Embed Size (px)
Citation preview
Global Economic Damages from Climate Changeand the Gains from Complying with the Paris
Accord
Tom Kompas
Centre of Excellence for Biosecurity Risk Analysis
School of Biosciences
School of Ecosystem and Forest Sciences
University of Melbourne
October 10, 2018
Introduction: Dimension Matters!
I It is not unusual to find numerical simulations that aredimensionally very large – and often impossible to solve –especially in studies that detail environmental impacts andclimate modelling or where optimization routines are required(the so-called ‘curse of dimensionality’).
I Objective: Solve a VERY large Intertemporal GTAP TradeModel to determine the local and global impacts of climatechange at various temperature levels, along with the benefitsof complying with the Paris Accord (at the 2oC target).
I Methodology: We employ an innovative approach to solving(otherwise unsolvable) large scale systems through the use ofparallel processing methods and an ordering of variables andequations in a ‘Nested Doubly Bordered Block Diagonal’(NDBBD) form.
Common Defects of Current CGE/GTAP Approaches
1. Static and comparative static modelling approaches: largerscale, but not dynamic or intertemporal.
2. Recursive CGE models: intertemporal with static and adaptiveexpectations, but not forward-looking and non-optimal.
3. Intertemporal models: Small in dimension, high degree ofaggregation, absence of sufficient regional or local effects.Static GGE models with less than 10 regions and 30commodities are common. The largest fully intertemporalCGE model has 10 regions/countries and 12 commodities, andmany climate models are even smaller (e.g., DICE model)
4. Bottom-line: Economists have largely and often badlyunderestimated the damages from climate change.
5. Standard climate change economic models are either verysmall dimensionally and/or are not forward-looking orintertemporal (Nordhaus, 1991, DICE model; Roson andMensbrugghe, 2012, EVISAGE model; Tol, 2002, 2012)
GTAP Model Database/Dimension
I Example: GTAP database v7: The full database includes 112(extendable to 139) regions (countries) and 57 commodities.
Table: GTAP model with different database/aggregation levels.
IDModel Size
Number ofendogenous
variables
Number ofexogenousvariables
Number ofnon-zeros
1 112 regions, 3 commodities,47 periods
13939336 7850081 58681307
2 112 regions, 4 commodities,47 periods
18592712 10450826 78830677
3 112 regions, 5 commodities,47 periods
23319784 13083155 97089894
4 112 regions, 26 commodities,47 periods
139612072 75657968 547479803
5 112 regions, 34 commodities,47 periods
192462632 103159736 749419439
6 112 regions, 57 commodities,47 periods∗
505112836 260697767 1.815*109
Source: Authors’ calculation. Note: ∗ Number of non-zeros is an approximation.
Current Solution Approaches
1. Even modern software packages such as GEMPACK or GAMS,which use a serial direct LU solver from HSL (2013) orLUSOL (Saunders et al., 2013), become incapable of solvingintertemporal CGE models of even a modest size.
2. Dixon et al. (2005) subsequently introduced a rationalexpectations approach, nevertheless it can be used to solvedynamic CGE models for only a handful of periods in thepractice and its numerical stability is also in doubt.
3. McKibbin (1987); McKibbin and Sachs (1991) propose anMSG algorithm to solve large scale intertemporal CGEmodels, however, the method is based only on a first-orderapproximation, it lacks of portability, and is dimensionallylimited.
Pre-ordered and Post-ordered NDBBD Matrices
Figure: The matrix withoutordering.
Source: Author’s calculation.
Figure: The reordered nestedmatrix.
Source: Authors’ calculation.The plot function is from PETSC (Balay et al., 1997, 2014, 2013).
Solution Time
Table: Calculation time in sec.
IDSBBD
reorderingNDBBD
reorderingPerformance
ratio
1 46.133858 115.809060 39.84%2 85.192744 149.345154 57.04%3 116.959756 205.926200 56.8%4 3043.344883 3984.619880 76.38%5 5352.380219 7500.384889 71.36%6 - 26934.744446 -
Source: Authors’ calculation.
The Effects of Climate Change on Growth and GDP
I Using Roson and Satori (2016) climate change damage (+/-)functions, we solve a large dimensional intertemporal GTAPtrade model to account for the some of the effects of globalwarming (e.g., loss in agricultural and labour productivity, sealevel rise, and human health effects) for 140 countries, bydecade and over the long term – where producers look forwardand adjust price expectations and capital stocks to accountfor future climate effects.
I Results are generated in terms of losses in GDP for eachcountry and by various RCPs and associated, examining the3oC and 4oC scenarios, with overall results for 1-4oC (SSP2‘business as usual’ baseline).
The Effects of Climate Change on Growth and GDP
I 57 commodity groups (with trade and spatial dimension),including paddy rice, wheat, cereal grains, vegetable, fruitsand nuts, bovine cattle, sheep, goats, horses, sugar cane, milk,wool, forestry, fishing, coal, oil, gas, meat products, vegetableoils and fats, dairy products, textiles, beverages and tobacco,wood products, paper products, chemical, rubber, leatherproducts, plastics, metal products, electronic equipment,machinery, manufactures, air transport, motor vehicles,electricity, construction, business services, defense, publicadministration, dwellings, communication, financial services,construction, transport, recreational and other services, etc.
I The goal (again) is to find total damages from climate changeand the economic gains from complying with the ParisClimate Accord.
Climate Change Impacts - 10 years; 3oC Path, %∆ GDP
Source: Authors’ calculation.
Climate Change Impacts - 20 years; 3oC Path, %∆ GDP
Source: Authors’ calculation.
Climate Change Impacts - 50 years; 3oC Path, %∆ GDP
Source: Authors’ calculation.
Climate Change Impacts – Long Run; 3oC Path, %∆ GDP
Source: Authors’ calculation.
Top Ten ‘Large and Small Effects’ (140 country databasewith no tourism and energy demand shocks); 3oC Path
Table: Dimension matters: Climate change impacts on world GDP in thelong run: intertemporal vs first-order/recursive estimation (% change).
Countries/regions GTAP-INT∗
Firstorder∗∗
Countries/regions GTAP-INT∗
Firstorder∗∗
Ghana -17.57 -13.19 Switzerland -0.35 1.42Cote dIvoire -17.53 -13.03 Sweden -0.35 1.72Nigeria -15.72 -13.93 Finland -0.25 1.48Rest of Western Africa -15.57 -6.07 Ukraine -0.24 0.92Philippines -14.80 -7.42 Belarus -0.25 0.18Indonesia -13.27 -6.80 Canada -0.22 1.27Senegal -13.00 -9.58 Rest of Europe -0.21 0.01India -10.35 -6.24 Slovakia - 0.47 1.23Rest of Southeast Asia -12.92 -9.83 USA -0.62 -0.18
Source: ∗ Authors’ calculation. ∗∗ From Roson, R. and Sartori, M. (2016), ‘Estimation of climate change
damage functions for 140 regions in the GTAP 9 database’, Journal of Global Economic Analysis, 1(2), 78–115.
Estimation of long term GDP loss per year from 2100under global warming scenarios ($US billion/year)
4oC 3oC 2oCWorld Total -23,149 -9,593 -5,659Sub-Saharan Africa -8,073.68 -2,889.66 -1,927.78India -4,484.96 -2,070.06 -1,149.36Southeast Asia -4,158.88 -2,073.09 -1,166.23China -1,716.91 -701.75 -394.59Latin America -1,371.81 -576.65 -259.82Rest of South Asia -1,157.92 -469.98 -283.78Middle East and North Africa -1,032.27 -451.96 -241.12United States of America -697.77 -223.83 -168.48Japan -253.18 -54.43 -23.02Mexico -127.70 -55.79 -20.88Australia -117.42 -36.87 -23.72South Korea -81.44 -14.72 -7.86Rest of Oceania -39.65 -14.97 -6.96Russian Federation -24.49 -10.88 -6.53United Kingdom 17.78 4.06 0.35Germany 23.85 5.38 2.46France 26.92 7.11 1.80
What do these BIG Numbers Mean?
I Global long term economic damages in 2100 (albeit withlimited damage functions) at 3oC are $US 9.5+ trillion peryear and at 4oC losses are $US 23+ trillion per year.
I $US 23+ trillion is (at least) 3 or 4 GFCs (2008) each year.
I Some country losses are especially severe. GDP losses, forexample, at 4oC , for Cambodia, Sri Lanka, and Nicaragua areover 17%, for Indonesia 19%, for India 14%, Thailand 17%,Singapore 16%, the Philippines 20%, and for much of Africathe losses range from 18 to over 26% of GDP. Global losses inGDP during the Great Depression (1930s) were 15%. (China4.6%, USA 0.9%)
I At 4oC , damages per person in Australia are projected to beAUD$4,886 or roughly AUD$13,945 per household, per year,every year, in the long run.
Gains from Complying with the Paris Accord
I The successful achievement of the Paris Accord, which aimsto keep global warming at roughly 2oC (or RCP4.5), or less,allows us to calculate the potential benefit of the Accord asthe difference in losses between various temperature changesand 2oC scenario.
I For example, comparing the 3oC and 2oC scenario: Thepotential gains of following the Paris Climate Accord foravoided global GDP losses are $US 3,934 trillion a year in thelong term.
I For the comparative case of RCP 8.5 (4oC ), the global gainsfrom complying with the 2oC target (RCP 4.5), areapproximately $US 17,489 trillion a year over the long term.
Cumulative GDP (Income Level) Impacts at 4oC
Figure: Cumulative GDP impacts at 4oC.
Sources: Authors’ calculation.
Cumulative Losses in GDP from 2017-2100 (bill. USD)
Impacts (GDP)4oc 3oc 2oc
World Total -604460.42 -271250.18 -171745.14Sub-Saharan Africa -177398.70 -67745.57 -49231.04India -131574.85 -65495.65 -39665.53Southeast Asia -118076.85 -62233.61 -37692.25China -64024.08 -28239.51 -16947.87Latin America -39444.52 -17240.66 -8529.39Rest of South Asia -29243.05 -11482.45 -8357.61Middle East and North Africa -25582.51 -12400.73 -7021.93United States of America -14401.80 -5699.37 -4334.33Japan -6625.19 -1716.01 -624.83Mexico -3133.90 -1289.18 -486.12Australia -2898.86 -1097.39 -695.97
Source: Authors’ calculation.
Distributional Effects of Climate Change at 4oC
Figure: Climate change impacts by country against income andimpact/damages (circle area = country population size).
Source: Authors’ calculation.
Final Thoughts (1)
I Model dimension matters: Averaging across countries andextremes in impacts distorts overall and country-specificdamages; severe damages occur even though standard damagefunctions are very limited in scope and impact (e.g., severeweather effects and ‘natural disasters’ that are climate changeinduced are excluded).
I The severe falls in GDP in the long term will put manygovernments in fiscal stress. Tax revenues will fall dramaticallyand increases in the frequency and severity of weather eventsand other natural disasters, which invoke significantemergency management responses and expenditures, indicatethat pressure on government budgets will be especially severe.
Final Thoughts (2)
I ’Three Great Myths in the Economics of Climate Change’:
1. Total damages, long term and cumulative, are relatively smallboth in income levels and as a percentage of GDP. The severedamages in the Stern Report are mainly due to the lowdiscount rate. (In our work the discount rate is standard at6%)
2. Poorer countries are not differentially affected, except perhapsat extreme parameter values and, in any case, they can bemore than compensated by countries on high emissionspathways with BAU.
3. The costs of emissions reduction far outweigh the relativedamages from climate change.
References
Balay, S., Adams, M. F., Brown, J., Brune, P., Buschelman, K., Eijkhout, V., Gropp, W. D., Kaushik, D., Knepley, M. G.,McInnes, L. C., Rupp, K., Smith, B. F. and Zhang, H. (2013), PETSc users manual, Technical Report ANL-95/11 -Revision 3.4, Argonne National Laboratory.URL: http://www.mcs.anl.gov/petsc
Balay, S., Adams, M. F., Brown, J., Brune, P., Buschelman, K., Eijkhout, V., Gropp, W. D., Kaushik, D., Knepley, M. G.,McInnes, L. C., Rupp, K., Smith, B. F. and Zhang, H. (2014), ‘PETSc Web page’, http://www.mcs.anl.gov/petsc.URL: http://www.mcs.anl.gov/petsc
Balay, S., Gropp, W. D., McInnes, L. C. and Smith, B. F. (1997), Efficient management of parallelism in object orientednumerical software libraries, in E. Arge, A. M. Bruaset and H. P. Langtangen, eds, ‘Modern Software Tools in ScientificComputing’, Birkhauser Press, pp. 163–202.
Dixon, P. B., Pearson, K., Picton, M. R. and Rimmer, M. T. (2005), ‘Rational expectations for large CGE models: Apractical algorithm and a policy application’, Economic Modelling 22(6), 1001 – 1019.URL: http://www.sciencedirect.com/science/article/pii/S0264999305000490
HSL (2013), ‘A collection of fortran codes for large scale scientific computation’, http://www.hsl.rl.ac.uk.
McKibbin, W. J. (1987), Numerical Solution of Rational Expectations Models With and Without Strategic Behaviour,RBA Research Discussion Papers rdp8706, Reserve Bank of Australia.
McKibbin, W. J. and Sachs, J. D. (1991), Global linkages: Macroeconomic interdependence and cooperation in the worldeconomy, Washington, D.C : Brookings Institution.
Saunders, M., Gill, P., Murray, W., Wright, M., O’Sullivan, M., Eikland, K., Zhang, Y. and Henderson, N. (2013), ‘Lusol:Sparse lu for ax = b’, Dept of Management Science and Engineering (MS&E), Stanford University, Online at:http://www.stanford.edu/group/SOL/ software/lusol.html, Accessed: 6 April 2013.
