THE CYPRUS COMPOSITE LEADING

ECONOMIC INDEX (CCLEI)

DECEMBER 12, 2019

A project funded by Hellenic Bank

Economic Research Centre (ERC) – Department of Economics, University of Cyprus (UCY)

Authors

Professor Elena Andreou, Director, Economic Research Center

Magdalini Tofini, Researcher, Economic Research Center

1 Project funded by Hellenic Bank

The Cyprus Composite Leading Economic Index (CCLEI)

Declined in October 2019 pointing to slow but still expanding

economy

What is a Composite Leading Economic Index (CLEI)? The index that is designed to provide early signals of turning points in business cycles i.e., early evidence of the turns in economic activity. This index comprises of a number of leading economic activity variables which tend to lead changes in the overall economic activity.

What are the components of the CCLEI? The CCLEI is the combination of multiple leading indicators which have been carefully selected from a large number of international and local variables. Currently, the components are the Brent Crude Oil price, the Euro Area Economic Sentiment Indicator, the tourists’ arrivals, the value of visa card transactions, the retail trade sales turnover volume index, the volume index of electricity production, and the number of authorized building permits. The leading properties of these variables will be assessed on a regular basis.

Performance of the Index in October 2019

The Cyprus Composite Leading Economic Index (CCLEI) based on the Aruoba, Diebold, and Scotti (ADS) (2009) model approach (CCLEI_ADS), as presented in the graph below, exhibited a year-on-year decrease of 0.71% in October 2019, following decreases of 0.69% in September, and 0.66% in August, signalling downward pressures on economic growth.

The downward pressures on the CCLEI is mainly due to a reduction in the Euro Area Economic Sentiment Indicator. This reflects the deterioration of the international economic environment due to, among others, trade conflicts and the prolonged uncertainty over the Brexit process. In contrast, the decline in oil prices as well as the positive performance of domestic indicators, in particular, retail sales volume, credit card transactions and tourist arrivals, have contributed to a smoother reduction of the Index.

In summary, recent downward trends in the CCLEI are due to the deteriorating external environment. The negative developments in the external environment, however, in conjunction with the positive performance of domestic variables, are in line with the forecasts of international and domestic organizations for the Cypriot economy, which is expected to continue expanding but at a slower pace.

Figure 1: The Cyprus Composite Leading Economic Index (CCLEI) declined in October 2019

-4

-3

-2

-1

0

1

2

3

04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19

The quarterly standardized YoY GDP growth rateThe monthly standardized YoY CCLEI_ADS growth rate

Peak: Trough:

Dec ’07 Sep ’09

Sep ‘10 Jun ‘13

Stan

dar

diz

ed Y

ear-

ov

er-Y

ear

(Yo

Y)

gro

wth

rat

es

Year -Quarterly standardized YoY GDP growth rate vis-à-vis the monthly standardized YoY CCLEI_ADS growth rate. Shade areas refer to recession periods defined following the CERP Euro Area Business Cycle Dating Committee and the conventional recession definition of at least two consecutive quarters of negative YoY GDP growth rate (2008M01-2009M12 & 2010M10-2014M12). -Source: Economic Research Centre (ERC) - Department of Economics @ University of Cyprus (UCY).

2 Project funded by Hellenic Bank

Business Cycle Phases of the Cyprus Economy

Monitoring and forecasting economic and business cycles is an important task for policymakers and researches. The GDP growth is the key indicator that measures the overall economic performance and business cycle behaviour.

The Business Cycle dating methodology combines two well-known recession definitions. The

first recession definition is applied by the Centre for Economic Policy Research (CERP) Euro

Area Business Cycle Dating Committee, which states that a recession starts just after the

economy reaches a peak and ends when it reaches a trough of activity. The CERP Recession is

defined as a substantial decline in the level of economic activity and it is based on the trough

method used by the FRED to compute NBER Recession periods for the U.S. The second recession

definition is a conventional definition of economic growth downturn which states that the

economy enters a recession when at least two consecutive quarters of negative year-over-year

GDP growth rate are recorded. Combining the above recession definitions, the Cyprus recession

periods are defined as 2008Q1-2009Q4 (24 months) and 2010Q4-2014Q4 (51 months).

Composing a Leading Economic Index for Cyprus

We construct a Composite Leading Economic Index (CLEI) that comprises a number of financial and economic indicators which have been tested for their leading ability. Following the literature (e.g. Massimiliano (2006), Aruoba, Diebold, and Scotti (2013), Stock and Watson (1990) etc.) and taking into account Cyprus’s specific economic characteristics, we have considered numerous indicators reported in Table 1. Using preliminary tests, we focus on a smaller but significant number of components with extensive availability of data to construct the CCLEI. The indicators cover certain categories representing the macroeconomic activity of Cyprus (Appendix) which combine both hard and survey data. We consider the monthly frequency for all indicators, except the Brent Crude Oil (€) – Commodity price which is available at both weekly and monthly frequency and the GDP which is available at low, quarterly frequency. The data for all series are adjusted for seasonal effects and potential outliers. Data sources and seasonal adjustment methodology can be found in the Appendix.

The Composite Leading Economic Indices (CLEIs) can be constructed using either model-based methods (e.g. Aruoba, Diebold and Scotti (ADS) developed in 2009, Stock and Watson (1990), and Massimiliano (2006)) or simple non-parametric approaches (e.g. Conference Board (CB) developed in 1995, and the OECD system of composite leading indices developed in 1970). We apply the two alternative methods to construct the CCLEIs; the ADS and the CB. For both methods, the final CCLEI comprises of the following significant leading variables for the Cyprus economy (Table 2): the Brent Crude Oil (OIL) in euro prices (€), the Euro Area Economic Sentiment Indicator (EAESI), the number of tourists’ arrivals (TOURA), the value of visa card transactions (CARDS), the retail trade sales turnover volume index (RETS), the volume index of electricity production (ELECT), and the number of authorized building permits (BUILD). All components refer to the most recent month they are available for and the dataset avoids any errors inevitably associated with forecasting since only actual data-no forecasts are used.

Key Economic Indicator Release Date Reference Period Gross Domestic Product (GDP) 9/12/2019 3rd Quarter 2019

Ordering Frequency Acronym Description Release Date Reference Period 1 Weekly OIL Brent Crude Oil (€) - Commodity Prices 31/10/2019 October 2019 2 Monthly EAESI Euro Area Economic Sentiment Indicator 30/10/2019 October 2019 3 Monthly TOURA Tourists’ Arrivals 18/11/2019 October 2019 4 Monthly CARDS Value of Visa Card Transactions 10/11/2019 October 2019 5 Monthly RETS Retail Trade, except of motor vehicles Turnover Volume Index 25/11/2019 September 2019 6 Monthly ELECT Volume Index of Electricity Production 28/11/2019 September 2019 7 Monthly BUILD Number of Authorized Building Permits 15/11/2019 August 2019

Table 1: Leading Indices tested

1 Number of Authorized Building Permits

2 Total Number of Registered Contract of Sales

3 Total Local Sales of Cement 4 Residential Property Price Index 5 Volume Index of Manufacturing

Production 6 Volume Index of Electricity

Production 7 Tourists’ Arrivals 8 Tourists’ Revenues 9 Registration of Motor Vehicles 10 Registration of Passenger Saloon

Cars 11 Value of Visa Card Transactions 12 Retail Trade, except of motor

vehicles Turnover Value Index 13 Retail Trade, except of motor

vehicles Turnover Volume Index 14 VAT Receivable 15 Loans to non-MFIs Domestic

Residents 16 New Companies Registration 17 CY Consumer Confidence

Indicator 18 CY Services Confidence Indicator 19 CY Construction Confidence

Indicator 20 CY Retail Confidence Indicator 21 CY Industry Confidence Indicator 22 CY Economic Sentiment Indicator 23

CY Services Business Situation Development over the past 3 months

24

CY Services Expectation of Demand over the next 3 months

25 CY Consumer Financial Situation over the next 12 months

26 Brent Crude Oil (€) - Commodity Prices

27 Euro Area Economic Sentiment Indicator

28 Total Number of People Employed

Source: CyStat, Eurostat, ECFIN, CBC, JCC, GFD, CDLS, DRCORRC.

Table 2: Components of the Cyprus Composite Leading Economic Index (CCLEI)

Source: Economic Research Centre (ERC) - Department of Economics @ University of Cyprus (UCY).

3 Project funded by Hellenic Bank

The Cyprus Composite Leading Economic Index based on the Aruoba, Diebold and Scotti model-based approach (CCLEI_ADS)

The Aruoba, Diebold, and Scotti (ADS) methodology has been used by the Philadelphia Fed for estimating the Business Conditions

Index in the U.S. Economy (Aruoba, Diebold, and Scotti (2009)) using a variety of mixed-frequency stock and flow data which are

available at very high frequencies (e.g. daily and weekly). This methodology assumes that the index is a function of a small-data

dynamic factor model stating that the business cycle is not about any single variable but is about the dynamics and co-movements

of many variables. The model recognizes the ability of the business conditions indicators to arrive at a diversity of frequencies,

encompasses them, and thus allows them to provide unremittingly-updated high frequency information. Moreover, it extracts and

forecasts latent business conditions using linear yet statistically optimal techniques, which are model-based and involve no

approximations. Since this methodology is founded on the basis of a dynamic model, it is necessary to assume a particular ordering

of the variables (Table 2). All seasonally and outlier adjusted variables used in the model are initially converted to annualized

weekly/monthly growth rates except of the Euro Area Economic Sentiment Indicator (EAESI) which is just divided by 100,

something analogous to the “Initial Claims” series used in the original ADS Real Activity Index in 2009. Finally, the Kalman filter and

smoother is used to obtain optimal extractions of the CCLEI_ADS monthly index. More details about the modeling framework of the

ADS approach are provided in the Appendix.

The Cyprus Composite Leading Economic Index based on the Conference Board non-parametric approach (CCLEI_CB)

The Conference Board (CB) methodology was developed in 1995 by the Bureau of Economic Analysis of the U.S. Department of Commerce (BCI Handbook (2001)). The CB Cyprus Composite Leading Economic Index (CCLEI_CB) is constructed by first computing month-to-month changes for each seasonally and outlier adjusted component where the signs of the month-to-month changes for the oil prices series are reversed because of their negative correlation with GDP. Next, monthly contributions of the components are adjusted to equalize the volatility of each component and then are added to obtain the growth rate of the index for each month. Finally, the monthly level of the index is computed using the symmetric percent change formula [200 * (Xt - Xt-1)/(Xt + Xt-1)] which is then rebased to average 100 in the base year (currently 2015). Since this is not a model-based approach, a simple moving average of three periods is used to obtain optimal and smooth extractions of the CCLEI_CB monthly index. The steps are provided with more details in the Appendix. The leading behaviour of the CCLEI and its components

Comparing the resulting indices from the above two methodologies, it seems that the two leading indices have similar cyclical phases with the CCLEI_ADS index having some relatively earlier leading indicator behavior. To determine the statistical relationship between the CCLEI index and the GDP, Pearson’s correlation coefficient test with backward shifts is used (Tkacova, Gavurova, and Behun (2017)). The CCLEI index based on the ADS approach is found to be statistically significant for up to five lags, while the CCLEI index based on the CB approach for up to seven lags. However, Pearson’s coefficient shows the association between the GDP and the CCLEI index for each lag length separately, whereas we want to assess their statistical relationship for the entire period until each specific lag length. Therefore, the following Distributed Lag (DL) models with maximum number of lags up to three years (12 quarters) where the dependent variable is the quarterly standardized (year-over-year) GDP growth rate, and the independent variables are lags of the quarterly standardized (year-over-year) CCLEI growth rate have been estimated:

𝐺𝐷𝑃𝑡 = 𝛼0 + ∑ 𝛾𝑗𝐶𝐶𝐿𝐸𝐼𝑡−𝑗𝑝𝑗=1 + 휀𝑡 , 𝑝 = 1,2, … 12.

The optimal number of lags (𝑝∗ ) for each leading index is based on the Akaike information criterion (AIC) which is an estimator of the relative quality of statistical models for a given set of data. Given a collection of models for the data, AIC estimates the quality of each model, relative to each of the other models and the model with the lowest AIC is preferred. The AIC criterion provides the lowest value for the DL model with five lags and eight lags for the CCLEI_ADS and CCLEI_CB indices respectively. Below, we present the estimated coefficients based on the Distributed Lag (DL) models chosen by the AIC criterion for each CCLEI index. The precondition for cyclical indicators is the position of the lengthiest statistically significant cross-correlation value at time t-1 to t-𝑝∗.

Optimal Distributed Lag model, DL(𝒑∗ ), using lags of the CCLEI_ADS index

t-8 t-7 t-6 t-5 t-4 t-3 t-2 t-1

DL(5) 0.814** -1.193*** 1.282*** -0.850 0.748*

Optimal Distributed Lag model, DL(𝒑∗ ), using lags of the CCLEI_CB index

DL(8) 0.094 0.267*** -0.141 0.312** -0.061 0.256** -0.047 0.573**

-*, **, *** denote statistical significance at 10%, 5%, and 1% respectively based on Bai, J., and S. Ng. (2008) hard thresholding. Cross-correlations values correspond to estimated coefficients from a Distributed Lag model with Andrews Automatic robust standard errors. -Source: Economic Research Centre (ERC) - Department of Economics @ University of Cyprus (UCY).

Table 3: Estimated coefficients of the CCLEIs using the Distributed Lag (DL) models chosen by the AIC criterion

4 Project funded by Hellenic Bank

The table shows that the CCLEI index based on the Aruoba, Diebold and Scotti method (CCLEI_ADS) is able to predict five quarters in advance an upcoming recession with statistically significant cross-correlation value 0.814, whereas the CCLEI index based on the Conference Board method (CCLEI_CB) is able to predict seven quarters in advance an upcoming recession with statistically significant cross-correlation value 0.267. The results are in line with the Pearson correlation coefficient results which also showed predictive abilities of five and seven quarters for the two indices respectively. Although the CCLEI_CB index seems to provide earlier signs of turning points in the economy, the impact of it is not as big as of the CCLEI_ADS index. The impact of the CCLEI_ADS index on the GDP is equal to 0.814 in the fifth lag which is almost three times bigger than the impact of the CCLEI_CB index (0.267) in the seventh lag. Moreover, by considering the precondition for cyclical indicators as the position of the highest statistically significant cross-correlation value at time t-1 to t-𝑝∗, the prediction abilities for the CCLEI_ADS and CCLEI_CB indices would have been three and one quarters respectively with cross-correlation values 1.282 and 0.573. Therefore, the CCLEI index based on the ADS method is a relatively earlier and more reliable leading indicator of turning points in the economy.

Furthermore, to check the robustness of our results, additional models that include lags of the GDP growth rate have been estimated. The objective of these additional estimations is to assess whether the prediction abilities of our constructed CCLEI index remain robust when previous values of the GDP growth rate are added in the model. Therefore, the following Autoregressive Distributed Lag (ADL) models with maximum number of lags up to three years (12 quarters) where the dependent variable is the quarterly standardized (year-over-year) GDP growth rate, and the independent variables are lags of the quarterly standardized (year-over-year) GDP growth rate and the quarterly standardized (year-over-year) CCLEI growth rate have been estimated:

𝐺𝐷𝑃𝑡 = 𝛼0 + ∑ 𝛽𝑗𝐺𝐷𝑃𝑡−𝑖𝑝𝑖=1 + ∑ 𝛾𝑗𝐶𝐶𝐿𝐸𝐼𝑡−𝑗

𝑝𝑗=1 + 휀𝑡, 𝑝 = 1,2, … 12.

For both the Distributed Lag (DL) model and the Autoregressive Distributed Lag (ADL) model, we specify a quadratic spectral kernel based HAC covariance estimation using prewhitened residuals. The kernel bandwidth is determined automatically using the Andrews AR(1) method. The optimal number of lags , 𝑝∗, is once more based on the Akaike information criterion (AIC) which provides the lowest value for the ADL model with six lagging values when lags of the CCLEI_ADS index are used and ten lagging values when lags of the CCLEI_CB index are used. By regressing the ADL(𝑝∗) models, the estimated coefficients of the Composite Leading Economic Indices (CLEIs) found to be statistically significant in the DL(𝑝∗) models remain statistically significant proving the robustness of our results. Comparing the monthly observations of the indices, the advantage of the CCLEI_ADS index of having some relatively earlier leading indicator behavior than the CCLEI_CB index is even more observable (Table 4). Regarding the global financial crisis of 2008-2009, although both indices reached bottom in March 2009, the CCLEI_ADS index reached a peak in July 2006, whereas the CCLEI_CB index reached a peak in August 2007. Following the business cycle methodology mentioned earlier, the first recession period (2008M01-2009M12) according to the CCLEI_ADS index would have been 2006M08-2009M10, whereas according to the CCLEI_CB index, it would have been 2007M09-2009M11. Similarly, the second recession period (2010M10-2014M12) would have been 2010M05-2013M06 and 2010M06-2013M09 according to the CCLEI_ADS index and CCLEI_CB index respectively. Moreover, Table 4 shows the ability of the CCLEI_ADS index to reach a peak during earlier monthly or quarterly observations than the CCLEI_CB index. More precisely, the CCLEI_ADS index reaches its peak during the first monthly observation of each quarter. Regarding the first recession period (2008M01-2009M12), it reached its peak in the first month of the third quarter in 2006 (July 2006) and regarding the second recession period (2010M10-2014M12), it reached its peak during the first month of the second quarter in 2010 (April 2010). On the other hand, regarding the first recession period (2008M01-2009M12), the CCLEI_CB index reached its peak in the second month of the third quarter in 2007 (August 2007) and regarding the second recession period (2010M10-2014M12), it reached its peak during the second month of the second quarter in 2010 (May 2010) which shows the ability of the CCLEI_ADS index to be a relatively earlier leading indicator of turning points in the economy.

First Recession period (2008M01-2009M12) based on the CCLEIs Recession Period Peak Trough CCLEI_ADS 2006M08-2009M10 2006M07 2009M03 CCLEI_CB 2007M09-2009M11 2007M08 2009M03 Second Recession period (2010M10-2014M12) based on the CCLEIs Recession Period Peak Trough CCLEI_ADS 2010M05-2013M06 2010M04 2012M06 CCLEI_CB 2010M06-2013M09 2010M05 2013M03

Table 4: Recession periods and turning points based on the Cyprus Composite Leading Economic Index (CCLEI)

Source: Economic Research Centre (ERC) - Department of Economics @ University of Cyprus (UCY).

5 Project funded by Hellenic Bank

Turning points based on individual components of the CCLEI

Turning points determine the time at which the economy turns from recession to recovery or from growth to recession. Many tools

have been developed for computing the turning points in economic activity, such as the Diffusion Indices (e.g. Stock and Watson

(1998)). Diffusion indices measure the proportion of the component series that contribute positively to the index. Following the

Conference Board methodology (BCI Handbook (2001)), components that rise more than 0.05 percent are given a value of 1,

components that change less than 0.05 percent are given a value of 0.5, and components that fall more than 0.05 percent are given

a value of 0. The corresponding monthly changes are changes calculated by comparing months with the same months of the previous

year. The components of our constructed Composite Leading Economic Index contribute all positively to the GDP growth rate except

of the Brent Crude Oil (€) price series and thus opposite values are assigned to them. For this reason, a value of “1” is assigned to

this series in the diffusion index computation instead of “0” when the price declines for a specific month. The Diffusion Index is

finally computed as the average of the values of the components for each month, multiplied by 100.

The Diffusion index complements the turning points methodology by focusing on the behaviour of the individual

components/indicators that comprise the CCLEI index. Following the diffusion index methodology, if the index is above 50, then

this is a sign that the economy is probably expanding, or at least moving in that direction and if the index is below 50, then this

suggests that the economy is probably in a recession, or at least moving in that direction. The Diffusion Index calculated based on

changes of the components of our CCLEI reached a peak in June 2007 (100, i.e. all components contributed positively to the index)

and it was below 50 during the last months of 2007 signalling downward pressures of economic growth and thus portending the

beginning of the Global Financial Crisis of 2008-2009. Moreover, after it hit bottom in September 2009, it started rising gradually

indicating that the economy would emerge from the crisis soon (Fig. 2). Similar conclusions can be derived from the most recent

financial crisis. These results show that the Diffusion Index computed based on year-over-year monthly changes of the components

of our CCLEI can consistently determine turning points in the economy.

The Diffusion index calculated based on the components of our constructed CCLEI increased during the most recent months. This is a result of the decline in oil prices as well as the positive performance of domestic indicators, in particular, retail sales volume, credit card transactions and tourist arrivals. In contrast, the Euro Area Economic Sentiment Indicator declined, reflecting the deterioration of the international economic environment due to, among others, trade conflicts and the prolonged uncertainty over the Brexit process.

Although the Diffusion Index increased during the most recent months, the CCLEI has exerted reductions since November 2018. Therefore, the negative developments in the external environment in conjunction with the positive performance of domestic variables, are in line with the forecasts of international and domestic organizations for the Cypriot economy, which is expected to continue expanding but at a slower pace.

Year

Figure 2: The Diffusion Index Level (and its 3-period moving average) based on YoY monthly changes of the CCLEI’s components

-Shaded areas refer to recession periods determined following the CERP Euro Area Business Cycle Dating Committee and the conventional recession

definition of two consecutive quarters of negative YoY GDP growth rate (2008M01-2009M12 & 2010M10-2014M12).

-Source: Economic Research Centre (ERC) - Department of Economics @ University of Cyprus (UCY).

Peak: Trough:

Dec ’07 Sep ’09

Sep ’10 Jun ‘13

0

20

40

60

80

100

120

04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19

Dif

fusi

on

Ind

ex

Leve

l

6 Project funded by Hellenic Bank

ACKNOWLEDGMENTS

The authors benefited from discussions with Andreas Assiotis, Maria Demetriadou, Niki Demosthenous, Tom Stark, and George

Syrichas.

REFERENCES

Andrews, Donald W.K. 1993. Tests for Parameter Instability and Structural Change with Unknown Change Point. Econometrica 61: 821-856.

Aruoba, S. B., and C. Sarikaya. 2013. A Real Economic Activity Indicator for Turkey. Central Bank of the Republic of Turkey 13: 15-29.

Aruoba, S. B., F. X. Diebold, and C. Scotti. 2009. Real-Time Measurement of Business Conditions. Journal of Business & Economic Statistics 27: 417-427.

Bai, J., and S. Ng. 2008. Forecasting economic time series using targeted predictors. Journal of Econometrics 146: 304-317.

Federal Reserve Bank of Philadelphia Official Website. Retrieved from http: https://www.philadelphiafed.org/research-and-

data/real-time-center/business-conditions-index.

Garnitz, J., R. Lehmann, and K. Wohlrabe. 2019. Forecasting GDP all over the world using leading indicators based on

comprehensive survey data. Journal of Applied Economics 51: 5802-5816.

Massimiliano, M. 2006. “Leading Indicators”. In G. Elliott, C.W.J. Granger, and A. Timmermann (eds.), Handbook of Economic

Forecasting, vol. 1. Elsevier B.V. Publishing.

McGuckin, R.H., Ozyildirim A. and Zarnowitz V. 2007. A More Timely and Useful Index of Leading Indicators. Journal of Business &

Economic Statistics 25: 110-120.

Phillips, K. R, L. Vargas, and V. Zarnowitz. 1996. New tools for analyzing the Mexican economy: indexes of coincident and leading

economic indicators. Economic and Financial Policy Review issue Q II, Federal Reserve Bank of Dallas.

Stock, J.H., and M.W. Watson. 1990. New Indexes of Coincident and Leading Economic Indicators. NBER Working Paper No. R1380.

Stock, J.H., and M.W. Watson. 1998. Diffusion Indexes. NBER Working Paper No. w6702.

Stock, J.H., and M.W. Watson. 2002. Macroeconomic Forecasting Using Diffusion Indexes. Journal of Business & Economic Statistics

20: 147-162.

The Conference Board. 2001. Business Cycle Indicators (BCI) Handbook.

The Conference Board Official Website. Calculating the Composite Indexes. Retrieved from http: https://www.conference-

board.org/data/bci/index.cfm?id=2154.

The Conference Board Official Website. How to Compute Diffusion Indexes. Retrieved from http: https://www.conference-

board.org/data/bci/index.cfm?id=2180.

Tkacova, A., B. Gavurova, and M. Behun. 2017. The Composite Leading Indicator for German Business Cycle. Journal of

Competitiveness 9: 114-133.

7 Project funded by Hellenic Bank

APPENDIX

Table: Description of Economic Leading Indicators

Category 1: Housing and Building Frequency Acronym Description Data Source 1 Monthly BUILD Number of Authorized Building Permits CyStat

2 Monthly POL Total Number of Registered Contract of Sales Cyprus Department of lands and surveys (CDLS)

3 Monthly CEM Total Local Sales of Cement CyStat

4 Quarterly HOUS Residential Property Price Index Central Bank of Cyprus (CBC) Category 2: Manufacturing and Electricity

1 Monthly MANUF Volume Index of Manufacturing Production Eurostat

2 Monthly ELECT Volume Index of Electricity Production Eurostat

Category 3: Tourists 1 Monthly TOURA Tourists’ Arrivals CyStat

2 Monthly TOURR Tourists’ Revenues CyStat

Category 4: Consumption and Trade 1 Monthly MOTOR Registration of Motor Vehicles CyStat

2 Monthly SALOON Registration of Passenger Saloon Cars CyStat

3 Monthly CARDS Value of Visa Card Transactions (sum of purchases of Cypriots in Cyprus, purchases of Cypriots abroad, and purchases of tourists in Cyprus)

JCC

4 Monthly RETS Retail trade, except of motor vehicles Turnover Value Index Eurostat, CyStat 5 Monthly RETS Retail trade, except of motor vehicles Turnover Volume Index Eurostat, CyStat 6 Quarterly VAT VAT Receivable CyStat

Category 5: Loans in new companies 1 Monthly LOANS Loans to non-MFIs Domestic Residents CBC

2 Monthly COMP Registration of New Companies Department of the Registrar of Companies and Official Receiver of the Republic of Cyprus (DRCORRC)

Category 6: Survey Indicators 1 Monthly Cons_CI CY Consumer Confidence Indicator European Commission (ECFIN) 2 Monthly Serv_CI CY Services Confidence Indicator ECFIN 3 Monthly Const_CI CY Construction Confidence Indicator ECFIN 4 Monthly Retail_CI CY Retail Confidence Indicator ECFIN 5 Monthly Ind_CI CY Industry Confidence Indicator ECFIN 6 Monthly ESI_CI CY Economic Sentiment Indicator ECFIN 7 Monthly ServBSD CY Services Business Situation Development over the past 3 months ECFIN 8 Monthly ServEOD CY Services Expectation of Demand over the next 3 months ECFIN 9 Monthly ConsFS CY Consumer Financial Situation over the next 12 months ECFIN

Category 7: Foreign Indicators 1 Weekly OIL Brent Crude Oil (€) - Commodity Prices Global Financial Database (GFD) 2 Monthly EAESI Euro Area Economic Sentiment Indicator ECFIN

Category 8: Macroeconomic Indicators 1 Quarterly GDP Gross Domestic Product CyStat

2 Quarterly EMP Total Number of People Employed CyStat

8 Project funded by Hellenic Bank

Seasonal Adjustment

For the seasonal adjustment of the series, three approaches have been considered; the seasonal dummy approach, the X-13

approach using the X-11 ARIMA method and the X-13 approach using the TRAMO/SEATS ARIMA method. All three approaches that

this report applies, provide similar results with highly correlated seasonally adjusted series.

The Aruoba, Diebold, and Scotti (ADS) Model

The ADS approach of constructing composite indices is based on a dynamic factor model of stock and flow variables at very high frequency, i.e. daily. Let 𝑥𝑡 denote underlying business conditions at day t, which evolve daily with AR(p) dynamics:

𝑥𝑡 = 𝛼1𝑥𝑡−1 + 𝛼2𝑥𝑡−2 +⋯+ 𝛼𝜌𝑥𝑡−𝜌 + 𝑒𝑡,

where, 𝑒𝑡 is a white noise innovation with unit variance, and 𝑥𝑡 is a scalar.

Let 𝑦𝑖𝑡 denote the ith daily economic or financial variable at day t, which depends linearly on 𝑥𝑡 and possibly also on various

exogenous variables and lags of 𝑦𝑖𝑡:

𝑦𝑖𝑡= 𝑐𝑖 + 𝛽𝑖𝑥𝑡 + 𝛿𝑖1𝑤

1𝑡 +⋯+ 𝛿𝑖𝑘𝑤

𝑘𝑡 + 𝛾𝑖1𝑦

𝑖𝑡−𝐷𝑖

+⋯+ 𝛾𝑖𝑛𝑦𝑖𝑡−𝑛𝐷𝑖

+ 𝑢𝑖𝑡,

where the 𝑤𝑡 are exogenous variables and the 𝑢𝑖𝑡 are contemporaneously and serially uncorrelated innovations. The lags of the dependent variable 𝑦𝑖

𝑡 are introduced in multiples of 𝐷𝑖 , where 𝐷𝑖 > 1 is a number linked to the frequency of the observed 𝑦𝑖

𝑡.

However, because most variables, although evolving daily, are not actually observed daily, let �̃�𝑖𝑡 denote the same variable observed

at a lower frequency (call it the “tilde frequency”). The relationship between �̃�𝑖𝑡 and 𝑦𝑖

𝑡 depends crucially on whether 𝑦𝑖

𝑡 a stock

or flow variable. If 𝑦𝑖𝑡 is a stock variable measured at a nondaily tilde frequency, then the appropriate treatment is straightforward,

because stock variables are simply point-in-time snapshots. At any time t, either 𝑦𝑖𝑡 is observed, in which case �̃�𝑖

𝑡 =𝑦𝑖

𝑡, or it is not,

in which case �̃�𝑖𝑡 =NA, where NA denotes missing data (“not available”). Hence the stock variable measurement equation is:

�̃�𝑖𝑡= {

𝑐𝑖 + 𝛽𝑖𝑥𝑡 + 𝛿𝑖1𝑤1𝑡 +⋯+ 𝛿𝑖𝑘𝑤

𝑘𝑡 + 𝛾𝑖1𝑦

𝑖𝑡−𝐷𝑖

+⋯+ 𝛾𝑖𝑛𝑦𝑖𝑡−𝑛𝐷𝑖

+ 𝑢𝑖𝑡 , 𝑖𝑓 𝑦𝑖𝑡 𝑖𝑠 𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑

𝑁𝐴 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒. }

Now consider flow variables. Flow variables observed at nondaily tilde frequencies are intraperiod sums of the corresponding daily values,

�̃�𝑖𝑡= {

∑𝑦𝑖𝑡−𝑗

𝐷𝑖

𝑗=0

, 𝑖𝑓 𝑦𝑖𝑡 𝑖𝑠 𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑

𝑁𝐴 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒,

}

where, 𝐷𝑖 is the number of days per observational period (e.g., 𝐷𝑖 = 7 if 𝑦𝑖𝑡 is measured weekly). Combining this fact with Equation

(2), the flow variable measurement equation is:

�̃�𝑖𝑡=

{

∑ 𝑐𝑖

𝐷𝑖−1

𝑗=0

+ 𝛽𝑖 ∑ 𝑥𝑖𝑡−𝑗

𝐷𝑖−1

𝑗=0

+ 𝛿𝑖1 ∑ 𝑤1𝑡−𝑗

𝐷𝑖−1

𝑗=0

+ 𝛿𝑖𝑘 ∑ 𝑤𝑘𝑡−𝑗

𝐷𝑖−1

𝑗=0

+

𝛾𝑖1 ∑ 𝑦𝑖𝑡−𝐷𝑖−j

𝐷𝑖−1

𝑗=0

+ 𝛾𝑖𝑛 ∑ 𝑦𝑖𝑡−𝑛𝐷𝑖−j

𝐷𝑖−1

𝑗=0

+ 𝑢∗𝑖𝑡, 𝑖𝑓 𝑦𝑖𝑡 𝑖𝑠 𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑

𝑁𝐴 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒, }

where, ∑ 𝑦𝑖𝑡−𝐷𝑖−j

𝐷𝑖−1𝑗=0 is by definition the observed flow variable one period ago (�̃�𝑖

𝑡−𝐷𝑖), and 𝑢∗𝑖𝑡 is the sum of the 𝑢𝑖𝑡 over the tilde

period.

Note that in general 𝐷𝑖 is time varying, as, for example, some months have 28 days, some have 29, some have 30, and some have 31. To simplify the notation above, 𝐷𝑖 is assumed to be fixed. Additionally, note that although 𝑢∗𝑖𝑡 follows a moving average process of

(1)

(2)

(3)

(4)

(5)

9 Project funded by Hellenic Bank

order 𝐷𝑖−1, at the daily frequency, it nevertheless remains white noise when observed at the tilde frequency, due to the ( 𝐷𝑖−1)-dependence of an MA( 𝐷𝑖−1) process. Hence 𝑢∗𝑖𝑡 is appropriately treated as white noise in what follows, where var(𝑢∗𝑖𝑡)= 𝐷𝑖·var(𝑢𝑖𝑡).

The exogenous variables 𝑤𝑡 are the key to handling trend. In particular, in the important special case where the 𝑤𝑡 are simply deterministic polynomial trend terms [𝑤1

𝑡−𝑗 = 𝑡 − 𝑗, 𝑤2𝑡−𝑗 = (𝑡 − 𝑗)

2 and so on] we have that

∑ [𝑐𝑖 + 𝛿𝑖1(𝑡 − 𝑗) + ⋯+ 𝛿𝑖𝑘(𝑡 − 𝑗)𝑘] ≡ 𝑐∗𝑖

𝐷𝑖−1𝑗=0 + 𝛿∗𝑖1𝑡 + ⋯+ 𝛿

∗𝑖𝑘𝑡

𝑘.

Assembling the results, the stock variable measurement equation is

�̃�𝑖𝑡= {

𝑐∗𝑖 + 𝛽𝑖𝑥𝑖𝑡 + 𝛿

∗𝑖1𝑡 +⋯+ 𝛿

∗𝑖𝑘𝑡

𝑘 + 𝛾𝑖1�̃�𝑖𝑡−𝐷𝑖

+⋯+ 𝛾𝑖𝑛�̃�𝑖𝑡−𝑛𝐷𝑖

+ 𝑢∗𝑖𝑡, 𝑖𝑓 𝑦𝑖𝑡 𝑖𝑠 𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑

𝑁𝐴 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒, }

and the flow variable measurement equation,

�̃�𝑖𝑡=

{

𝑐∗𝑖 + 𝛽𝑖 ∑ 𝑥𝑖𝑡−𝑗

𝐷𝑖−1

𝑗=0

+ 𝛿∗𝑖1𝑡 + ⋯+ 𝛿∗𝑖𝑘𝑡

𝑘 +

𝛾𝑖1�̃�𝑖𝑡−𝐷𝑖

+⋯+ 𝛾𝑖𝑛�̃�𝑖𝑡−𝑛𝐷𝑖

+ 𝑢∗𝑖𝑡, 𝑖𝑓 𝑦𝑖𝑡 𝑖𝑠 𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑

𝑁𝐴 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒, }

which completes the specification of the model and has a natural state-space form.

The Conference Board (CB) Approach of constructing Leading Economic Indices

The Conference Board procedure for calculating the composite indexes has five distinct steps:

1) Month-to-month changes are computed for each component. If the component X is in percent change form or an interest rate, simple arithmetic differences are calculated: x t=X t - X t-1. If the component is not in percent change form, a symmetric alternative to the conventional percent change formula is used: x t = 200 * (X t - X t-1) / (X t + X t-1), and if the component X is a diffusion index or an interest rate spread the monthly level is used x t=X t. The conventional percent change formula 200 * (Xt - Xt-1)/(Xt + Xt-1), treats positive and negative changes symmetrically. When it shows a one percent increase followed by a one percent decrease, the level of X has returned to its original value. This is not true with the more conventional formula, 100 * (Xt - Xt-1)/Xt-1, in which the same percent increase and decrease would leave X at slightly lower value. Both formulas, as well as a third, increasingly popular alternative based on logarithmic differences, produce very similar cyclical patterns.

2) The monthly contributions of the components are adjusted to equalize the volatility of each component. Standard deviations vx of the changes in each component are computed. These statistical measures of volatility are inverted (wx = 1/vx), their sum is calculated (k={sum over [x]} wx), and they are restated so the index's component standardization factors sum to one (rx=(1/k) * wx). The adjusted contribution in each component is the monthly contribution multiplied by the corresponding component standardization factor (mt = rx * xt).

3) Add the adjusted monthly contribution across the components for each month to obtain the growth rate of the index. This step results in the sum of the adjusted contributions (it={sum over [x]} mx,t) which is the monthly growth rate of the index.

4) The level of the index is computed using the symmetric percent change formula. The index is calculated recursively starting from an initial value of 100 for the first month of the sample period (i.e. February 2000). The first month's value is I1=100. The second month's value I2 = I1 * (200+i2`)/(200-i2`) and this formula is used recursively to compute the index levels for each month that data are available.

5) The index is rebased to average 100 in the base year. The history of the index is multiplied by 100 and divided by the average for the twelve months of the based year, currently 2015.

(6)

(8)

(7)