 76

Chapter 7: Supplementary Document - S6002

unitedutilities.com

broadly static and the risk from recent changes allowing for Lane Rental charges (which would effectively become a permit cost) is not yet known within our region. There is a slight increase in costs, as we believe the inflationary cost pressures would be above CPIH in line with contractual cost increases. 1.6.3.6 Third party services Table 25 Summary totex adjustments because of Third party services (tables WS1/WWS1)

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus 1.278 1.274 1.283 1.295 1.309 6.438 Water Resources 0.000 0.000 0.000 0.000 0.000 0.000 Wastewater Network Plus 0.127 0.128 0.128 0.128 0.128 0.639 Bioresources 0.006 0.006 0.006 0.006 0.006 0.030 Total 1.411 1.408 1.417 1.429 1.443 7.107

This is the total expenditure for any third party services provided by the Water and Wastewater Network Plus businesses incurred because of correcting damages to United Utilities assets by a third party. Given that it is clearly difficult to predict the future levels for these costs, we have used historic averages for the purposes of the business plan. 1.6.3.7 Statutory water softening United Utilities has no requirements to soften water and so no adjustment is required for this policy item. 1.6.3.8 Pension deficit repair costs As set out within IN 13/17 (Ofwat, 2013), companies have a predetermined amount of time over which they are permitted to recover a proportion of their pension deficit repair costs from customers, after which the liability is borne solely by the company. Whilst some companies’ recovery period extend into future periods, the final assumed payment year for United Utilities is 2019-20 and as such, no adjustment is required to the baseline and any future payments from 2020 onwards will be at the company’s expense. 1.6.3.9 RAG changes – Bioresources liquor treatment recharges Table 26 Summary totex adjustments because of RAG changes – Bioresources liquor treatment recharges

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Wastewater Network Plus (9.234) (9.301) (9.387) (9.506) (9.663) (47.091) Bioresources 9.234 9.301 9.387 9.506 9.663 47.091 Total 0.000 0.000 0.000 0.000 0.000 0.000

Appendix 6 of the final methodology confirmed, “When sludge liquors are returned to a wastewater treatment works, the activity of treating the liquors is a network plus wastewater activity. The cost of treating liquors should be paid for by the Bioresources business.” (Ofwat, 2017) This is a new condition for companies to account for and therefore the costs are not present within the historically reported expenditure that develops the basic cost thresholds. As the Bioresources and Wastewater Network Plus price controls are fully separated with their own totex incentive rates, an adjustment to each control needs to be made to account for the payments between the two businesses. As set out within “995_Using_markets_and_innovation_Chapter6_UUW_IAP5”, we have calculated the expected recharges in AMP7 using relevant “Mogden” components of our Trade Effluent charges, to ensure that the recharges are cost-reflective, transparent and provide a level playing field with third parties and made these adjustments to each of the baselines set out above.

Copyright © United Utilities Water Limited 2018 77

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.6.3.10 RAG changes – principal use allocation and opex recharges Table 27 Summary totex adjustments because of RAG changes – principal use allocation and opex recharges

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus 13.841 11.982 11.931 10.221 9.575 57.549 Water Resources 1.617 1.367 1.264 1.071 1.023 6.341 Wastewater Network Plus (18.310) (15.888) (15.807) (13.755) (12.965) (76.724) Bioresources 2.852 2.539 2.612 2.462 2.367 12.833 Total 0.000 0.000 0.000 0.000 0.000 0.000

Regulatory accounting guideline (RAG) 2.07 requires that, where possible, companies allocate capital expenditure and associated depreciation to one of the price control units. For shared assets (where this is not possible), a single price control owns the asset and makes principal use recharges to other price controls to prevent any cross-subsidisation. Consequently, for annual regulatory reporting and the cost assessment information request (Ofwat, 2017 July Cost Assmt Information Request, 2018), on which botex models are developed, these principal use recharges are made as depreciation recharges and therefore they are not included as totex within the Wholesale business units. Clarification received in responses to PR19 queries48 (Ofwat, 2018) requires companies to account for these recharges as part of totex (rather than other operational costs, outside of totex), within each of the Wholesale business plans. On this basis, we derive the basic cost thresholds on a different accounting standard to that which companies submit their business plans making it necessary to adjust the baseline to correct for allocative differences. It is important that baselines and business plan assumptions be set on a comparable basis as differing totex sharing incentive rates between price controls could lead to outperformance/underperformance against a baseline solely as a result of accounting differences. As the sole aim of this adjustment is to enable comparability between totex assumptions, the adjustment can be to the baseline (which we have done) or the inverse of the proposed could be made to the business plans for the Wholesale controls. Ultimately, the ‘correct’ method by which to make the adjustment will be that which aligns with the future reporting requirements and the basis on which Ofwat will reconcile AMP7 totex performance at PR24. 1.6.3.11 RAG changes – IFRS16 Table 28 Summary totex adjustments because of RAG changes – IFRS16

£m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus (0.104) (0.128) (0.106) (0.104) (0.128) (0.569) Water Resources (0.009) (0.009) (0.009) (0.009) (0.009) (0.043) Wastewater Network Plus (1.428) (1.426) (1.476) (1.289) (1.426) (7.044) Bioresources (0.070) (0.040) (0.045) (0.070) (0.040) (0.264) Total (1.610) (1.602) (1.635) (1.471) (1.602) (7.921)

The International Accounting Standards Board (IASB) issued IFRS 16 Leases in January 2016. Changes to accounting standards mean that operating leases, whose costs were previously expensed, will be brought onto company balance sheets for statutory accounting purposes. The rationale for the change is to improve comparability between companies irrespective of the way assets are reported and to increase transparency of future liabilities. The new accounting standard requires companies to capitalise leased assets and record both an asset and a liability in the balance sheet. Water companies with existing operating leases that remain in place will see an increase in debt liabilities once the new accounting standard takes effect and a reduction in operating expenses. As at the time of the information request (Ofwat, 2017 July Cost Assmt Information Request, 2018), on which botex models are developed, these leases were expensed, the resulting basic cost thresholds will be derived

48 Query reference numbers 12, 250, 549 and 576 : PR19 final methodology queries and answers – 25 June 2018

Copyright © United Utilities Water Limited 2018 78

Chapter 7: Supplementary Document - S6002

unitedutilities.com

on a different accounting standard to that which companies submit their business plans making an adjustment to the baseline necessary to correct for accounting differences. We have summarised the adjustments to both capex and opex due to adopting this new standard, with the net reduction to totex being £7.921m across all Wholesale price controls. Table 29 Opex adjustments because of RAG changes – IFRS16 (table WS1a/WWS1a/App33)

Opex £m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus (0.206) (0.206) (0.206) (0.206) (0.206) (1.031) Water Resources (0.009) (0.009) (0.009) (0.009) (0.009) (0.043) Wastewater Network Plus (1.874) (1.874) (1.874) (1.874) (1.874) (9.369) Bioresources (0.070) (0.070) (0.070) (0.070) (0.070) (0.348) Opex total (2.158) (2.158) (2.158) (2.158) (2.158) (10.792)

Table 30 Capex adjustments because of RAG changes – IFRS16 (table WS1a/WWS1a/App33)

Capex £m 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Water network plus 0.102 0.078 0.101 0.102 0.078 0.462 Water Resources 0.000 0.000 0.000 0.000 0.000 0.000 Wastewater Network Plus 0.446 0.448 0.398 0.585 0.448 2.326 Bioresources 0.000 0.030 0.025 0.000 0.030 0.084 Capex total 0.548 0.556 0.523 0.687 0.556 2.872

1.6.3.12 The impact of regional wages To varying degrees across the Water industry, gross labour (i.e. including that which is capitalised) costs can form a significant proportion of total expenditure requirements and to some extent, are determined by market forces, which may be viewed as exogenous and outside of management control. Given the poor and irregular performance of variables to represent regional wage differences within econometric models both at PR14 and within the consultation, we have looked at whether an ex-post adjustment is required to correct for any misallocations between companies within the BCT generation. In assessing the potential impact that regional market pressures have on a company’s expenditure requirements, we first need to understand the metrics often used for comparative purposes. The most commonly used measures utilise the ONS’ Annual Survey of Hours and Earnings (ASHE), which reports the hourly wages and number of jobs across the different regions of the country. Standard Occupation Classification (SOC) or Standard Industrial Classification (SIC) can disaggregate this information further so that we can develop a more granular assessment. Previously, Ofwat have utilised the SOC as well as their own mapping of company boundaries to geographic regions to derive a regional wage for each company. The PR14 regional wage developed by CEPA (CEPA, 2014) utilised only two (two-digit) SOC codes, Science, research, engineering and technology professionals (SOC 21); and Skilled Construction and Building Trades (SOC 53), with a weighting of 60/40 respectively to reflect the expected higher utilisation of skilled labour across the industry. Since then, Ofwat has sought to refine its approach to calculating a regional wage by expanding the assessment to include more SOC codes as well as developing it at both the one-digit and two-digit SOC codes. The revised approach now utilises 12 two-digit SOC codes (7 one-digit SOC codes) with weightings informed, in part, by company submissions. Notably, the SOC codes utilised for the PR14 constructed variable only form 19% of the overall weighting in Water and 16% in Wastewater within the new indices. In addition to applying a more granular assessment of occupational weightings, Ofwat has made a further improvement that assumes that 70% of a company’s wages are subject to regional pressures, with the remaining 30% sourced from ‘national’ labour markets and therefore subject to the same inflationary pressures for all companies. However, this assumption applies to all occupations equally rather than weighting more toward high-skilled labour as you might expect given the unpinning logic (higher mobility of labour amongst higher skilled workers) to the inclusion of this adjustment factor. Given that we would expect to see a larger differential in wages between the high skilled labour (across the various sectors rather than regions) than you would for low skilled labour, then this will eliminate less of the regional variation than what was originally intended due to e.g. the impact of the financial sector being focused around one region (London).

Copyright © United Utilities Water Limited 2018 79

Chapter 7: Supplementary Document - S6002

unitedutilities.com

We can derive the resulting indices for each company; typically displaying the mean or the median of the reported values within the ASHE, so that comparative assessments can be made across the industry as within Figure 29. Figure 29 2016 SOC1 mean for Water, ASHE, Table 15.6a Hourly pay - Excluding overtime (£) - For all employee jobs, using Ofwat percentage allocations for company regions and SOC1 weightings.

Simply looking at the different average hourly wages for each company in this format might lead the reader to jump to the conclusion that regional wage differences do exist between companies and that an adjustment is required in order to correct for the differences that are exogenous in nature. However, we need to address several key issues before we can draw a definitive conclusion as to the relative differences between regions for wages and therefore if an adjustment to the baselines are required. Firstly, utilising SOC data from the ASHE has the disadvantage of including all sectors in the economy within the derivation rather than using wage information that is more relative to the Water industry. This means that certain (higher paying) industries will artificially bias the results for some regions, notably the finance and insurance sectors that are centralised around London and the South East of England. It is unlikely that these sectors will share the same pool of labour for most roles (particularly highly skilled roles) and therefore will not be in direct competition with one another for labour, meaning that wages will not be comparable or correlated between these sectors and the Water industry. In their investigation into the potential to use SIC data within the assessment, Vivid (T6005 - Arup & Vivid Economics, 2018) calculated that the Finance sector drove 16% of the regional wage difference that one would measure for Thames, using regional wage data from the ASHE, when compared to the average. Whilst there are undoubtedly some sectors that will derive labour from the same pools as the Water industry, for example the construction industry, it is vital that we exclude those sectors that do not utilise the same pools of labour and therefore skew the results. Additionally, high skilled labour is more mobile than low skilled labour meaning that in order to attract and retain high skilled labour, companies must offer competitive wages irrespective of geographic location. This result of this is that company’s then source from a national labour market for occupations rather than a regional one with no regional variation between companies. Ofwat has attempted to account for this in its 70/30 weighting between regional and national wages but we believe that this underestimates the adjustment factor for two reasons. The adjustment applies to all occupations whether

Copyright © United Utilities Water Limited 2018 80

Chapter 7: Supplementary Document - S6002

unitedutilities.com

they are low or high skilled labour. The reality is that low skilled labour is less mobile and therefore more subject to regional differences in wages meaning that this adjustment factor is less appropriate. Whilst the percentage weighting of high skilled to low skilled labour within the company allocations is broadly even, the monetary value associated with each is not and therefore applying the adjustment evenly will over remunerate those companies that operate in London and the South East as they retain a greater proportion of the bias caused by the e.g. financial sectors. The second issue, which needs addressing, is what statistic to choose in reporting the differences between companies? Previously, CEPA used the mean as they believed “it better captures the distribution of earnings within the occupation category” (CEPA, 2014), but given that it includes sectors that are unrelated to the Water industry, it may be more appropriate to choose another level as the basis for comparisons. In reporting the base data for SOC wages, the ONS report not only the mean and median values for each SOC but also the percentiles49 within the distribution. We can incorporate this data into the assessment by deriving the percentiles for each company. Whilst Figure 29 seemingly showed a clear picture of the relative differences between companies, Figure 30 shows the range of potential results for each company given the distribution of wages within each SOC code. These ranges are wide reducing the confidence that the mean or median are truly an accurate reflection of regional wage pressures faced by the company. Figure 30 2016 SOC1 percentiles for Water, ASHE, Table 15.6a Hourly pay - Excluding overtime (£) - For all employee jobs, using Ofwat percentage allocations for company regions and SOC1 weightings.

To use the mean for comparative purposes implicitly assumes that each company is the ‘average’ company within that region. For some regions, this assumption may not be appropriate. Indeed, it might be more appropriate for companies that operate in relatively ‘poorer’ regions with limited financial sector influence, they are assessed as ‘higher than average’ for their region given the relative skill sets required for the labour employed meaning the P70 wage could be more reflective of the costs faced when hiring and vice versa. There is nothing to say that all companies have to use the same percentile or statistic when deriving the industry regional wage index but placing undue weight on a single result such as the mean, without

49 10, 20, 25, 30, 40, 60, 70, 75, 80, 90th percentiles are all reported in addition to the mean and the median to varying degrees of confidence, referred to as the CV (coefficient of variation).

Copyright © United Utilities Water Limited 2018 81

Chapter 7: Supplementary Document - S6002

unitedutilities.com

acknowledgement of the potential for error within the estimate risks over remunerating companies, particularly if irrelevant sectors are included within the estimate. This issue alone calls into question the validity of using any form of regional wage data to predict regional wage variations between companies, as: • It overstates the costs in high average wage regions (such as London), which include irrelevant industries,

and excludes opportunities for mitigation (such as relocating office based staff to less costly locations, greater use of remote monitoring and control technologies etc.)

• It understates the costs in lower average wage locations (such as the North), which may have higher than average skill requirements for the region and/or need to pay more to attract skilled staff to relocate into the region.

The final aspect to consider when deciding on the appropriateness of a regional wage adjustment is what proportion of costs the adjustment applies. Within the datashare (Ofwat, 2017 July Cost Assmt Information Request, 2018) companies provided their gross labour costs for all value chains against both direct and indirect labour employed. We can use this information to derive an adjustment for each company to correct for regional wage differences but as illustrated by Table 31, the materiality of these adjustments are immaterial, even for those companies with the largest differential adjustment. This calculation has been made assuming the mean regional wage for every region/company without adjustments for unrelated sectors (finance etc) and so is an overestimation of the pressures faced by companies given the uncertainties highlighted above. Table 31 Impact of regional wages/forces on average Water network plus historic expenditure for employment as a proportion of average totex (2011/12 - 2016/17)

Company Average employme

nt costs (£m)

Average totex (£m)

Average SOC1

(Mean) £/hr

Variance to

industry average

Expenditure for 'average' employment

unit costs (£m)

Increased costs due

to regional factors

(£m)

Materiality

a b c d = [c-ave(c)] /

ave(c)

e = a / (1+d) f = a - e g = f / b

ANH 41.47 311.94 £14.87 -2.21% 42.41 -0.94 -0.30% NES 29.84 215.74 £15.15 -0.35% 29.95 -0.11 -0.05% NWT 53.85 455.45 £14.76 -2.87% 55.44 -1.59 -0.35% SRN 8.51 144.27 £15.65 2.95% 8.26 0.24 0.17% SVT 30.64 510.84 £14.66 -3.55% 31.77 -1.13 -0.22% SWT 9.72 116.15 £14.61 -3.88% 10.11 -0.39 -0.34% TMS 43.71 625.66 £17.24 13.40% 38.54 5.17 0.83% WSH 26.81 260.18 £14.31 -5.86% 28.48 -1.67 -0.64% WSX 4.67 146.91 £14.61 -3.88% 4.86 -0.19 -0.13% YKY 24.10 271.08 £14.50 -4.61% 25.26 -1.16 -0.43% AFW 11.82 208.03 £16.16 6.34% 11.12 0.70 0.34% BRL 5.34 88.77 £14.61 -3.88% 5.56 -0.22 -0.24% DVW 1.79 16.12 £14.45 -4.94% 1.88 -0.09 -0.58% PRT 3.57 25.79 £15.65 2.95% 3.47 0.10 0.40% SES 2.94 42.79 £16.74 10.12% 2.67 0.27 0.63% SEW 51.75 719.62 £16.15 2.99% 50.25 1.50 0.21% SSC 39.01 357.49 £15.38 -1.96% 39.79 -0.78 -0.22%

As Table 31 illustrates, even when assessing the impact of regional wage differences using the mean, which has been shown the be unreliable and overly skewed, the cost differential for all companies is immaterial for

Copyright © United Utilities Water Limited 2018 82

Chapter 7: Supplementary Document - S6002

unitedutilities.com

the Wholesale business. We have only shown the materiality for the Water network plus price control above but the same results hold for all other Wholesale price controls where other controls are even less material50. For all of the reasons discussed above, we do not believe that there is any definitive evidence to support the application of a regional wage adjustment to any company within the industry. 1.6.3.13 Cost adjustment claims After careful consideration, we are proposing four cost adjustment claims relating to our Wholesale businesses (and one further cost adjustment claim in retail, referred to in section 2 below), to reflect the unique characteristics of operating in the North West. Depending upon the final benchmarking models selected, we believe the resulting basic cost thresholds may not fully capture these characteristics. The net values of these claims are below and the detail supporting each claim including the calculation of the implicit allowance is set out within each claim. Table 32 Wholesale proposed cost adjustment claims

Net adjustment 2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Age and number of reservoirs (WR) 5.791 8.860 12.713 11.530 12.293 51.187

Manchester and Pennines resilience (WN+) 53.312 7.590 2.615 1.832 7.332 72.681

Distance to landbank (Bio) 6.373 6.365 6.301 6.194 6.216 31.449 Surface water run-off (WwN+) 17.324 17.530 17.489 17.643 17.731 87.717

Total 82.800 40.345 39.118 37.199 43.572 243.034 Where applicable, we have calculated the implicit allowance and resulting net claim value for each cost adjustment claim using our proposed model suite, triangulation and efficiency adjustment. For the purpose of cost assessment, our proposed econometric botex models include factors that should adequately capture the unique attributes for the Bioresources and Wastewater Network Plus cost adjustment claims. As such, we have not added these values to our view of the baseline when using these models in order to prevent obvious double counting. Should the final PR19 benchmarking models fail to capture these attributes then these claims would be required in their full value.

50 We would expect Thames to have the largest materiality given their location but the results show materiality values of WR = 0.70%, WwN+ = 0.35% and Bioresources = 0.37%.

Copyright © United Utilities Water Limited 2018 83

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.6.4 Water (Network plus & Resources) totex baselines We have derived totex baselines for each of the AMP7 Water price controls, combining our proposed benchmarking models and approaches to setting an effective baseline set out in the preceding sections. These individually derived baselines provide further evidence that our business plans are stretching and pushing the efficiency frontier beyond its current level. These baselines and the ratio between that and the business plan assumptions form the basis for setting the incentive rates for the combined Water control.

Water

Resources Water

Network plus Water total Modelled Botex (comparable accounting)51 150.802 1,667.994 1,818.796

SDB Lead Metering New development Modelled Enhancement 0.000 0.000 0.000

Assessed enhancement 25.240 367.140 392.380

Net effect of RPEs and productivity improvements 0.907 10.035 10.942 Grants & contributions (enhancements) (14.456) (14.456) Grants & contributions (diversions) (36.399) (36.399) Costs associated with Traffic Management Act 0.000 16.889 16.889 Statutory water softening 0.000 0.000 0.000 Abstraction Charges / Discharge consent 82.818 1.514 84.332 Local authority and Cumulo rates 89.514 234.632 324.146 Third party services 0.030 6.438 6.468

Cost adjustment claim 51.187 0.00052 51.187

Net Totex 400.499 2,253.787 2,654.286

Gap to business plan 26.068 179.994 206.061

PR19 business plan 374.431 2,073.793 2,448.224 The resulting ratio between the baseline and business plan is 92.24% (i.e. our proposed business plan totex is 7.7% below our assessment of efficient costs), equating to a cost-sharing rate for outperformance of 57.8% and an incentive payment of £119.03m53.

51 Includes Principal use and IFRS16 adjustments 52 Claim for Manchester & Pennines resilience not required as enhancement expenditure assessed through alternative approach. 53 Using the proposed cost-sharing model provided alongside the PR19 final methodology (Ofwat, 2017).

Copyright © United Utilities Water Limited 2018 84

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.6.5 Wastewater (Network plus & Bioresources) totex baselines We have derived totex baselines for each of the AMP7 Wastewater price controls, combining our proposed benchmarking models and approaches to setting an effective baseline set out in the preceding sections. These individually derived baselines provide further evidence that our business plans are stretching and pushing the efficiency frontier beyond its current level. These baselines and the ratio between that and the business plan assumptions form the basis for setting the incentive rates for the Wastewater Network Plus price control. As confirmed within the Final Methodology for PR19 (Ofwat, 2018), the Bioresources control will not have any cost sharing mechanism between companies and customers.

Bioresources Wastewater

Network Plus Total Modelled Botex (comparable accounting)54 340.905 1,896.416 2,237.321

New development FTS EDM Storage Flooding Private sewers NEP Sanitary NEP P STW growth CIP Modelled Enhancement 0.000 0.000 0.000

Assessed enhancement 0.000 888.480 888.480

Net effect of RPEs and productivity improvements 2.051 11.409 13.460 Grants & contributions (enhancements) (49.352) (49.352) Grants & contributions (diversions) (27.283) (27.283) Costs associated with Traffic Management Act 0.000 3.980 3.980 Statutory water softening 0.000 Abstraction Charges / Discharge consent 0.000 Local authority and Cumulo rates 35.733 121.960 157.693 Costs associated with Industrial Emissions Directive 8.673 0.892 9.564 Third party services 0.639 0.639

Cost adjustment claim 0.000 0.000 0.000 Net Totex 387.362 2,847.142 3,234.504

Gap to business plan 14.911 233.918 248.829

PR19 business plan 372.450 2,613.224 2,985.674 The resulting ratio between the baseline and business plan for Wastewater network plus is 91.78% (i.e. our proposed business plan totex is 8.2% below our assessment of efficient costs), equating to a cost-sharing rate for outperformance of 58.2% and an incentive payment of £136.18m55.

54 Includes Principal use, Liquor treatment recharges and IFRS16 adjustments 55 Using the proposed cost-sharing model provided alongside the PR19 final methodology (Ofwat, 2017).

Copyright © United Utilities Water Limited 2018 85

Chapter 7: Supplementary Document - S6002

unitedutilities.com

1.6.6 Alternative approach to Bioresources cost assessment In a similar manner to how Non-household Retail was assessed at PR14, an argument could be made for cost assessment for Bioresources being less based on model results and more driven by market forces. For Non-household Retail, Ofwat used a ‘tariff corridor’ approach for 2015-20 whereby it allowed companies to set default tariffs within a range of allowed average revenue per customer type for each company using their internal cost projections for the future periods. This struck the best balance between intervention to protect customers and allowing the market to develop and companies to respond flexibly. Given the similar desire to increase competition within the Bioresources value chains, this may be an equally legitimate approach to adopt, either now or at future reviews in order to reduce the regulatory burden and stimulate the market. Allowing companies to set a range of allowed average revenue per tonne of dry solid may be appropriate if there is sufficient market pressures preventing inefficient behaviours from occurring. As the Wholesale business will still be required to trade with the incumbent Bioresources business in AMP7, then there may not be sufficient market forces to rely completely on competition to derive an appropriate average revenue and adopt this approach at PR19. That being said, if a company submits an average revenue requirement that is within a prescribed ‘corridor’ around a central estimate, it may still be appropriate that this is viewed as being acceptable given the relative accuracy within which the models can make their estimations. If a true view of the regional costs of operation based on the actual company requirements are assumed within the revenue build up, this will then enable the correct price signals to be sent to the market, helping potential new entrants best identify opportunities for competition and facilitating the growth of the market more readily in future periods.

Copyright © United Utilities Water Limited 2018 86

Chapter 7: Supplementary Document - S6002

unitedutilities.com

2 Residential retail cost assessment 2.1 Introduction The objective of this report is to present United Utilities’ approach to retail cost assessment. We set out United Utilities’ preferred approach to developing a robust retail model suite, and discuss how we have used these models to build an appropriate assessment of an efficient retail cost threshold. 2.2 A Background to retail cost assessment at PR19 At PR14, Ofwat assessed retail costs separately from wholesale costs for the first time, using a simple cost to serve approach. The drawback of this approach was that it was unable to account for variations in company operational and demographic characteristics. As a result, Ofwat allowed significant extra revenue in response to United Utilities’ cost adjustment claim relating to regional deprivation and its impact on cost to serve. We are encouraged by the evolution of Ofwat’s approach to retail cost assessment at PR19. We consider that econometric models represent a transparent and evidence-based approach to capture variation in costs across the industry. However, econometric modelling is not an end in itself. As Ofwat noted in its March consultation, it is a challenging task requiring deep knowledge of the area being modelled and careful use of statistical techniques. 2.3 Prior expectation of retail cost drivers United Utilities, along with Northumbrian, commissioned Economic Insight to assess options relating to retail cost assessment (Economic Insight, Options for Retail Cost Assessment at PR19, 2016). In this report, Economic Insight considered what cost drivers are ex ante likely to be relevant to a retail cost assessment. Their findings, along with our own further observations and insight, are summarised in Table 33: Table 33 - Economic Priors in Retail Cost Assessment, as Identified by Economic Insight

Report Section

Area of Interest as Identified by

Economic Insight

Economic Insight’s Findings and United Utilities’

observations

Associated Sign Linear or non-linear?

3.2.1 Bill size is likely to be a substantive driver of bad debt and debt management costs.

It is reasonably expected that as average household charges increase levels of bad debt will increase. It is also to be expected that more expensive forms of debt management become economically viable as the magnitude of average individual household debt increase. Operational experience confirms this to be the case. We expect average household charges to be positively correlated with retail costs.

Positive Non-linear – we expect, all else equal, a £1 increase in bill size to have an increasingly larger impact on cost as bill size increases.

3.2.1 Socioeconomic deprivation is likely to be substantive driver of bad debt and debt management costs.

Arrears risk varies from customer to customer. It is reasonable to assume that customers on lower incomes and higher levels of social deprivation are likely to carry higher arrears risk. Operational experience

Positive Non-linear – see Figure 7

Copyright © United Utilities Water Limited 2018 87

Chapter 7: Supplementary Document - S6002

unitedutilities.com

confirms that more deprived areas, as measured by government indices of deprivation have a higher incidence of bad debt. We expect increasing levels of deprivation to be positively correlated with retail costs.

3.2.2 Companies that provide a good standard of customer service or face a higher demand for customer services could incur higher costs.

An internal UU investigation of the relationship between customer service performance and cost showed an inverse correlation, indicating the “right first time” service objectives deliver simultaneous cost efficiencies and service improvements. In addition, analysis of customer service performance (as measured by SIM) and companies’ retail costs showed no observable correlation between the two. We consider that separate regulatory mechanisms provide strong incentives for good customer service. We do not expect levels of customer service to appear as material driver of retail costs. Given pre-existing regulatory incentives we do not believe that measures of customer service should be used in a cost assessment model.

Uncertain, but we consider existing regulatory incentive mechanisms provide a more suitable method to improve quality of service than cost assessment.

Uncertain

3.2.3 Meter reading is a driver of costs. Numbers of household meters fitted could be used as a proxy for numbers of meter reads undertaken

Market tested costs provided by an independent metering provider suggest that the number of meters is a relevant cost driver. Around 5% of retailer costs are associated with meter reading, indicated this is likely a small but relevant element of overall costs. We expect levels of metering to be positively correlated with retail costs.

Positive Linear – internal evidence demonstrates that reading an additional meter is associated with a constant increase in cost

3.2.3 Ease of reading a meter could also be a cost driver. Factors such as property density

Market tested costs obtained from an independent metering provider showed that ease of meter reads and changes in population density

Immaterial Uncertain

Copyright © United Utilities Water Limited 2018 88

Chapter 7: Supplementary Document - S6002

unitedutilities.com

and meter location may influence this

drive an immaterial difference in metering costs. The metering provider’s cost comparison also indicated that Automated Meter Reading technologies, (the installation of which is within management control) is a greater factor in meter reading cost differentials than meter location or density. We do not expect measures related to ease of meter reading to emerge in cost models as material.

3.4.1 Economies of scale may be present, and can be tested for through empirical analysis.

Economic Insight considered the question of economies of scale and concluded there are arguments both for and against it. As a result, a prior assumption on the impact of scale is not felt to be required. Empirical analysis can be used to test for it Reckon tested for economies of scale but could find little evidence to support that scale economies are currently a significant driver of cost in retail.

Immaterial Theoretically non-linear, but there is little empirical evidence to suggest scale economies are present

3.4.2 Economies of scope (single service water or wastewater customers vs dual service customers)

Operational experience confirms there is a difference in retail costs for single and dual service customers. UU analysis of our own retail costs provides evidence to support this view. We expect higher levels of dual service provision to drive additional cost, relative to an otherwise equivalent company.

Positive Linear

2.3.1 Measures of deprivation: average vs extreme Given the strong evidence base from PR14 suggesting socioeconomic deprivation is a significant cost driver, United Utilities engaged Reckon LLP to assist our understanding of how bad debt and deprivation affect retail costs. Deprivation and arrears risk are not evenly distributed across England and Wales, with some areas of the country showing markedly higher levels of extreme deprivation than others do. It is therefore important to control for the cost impacts of such variations.

Copyright © United Utilities Water Limited 2018 89

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 31 - Deprivation of 10% most deprived areas (Equifax index 2016/17)

The Department for Work and Pensions (DWP) report ‘Households Below Average Income’ (HBAI) (Department for Work and Pensions) found at a national level that household income is a significant factor in relation to the incidence of non-payment of household bills of all kinds. Table 34 shows that the bottom two quintiles by household income contain smaller proportions of households with no bills in arrears and more than three quarters of those having one or more bills in arrears. Table 34 - Household income and arrears (Source: HBAI 2015/16) Household income quintiles

Bottom Quintile

Second Quintile

Third Quintile

Fourth Quintile

Top Quintile

Working age adults (m)

Zero bills in arrears 16% 16% 19% 23% 26% 32.7 One or more bills in arrears

43% 27% 19% 8% 3% 2.9

The poorest 20% of households are more than ten times as likely to have one or more bills in arrears as the richest 20%. The relationship is not linear – with the very poorest substantially more likely to be in arrears and little apparent difference between those in the top two better-off income quintiles. This is suggestive of an exponential relationship and demonstrates that the propensity for being in bad debt ramps up substantially at the poorest end of the socio-economic spectrum. Equally, whilst arrears risk at the median level is substantially less than levels seen amongst the lowest quintile it is still materially greater than that seen amongst the more affluent. Whilst the overall trend is not surprising, with the poorest most likely to be in arrears, it is the scale of variation between the extreme ends of the socio-economic groups that is especially striking. Looking at average levels of deprivation tends to presume that overpayment in affluent areas is as pronounced as underpayment/arrears in deprived areas. Clearly, this does not make a lot of practical sense, which is why it is important to take account of extreme deprivation. This means that indicators of extreme deprivation should be considered alongside average measures when looking at the external drivers of bad debt – simply examining the average level of deprivation in any given area is likely to understate the expected impact on bad debt. Including a model that captures average deprivation levels alongside one that captures extreme deprivation ensures that cost allowances adequately represent the complex relationship between socioeconomic deprivation and retail cost. Reckon’s 2017 report (Reckon, Capturing deprivation and arrears risk in household retail cost assessment, 2017) explored how data from credit rating agencies and government statistics on deprivation could be used

Copyright © United Utilities Water Limited 2018 90

Chapter 7: Supplementary Document - S6002

unitedutilities.com

to predict geographic propensity to default on water bills. Equifax provided around 400 variables, with Reckon narrowing this selection down to a shortlist of 29 candidates. Reckon ran further analysis that demonstrated six variables possessed strong predicted power for arrears risk, justifying their inclusion in company level models of bad debt. Reckon then incorporated the best performing Equifax variables into company level models of cost in its 2018 report (Reckon, Econometric models for residential retail cost assessment, 2018). 2.3.2 Transiency We note that a number of companies have proposed transiency as a cost driver. United Utilities observes that house moves do cost money through the mechanism of opening, closing, and moving accounts. However, the cost of this is under management control; efficient management practices such as the creation of digital interfaces allow customers to open, close and switch accounts remotely with minimal cost. We do recognise that a household could move and leave unpaid debt behind. However, in our view, the underlying driver is not transiency, but deprivation. To test this view, we looked at internal data on account changes and on deprivation by Lower Super Output Area (LSOA) within United Utilities’ area of appointment for one year. The number of account changes per LSOA ranges from one to 474. The dataset has 4,480 observations, with a mean of 71 account changes. To test our hypothesis, we used an OLS model with standard errors adjusted for heteroscedascity. The dependent variable was retail cost to serve in each LSOA, calculated as the total cost to serve in the LSOA, divided by the number of households in that LSOA. Two independent variables were included across different models – transiency and deprivation. The model had a log-log specification. This is because we consider there to be a multiplicative relationship between these factors, as opposed to an additive one. Under a simple linear regression of cost to serve on transiency (and omitting any explanatory factors for deprivation), the model implied that a 1% increase in transiency increases cost to serve by 0.49%, with a t score of 22.5 and r2 of 0.13. However, when we added deprivation to the model, as measured by the average IMD score for an LSOA, the impact of transiency drops to 0.087%, with a t score of 4.84 and r2 of 0.75. This result indicates that the impact of cost to serve on transiency is negligible once deprivation is accounted for. Table 35 below shows the implied cost of transiency across both models. Note that we based these workings on a company with United Utilities’ characteristics. Monetary impacts may well change if company characteristics change. Table 35 – Estimates of transiency’s impact on cost

Independent Variables Included

Implied Increase in CTS per 1% Increase in

Transiency56

Implied Overall Cost of Transiency57

Transiency Cost as % of UU 2015/16

Total Retail Cost58

Transiency £0.19 £570,000 0.49%

Transiency and Deprivation £0.03 £101,000 0.09%

This leads to an important implication; transiency itself has a weak effect on retail cost. Instead, measures of transiency seem to be picking up the effect of deprivation, such as residual unpaid bad debt. In this context, using transiency to explain retail cost does not seem optimal, given that deprivation explains a much larger portion of variation in cost to serve.

56 The average cost to serve per property across all UU LSOAs in 2015/16 was c.£40. 57 There were c.320k account changes in the year. 58 UU total retail cost in 2015/16 was £116m.

Copyright © United Utilities Water Limited 2018 91

Chapter 7: Supplementary Document - S6002

unitedutilities.com

The weakness of the effect also seems to suggest why transiency is unstable in company level models of retail cost. For example, we note that Economic Insight (Economic Insight, Population transience as a driver of household retail costs, 2018) demonstrate that 24% of transiency coefficients are negative across model suite A, while 43% are negative across model suite B. These occurrences of counter-intuitive results cast doubt on the ability of the econometric models to give reasonable estimates of the economic relationship between transiency and retails. The analysis above suggests that transiency costs are immaterial and as such, we have disregarded them for inclusion in econometric models. However, we do recognise that central London is subject to a different challenge to the rest of the country. However, we consider that Ofwat would find it difficult to reflect this in an econometric model. Therefore, we consider that an off-model adjustment may be appropriate for this specific area. 2.3.3 Regional wages As Ofwat note (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018), wages paid to staff are within management control. Therefore, we consider that including regional wages could lead to inefficient management action. As such, we have disregarded regional wages for inclusion in cost assessment models. 2.4 United Utilities’ approach to econometric modelling of retail costs In conjunction with Reckon LLP, United Utilities has developed a set of retail cost models, which we have used to form our view of an efficient cost allowance, at each level of the retail value chain. Ofwat has guided our thinking on the retail value chain, and as a result, we have developed models of bad debt and remaining retail costs, as well as top down total cost models. Additionally, we followed Ofwat’s approach to dependent variable specification, as we utilise a cost per household approach. We refer to the models using the references Reckon set out in its final report. Reckon sought to explore a variety of modelling approaches, which we have drawn on for our recommended model suite. However, our focus has been on the most useful and relevant approaches for PR19 cost assessment. We have selected six preferred models from our overall model suite of 30: two preferred bad debt models (coded BD1_3 and BD1_d5 in the report); two preferred remaining retail cost models (RR2 and RR3); and two preferred total cost models (RT4_d2 and RT4_d4). On balance, we consider six models to strike an appropriate balance between comprehensiveness and simplicity/practicality. Crucially, we note that our model suite provides coverage of each salient economic prior set out in Section 2.3. We consider our model choices to fit well against Ofwat’s stated criteria for good models (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018), in particular: • Models that are motivated using engineering, operational and economic logic. • Coefficients that are the right sign and of plausible magnitude. • Coefficients that are robust. • Cost drivers free of management control and the risk of perverse incentives. • Good performance in diagnostic tests. We sought to include separate factors to capture average deprivation and extreme deprivation within our selection for both the bad debt and total cost value chains. This is because average and extreme deprivation affect costs in different ways, as set out in Section 2.3.1. We now give a brief overview of how our chosen models satisfied these stated criteria.

Copyright © United Utilities Water Limited 2018 92

Chapter 7: Supplementary Document - S6002

unitedutilities.com

2.4.1 Bad debt and debt management model choice Our choice of bad debt and debt management models is restricted to the BD1 model range, as we do not consider the range of BD2 models developed by Reckon to be realistic candidates for retail cost assessment as they impose an unnecessary constraint namely, that bad debt has a one-to-one relationship with bill size. Table 36 sets out the range of different deprivation measures we explored for use in bad debt models. Reckon constructed predictive measures of arrears risk as part of its 2017 report. Table 36 - Key to deprivation measures in bad debt models

Suffix Deprivation measure Description _d1 Proportion of revenue from household served

in LSOAs within top-20 percent, as measured by RGC102

This measure is calculated as the proportion of households in the LSOAs served by a company that are within the 20 percent most deprived

LSOAs in England and Wales, according to their RGC102 score. The number of households in an

LSOA is weighted by the average bill of the services provided by the company in that LSOA, so that the measure can be interpreted as a proxy for

the proportion of a company’s revenue that is from LSOAs that are within the 20 percent most

deprived. _d2 Proportion of revenue from household served

in LSOAs within top-10 percent, as measured by RGC102

As above but based on revenue from the LSOAs that are within the 10 percent most deprived

LSOAs in England and Wales. _d3 Proportion of revenue from household served

in LSOAs within top-20 percent, as measured by IMD (predicted)

As above, but using the IMD (predicted) score to identify the 20 percent most deprived LSOAs.

_d4 Proportion of revenue from household served in LSOAs within top-10 percent, as measured

by IMD (predicted)

As above, but using the IMD (predicted) score to identify the 10 percent most deprived LSOAs.

_d5 Average IMD score (predicted) IMD (predicted) is a proxy for the Indices of Multiple Deprivation, which Reckon developed in its 2017 report. The measure is estimated using

local-level Equifax data, which allows it to be calculated on a consistent basis across England

and Wales. _d6 Average RGC102 score (predicted) RGC102 is an Equifax proprietary measure of

arrears risk. It has an inverse relationship with deprivation.

While we consider all these measures capture the impact of deprivation on costs appropriately, we selected BD1_d3 and BD1_d5 as our preferred models. This is because robustness tests indicated that these models were associated with the more stable coefficients on the deprivation measure, while also satisfying our criteria of including one extreme measure and one average measure of deprivation. Figure 32Error! Reference source not found. illustrates how the deprivation measures’ coefficients changed as we altered the dataset by dropping observations. Given we are building predictive models, we consider the low relative variance associated with BD1_d3 and BD1_d5 to provide strong support for their selection59. These models also performed well on other diagnostic tests

59 Note that we are only able to compare visually the variance associated with models BD1_d1-BD1_d4 as the deprivation variable in these models are specified as percentages.

Copyright © United Utilities Water Limited 2018 93

Chapter 7: Supplementary Document - S6002

unitedutilities.com

2.4.2 Remaining retail cost model choice We selected two models of remaining retail cost. One includes two cost drivers, one for metering and one to capture the cost differential associated with dual service customers. The other model contains the dual service cost driver only. Modelling evidence did not suggest that metering drives enough variation in cost to justify the inclusion of a model with metering as the sole cost driver. For this reason, we selected models RR2 and RR3. While these models do not perform well in terms of statistical significance or r-squared, we consider them to be economically and operationally intuitive. We also note that the models predict costs better than an equivalent aggregate cost model60 (we specify all our models as unit cost models). Given we are concerned with predictive models and our proposed models predict costs well, on balance, we do not consider the low r-squared to represent a significant drawback to the model suite. In developing cost models, Reckon adopted a different functional form for remaining retail cost models than that used for total cost and bad debt models. Where models have factors that are likely to show a degree of interdependence in their effect on the dependent variable, log-linear models offer benefits over simpler linear models. We consider factors for bill size and deprivation to display some element of interdependence, and therefore believe it is appropriate to specify models that include derivation and bill size as cost drivers as log-linear models. Unlike bad debt and total cost models, we have specified remaining cost models as level-level models, meaning no logarithmic transformations are carried out prior to modelling. Intuitively, this assumes the cost drivers included in the model have an additive impact on cost. We consider this a reasonable

60 See Reckon (2018) paragraph 127 for example.

Figure 32 - Histograms of estimated coefficients for each deprivation measure as dataset is varied in bad debt models

Copyright © United Utilities Water Limited 2018 94

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 33 - Histograms of estimated coefficients for each deprivation measure as the dataset is varied in total cost models

assumption in the case of economies of scope and metering costs, which would appear to display little interaction with one another. 2.4.3 Total cost model choice We have picked two models from the RT4 set of models. These models include as cost drivers bill size and the proportion of dual service customers alongside a deprivation measure. We consider this set of variables to strike the right balance between explaining sufficient variation in costs across the industry and model parsimony. While we would also advocate the use of the RT2 and RT3 model set, these criteria led us to choose the RT4 set as our preferred set. Table 37 sets out the range of different deprivation measures we explored for use in total cost models. Reckon constructed predicted IMD measures as part of its 2017 report. Table 37 - Key to deprivation measures used in total cost models Suffix Deprivation measure

_d1 Proportion of revenue from household served in LSOAs within top-20 percent, as measure by RGC102

_d2 Proportion of revenue from household served in LSOAs within top-20 percent, as measure by IMD (predicted)

_d3 Average RGC102 score (predicted)

_d4 Average IMD score (predicted)

From the RT4 model set, we selected RT4_d2 and RT4_d4 as the deprivation measures in these models performed relatively well in robustness testing, and the models performed well in wider diagnostic tests. For more details on the wider set of diagnostic tests utilised, please refer to the Reckon report.

Copyright © United Utilities Water Limited 2018 95

Chapter 7: Supplementary Document - S6002

unitedutilities.com

2.4.4 United Utilities’ Preferred Model Suite Table 38 - United Utilities' preferred model suite

Model code Cost drivers Bad debt models BD1_d3 • Bill size

• Proportion of revenue from households served in LSOAs that are among the 20 percent most deprived, as measured by fitted indices of multiple deprivation (IMD)

BD1_d5 • Bill size • Average IMD (predicted) score

Remaining retail cost models RR2 • Number of metered services per household

• Proportion of customers that are dual service customers RR3 • Proportion of customers that are dual service customers

Total cost models RT4_d2 • Proportion of customers that are dual service customers

• Bill ratio (measure of average bill adjusted to normalise for extent of dual service customers)

• Proportion of revenue from households served in LSOAs that are among the 20 percent most deprived, as measured by fitted indices of multiple deprivation (IMD)

RT4_d4 • Proportion of customers that are dual service customers • Bill ratio (measure of average bill adjusted to normalise for

extent of dual service customers) • Average IMD (predicted) score

All models generate results that compare favourably with expectations from the economic priors set out in Section 2.3. Modelled cost drivers cover the full range of expected retail cost factors and display both expected signs and scale. We consider this to provide validation that our models conform to priori expectations of operational realities. Additionally, our chosen models perform well on diagnostic tests of functional form, passing both the Ramset RESET test and the Linktest. While these two tests are similar, formally, there are slight differences; the Linktest tests explicitly for dependent variable misspecification, while the RESET test is a more general specification test. Therefore, we considered both tests in our model selection process, whilst recognising that there is a good degree cross over between the two tests. We present model results in Section 0.

Copyright © United Utilities Water Limited 2018 96

Chapter 7: Supplementary Document - S6002

unitedutilities.com

2.4.5 Model results 2.4.5.1 Bad Debt Models

BD1_d3 BD1_d5

Dependent variable ln (bad debt costs) ln (bad debt costs)

ln (Bill size) 1.142*** 1.115*** (12.43) (12.98)

Proportion of revenue from households served in LSOAs classed as within 20 percent most deprived (as measured by IMD predicted)

1.204

(1.85)

Average IMD (predicted) score 3.001* (2.46)

Dummy (2014 = 1) 0.157 0.160 (1.88) (1.91)

Dummy (2015 = 1) 0.204* 0.195* (2.24) (2.19)

Dummy (2016 = 1) 0.136 0.121 (1.68) (1.55)

Constant -4.377*** -4.658***

(-8.47) (-8.62) R2 0.78 0.79 VIF 1.35 1.37 Reset test 0.428 0.393 Estimation method (e.g. OLS or RE) OLS OLS N (sample size) 71 71

2.4.5.2 Remaining Retail Cost Models

RR2 RR3

Dependent variable Other retail costs Other retail costs

Number of metered services per household 3.87

(1.04)

Proportion of dual service households 0.753 2.714 (0.39) (1.73)

Dummy (2014 = 1) 0.854 0.561 (1.34) (1.00)

Dummy (2015 = 1) 1.541** 1.379* (3.11) (2.56)

Dummy (2016 = 1) 0.234 0.183 (0.54) (0.36)

Constant 12.54*** 14.68***

(5.77) (13.61) R2 0.18 0.11 VIF 1.6 1.4 Reset test 0.857 0.942 Estimation method (e.g. OLS or RE) OLS OLS N (sample size) 71 71

Copyright © United Utilities Water Limited 2018 97

Chapter 7: Supplementary Document - S6002

unitedutilities.com

2.4.5.3 Total Cost Models RT4_d2 RT4_d4

Dependent variable ln (total retail costs) ln (total retail costs)

Proportion of revenue from households served in LSOAs classed as within 20 percent most deprived (as measured by IMD predicted)

0.606

(1.09)

Average IMD (predicted) score 1.62 (1.41)

Proportion of dual service households 0.323 0.278 (2.01) (1.70)

Bill ratio 0.81** 0.859** (3.01) (3.27)

Dummy (2014 = 1) 0.0651* 0.0675* (2.22) (2.27)

Dummy (2015 = 1) 0.119** 0.115** (3.70) (3.72)

Dummy (2016 = 1) 0.042 0.0346 (1.86) (1.65)

Constant 2.222** 1.954*** (8.49) (5.05)

R2 0.675 0.6863 VIF 1.89 1.96 Reset test 0.208 0.615 Estimation method (e.g. OLS or RE) OLS OLS N (sample size) 71 71

2.4.5.4 How the model results compare to our prior expectations Table 39 demonstrates that the model results conform to our prior expectations, which we based on operational experience and economic intuition. We used the Ramsey RESET test to assess whether our functional form was correctly specified. As discussed in Section 2.4.2, we have used logarithms in model where we expect a non-linear relationship. The RESET test confirms whether this is appropriate. Table 39 - Validation of economic priors

Report Section

Area of Interest as Identified by Economic Insight

Modelled Sign Linear or non-linear? (RESET test

pass or fail)

3.2.1 Bill size is likely to be a substantive driver of bad debt and debt management costs. Positive Pass

3.2.1 Socioeconomic deprivation is likely to be substantive driver of bad debt and debt management costs.

Positive Pass

3.2.2 Companies that provide a good standard of customer service or face a higher demand for customer services could incur higher costs.

We did not seek to model customer

service NA

3.2.3 Meter reading is a driver of costs. Numbers of household meters fitted could be used as a proxy for numbers of meter reads undertaken

Positive Pass

3.2.3 Ease of reading a meter could also be a cost driver. Factors such as property density and meter location may influence this

We did not seek to this NA

Copyright © United Utilities Water Limited 2018 98

Chapter 7: Supplementary Document - S6002

unitedutilities.com

3.4.1

Economies of scale may be present, and can be tested for through empirical analysis.

Empirical analysis indicated economies

of scale are not present (see Section

2.4.7)

NA

3.4.2 Economies of scope (single service water or wastewater customers vs dual service customers)

Positive Pass

2.4.6 Validation of model results We have sought to validate our modelled predictions for the impact of deprivation on cost to serve, using granular local-level cost data that is available to us over our entire region. As Reckon notes61: “a change in the deprivation level from the industry average to the highest value would increase unit bad debt costs by 19 to 26 percent, depending on the model”. In order to test whether this is a reasonable range, we corroborated this estimate with an estimate derived from a UU model based on cost to serve at a Local Super Output Area (LSOA) level. This model compares Index of Multiple Deprivation (IMD) scores to average cost to serve at a LSOA level for all LSOAs in the UU region. The model uses simple regressions of deprivation against average cost to serve per household in each LSOA. The model uses government data on 2015 deprivation and 2016/17 UU customer level data on retail cost to serve. Each LSOA included in the model dataset is served by UU and therefore subject to the same set of charges, tariffs and debt management processes. Each LSOA is all subject to broadly comparable levels of meter penetration and average bills. We have taken the industry average deprivation and industry maximum deprivation, as measured by IMD an observed the impact on cost to serve as generated by the UU LSOA model. The LSOA model shows that an increase from industry average to industry maximum IMD would see an increase in cost to serve of 28%. This is very close to the 19 to 26 percent range implied by Reckon retail cost models. Additionally, we note that the sample size in our granular model is large (4,496), which further supports the estimates from our company level models of cost. Table 40 - UU LSOA model results Coefficient Standard error t-stat R-squared

Intercept 0.9273 0.0257 36.14 0.6398

IMD coefficient 0.7372 0.0083 89.38

Regression formula prediction ln(retail cost/props) = 0.9273 + 0.7372ln(IMD) Table 41 - UU LSOA model implied cost to serve IMD score Estimated CtS

Industry average IMD score 21.12 23.95

Industry maximum IMD score 29.49 30.64

61 In paragraph 98 of its 2018 report (see references for more details)

Copyright © United Utilities Water Limited 2018 99

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 34 - Evidence of constant returns to scale in retail models

2.4.7 Ofwat’s draft retail cost models In March 2018, Ofwat published a consultation on the use of econometric models in cost assessment at PR19. Overall, we were encouraged by the majority of Ofwat’s proposals: • We support the use of econometric models to assess retail costs. • We support Ofwat’s specification of the dependent variable, as the cost per household approach reduces

the potential for statistical issues like heteroscedasticity and multicollinearity to impact results, and our work with Reckon found that this specification produces estimates with greater alignment to observed industry costs than an equivalent aggregate cost model.

• We are encouraged that Ofwat has spotted the potential for credit ratings agencies to predict arrears risk, and are supportive of Ofwat’s work in this area so far.

• We consider that Ofwat is justified in using bill size as a cost driver; its inclusion makes sense both logically and intuitively.

• We support Ofwat’s inclusion of dual service customers. We consider it preferable to allow an econometric to estimate a relationship, as opposed to making ex ante assumptions.

However, we were concerned with the use of economies of scale in retail models. We do not consider economies of scale to be a factor in the water and wastewater retail industry. We note that management is able to take action to benefit from economies of scale in retail. These can be achieved through joint ventures and joint billing arrangements, as evidenced by actual company strategic operating decisions in the sector. We are concerned that including a measure of economies of scale would disincentivise efficient management activity, and be contrary to one of Ofwat’s stated criteria in its consultation document - to minimise endogeneity within cost assessment models. We also note that regulatory focus on retail cost and incentives is relatively new. This could mean that models seeking to capture economies of scale might instead capture historical arrangements, where companies are yet to realise the opportunities for greater efficiency, which might lead to models overstating the effects of scale on efficient costs. Therefore, on balance, we consider it reasonable not to include economies of scale as a cost driver. Our work with Reckon has tested whether economies of scale exist for water and wastewater retailers. The figure below shows the histogram for the estimated coefficients on the logarithm of households that are obtained when a simple model that regresses the logarithm of remaining operating costs on the logarithm of the number of household customers is run across variations62 to the dataset utilised by Reckon in its report.

As shown in Figure 34, the estimated coefficient on the variable relating to number of households is consistently very close to one. This supports the use of models with no driver relating to economies of scale.

62 We varied the dataset by sequentially dropping companies and years.

Copyright © United Utilities Water Limited 2018 100

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Additionally, the magnitude of the coefficients associated with economies of scale in Ofwat’s models seem to point to scale differences across the industry that do not seem credible. Table 42 presents implied economies of scale from Ofwat’s models when comparing two companies, one of which serves five times as many customers as the other.

Table 42 - Economies of scale implied by Ofwat's draft models Model Coefficient on ln(number of

households) Implied economies of scale

ORDC3 -0.128 18.6%

ORDC4 -0.032 5.0%

OROC3 -0.08 12.1%

OROC4 -0.068 10.4%

ORTC4 -0.119 17.4%

Note: this table presents implied economies of scale from Ofwat's models when comparing two companies, the larger of which serves five times more customers than the smaller.

For example, Ofwat’s ORTC4 model implies that costs per customer would be 17.5 percent lower at the larger company. We do not consider that this ties in with economic reality, and are concerned that the inclusion of this variable could undermine Ofwat’s stated aim of promoting models that are credible from an operational and economic viewpoint, given the large variation in scale across the industry. However, aside from this, we are generally supportive of Ofwat’s approach to econometric assessment of retail costs at PR19. 2.4.8 Differences in approach between wholesale and retail Our approach to the development of retail models has been much the same as the development of wholesale models set out elsewhere in this chapter, and we have largely imitated the choices made in the creation of wholesale models and wholesale cost thresholds. For example: • Basing models on engineering and economic logic. • Triangulation between models to minimise the risk that bias in individual models adversely affects overall

cost assessment. • The use of diagnostic testing to assess statistical validity. However, there are two areas where our approach diverged; our use of time fixed effects in retail, and a simpler method of triangulation between value chain splits. This section discusses the motivation behind our use of time fixed effects. Additionally, we briefly comment on data quality. Section 2.5.4 sets out our approach to triangulation for retail. 2.4.8.1 Use of Time Fixed Effects Time fixed effects are generally used to control for the impact of a dynamic change in a panel dataset. When unaccounted for, dynamic changes can lead to a statistical estimator under/over weighting the cost drivers in the model. The introduction of bias in this way could lead to a sub-optimal outcome. There are three main options for modelling temporal effects:

1. Assume that retail costs are constant over the data sample period.

Copyright © United Utilities Water Limited 2018 101

Chapter 7: Supplementary Document - S6002

unitedutilities.com

2. Apply a constant annual time trend as an explanatory variable. This implies that costs change by the same magnitude across the industry in each year.

3. Use time fixed effects to allow for changes in industry-wide retail costs from one year to the next, with flexibility for the change to be different across each year.

We consider that retail businesses over the last four years have been subject to a series of dynamic changes: • Technological changes, including growth in digital channels and the use of social media; advances in

customer data management and analytical tools which enable the cost effective use of third party datasets e.g. credit reference agency data.

• Substantial macroeconomic developments and reforms to the benefits system over the last decade have resulted in substantial pressure on household incomes and subsequent pressure on retailers’ arrears risk.

• Separation of retail price controls and introduction of non-household retail completion has exposed new cost and performance benchmarks for the water retail sector.

We do not consider options one and two are viable options to capture these dynamic changes, given our operational understanding of retail costs across the industry over the last four years. Therefore, option three should allow for a more accurate overall assessment. Given these considerations, we consider that the use of time fixed effects in retail models of cost assessment is reasonable. 2.4.8.2 Residential retail data quality and comparability Residential retail cost allocations have only been completed on a consistent basis for a limited period, creating some data limitations that have had to be considered as part of cost assessment modelling. Notably, cost allocations to retail activities prior to 2012/13 did not separate residential and non-household costs, greatly limiting the value of retail cost information prior to 2012/13. However, we do not consider these challenges to be insurmountable or to provide serious detriment to the cost assessment process. In another example, there are indications that the approach companies have taken to allocating costs between sub elements of the residential retail value chain are different. We note that some companies have allocated a large proportion of total retail costs to the “other” category, whilst United Utilities and some other companies have allocated a larger percentage to costs such ‘Customer Service’ or ‘Debt management’. We believe the chosen model structures, and in particular, the careful and limited use of sub-models helps to minimise the impact of these deferent approaches to cost allocation across the industry. 2.5 Building the Cost Threshold This section discusses how we have developed the results from our preferred econometric models into a cost threshold. 2.5.1 Incorporating time fixed effects into the cost threshold Our favoured approach of including time fixed effects in econometric models of retail cost assessment raises the question of how these time fixed effects should be incorporated into the cost threshold. As stated above, we consider the time dummies to capture a combination of factors that can be thought of as structural breaks. In any given year, the impact of each different factor is difficult to forecast in advance and experience has shown year on year movements can be volatile. Therefore, we advocate incorporating an average of the time dummies into the cost baseline. This has the effect of averaging out fluctuations in the panel period. However, we stress that this choice was made in conjunction with the choice of efficiency challenge.

Copyright © United Utilities Water Limited 2018 102

Chapter 7: Supplementary Document - S6002

unitedutilities.com

2.5.2 Efficiency Adjustment We accept that our preference for incorporating time fixed effects in the cost threshold may not, on its own, strike the right balance of risk and return between companies and customers. For that reason, we would propose that Ofwat could base the static efficiency challenge on companies’ performance in 2017 alone. We prefer this approach because observations of companies’ retail investment over last few years indicate there has been a year on year reduction in average retail costs. It is less clear that a similar effect has taken place for wholesalers. In addition, capital investment represents a far higher proportion of wholesale totex than equivalent investment in retailers. Therefore, it is more reasonable for Ofwat to use a point estimate of efficiency in retail models, whereas a long-term average would seem more appropriate for wholesale investment profiles. Therefore, an efficiency challenge based on the year 2017 is more likely to be represent a stretching challenge for retail cost assessment. We consider this to represent a cleaner approach to imposing an efficiency challenge; there is little risk of conflating inefficiency with high input prices, and the efficiency assessment is more likely to be based on efficient performance. This protects customers from paying for an inefficient service, and protects companies from unduly low estimates of input costs. We recognise that bad debt charges can display volatility for individual companies in any given year, and that an averaged efficiency adjustment may help to address this. However in general the industry has displayed less volatility in bad debt charges in recent years and so a point estimate could still be appropriate. Again, we stress that we have taken this decision in conjunction with the choice of how to incorporate time dummies into the basic cost threshold. 2.5.2.1 Selecting an Efficiency Benchmark We consider an upper-quartile challenge to be an appropriate static efficiency challenge given the spread of residuals across our models, illustrated in Figure 35 (we indicate the upper quartile by the large dot63):

63 Where the upper-quartile lies between two companies, we have used the interpolated value. We consider this a more transparent method of setting an efficiency benchmark.

RT4 d

RT4 d

RR2 RR3 BD1 d

BD1 d

Figure 35 - Unexplained Variation in United Utilities' Retail Model Suite

Copyright © United Utilities Water Limited 2018 103

Chapter 7: Supplementary Document - S6002

unitedutilities.com

There is clearly a large residual spread; the smallest minimum-maximum range is 62 percentage points, while the largest is 95 percentage points. We consider it unlikely that this variance solely reflects differences in efficiency. It is more likely that the unexplained variation in our models is a composite of efficiency differences and other factors, including company-specific noise and model biases resulting from measurement error, overfitting and omitted variables. This means that any efficiency challenge imposed by Ofwat needs to balance asymmetric and imperfect information on company efficiency with ensuring companies are sufficiently stretched. We consider that an upper quartile challenge applied to cost allowances produced by our model suite strikes this balance. While some companies outperform the upper quartile, a majority do not. Importantly whilst the frontier company’s implied efficiency varies substantially between models, the upper quartile challenge shows a much higher degree of stability across cost models as demonstrated in Table 43 - Variation in the efficiency challenge. Table 43 - Variation in the efficiency challenge

RT4_d2 RT4_d4 RR2 RR3 BD1_d3 BD1_d5 Upper quartile 0.88 0.89 0.89 0.86 0.85 0.84 Frontier 0.80 0.79 0.67 0.73 0.62 0.62

We note that the combination of effects including company-specific noise and model biases resulting from measurement error, overfitting and omitted variables mean that the efficiency of the frontier company is very likely to be overstated. Therefore, the frontier company will tend to be the company for which modelling noise and limitations has a favourable impact. Table 44 sets out the companies who outperform the upper quartile benchmark and the frontier company. Table 44 - Upper quartile and frontier companies by model

RT4_d2 RT4_d4 RR2 RR3 BD1_d3 BD1_d5 UQ outperformer BRL BRL ANH ANH NES NES UQ outperformer NES NES NES NES SES SBW UQ outperformer SEW SEW PRT PRT SEW SEW UQ outperformer WSX WSX WSX WSX SVT SVT UQ outperformer YKY YKY - - YKY YKY Frontier YKY NES ANH WSX YKY YKY

Therefore, we consider the upper quartile level to represent a stretching and transparent efficiency target. In addition, we note that the upper quartile obtained using our suite of parsimonious models is significantly more stretching than that used by Ofwat at PR14. Given our choice of models, the upper quartile represents a 13.4% cost challenge. We consider this to represent a stretching target, but one that is appropriate for a maturing industry. Additionally, we consider a more stretching challenge would not recognise the trade-off that exists between service quality and cost reduction. 2.5.3 Real price effects and dynamic efficiency We have considered the impact of future cost pressures on the household retail cost base. Ofwat’s PR19 final methodology proposes that retail revenue allowances will not adjust for inflation, but instead the impacts of Input Price Pressures (IPP) will form part of the totex allowance process. As part of a recent workshop on innovation and efficiency gains, KPMG presented the results of work on behalf of Ofwat looking at long-term trends in retail frontier shift. This result stated that an annual frontier shift of 0.8% to 1.8% might be possible. Whilst not explicitly stated in the documentation, our understanding is that this frontier shift assessment is RPI based, whilst retail price controls will be made on a nominal basis. Adjusting efficiency factors to a nominal

Copyright © United Utilities Water Limited 2018 104

Chapter 7: Supplementary Document - S6002

unitedutilities.com

basis indicates that frontier company outturn totex would under the most aggressive KPMG assumptions still be expected to increase between 1.2% and 2.0% per year. Across the forecast period 2017/18 to 2024/25 this would result in AMP7 average nominal cost inflation of around 9.1%. Table 45 - Dynamic efficiency in retail

2017/18

2018/19

2019/20

2020/21

2021/22

2022/23

2023/24

2024/25

AMP7 average

RPI 3.8% 3.6% 3.2% 3.0% 3.0% 3.0% 3.0% 3.0% 3.0%

KPMG Max annual efficiency (RPI base)

-1.8% -1.8% -1.8% -1.8% -1.8% -1.8% -1.8% -1.8% -1.8%

Annual nominal cost inflation 2.0% 1.8% 1.4% 1.2% 1.2% 1.2% 1.2% 1.2% 1.2%

Cumulative nominal inflation/(efficiency) factors

2.0% 3.8% 5.2% 6.5% 7.8% 9.1% 10.4% 11.7% 9.1%

We support Ofwat’s decision in the PR19 final methodology to apply any future cost pressure allowances symmetrically across the industry. Ofwat state in the Draft PR19 Methodology that only 5% of retail household employment costs are region specific. As such, any allowance made for future cost pressures allowed for one company should be viewed as applying nationally and therefore to all companies. In addition, our views on our ability to absorb IPPs are closely linked to prevailing views on future CPIH inflation. Currently we forecast CPIH inflation of 2% over AMP7, however if between September 2018 and the setting of final determinations in December 2019 forecasts of inflation were to materially increase we would need to reassess the impact of input price pressures. 2.5.4 Triangulation Triangulation is an opportunity to alter the relative weights models have in the final cost threshold. One might choose to do this because of uncertainty of model veracity. For example, if a model has a large residual spread relative to other models in the suite, then we could choose to place less weight on that model in triangulation. We could also choose to place differing weight on certain value chain models for similar reasons. For wholesale we proposed an approach which placed greater weight on model suites that best predict costs for an individual company. This is important for wholesale cost assessment given the complexity of interaction between numerous cost drivers and the large differences in external influences between regions. This is less of an issue for retail, so we have chosen to place equal weight on each model. We have also placed equal weight on top-down and bottom-up modelling[1]. There are a number why a simpler approach is more justified for retail: • Retail has fewer cost drivers and a simpler cost structure. Alongside Ofwat and United Utilities, a number

of companies have shared work (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018) that suggests there are fewer cost drivers considered material present in retail businesses. Therefore, we do not consider it necessary to seek to guard against certain cost drivers unduly affecting certain companies or splits of the value chain in different ways. The fact that we have followed Ofwat’s methodology by splitting retail into two value chains, as opposed to the larger value chains utilised in wholesale, reinforces this effect.

[1] The formula we applied is as follows: ( ( ( 0.5 x BD1_d3 ) + ( 0.5 x BD1_d5 ) ) + ( ( 0.5 x RR2 ) + ( 0.5 x RR3 ) ) x 0.5 ) + ( ( ( 0.5 x RT4_d2 ) + ( 0.5 x RT4_d4 ) ) x 0.5 )

Copyright © United Utilities Water Limited 2018 105

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Figure 36- Historical Trends in Deprivation Data

• Retail business are more homogenous. A myriad of different geographical, operational and economic factors impact the provision of a water and wastewater wholesale service, the effect of which needs to be mitigated through a careful cost assessment process, as set out in Section 1. Common themes across work shared by Ofwat and the industry on retail cost assessment to date indicate the factors underlying variations in retail cost are more easily reflected in retail models.

2.6 Additional Methodological Considerations 2.6.1 Forecast method We use forecasts to build the basic cost threshold for the next price control period. The historical relationships of cost and cost driver estimated in our econometric models are combined with future forecasts for each cost driver. This provides an estimate for the level of cost an efficient company can expect to incur over the price control period. Table 46- Forecast method by cost driver

Cost Driver Forecast Method Bill size United Utilities’ financial model produces a bill profile to 2025 that forms

our forecasts for bill size. Deprivation factors Trend analysis suggests there is an upwards trend across all deprivation

measures, which suggests companies can expect to serve an increasingly deprived customer base. However, economic forecasts indicate that GDP

will continue to grow, albeit at a slowing pace. Lacking any further evidence on how this economic growth will be distributed within society, we have adopted a conservative approach to forecasting deprivation, and

used the historical average. We consider this to be in the customer interest. Historical trends in deprivation data are illustrated in Figure 36.

Metered services provided Our customer forecasts use future housing growth evidence from the Local Development Plans of just over 50 Local Authority Districts and

Unitary Authorities, as well as three National Park Authorities. We augment this data with evidence gathered from the house building

industry. We use these forecasts to define different customer groups, like number of metered customers.

Percentage of dual service customers

Our customer forecasts use future housing growth evidence from the Local Development Plans of just over 50 Local Authority Districts and

Unitary Authorities, as well as three National Park Authorities. We augment this data with evidence gathered from the house building

industry. We use these forecasts to define different customer groups, like the percentage of dual service customers.

Customer numbers Our customer forecasts use future housing growth evidence from the Local Development Plans of just over 50 Local Authority Districts and

Unitary Authorities, as well as three National Park Authorities. We augment this data with evidence gathered from the house building

industry.

Copyright © United Utilities Water Limited 2018 106

Chapter 7: Supplementary Document - S6002

unitedutilities.com

2.6.2 Separated cost to serve for metered and unmetered customers We do not consider that there is a need to establish a separate cost to serve for metered/unmetered and dual/single service customers. Our proposed cost models account for variation in the cost to serve across these customer types. This means that, for ex-ante cost allowance and price setting purposes, a split cost to serve will produce the same outcome. While the impact for price setting purposes is neutral, this is not necessarily the case for future revenue correction purposes. However, experience shows that our forecasts for customer numbers and metering rates have generally tracked outturn customer numbers. Given this, on balance we consider there is no need to add extra complexity into the regulatory system by setting a range of different cost to serve values for different customer types, and so propose a single cost to serve allowance for all residential retail customers. 2.7 Retail Totex Baselines We have derived a totex baseline for the retail price control, combining our proposed benchmarking models and approaches to setting an effective baseline set out in the preceding sections. We compare this baseline to our PR19 Business Plan in Table 47. This individually derived baseline provides further evidence that our plan is stretching. Table 47 - How our plan compares to the modelled baseline

2020-21 2021-22 2022-23 2023-24 2024-25 AMP7 Modelled totex (16/17 CPI price base) 97.01 97.64 98.29 98.96 99.67 491.57

Cumulative nominal inflation (as per Table 45)

6.5% 7.8% 9.1% 10.4% 11.7% 9.1%

Cost pressure (outturn) 6.31 7.62 8.94 10.29 11.66 44.82

Totex gross of cost pressure (outturn) 103.32 105.25 107.23 109.26 111.33 545.7

PR19 Business Plan (outturn) 98.0 98.1 97.5 98.0 97.9 489.6

These values exclude pension deficit repair costs, and include depreciation on pre-2015 assets.

Copyright © United Utilities Water Limited 2018 107

Chapter 7: Supplementary Document - S6002

unitedutilities.com

References CEPA. (2014, March 20). Cost assessment – advanced econometric models. Retrieved from

http://webarchive.nationalarchives.gov.uk/20150603214107/http://www.ofwat.gov.uk/pricereview/pr14/pap_tec1402feederbasiccostappb.pdf

CEPA. (2018). PR19 Econometric benchmarking models Ofwat. Cambridge Economic Policy Associates Ltd (CEPA). Retrieved from https://www.ofwat.gov.uk/wp-content/uploads/2018/03/CEPA-cost-assessment-report.pdf

CMA. (2015, October 6). Bristol Water plc: A reference under section 12(3)(a) of the Water Industry Act 1991. Competition and Markets Authority. Retrieved from Competition and Markets Authority: https://www.gov.uk/cma-cases/bristol-water-plc-price-determination

Department for Environment Food & Rural Affairs. (2016, January 15). Water abstraction management reform in England. Retrieved from https://www.google.co.uk/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&ved=0ahUKEwiYx4vd0ajcAhXERMAKHcWdA-4QFggoMAA&url=https%3A%2F%2Fassets.publishing.service.gov.uk%2Fgovernment%2Fuploads%2Fsystem%2Fuploads%2Fattachment_data%2Ffile%2F492414%2Fabstraction-ref

Department for Work and Pensions. (n.d.). Households Below Average Income. Retrieved from https://www.gov.uk/government/statistics/households-below-average-income-199495-to-201516

Economic Insight. (2016). Options for Retail Cost Assessment at PR19. Retrieved from https://www.unitedutilities.com/globalassets/z_corporate-site/about-us-pdfs/looking-to-the-future/options-for-retail-cost-assessment.pdf

Economic Insight. (2018). Population transience as a driver of household retail costs. Retrieved from http://www.economic-insight.com/2018/03/20/report-transience-as-a-driver-of-household-retail-costs/

First Economics. (2008, June). The rate of frontier shift affecting water industry costs. Jacobs. (2014, March 24). PR14 Forecast of exogenous variables. Retrieved from Setting price controls for 2015-20 –

wholesale cost assessment: http://webarchive.nationalarchives.gov.uk/20150603214107/http://www.ofwat.gov.uk/pricereview/pr14/pap_tec1402feederbasiccostappd.pdf

KPMG. (2018, March 21). PR19 - Innovation and Efficiency Gains from the Totex Framework - KPMG and Aqua Consultants . Retrieved from https://www.ofwat.gov.uk/wp-content/uploads/2018/03/Ofwat-totex-efficiency_workshop-pack_FINAL.pdf

Ofwat. (2013). Calculation of efficiency scores and efficiency adjustment factors. Retrieved from Setting price controls for 2015-20 – wholesale cost assessment: http://webarchive.nationalarchives.gov.uk/20150603214107/http://www.ofwat.gov.uk/pricereview/pr14/pap_tec1408uqwholesale.xlsx

Ofwat. (2013, October 31). IN 13/17: Treatment of companies’ pension deficit repair costs at the 2014 price review. Retrieved from https://www.ofwat.gov.uk/wp-content/uploads/2015/11/prs_in1317pr14pension.pdf

Ofwat. (2014, March 31). Setting price controls for 2015-20 – wholesale cost assessment. Retrieved from Price Review 2014: http://webarchive.nationalarchives.gov.uk/20150603202623/http:/www.ofwat.gov.uk/pricereview/pr14/pap_tec1402feederbasiccostappasewerage.xlsx

Ofwat. (2014, March 27). Setting price controls for 2015-20 – wholesale cost assessment. Retrieved from Price Review 2014: http://webarchive.nationalarchives.gov.uk/20150603202625/http:/www.ofwat.gov.uk/pricereview/pr14/pap_tec1402feederbasiccostappawater.xlsx

Ofwat. (2017, July 11). 2019 price review: Draft methodology. Retrieved from https://www.ofwat.gov.uk/consultation/delivering-water2020-consulting-on-our-methodology-for-the-2019-price-review/

Ofwat. (2017, December 13). 2019 price review: Final methodology. Retrieved from https://www.ofwat.gov.uk/regulated-companies/price-review/2019-price-review-final-methodology/pr19-final-methodology/

Ofwat. (2017, December 15). Constructed data. Retrieved from PR19: https://ofwat.sharepoint.com/:f:/r/sites/dropbox/PR19/Shared%20Documents/2017%20July%20Cost%20Assmt%20Information%20Request/Ofwat_adjusted_working_files_070218?csf=1

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Ofwat. (2017, Decemeber 13). Cost sharing rates spreadsheet. Retrieved from 2019 price reivew: Data tables and models: https://www.ofwat.gov.uk/wp-content/uploads/2017/12/Cost-sharing-model-for-publication-FAST.xlsx

Ofwat. (2017, November 02). New connections charges rules from April 2020 – England: Decision Document. Retrieved from https://www.ofwat.gov.uk/wp-content/uploads/2017/11/New-connections-charges-rules-from-April-2020-–-England-Decision-Document.pdf

Ofwat. (2018, February 7). 2017 July Cost Assmt Information Request. Retrieved from PR19: https://ofwat.sharepoint.com/:f:/r/sites/dropbox/PR19/Shared%20Documents/2017%20July%20Cost%20Assmt%20Information%20Request/Ofwat_adjusted_working_files_070218?csf=1

Ofwat. (2018, June 25). 2019 price review: Final methodology Q&A. Retrieved from https://www.ofwat.gov.uk/publication/pr19-final-methodology-queries-answers-25-june-2018/

Ofwat. (2018, March 29). Cost Assessment for PR19 – a consultation on econometric cost modelling. Retrieved from Ofwat: https://www.ofwat.gov.uk/consultation/cost-assessment-pr19-consultation-econometric-modelling/

Ofwat. (2018, June 12). IN 18/11: Enhancement expenditure - setting expectations for well-evidenced proposals and clarifying interaction with cost adjustment claims. Retrieved from https://www.ofwat.gov.uk/wp-content/uploads/2018/06/IN-1811-Enhancement-expenditure-setting-expectations-for-well-evidenced-proposals-and-clarifying-interaction-with-cost-adjustment-claims.pdf

Ofwat. (2018, March 1). Jonson Cox speech at Water Industry City Conference. Retrieved from https://www.ofwat.gov.uk/wp-content/uploads/2018/03/Jonson-Cox-speech-at-Water-City-UK-1-March-2018.pdf

Reckon. (2017). Capturing deprivation and arrears risk in household retail cost assessment. Retrieved from www.unitedutilities.com: https://www.unitedutilities.com/corporate/about-us/our-future-plans/looking-to-the-future/

Reckon. (2018). Econometric models for residential retail cost assessment. T6001 - Arup & Vivid Economics. (2017). Understanding the exogenous drivers of wholesale wastewater. Retrieved from

https://www.unitedutilities.com/globalassets/z_corporate-site/about-us-pdfs/looking-to-the-future/understanding-the-exogenous-drivers-of-wholesale-wastewater-costs-in-eng....pdf

T6005 - Arup & Vivid Economics. (2018). Use of econometric models for cost assessment at PR19. Retrieved from https://www.unitedutilities.com/globalassets/z_corporate-site/about-us-pdfs/looking-to-the-future/use-of-econometric-models-for-cost-assessment-at-pr19--vivid-arup-feb-2....pdf

United Utilities. (2018, May 4). Cost Assessment for PR19 – a consultation on econometric cost modelling response. Retrieved from https://www.ofwat.gov.uk/wp-content/uploads/2018/03/UU-consultation-response-sheet.xlsx

WIK-Consult. (2011). Cost Benchmarking in Energy Regulation in European Countries.

Copyright © United Utilities Water Limited 2018 109

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Appendix Model assessment framework – consultation results/application In selecting our final suite of benchmarking models, we applied our model assessment framework to all models proposed within the 2018 consultation (Ofwat, Cost Assessment for PR19 – a consultation on econometric cost modelling, 2018) as well as those developed internally by United Utilities. The results of this framework are set out in the following sections with justifications given where models failed or were not used within the final suite. We have not included a full review of each model here but this is within our response to the consultation (United Utilities, 2018). In splits where there were many models proposed by companies (e.g. Network plus) we could be more rigid in our application of the framework as there is a much bigger pool of models to select from. Where there were fewer models within a particular split that we are looking to utilise, we have been less prescriptive. Rather than simply discounting models that may have some issues, we have in some instances sought to correct these deficiencies where possible in order to develop a credible model that aligns with expectations. We have used the template below for the results of the model assessment framework, summarising each model and the results of the three tests. Where a model passes a test we colour it green, where it fails we have coloured it red and added comments on why it has failed. Where we have concerns over aspects of a model with respect to the test (which are not severe enough to result in a fail) we have also made comments.

Model ID Test Comment

Model ID Engineering/Economic

Transparency

Statistical

Water Resources For water resources, there were a number of potentially acceptable models within the consultation responses but for our final suite, we have selected OWR1 and a model that we have developed combining the various cost drivers associated with the value chain and the findings from the assessment of each model within the consultation.

Model ID Test Comment

OWR1 Engineering/Economic

Transparency Statistical Used in final suite

OWR2 Engineering/Economic Only capturing pumping head and no other factor to distinguish between source

types

ANHWR1 Engineering/Economic We do not agree that river and pumped storage sources would (typically) have

comparable costs at an industry level and so should not be grouped together.

ANHWR2

Engineering/Economic

Transparency Use of endogenous factor in scale variable may lead to perverse incentives and reward inefficient behaviour e.g. leakage.

ANHWR3

Engineering/Economic

Transparency Use of endogenous factor in scale variable may lead to perverse incentives and reward inefficient behaviour e.g. leakage.

ANHWR4

Engineering/Economic

Transparency Use of endogenous factor in scale variable may lead to perverse incentives and reward inefficient behaviour e.g. leakage.

Copyright © United Utilities Water Limited 2018 110

Chapter 7: Supplementary Document - S6002

unitedutilities.com

SRNWR1

Engineering/Economic Unsure as to the relevance of having reservoir and capacity and % DI from IR within the same model

Transparency

Statistical Potential model but we have developed a new WR model to better align to the engineering priors

SRNWR2

Engineering/Economic Unsure as to the relevance of having reservoir and capacity and % DI from IR within the same model

Transparency

Statistical Potential model but we have developed a new WR model to better align to the engineering priors

SRNWR3

Engineering/Economic Transparency

Statistical Potential model but we have developed a new WR model to better align to the engineering priors

YKYWR1

Engineering/Economic Unsure as to the relevance of having reservoir and capacity and % DI from IR within the same model

Transparency

Statistical Potential model but we have developed a new WR model to better align to the engineering priors

YKYSSCWR2

Engineering/Economic Transparency

Statistical Potential model but we have developed a new WR model to better align to the engineering priors

YKYWR3

Engineering/Economic Unsure as to the relevance of having reservoir and capacity and % DI from IR within the same model

Transparency

Statistical Potential model but we have developed a new WR model to better align to the engineering priors

YKYWR4

Engineering/Economic Transparency

Statistical Potential model but we have developed a new WR model to better align to the engineering priors

BRLWR1

Engineering/Economic

Transparency Definition of botex does not exclude abstraction costs, which we do not believe can be predicted by an econometric model. Time trend lacks validity in the context of an ex-post dynamic efficiency challenge.

BRLWR2

Engineering/Economic

Transparency Definition of botex does not exclude abstraction costs, which we do not believe can be predicted by an econometric model. Time trend lacks validity in the context of an ex-post dynamic efficiency challenge.

Water Treatment We have not utilised Water treatment models within our proposed suite and therefore these models did not go through the model assessment framework. We have concerns about the validity of a Water Treatment model in isolation and the ability of them to be credibly predict efficient expenditure created given the available information and the quality of that data. The models are highly susceptible to extreme outliers which indicates that there are omitted variables driving significant variation between companies which will incorrectly be associated with (in)efficiency differences. Water Resources (plus) We agree that a Water Resources (plus) suite is a beneficial split to utilise as it has the potential to capture the substitution effects that exist between the value chains and could therefore be more capable of explaining cost differences between companies than single value chain models if specified correctly. However, whilst four models passed all criteria (OWRP1-4), we have developed a new model for use within the final suite in an attempt to capture all drivers within a single model. Our proposed model takes into account the required cost drivers for these value chains building on the findings from the assessment framework.

Copyright © United Utilities Water Limited 2018 111

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Model ID Test Comments

OWRP1

Engineering/Economic No explanatory factors to capture variations in treatment complexity that drives a significant variation in expenditure within this split but acceptable model if used in conjunction with other models that capture these impacts.

Transparency

Statistical

OWRP2

Engineering/Economic No explanatory factors to capture variations in treatment complexity that drives a significant variation in expenditure within this split but acceptable model if used in conjunction with other models that capture these impacts.

Transparency

Statistical

OWRP3

Engineering/Economic No explanatory factors to capture variations in treatment complexity that drives a significant variation in expenditure within this split but acceptable model if used in conjunction with other models that capture these impacts.

Transparency

Statistical

OWRP4

Engineering/Economic No explanatory factors to capture variations in treatment complexity that drives a significant variation in expenditure within this split but acceptable model if used in conjunction with other models that capture these impacts.

Transparency

Statistical

OWRP5

Engineering/Economic The size of the coefficient on average pumping head is too large to be capturing only the costs associated with pumping for this allocation of cost and therefore will over remunerate those companies that have higher pumping.

OWRP6

Engineering/Economic The size of the coefficient on average pumping head is too large to be capturing only the costs associated with pumping for this allocation of cost and therefore will over remunerate those companies that have higher pumping.

OWRP7

Engineering/Economic The size of the coefficient on average pumping head is too large to be capturing only the costs associated with pumping for this allocation of cost and therefore will over remunerate those companies that have higher pumping.

OWRP8

Engineering/Economic The size of the coefficient on average pumping head is too large to be capturing only the costs associated with pumping for this allocation of cost and therefore will over remunerate those companies that have higher pumping.

WSXWRP1 Engineering/Economic

The size of the coefficient on average pumping head is too large to be capturing only the costs associated with pumping for this allocation of cost and therefore will over remunerate those companies that have higher pumping. We do not believe that dummy variables for companies should be used within industry models (proposed density measure is a dummy variable for Thames)

Copyright © United Utilities Water Limited 2018 112

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Transparency Use of endogenous factor in scale variable may lead to perverse incentives and reward inefficient behaviour e.g. leakage.

WSXWRP2

Engineering/Economic

The size of the coefficient on average pumping head is too large to be capturing only the costs associated with pumping for this allocation of cost and therefore will over remunerate those companies that have higher pumping. We do not believe that dummy variables for companies should be used within industry models (proposed density measure is a dummy variable for Thames)

Transparency Use of endogenous factor in scale variable may lead to perverse incentives and reward inefficient behaviour e.g. leakage.

WSXWRP3

Engineering/Economic

The size of the coefficient on average pumping head is too large to be capturing only the costs associated with pumping for this allocation of cost and therefore will over remunerate those companies that have higher pumping. We do not believe that dummy variables for companies should be used within industry models (proposed density measure is a dummy variable for Thames)

Transparency Use of endogenous factor in scale variable may lead to perverse incentives and reward inefficient behaviour e.g. leakage.

WSXWRP4

Engineering/Economic

The size of the coefficient on average pumping head is too large to be capturing only the costs associated with pumping for this allocation of cost and therefore will over remunerate those companies that have higher pumping. We do not believe that dummy variables for companies should be used within industry models (proposed density measure is a dummy variable for Thames)

Transparency Use of endogenous factor in scale variable may lead to perverse incentives and reward inefficient behaviour e.g. leakage.

Treated water distribution For treated water distribution, there were a number of potentially acceptable models within the consultation responses but for our final suite, we have selected a model (same model with different scale drivers) that we have developed combining the various cost drivers associated with the value chain and the findings from the assessment of each model within the consultation. We prefer using fully exogenous scale drivers in models but within treated water distribution, it may be appropriate to use the length of mains given the direct impact and relationship with expenditure. Model ID Test Comments

OTWD1

Engineering/Economic

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Acceptable model for internal validity but endogenous factor makes it less appropriate for predictions

OTWD2

Engineering/Economic

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Acceptable model for internal validity but endogenous factor makes it less appropriate for predictions

Copyright © United Utilities Water Limited 2018 113

Chapter 7: Supplementary Document - S6002

unitedutilities.com

OTWD3

Engineering/Economic Engineering and economic evidence supports the use of a variable to capture the percentage of old mains rather than assuming that all mains other than new mains have the same failure rates.

OTWD4

Engineering/Economic Engineering and economic evidence supports the use of a variable to capture the percentage of old mains rather than assuming that all mains other than new mains have the same failure rates.

OTWD5

Engineering/Economic Engineering and economic evidence supports the use of a variable to capture the percentage of old mains rather than assuming that all mains other than new mains have the same failure rates.

OTWD6

Engineering/Economic Engineering and economic evidence supports the use of a variable to capture the percentage of old mains rather than assuming that all mains other than new mains have the same failure rates.

OTWD7

Engineering/Economic Engineering and economic evidence supports the use of a variable to capture the percentage of old mains rather than assuming that all mains other than new mains have the same failure rates.

OTWD8

Engineering/Economic Engineering and economic evidence supports the use of a variable to capture the percentage of old mains rather than assuming that all mains other than new mains have the same failure rates.

TMSTWD1

Engineering/Economic

Economic priori does not support the idea notion that regional wages are a significant driver of expenditure requirements across the industry. Selection of network variables for age and diameter not consistent with engineering priori. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

TMSTWD2

Engineering/Economic

Economic priori does not support the idea notion that regional wages are a significant driver of expenditure requirements across the industry. Selection of network variables for age and diameter not consistent with engineering priori. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

Copyright © United Utilities Water Limited 2018 114

Chapter 7: Supplementary Document - S6002

unitedutilities.com

TMSTWD3

Engineering/Economic

Economic priori does not support the idea notion that regional wages are a significant driver of expenditure requirements across the industry. Selection of network variables for age and diameter not consistent with engineering priori. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

TMSTWD4

Engineering/Economic

Economic priori does not support the idea notion that regional wages are a significant driver of expenditure requirements across the industry. Selection of network variables for age and diameter not consistent with engineering priori. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

TMSTWD5

Engineering/Economic

Economic priori does not support the idea notion that regional wages are a significant driver of expenditure requirements across the industry. Selection of network variables for age and diameter not consistent with engineering priori. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

WSXTWD1

Engineering/Economic

Transparency

Statistical Squared term for density is not stable to replication making it less appropriate for use in cost assessment

WSXTWD2

Engineering/Economic

Transparency

Statistical Squared term for density is not stable to replication making it less appropriate for use in cost assessment

Water network plus There was a large number of Water Network plus models submitted as part of the consultation, which meant that we have been more firm in our application of the assessment framework and deciding whether to fail a model. We have rationalised down to three final models, BRLNPW3, SVTNPW1 and a model that we have developed which takes into account the strengths of those models that were deemed acceptable but did not get included in the final suite. Model ID Test Comments

ONPW1

Engineering/Economic Engineering and economic priori does not support including only water treatment pumping head within a network plus model

ONPW2 Engineering/Economic

Engineering and economic priori does not support including only water treatment pumping head within a network plus model Engineering and economic priori supports the use of using metrics for the percentages of older mains to capture asset failure rates and not post 1981.

Copyright © United Utilities Water Limited 2018 115

Chapter 7: Supplementary Document - S6002

unitedutilities.com

ONPW3

Engineering/Economic

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Acceptable model for internal validity but endogenous factor makes it less appropriate for predictions

ONPW4

Engineering/Economic Engineering and economic priori supports the use of using metrics for the percentages of older mains to capture asset failure rates and not post 1981.

ONPW5

Engineering/Economic

Engineering and economic priori does not support including only water treatment pumping head within a network plus model Engineering and economic priori supports the use of using metrics for the percentages of older mains to capture asset failure rates and not post 1981.

ONPW6

Engineering/Economic Engineering and economic priori supports the use of using metrics for the percentages of older mains to capture asset failure rates and not post 1981.

ONPW7

Engineering/Economic

Engineering and economic priori does not support including only water treatment pumping head within a network plus model Engineering and economic priori supports the use of using metrics for the percentages of older mains to capture asset failure rates and not post 1981.

ONPW8

Engineering/Economic Engineering and economic priori supports the use of using metrics for the percentages of older mains to capture asset failure rates and not post 1981.

ANHNPW1

Engineering/Economic

Time trend supports internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied. Economic priori does not support the notion that regional wages are a significant driver of expenditure requirements across the industry.

ANHNPW2

Engineering/Economic

Time trend supports internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied. Economic priori does not support the notion that regional wages are a significant driver of expenditure requirements across the industry.

ANHNPW3 Engineering/Economic Time trend supports internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied.

Copyright © United Utilities Water Limited 2018 116

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Economic priori does not support the notion that regional wages are a significant driver of expenditure requirements across the industry.

ANHNPW4

Engineering/Economic

Time trend supports internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied. Economic priori does not support the notion that regional wages are a significant driver of expenditure requirements across the industry.

ANHNPW5

Engineering/Economic

Time trend supports internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied. Economic priori does not support the notion that regional wages are a significant driver of expenditure requirements across the industry.

SRNNPW1

Engineering/Economic Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. Length of main a less appropriate scale driver at the Network plus level

SRNNPW2

Engineering/Economic Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

SRNNPW3

Engineering/Economic Simple (aggregate company) measures of density are inferior to granular metrics and are unlikely to capture the true relationship with expenditure requirements.

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

SVTNPW1

Engineering/Economic Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Transparency

Statistical High correlation between number of treatment works and scale variable. Model is in the final suite but with the time trend and number of works variables removed.

SVTNPW2

Engineering/Economic Inclusion of squared term for scale driver (length) with no engineering priori as density is captured through other variables

SVTNPW3 Engineering/Economic Inclusion of squared term for scale driver (length) with no engineering priori as density

is captured through other variables

Copyright © United Utilities Water Limited 2018 117

Chapter 7: Supplementary Document - S6002

unitedutilities.com

SVTNPW4

Engineering/Economic

Transparency Use of a more endogenous factor for the scale variable may lead to perverse incentives and reward inefficient behaviour on e.g. leakage.

SVTNPW5

Engineering/Economic Inclusion of squared term for scale driver (DI) with no engineering priori as density is captured through other variables

SVTNPW6

Engineering/Economic Inclusion of squared term for scale driver (length) with no engineering priori as density is captured through other variables

SVTNPW7

Engineering/Economic Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Transparency

Statistical Acceptable model but inferior to SVTNPW1 in statistical & robustness testing and so rejected from final suite

SWBNPW1

Engineering/Economic Measure of density (mains/property) would suggest that increased density i.e. smaller mains/property, would result in smaller expenditure requirements which is contrary to engineering priori.

SWBNPW2

Engineering/Economic Measure of density (mains/property) would suggest that increased density i.e. smaller mains/property, would result in smaller expenditure requirements which is contrary to engineering priori.

SWBNPW3

Engineering/Economic

Transparency More appropriate to assess enhancement expenditure separately due to the ex post adjustments that need to be made to the BCT for grants and contributions

SWBNPW4

Engineering/Economic

Transparency More appropriate to assess enhancement expenditure separately due to the ex post adjustments that need to be made to the BCT for grants and contributions

SWBNPW5

Engineering/Economic

Transparency More appropriate to assess enhancement expenditure separately due to the ex post adjustments that need to be made to the BCT for grants and contributions

SWBNPW6

Engineering/Economic

Transparency More appropriate to assess enhancement expenditure separately due to the ex post adjustments that need to be made to the BCT for grants and contributions

Copyright © United Utilities Water Limited 2018 118

Chapter 7: Supplementary Document - S6002

unitedutilities.com

TMSNPW1

Engineering/Economic

Economic priori does not support the notion that regional wages are a significant driver of expenditure requirements across the industry. Selection of network variables for age and diameter not consistent with engineering priori. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

TMSNPW2

Engineering/Economic

Economic priori does not support the notion that regional wages are a significant driver of expenditure requirements across the industry. Selection of network variables for age and diameter not consistent with engineering priori. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

TMSNPW3

Engineering/Economic

Economic priori does not support the notion that regional wages are a significant driver of expenditure requirements across the industry. Selection of network variables for age and diameter not consistent with engineering priori. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

TMSNPW4

Engineering/Economic

Economic priori does not support the notion that regional wages are a significant driver of expenditure requirements across the industry. Selection of network variables for age and diameter not consistent with engineering priori. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

WSHNPW1

Engineering/Economic

Inclusion of two treatment complexities adds additional validity to better distinguish between requirements. No measure of network complexity or pumping requirements will cause some unintended outliers

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Model is acceptable but not selected in final suite due to similarities with WSHWW1 that has been included.

WSHNPW2

Engineering/Economic

Inclusion of two treatment complexities adds additional validity to better distinguish between requirements. No measure of network complexity or pumping requirements will cause some unintended outliers

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Model is acceptable but not selected in final suite due to similarities with WSHWW1 that has been included.

Copyright © United Utilities Water Limited 2018 119

Chapter 7: Supplementary Document - S6002

unitedutilities.com

YKYNPW1

Engineering/Economic

Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

YKYNPW2

Engineering/Economic

Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

YKYNPW3

Engineering/Economic

Aggregate measures of density are inferior to granular metrics that look at e.g. population densities. The level of treatment within the proposed treatment complexity measure is too low and will therefore not fully capture the differences between surface and ground water sources.

Transparency

Statistical

BRLNPW1

Engineering/Economic Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical Acceptable model but BRLNPW3 more appropriate as engineering priori supports surface water percentage in favour of a tight (>=band 5) treatment complexity measure

BRLNPW2

Engineering/Economic Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical Acceptable model but BRLNPW3 more appropriate as engineering priori supports surface water percentage in favour of a tight (>=band 5) treatment complexity measure

BRLNPW3

Engineering/Economic Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical Used in proposed model suite without time dummies

SEWNPW1

Engineering/Economic No measure of network complexity or pumping requirements will cause some unintended outliers

Transparency

Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Potential to use as part of suite of suitable complementary models are found that can account for omitted variables within model.

SEWNPW2 Engineering/Economic No measure of network complexity or pumping requirements will cause some unintended outliers

Copyright © United Utilities Water Limited 2018 120

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Transparency

Use of endogenous factor in scale variable may lead to perverse incentives and reward inefficient behaviour e.g. leakage. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

SEWNPW3

Engineering/Economic No measure of network complexity or pumping requirements will cause some unintended outliers

Transparency

Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Potential to use as part of suite of suitable complementary models are found that can account for omitted variables within model.

SEWNPW4

Engineering/Economic No measure of network complexity or pumping requirements will cause some unintended outliers

Transparency

Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Potential to use as part of suite of suitable complementary models are found that can account for omitted variables within model.

SEWNPW5

Engineering/Economic No measure of network complexity or pumping requirements will cause some unintended outliers

Transparency

Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Potential to use as part of suite of suitable complementary models are found that can account for omitted variables within model.

SSCNPW1

Engineering/Economic

No measure of network complexity or pumping requirements will cause some unintended outliers Aggregate measures of density are inferior to granular metrics that look at e.g. population densities. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements.

SSCNPW2 Engineering/Economic No measure of network complexity or pumping requirements will cause some unintended outliers

Copyright © United Utilities Water Limited 2018 121

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Aggregate measures of density are inferior to granular metrics that look at e.g. population densities. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements.

Wholesale Water There was a large number of Wholesale Water models submitted as part of the consultation, which meant that we have been more firm in our application of the assessment framework and deciding whether to fail a model. We have rationalised down to three final models, BRLNPW3, SVTNPW1 and a model that we have developed which takes into account the strengths of those models that were deemed acceptable but did not get included in the final suite. Model ID Test Comment

OWW1

Engineering/Economic

Model uses a split of average pumping head that does not align with the dependent variable even though there is significant variation between companies in treated water distribution pumping head. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW2

Engineering/Economic

Model uses a split of average pumping head that does not align with the dependent variable even though there is significant variation between companies in treated water distribution pumping head. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW3

Engineering/Economic

Model uses a split of average pumping head that does not align with the dependent variable even though there is significant variation between companies in treated water distribution pumping head. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW4 Engineering/Economic Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements.

Copyright © United Utilities Water Limited 2018 122

Chapter 7: Supplementary Document - S6002

unitedutilities.com

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW5

Engineering/Economic

Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW6

Engineering/Economic

Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW7

Engineering/Economic

Model uses a split of average pumping head that does not align with the dependent variable even though there is significant variation between companies in treated water distribution pumping head. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW8

Engineering/Economic

Model uses a split of average pumping head that does not align with the dependent variable even though there is significant variation between companies in treated water distribution pumping head. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW9 Engineering/Economic

Model uses a split of average pumping head that does not align with the dependent variable even though there is significant variation between companies in treated water distribution pumping head. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Copyright © United Utilities Water Limited 2018 123

Chapter 7: Supplementary Document - S6002

unitedutilities.com

OWW10

Engineering/Economic

Model uses a split of average pumping head that does not align with the dependent variable even though there is significant variation between companies in treated water distribution pumping head. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW11

Engineering/Economic

Model uses a split of average pumping head that does not align with the dependent variable even though there is significant variation between companies in treated water distribution pumping head. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

OWW12

Engineering/Economic

Model uses a split of average pumping head that does not align with the dependent variable even though there is significant variation between companies in treated water distribution pumping head. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

SRNWW1

Engineering/Economic

Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

SRNWW2 Engineering/Economic

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities. Time trend supports internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Copyright © United Utilities Water Limited 2018 124

Chapter 7: Supplementary Document - S6002

unitedutilities.com

SRNWW3

Engineering/Economic

Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities. Time trend supports internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SWBWW1

Engineering/Economic

Some issues but potential to use as part of a suite if complementary models are found. Missing variable(s) that capture network characteristics e.g. age, size etc. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

SWBWW2

Engineering/Economic

Some issues but potential to use as part of a suite if complementary models are found. Missing variable(s) that capture network characteristics e.g. age, size etc. Aggregate measures of density are inferior to granular metrics that look at e.g. population densities.

SWBWW3

Engineering/Economic

Transparency More appropriate to assess enhancement expenditure separately due to the ex post adjustments that need to be made to the BCT for grants and contributions

SWBWW4

Engineering/Economic

Transparency More appropriate to assess enhancement expenditure separately due to the ex post adjustments that need to be made to the BCT for grants and contributions

SWBWW5

Engineering/Economic

Transparency We do not support the notion that Totex models can accurately predict required expenditure

SWBWW6

Engineering/Economic

Transparency We do not support the notion that Totex models can accurately predict required expenditure

WSHWW1

Engineering/Economic Some issues but potential to use as part of a suite if complementary models are found. Captures treatment complexity at both the simple and complex forms.

Transparency

Statistical

Copyright © United Utilities Water Limited 2018 125

Chapter 7: Supplementary Document - S6002

unitedutilities.com

WSHWW2

Engineering/Economic

Transparency Same as WSHWW1 but uses endogenous variable for % renewed/relined and so is inferior from a cost assessment and does not include two metrics to capture treatment complexities and so is inferior for use within a suite.

YKYWW1

Engineering/Economic

Length of mains less appropriate as a scale driver at the aggregate level. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements. Lacks adequate treatment complexity measures, source types are insignificant

YKYWW2

Engineering/Economic Length of mains less appropriate as a scale driver at the aggregate level.

YKYWW3

Engineering/Economic Length of mains less appropriate as a scale driver at the aggregate level. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements.

YKYWW4

Engineering/Economic

Uses simple metric for density which is inferior but use of quadratic term can be beneficial in explaining additional costs driven by sparsity. Treatment complexity for simple treatment included but variable for complex treatment is omitted. Consider for use in suite of complementary model and be found

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Model is acceptable but considered inferior to WSHWW1 and so not selected for inclusion

YKYWW5

Engineering/Economic

Uses simple metric for density which is inferior but use of quadratic term can be beneficial in explaining additional costs driven by sparsity. Treatment complexity for simple treatment included but variable for complex treatment is omitted. Consider for use in suite of complementary model and be found

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Model is acceptable but considered inferior to WSHWW1 and so not selected for inclusion

YKYWW6

Engineering/Economic

Uses simple metric for density which is inferior but use of quadratic term can be beneficial in explaining additional costs driven by sparsity. Treatment complexity for simple treatment included but variable for complex treatment is omitted. Consider for use in suite of complementary model and be found

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Copyright © United Utilities Water Limited 2018 126

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Statistical Model is acceptable but considered inferior to WSHWW1 and so not selected for inclusion

AFWWW1

Engineering/Economic

Transparency

Statistical Wide spread of residuals that appear to be clustered into two separate types. Consider including if corresponding model can be found to complement this model and prevent this issue

AFWWW2

Engineering/Economic

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied.

Statistical Significance of source type and time dummies

AFWWW3

Engineering/Economic No explanatory factor for source type but uses treatment complexity (which increases when surface water variables are not included) and so is likely picking up some of the variation that is driven by this.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical Wide spread of residuals that appear to be clustered into two separate types. Consider including if corresponding model can be found to complement this model and prevent this issue

AFWWW4

Engineering/Economic

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical Wide spread of residuals that appear to be clustered into two separate types. Consider including if corresponding model can be found to complement this model and prevent this issue

BRLWW1

Engineering/Economic

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical 50% of variables insignificant

BRLWW2

Engineering/Economic

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical Correlation between Surface water treated / Total water treated (%) and Share of water from reservoirs (%) 0.636 which is too high

BRLWW3

Engineering/Economic

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Copyright © United Utilities Water Limited 2018 127

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Statistical >50% of variables insignificant

SEWWW1

Engineering/Economic

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Model is acceptable but inferior to SEWWW4 and so not selected for inclusion

SEWWW2

Engineering/Economic

Transparency

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Statistical Model is acceptable but inferior to SEWWW4 and so not selected for inclusion

SEWWW3

Engineering/Economic

Transparency

Use of endogenous factor in scale variable may lead to perverse incentives and reward inefficient behaviour e.g. leakage. % of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network. Time trend lacks validity in the context of an ex-post dynamic efficiency challenge.

SEWWW4

Engineering/Economic

Transparency Time trend lacks validity in the context of an ex-post dynamic efficiency challenge.

Statistical Used in final suite

SSCWW1

Engineering/Economic Length of mains less appropriate as a scale driver at the aggregate level. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements.

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

SSCWW2

Engineering/Economic Length of mains less appropriate as a scale driver at the aggregate level. Engineering priori does not support the use of metrics for the percentages pre 1981 to explain failure rates and resulting maintenance requirements.

% of mains length refurbished and relined is an endogenous activity based measure that does not capture the underlying need and therefore risks not adequately remunerating companies that take more innovative approaches to solving structural issues on the network.

Wholesale Water (plus enhancement) Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Copyright © United Utilities Water Limited 2018 128

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Bioresources For Bioresources, there were a number of potentially acceptable models within the consultation responses but for our final suite, we have selected WSHBR1 and UUBR2 as we feel that these models complement one another when used together which should give a more fair distribution between companies. Model ID Test Comment

OBR1

Engineering/Economic

Transparency Bioresources models should use sludge produced as the scale driver to maintain consistency with the volume forecasting incentive mechanism.

OBR2

Engineering/Economic

Most companies have >90% of sludge disposed to farmland apart from United Utilities which uses incineration. This means that one company drives a significant amount of variation as the % of sludge disposed to farmland will act as a proxy dummy variable that will also have impacts on the % intersiting variable.

Transparency

Whilst measures of work done will inevitably represent the actual activity that a company undertakes and add to the internal validity of the model, given its endogeneity, it risks remunerating inefficient behaviours or over remunerating companies that travel greater distances in order to sell sludge .

OBR3

Engineering/Economic

Most companies have >90% of sludge disposed to farmland apart from United Utilities which uses incineration. This means that one company drives a significant amount of variation as the % of sludge disposed to farmland will act as a proxy dummy variable that will also have impacts on the % intersiting variable.

Transparency

Whilst measures of work done will inevitably represent the actual activity that a company undertakes and add to the internal validity of the model, given its endogeneity, it risks remunerating inefficient behaviours or over remunerating companies that travel greater distances in order to sell sludge .

ANHBR1

Engineering/Economic

Measures of a company’s area are not an appropriate driver of cost as it ignores the population density within that region i.e. a company could have a large area but all of its customers centred within a single city, as well as not accounting for the availability of suitable landbank at which sludge can be disposed within that region. We cannot see any evidence to support a positive time trend of that magnitude.

ANHBR2

Engineering/Economic

Measures of a company’s area are not an appropriate driver of cost as it ignores the population density within that region i.e. a company could have a large area but all of its customers centred within a single city, as well as not accounting for the availability of suitable landbank at which sludge can be disposed within that region. We cannot see any evidence to support a positive time trend of that magnitude.

ANHBR3

Engineering/Economic

We cannot see any evidence to support a positive time trend of that magnitude. The calculation of ttds generated by different sized works ignores both the treatment process within Network plus (and therefore how much sludge is produced) as well as the amount of dewatering at each site (and therefore the volume that is transported/intersited)

ANHBR4 Engineering/Economic We cannot see any evidence to support a positive time trend of that magnitude.

Copyright © United Utilities Water Limited 2018 129

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

ANHBR5

Engineering/Economic

We cannot see any evidence to support a positive time trend of that magnitude. Whilst the engineering priori supports that sparsity drives increased intersiting costs there is not a variable to capture landbank availability and so this will result in skewed predictions. It is logical to hypothesise that companies which operate within more sparse regions have a greater (potential for) available landbank and therefore will also have lower disposal costs which will in part, offset the increased intersiting requirements.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

ANHBR6

Engineering/Economic We cannot see any evidence to support a positive time trend of that magnitude. Combined coefficient on the scale drivers is significantly greater than 1 (1.260), which is not credible.

SRNBR1

Engineering/Economic Colocation can act as a proxy for the amount of intersiting but there is no variable to account for disposal requirements or landbank availability that do vary significantly across the industry.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

SRNBR2

Engineering/Economic No variable to account for disposal requirements or landbank availability that do vary significantly across the industry.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

SRNBR3

Engineering/Economic

Transparency % tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest

Copyright © United Utilities Water Limited 2018 130

Chapter 7: Supplementary Document - S6002

unitedutilities.com

in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

SRNBR4

Engineering/Economic The % of load treated in small WwTW will act as a proxy for sparsity (and therefore intersiting requirements) and so it is not clear why an additional measure of density is required.

Transparency

Statistical

SVTBR1

Engineering/Economic

All companies have <9% of sludge intersited by pipelines apart from United Utilities (37%), with many companies having 0% transported using this method. This means that one company drives a significant amount of variation and the distance intersited by pipeline will act as a proxy dummy variable for that company. We cannot see any evidence to support a positive time trend of that magnitude.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

SVTBR2

Engineering/Economic

Coefficient on main scale driver is significantly greater than 1 and so not credible. Further to this, the squared term on the scale driver is positive and so this will exacerbate this issue. All companies have <9% of sludge intersited by pipelines apart from United Utilities (37%), with many companies having 0% transported using this method. This means that one company drives a significant amount of variation and the distance intersited by pipeline will act as a proxy dummy variable for that company. We cannot see any evidence to support a positive time trend of that magnitude.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

SWBBR1

Engineering/Economic No variable to account for disposal activity or available landbank

Transparency

Statistical Acceptable model to use as part of a suite. WSHBR1 preferred in final suite.

SWBBR2

Engineering/Economic No variable to account for disposal activity or available landbank

Transparency

Statistical Acceptable model to use as part of a suite. WSHBR1 preferred in final suite.

SWBBR3

Engineering/Economic No variable to account for disposal activity or available landbank

Transparency

Statistical Acceptable model to use as part of a suite. WSHBR1 preferred in final suite.

SWBBR4 Engineering/Economic

Copyright © United Utilities Water Limited 2018 131

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Transparency The final methodology confirmed that enhancement expenditure within Bioresources will be assessed separately and therefore these models contradict the guidance.

SWBBR5

Engineering/Economic

Transparency The final methodology confirmed that enhancement expenditure within Bioresources will be assessed separately and therefore these models contradict the guidance.

SWBBR6

Engineering/Economic

Transparency The final methodology confirmed that enhancement expenditure within Bioresources will be assessed separately and therefore these models contradict the guidance.

SWBBR7

Engineering/Economic

Transparency The final methodology confirmed that enhancement expenditure within Bioresources will be assessed separately and therefore these models contradict the guidance.

SWBBR8

Engineering/Economic

Transparency The final methodology confirmed that enhancement expenditure within Bioresources will be assessed separately and therefore these models contradict the guidance.

SWBBR9

Engineering/Economic

Transparency The final methodology confirmed that enhancement expenditure within Bioresources will be assessed separately and therefore these models contradict the guidance.

UUBR1

Engineering/Economic

Transparency

Statistical Acceptable model but inferior to UUBR2

UUBR2

Engineering/Economic

Transparency

Statistical Used in final model suite

WSHBR1

Engineering/Economic

Transparency

Statistical Used in final model suite

WSXBR1

Engineering/Economic

Simple model but Ofwat measure of highly dense areas is a dummy variable for Thames and should not be utilised in an industry model. Resulting model for the industry would in effect be a simple unit cost model, which would be more preferable and could be considered for validation purposes.

WSXBR2 Engineering/Economic

Simple model but Ofwat measure of highly dense areas is a dummy variable for Thames and should not be utilised in an industry model. Resulting model for the industry would in effect be a simple unit cost model, which would be more preferable and could be considered for validation purposes.

Copyright © United Utilities Water Limited 2018 132

Chapter 7: Supplementary Document - S6002

unitedutilities.com

WSXBR3

Engineering/Economic

Simple model but Ofwat measure of highly dense areas is a dummy variable for Thames and should not be utilised in an industry model. Resulting model for the industry would in effect be a simple unit cost model, which would be more preferable and could be considered for validation purposes.

WSXBR4

Engineering/Economic

Simple model but Ofwat measure of highly dense areas is a dummy variable for Thames and should not be utilised in an industry model. Resulting model for the industry would in effect be a simple unit cost model, which would be more preferable and could be considered for validation purposes.

YKYBR1

Engineering/Economic No variable to capture land bank constraints and therefore % tds treated by conventional or advanced anaerobic digestion will only account for one element of the issue.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

YKYBR2

Engineering/Economic No variable to capture land bank constraints and therefore % tds treated by conventional or advanced anaerobic digestion will only account for one element of the issue.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

YKYBR3

Engineering/Economic No variable to capture land bank constraints and therefore % tds treated by conventional or advanced anaerobic digestion will only account for one element of the issue.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

YKYBR4

Engineering/Economic No variable to capture land bank constraints and therefore % tds treated by conventional or advanced anaerobic digestion will only account for one element of the issue.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

Copyright © United Utilities Water Limited 2018 133

Chapter 7: Supplementary Document - S6002

unitedutilities.com

YKYBR5

Engineering/Economic

No variable to capture land bank constraints and therefore % tds treated by conventional or advanced anaerobic digestion will only account for one element of the issue. Measures of a company’s area are not an appropriate driver of cost as it ignores the population density within that region i.e. a company could have a large area but all of its customers centred within a single city, as well as not accounting for the availability of suitable landbank at which sludge can be disposed within that region.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

YKYBR6

Engineering/Economic

No variable to capture land bank constraints and therefore % tds treated by conventional or advanced anaerobic digestion will only account for one element of the issue. Measures of a company’s area are not an appropriate driver of cost as it ignores the population density within that region i.e. a company could have a large area but all of its customers centred within a single city, as well as not accounting for the availability of suitable landbank at which sludge can be disposed within that region.

Transparency

% tds treated by conventional or advanced anaerobic digestion is largely a response to landbank availability and power costs and therefore you would expect companies that had lower percentages to have easy/cheap access to land e.g. liming. Companies invest in aerobic digestion activities as it is the most efficient option for them to pursue and so this variable will only address one side of the balance (over remunerating companies that have adopted simple treatment types or have cheap/easy access to suitable landbank)

Sewage treatment We have not utilised Wastewater treatment models within our proposed suite and therefore these models did not go through the model assessment framework. Bioresources plus/Treatment and sludge We agree that a Bioresources (plus) suite is a beneficial split to utilise as it has the potential to capture the substitution effects that exist between the value chains and could therefore be more capable of explaining cost differences between companies than single value chain models if specified correctly. There were not as many models submitted covering this aggregation of the value chains and so we have adopted two of the models proposed by Ofwat as well as one of the models developed by Vivid in their report (T6005 - Arup & Vivid Economics, 2018). Our proposed model takes into account the required cost drivers for these value chains building on the findings from the assessment framework. Model ID Test Comments

OBRP1

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale

OBRP2

Engineering/Economic

Omitted variables to capture treatment complexity or economies of scale Most companies have >90% of sludge disposed to farmland apart from United Utilities which uses incineration. This means that one company drives a significant amount of variation as the % of sludge disposed to farmland will act as a proxy dummy variable

Copyright © United Utilities Water Limited 2018 134

Chapter 7: Supplementary Document - S6002

unitedutilities.com

OBRP3

Engineering/Economic

Omitted variables to capture treatment complexity or economies of scale Most companies have >90% of sludge disposed to farmland apart from United Utilities which uses incineration. This means that one company drives a significant amount of variation as the % of sludge disposed to farmland will act as a proxy dummy variable

OBRP4

Engineering/Economic Omitted variable to capture treatment complexity

Transparency

Statistical Acceptable model but considered inferior to OBRP5 as it uses a more exogenous scale variable with greater certainty around measurement

OBRP5

Engineering/Economic Omitted variable to capture treatment complexity

Transparency

Statistical Used in final suite, OBRP5 and OBRP7 complement one another correcting for omitted variables

OBRP6

Engineering/Economic Omitted variable to capture economies of scale

Transparency

Statistical Acceptable model but considered inferior to OBRP7 as it uses a more exogenous scale variable with greater certainty around measurement

OBRP7

Engineering/Economic Omitted variable to capture economies of scale

Transparency

Statistical Used in final suite, OBRP5 and OBRP7 complement one another correcting for omitted variables

Sewage collection For sewage collection, we have selected two models from those that were proposed OSWC2 and UUSWC1 as we feel that these capture the relevant cost drivers when used together as part of an overall suite.

Model ID Test Comment

OSWC1

Engineering/Economic Transparency

Statistical Acceptable model, considered for use in final suite but OSWC2 has a narrower spread of residuals across the industry (90th percentile)

OSWC2 Engineering/Economic Omitted variable for drainage/run-off requirements as properties will only

capture household waste. Transparency Statistical Used in final suite

OSWC3

Engineering/Economic Omitted variable for density Transparency

Statistical Acceptable model, considered for use in final suite but OSWC2 has a narrower spread of residuals across the industry (90th percentile)

OSWC4

Engineering/Economic Omitted variable for drainage/run-off requirements as properties will only capture household waste. Omitted variable for density

Transparency

Statistical Acceptable model, considered for use in final suite but OSWC2 has a narrower spread of residuals across the industry (90th percentile)

OSWC5 Engineering/Economic Omitted variable for drainage/run-off requirements as properties will only capture household waste. Omitted variable for density

Copyright © United Utilities Water Limited 2018 135

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Transparency

Statistical Acceptable model, considered for use in final suite but OSWC2 has a narrower spread of residuals across the industry (90th percentile)

TMSSWC1 Engineering/Economic

Economic priori does not support the idea notion that regional wages are a significant driver of expenditure requirements across the industry. Metric to capture volume per household measures the volume received at WwTW and not the volume that enters the network (it ignores spills)

TMSSWC2 Engineering/Economic

Economic priori does not support the idea notion that regional wages are a significant driver of expenditure requirements across the industry. Metric to capture volume per household measures the volume received at WwTW and not the volume that enters the network (it ignores spills)

TMSSWC3 Engineering/Economic Economic priori does not support the idea notion that regional wages are a

significant driver of expenditure requirements across the industry.

UUSWC1

Engineering/Economic

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical Used in final suite without time dummies

UUSWC2

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process. Removal of the enhancement expenditure within this model simply in UUSWC1 being produced.

WSXSWC1 Engineering/Economic Transparency Statistical >50% of variables are insignificant

WSXSWC2 Engineering/Economic Economic priori does not support using a unit cost model when the

coefficient on the aggregate model is significantly less than 1.

Wastewater Network Plus There was a large number of Wastewater Network Plus models submitted as part of the consultation, which meant that we have been more firm in our application of the assessment framework and deciding whether to fail a model. We have rationalised down to two final models, ONPWW9 and UUNPWW1. For UUNPWW1, we have adjusted the treatment complexity measure to one that is more significant in order to improve model performance but this does not change the underlying cost drivers that are accounted by the model. Similar to other models selected, these two models work well when used together as part of a suite with each offering different drivers that a more specific to different companies. Model ID Test Comment

ONPWW1 Engineering/Economic Omitted variables to capture treatment complexity or economies of scale

Copyright © United Utilities Water Limited 2018 136

Chapter 7: Supplementary Document - S6002

unitedutilities.com

ONPWW2

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale

ONPWW3

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale

ONPWW4

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale

ONPWW5

Engineering/Economic Omitted variables to capture treatment complexity, economies of scale or density

ONPWW6

Engineering/Economic Omitted variables to capture treatment complexity, economies of scale or density

ONPWW7

Engineering/Economic Omitted variables to capture treatment complexity, economies of scale or density

ONPWW8

Engineering/Economic Omitted variables to capture treatment complexity, economies of scale or density

ONPWW9

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale

Transparency

Statistical Used in final suite with density measure included

ONPWW10

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale

Transparency

Statistical Acceptable model, ONPW9 preferred as density measure included uses property density.

ANHNPWW1

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW2 Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Copyright © United Utilities Water Limited 2018 137

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW3

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW4

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW5

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW6

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW7

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW8 Engineering/Economic

Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process.

Copyright © United Utilities Water Limited 2018 138

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW9

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW10

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

ANHNPWW11

Engineering/Economic Omitted variables to capture treatment complexity or economies of scale Too much emphasis on the impact of sparsity (1-sparsity) does not equate to density and so these variables have little basis.

Transparency

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SRNNPWW1

Engineering/Economic Negative coefficient on density measure does not align with engineering priori

SRNNPWW2

Engineering/Economic Negative coefficient on density measure does not align with engineering priori

SVTNPWW1

Engineering/Economic Omitted variable to capture economies of scale Engineering priori would suggest that <5mg/l for ammonia is too lax of a threshold and that <1mg/l is more appropriate.

Transparency Unable to replicate treatment quality variable. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SVTNPWW2

Engineering/Economic Omitted variable to capture economies of scale Engineering priori would suggest that <5mg/l for ammonia is too lax of a threshold and that <1mg/l is more appropriate.

Transparency Unable to replicate treatment quality variable. Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SVTNPWW3 Engineering/Economic Omitted variable to capture economies of scale

Copyright © United Utilities Water Limited 2018 139

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Logic for a squared term on the scale driver unknown as engineering logic would not suggest that this is a U-shape.

Overly complicated model form for scale drivers and cross product terms are unlikely to be supportive of wider cost assessment process Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Model unstable to replication causing inaccurate predictions

SVTNPWW4

Engineering/Economic

Omitted variable to capture economies of scale Engineering priori would suggest that <3mg/l for ammonia is too lax of a threshold and that <1mg/l is more appropriate. Logic for a squared term on the scale driver unknown as engineering logic would not suggest that this is a U-shape.

Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Model unstable to replication causing inaccurate predictions

SVTNPWW5

Engineering/Economic

Economic priori does not support using a unit cost model when the coefficient on the aggregate model is significantly less than 1. No of STW/load could act as a proxy for the number of small works and therefore fore economies of scale but metric considered inferior to bands 1-3 on an engineering basis. Uncertainty as to what length/load is designed to capture and so expectation of sign and magnitude of the coefficient not possible. Engineering priori would suggest that <3mg/l for ammonia is too lax of a threshold and that <1mg/l is more appropriate.

Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SWBNPWW1

Engineering/Economic

Negative coefficient on density measure does not align with engineering priori Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and/or transport run off and create perverse incentives.

SWBNPWW2

Engineering/Economic

Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and/or transport run off and create perverse incentives.

SWBNPWW3

Engineering/Economic Measure of economies of scale (nr works/property) is inferior to bands 1-3 on an engineering basis Omitted variable to capture density

Transparency

Statistical Considered for inclusion but deemed inferior to suite containing UUNPWW1 and ONPWW9

SWBNPWW4 Engineering/Economic Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure

Copyright © United Utilities Water Limited 2018 140

Chapter 7: Supplementary Document - S6002

unitedutilities.com

of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives. We accept that non-resident population is likely to be a significant driver of cost for South West due to holiday goers but we do not believe that the variation between other companies is significant enough to warrant its inclusion in an industry model.

Transparency

Statistical

SWBNPWW5

Engineering/Economic Negative coefficient on density measure does not align with engineering priori

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Statistical

SWBNPWW6

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Statistical

SWBNPWW7

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Statistical

SWBNPWW8

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Statistical

SWBNPWW9

Engineering/Economic Negative coefficient on density measure does not align with engineering priori

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Statistical

Copyright © United Utilities Water Limited 2018 141

Chapter 7: Supplementary Document - S6002

unitedutilities.com

SWBNPWW10

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Statistical

SWBNPWW11

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Statistical

SWBNPWW12

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Statistical

TMSNPWW1

Engineering/Economic

Economic priori does not support the idea notion that regional wages are a significant driver of expenditure requirements across the industry. Threshold for P within quality measure is too tight and therefore not appropriate as it ignores most companies (other than trials). Load capacity treatment works variable is an inferior metric to capture economies of scale at small works to using the proportion of load within bands 1-3 as it will not capture the actual amount of load that is treated in small works

TMSNPWW2

Engineering/Economic

Economic priori does not support the idea notion that regional wages are a significant driver of expenditure requirements across the industry. Threshold for P within quality measure is too tight and therefore not appropriate as it ignores most companies (other than trials). Load capacity treatment works variable is an inferior metric to capture economies of scale at small works to using the proportion of load within bands 1-3 as it will not capture the actual amount of load that is treated in small works

UUNPWW1

Engineering/Economic

Transparency

Statistical Used in final suite but with time dummies removed and treatment quality variable replaced with more significant variation.

WSHNPWW1 Engineering/Economic Omitted variable to capture the effects of density or economies of scale at treatment works.

Copyright © United Utilities Water Limited 2018 142

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives.

YKYNPWW1

Engineering/Economic

Omitted variable to capture the effects of density or economies of scale at treatment works. Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives.

YKYNPWW2

Engineering/Economic

Omitted variable to capture the effects of economies of scale at treatment works. Negative coefficient on density measure does not align with engineering priori. Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives.

YKYNPWW3

Engineering/Economic

Omitted variable to capture the effects of economies of scale at treatment works. Negative coefficient on density measure does not align with engineering priori. Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives.

YKYNPWW4

Engineering/Economic Omitted variable to capture the effects of density or economies of scale at treatment works.

Transparency

Statistical Considered for inclusion but deemed inferior to suite containing UUNPWW1 and ONPWW9

YKYNPWW5

Engineering/Economic Omitted variable to capture the effects of economies of scale at treatment works. Negative coefficient on density measure does not align with engineering priori.

Wholesale Wastewater There was a large number of Wholesale Wastewater models submitted as part of the consultation, which meant that we have been more firm in our application of the assessment framework and deciding whether to fail a model. We have rationalised down to two final models, YKYWWW2and UUWWW1. For UUWWW1, we have adjusted the treatment complexity measure to one that is more significant in order to improve model performance but this does not change the underlying cost drivers that are accounted by the model. Similar to other models selected, these two

Copyright © United Utilities Water Limited 2018 143

Chapter 7: Supplementary Document - S6002

unitedutilities.com

models work well when used together as part of a suite with each offering different drivers that a more specific to different companies. Model ID Test Comments

OWWW1

Engineering/Economic Omitted variables to capture treatment complexity or density

OWWW2

Engineering/Economic Omitted variables to capture treatment complexity or density

OWWW3

Engineering/Economic Omitted variables to capture treatment complexity or sewage collection variations in expenditure requirements.

OWWW4

Engineering/Economic

Omitted variables to capture treatment complexity or sewage collection variations in expenditure requirements. Most companies have >90% of sludge disposed to farmland apart from United Utilities which uses incineration. This means that one company drives a significant amount of variation

OWWW5

Engineering/Economic Omitted variables to capture treatment complexity

OWWW6

Engineering/Economic Omitted variables to capture treatment complexity or sewage collection variations in expenditure requirements.

OWWW7

Engineering/Economic

Omitted variables to capture treatment complexity or sewage collection variations in expenditure requirements. Most companies have >90% of sludge disposed to farmland apart from United Utilities which uses incineration. This means that one company drives a significant amount of variation.

OWWW8

Engineering/Economic Omitted variables to capture treatment complexity

SRNWWW1

Engineering/Economic Omitted variable to capture the effects of economies of scale at treatment works. Negative coefficient on density measure does not align with engineering priori.

Transparency Bioresources spend as a proportion of Wholesale Wastewater is not significant enough to warrant including Bioresource-specific variables and should be captured within density/sparsity and economies of scale drivers instead.

SRNWWW2 Engineering/Economic Omitted variable to capture the effects of economies of scale at treatment works.

Copyright © United Utilities Water Limited 2018 144

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Negative coefficient on density measure does not align with engineering priori.

Transparency Bioresources spend as a proportion of Wholesale Wastewater is not significant enough to warrant including Bioresource-specific variables and should be captured within density/sparsity and economies of scale drivers instead.

SRNWWW3

Engineering/Economic Omitted variable to capture the effects of economies of scale at treatment works. Negative coefficient on density measure does not align with engineering priori.

Transparency Bioresources spend as a proportion of Wholesale Wastewater is not significant enough to warrant including Bioresource-specific variables and should be captured within density/sparsity and economies of scale drivers instead.

SRNWWW4

Engineering/Economic Omitted variable to capture the effects of economies of scale at treatment works. Negative coefficient on density measure does not align with engineering priori.

Transparency Bioresources spend as a proportion of Wholesale Wastewater is not significant enough to warrant including Bioresource-specific variables and should be captured within density/sparsity and economies of scale drivers instead.

SVTWWW1

Engineering/Economic Omitted variables to capture economies of scale at an asset level. Engineering priori would suggest that <5mg/l for ammonia is too lax of a threshold and that <1mg/l is more appropriate.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical Considered for use in final suite but inferior to suite containing YKYWWW2 and UUWWW1

SVTWWW2

Engineering/Economic Omitted variables to capture economies of scale at an asset level. Logic for a squared term on the scale driver unknown as engineering logic would not suggest that this is a U-shape.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SVTWWW3

Engineering/Economic

Omitted variables to capture economies of scale at an asset level. Engineering priori would suggest that <5mg/l for ammonia is too lax of a threshold and that <1mg/l is more appropriate. Logic for a squared term on the scale driver unknown as engineering logic would not suggest that this is a U-shape.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SVTWWW4

Engineering/Economic

Omitted variables to capture economies of scale at an asset level. Engineering priori would suggest that <5mg/l for ammonia is too lax of a threshold and that <1mg/l is more appropriate. Logic for a squared term on the scale driver unknown as engineering logic would not suggest that this is a U-shape.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SVTWWW5 Engineering/Economic Omitted variables to capture economies of scale at an asset level.

Copyright © United Utilities Water Limited 2018 145

Chapter 7: Supplementary Document - S6002

unitedutilities.com

Engineering priori would suggest that <3mg/l for ammonia is too lax of a threshold and that <1mg/l is more appropriate. Logic for a squared term on the scale driver unknown as engineering logic would not suggest that this is a U-shape.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SVTWWW6

Engineering/Economic

Economic priori does not support using a unit cost model when the coefficient on the aggregate model is significantly less than 1. No of STW/load could act as a proxy for the number of small works and therefore fore economies of scale but metric considered inferior to bands 1-3 on an engineering basis.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SVTWWW7

Engineering/Economic

Economic priori does not support using a unit cost model when the coefficient on the aggregate model is significantly less than 1. Uncertainty as to what length/load is designed to capture and so expectation of sign and magnitude of the coefficient not possible.

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

SWBWWW1

Engineering/Economic

Negative coefficient on density measure does not align with engineering priori Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and/or transport run off and create perverse incentives.

SWBWWW2

Engineering/Economic

Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and/or transport run off and create perverse incentives.

SWBWWW3

Engineering/Economic Measure of economies of scale (nr works/property) is inferior to bands 1-3 on an engineering basis Omitted variable to capture density

Transparency

Statistical Considered for use in final suite but inferior to suite containing YKYWWW2 and UUWWW1

SWBWWW4 Engineering/Economic

Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives. We accept that non-resident population is likely to be a significant driver of cost for South West due to holiday goers but we do not believe that the variation between other companies is significant enough to warrant its inclusion in an industry model.

Copyright © United Utilities Water Limited 2018 146

Chapter 7: Supplementary Document - S6002

unitedutilities.com

SWBWWW5

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

SWBWWW6

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

SWBWWW7

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

SWBWWW8

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

SWBWWW9

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

SWBWWW10

Engineering/Economic

Transparency Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major

Copyright © United Utilities Water Limited 2018 147

Chapter 7: Supplementary Document - S6002

unitedutilities.com

capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

SWBWWW11

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

SWBWWW12

Engineering/Economic

Transparency

Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

UUWWW1

Engineering/Economic

Transparency Time dummies support internal validity but of less relevance when an ex-post dynamic efficiency challenge is to be applied

Statistical Used in final suite but with time dummies removed and treatment quality variable replaced with more significant variation.

WSHWWW1

Engineering/Economic

Omitted variable to capture the effects of density or economies of scale at treatment works. Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives.

Transparency

Statistical Not >50% of variables are significant

YKYWWW1

Engineering/Economic

Omitted variable to capture the effects of density or economies of scale at treatment works. Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives.

YKYWWW2 Engineering/Economic

Omitted variable to capture the effects of density or economies of scale at treatment works. Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of

Copyright © United Utilities Water Limited 2018 148

Chapter 7: Supplementary Document - S6002

unitedutilities.com

drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives.

Transparency

Statistical Used in final model suite

YKYWWW3

Engineering/Economic

Omitted variable to capture the effects of density or economies of scale at treatment works. Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives.

YKYWWW4

Engineering/Economic

Omitted variable to capture the effects of density or economies of scale at treatment works. Number of CSO’s are a potential solution to the underlying need (having to deal with increased flow) and result in spilling surface water rather than installing additional capacity to store and then transport it. Including this variable rather than a measure of drainage requirements will not remunerate those companies that seek to store and transport run off and create perverse incentives.

Wholesale Wastewater (plus enhancement) Whilst including enhancement expenditure within botex models may be internally valid in some instances it does not support the external validity as enhancement programmes are dependent upon individual interventions and can often involve major capital projects. Additionally, we do not feel that the inclusion of enhancement expenditure, particularly developer related activities, aligns well with the various incentive mechanisms in place for PR19 or supports an effective cost adjustment process.

Copyright © United Utilities Water Limited 2018 149