PWT 8.0 – a user guide
Contents
1. computes the GDP per capita level based on the World Bank 2008 PPPs and current National Accounts data the December 2012 release of the UN NA Main Aggregates Database These revisions do not just affect the level of real GDP but also the growth rates Johnson et al 2013 document how different versions of PWT can show notably 38 different growth patterns Especially the results of studies that rely on annual GDP growth rates were found to be sensitive to the PWT version that was used In part this will be due to the earlier PPP estimation methodology that relied on PPPs for a given benchmark year and replaced those after a new benchmark 31 But while this source of variation across PWT version will be much reduced by our newly adopted methodology NA revisions are another source of variation Table 7 illustrates this for two UN NA vintages the first with data up to 2009 released December 2010 and the second with data up to 2011 released December 2012 The table compares two sets of growth rates one for 2009 and one for 1995 2009 is the latest year that can be compared while 1995 is chosen as an earlier year for which revisions are presumably no longer as substantial For both years we compare the annual growth rate the five year average annual growth rate and the ten year average annual growth rate following the findings by Johnson et al 2013 that average growth over these longer time spans is more stable across PWT2. Beyond that changing basic national accounts data can and will lead to substantial differences in PWT versions over time In response to the work of Johnson et al 2013 we have changed the PWT methodology to reduce the likelihood of revisions over time but a complete elimination of this concern is impossible because of changes to the underlying data This implies that caution is in order when using the reported results and we discuss when to be cautious and provide practical suggestions on how to be cautious These recommendations will not apply to the same degree to every user and we will clarify this throughout In general users that are interested in specific numbers such as the relative price level of Tanzania in 2000 or the real GDP per capita ranking of Vietnam in 1995 will need to be most careful as limitations to the basic data and specific methodological choices have their largest impact on such individual observations These would typically be most relevant for the type of analysis done in country level studies such as OECD Economic Surveys If instead an econometric analysis is performed on a dataset based on PPPs e g Rodrik 2008 or real GDP per capita levels e g Ashraf and Galor 2013 some of these considerations may be less important as the broad cross country pattern of data is not severely affected by some of these choices Finally for those aiming to explain differences in cross country economic growth while accounti
3. The left hand panel of Figure 2 plots these correlations against the log of GDP per capita for each of the six global comparisons As expected expenditure shares in rich countries are more similar to those in the US than shares in poor countries If the aim is to compare each country s income level to that in the US then this correlation measure suggests that richer countries income levels can be more precisely estimated than poorer countries income levels The right hand panel plots these correlations against the percent difference in GDP PPPs across two popular index number methods the GEKS and the Geary Khamis GK methods 15 This shows that for countries with more highly correlated expenditure patterns the choice of PPP method matters less than for countries with much lower correlations To illustrate if the correlation is 0 7 or higher the average absolute difference between the two PPPs is only 4 5 percent if it is lower than 0 7 the difference is almost 15 percent More in general the fact that the difference in PPPs according to two widely used method can be as large as 50 percent illustrates that due care is needed when comparing living standards between rich and poor countries 14 The correlation of expenditure shares is chosen as it is an intuitive measure that is related to the computation of PPPs For a more rigorous discussion on similarity measures see Diewert 2002 15 See e g Diewert 2013 or Balk 2008 for
4. and Feenstra and Romalis 2012 has estimated how much of the observed differences in unit values is due to differences in product quality and how much is due to actual price differences After netting out the portion of unit value differences across countries that are due to quality the remainder is the quality adjusted price component These remainders are aggregated up to obtain the export and import PPP for each country and year Because the law of one price is closer to holding for traded goods than for non traded goods these export and import PPPs are closer to the nominal exchange rates of countries The resulting trade PPPs can then be used alongside the domestic PPPs from the ICP By dividing the trade PPPs by the nominal exchange rate we obtain the price levels of exports and imports for each country A country will have favorable terms of trade if it receives a relatively high price for its exports compared with prices received by other countries exporting the same product and pays a relatively low price for its imports again compared with prices paid by other countries importing the same product This will tend to make real GDPe higher than real GDPo The impact on real GDP will not only depend on the terms of trade however but also on the domestic prices obtained by dividing the domestic PPPs by the nominal exchange rate If domestic prices are lower than trade prices as would be typical for a developing countrie
5. 8 1960 1970 3 3 2 8 0 5 1970 1980 6 2 4 6 1 6 1980 1990 9 3 6 2 3 0 1990 2000 10 4 7 1 3 3 2000 2010 10 5 9 2 1 3 Note adjusted GDP growth is provided by Harry Wu based on Wu 2011 official data is from the UN NA December 2012 version The adjusted growth series are used in PWT8 0 Concluding remarks In this guide we have explained and motivated the choices we made in constructing PWT8 0 and discussed the implications of these choices for researchers using PWT To summarize we recommend the following 1 Use GDP and GDP series only as a measure of the relative level across countries For comparing GDP growth use the series of GDP at constant national prices from the National Accounts data RGDP4 2 Use GDP when interested in comparative well being use GDP when interested in an economy s productive capacity 3 Beware that observations in PWT that are directly based on PPP benchmark data or interpolations between PPP benchmarks are more reliable than observations based on extrapolations from benchmarks and can show differences in patterns such as the Penn effect 42 4 Beware that there is a greater margin of uncertainty when comparing countries with very different spending patterns 5 Beware that revisions to National Accounts data can have a substantial impact on the level of GDP and on GDP growth rates and that such revisions are typically the dominant reason for changing data between PWT versions Not
6. 9 2 13 44 McCarthy 2013 Extrapolating PPPs and Comparing ICP Benchmark Results in World Bank ed Measuring the Real Size of the World Economy World Bank Washington DC Chapter 18 Neary J Peter 2004 Rationalizing the Penn World Table True Multilateral Indices for International Comparisons of Real Income American Economic Review 94 5 1411 1428 Psacharopoulos George 1994 Returns to investment in education A global 46 update World Development 22 9 1325 1343 Rao D S Prasada 2005 On the equivalence of weighted country product dummy CPD method and the Rao system for multilateral price comparisons Review of Income and Wealth 51 4 571 580 Rodrik Dani 2008 The Real Exchange Rate and Economic Growth Theory and Evidence Brookings Papers on Economic Activity Fall 365 412 Samuelson Paul A 1994 Facets of Balassa Samuelson Thirty Years Later Review of International Economics 2 3 201 226 Seskin Eugene P and Shelly Smith 2009 Improved Estimates of the National Income and Product Accounts Results of the 2009 Comprehensive Revision Survey of Current Business September 15 35 Summers Robert and Alan Heston 1991 The Penn World Table Mark 5 An Expanded Set of International Comparisons 1950 1988 Quarterly Journal of Economics 106 2 327 68 van Veelen Matthijs 2002 An Impossibility Theorem Concerning Multilateral International Compa
7. Abroad indicates that prices are almost 50 percent higher The same range of estimates can also be found for rich countries such as the UK So while there may be a statistical relationship between PPPs and post adjustment indices it is hard to determine how far such a relationship can be trusted Both of these approaches have two further mostly practical drawbacks First of all to deal with non benchmark years we use relative prices for components of GDP rather than GDP as a whole see below for details Even if these two approaches would be useful for estimating economy wide price levels estimation for the GDP components would be required to fully incorporate these non benchmark countries into PWT The second and related drawback is a loss of transparency Because one or more additional estimations would be needed 18 for some countries but not for others we believe it would make PWT harder to interpret We have therefore opted for a third approach namely to omit these countries This is not ideal as it means a less rich dataset However less than 3 percent of the world population live in non benchmark countries Of these Myanmar Algeria Afghanistan and North Korea already account for two thirds of the non benchmark population Omitting non benchmark would thus not lead to a distorted view of global economic performance Furthermore with the even greater country coverage of ICP 2011 the category of non benchmark countrie
8. all recommendations will be equally relevant to all types of users For instance recommendation 4 would mostly be relevant when comparing a single country s GDP or GDP per capita or PPP to other countries Furthermore we provide the basic data and programs so that alternative PWT datasets can be constructed for instance based on a different index number method or on official rather than adjusted Chinese data This guide was explicitly aimed at a non technical audience the Feenstra et al 2013 paper discusses the more technical aspects and the main new insights from PWT8 0 while Inklaar and Timmer 2013c discuss the new capital and TFP data in more detail This guide could also not discuss some of the more detailed aspects of PWT To cover those aspects there is additional documentation available on e The technical details on how the different data sources are combined and how PWT is constructed in Stata including the underlying programs Technical guide to PWT8 0 e The differences in methodology results and variable naming between PWT7 1 and PWT8 0 Comparing PWT8 0 with 7 1 e How data on capital and TFP have been constructed Capital labor and TFP in PWT8 0 43 e The sources of NA data variable definitions and accounting rules National Accounts in PWT8 0 e The choice for a particular exchange rate series Exchange rates in PWT8 0 e The identification of outliers Outliers in PWTS8 0 While the appeal
9. exports and imports This allows us to distinguish two measures of real GDP one aimed at capturing relative living standards as before and one aimed at capturing relative productive capacity Researchers can thus choose the measure that is most appropriate to the research setting The second change is in how we estimate PPPs over time by using more of the historical price survey material This change has important implications for using PWT in research on cross country economic growth The third change is that we introduce measures of capital stock and total factor productivity based on newly developed basic data discussed in Inklaar and Timmer 2013c This broadens the type of research questions that can be answered directly using PWT such as models relying on the distance to the technological frontier e g Aghion and Howitt 2006 Here we discuss these changes in broad terms and focus on the implications of our choices The limitations inherent in comparing living standards or productive capacity of economies across countries are a recurring theme in this guide Whether due to the very nature of the exercise or the practical challenges encountered along the 3 GDP per capita data also requires data on the population of a country Such data is typically less prone to large revisions way it is not possible to be very precise in comparing countries at very different levels of economic development see also Deaton and Heston 2010
10. growth directly from the National Accounts and the change in real GDP from PWT7 1 and from PWT8 0 Column 1 shows that both real GDP measures are strongly correlated with GDP growth from the National Accounts but that this correlation is clearly 24 For growth of GDP at constant national prices the weights would be expenditure valued at national prices while for real GDP the weights would be expenditure valued at reference prices i e in PPP converted values 23 higher in PWT7 1 than in PWT8 0 Moreover when splitting the sample between observations based on benchmark or interpolated PPPs and extrapolated PPPs the difference becomes starker For PWT7 1 there is little change in the correlation which is as expected since real GDP is computed by extrapolating backwards or forwards from the benchmark year in the same way for all observations For PWT8 0 there is a difference between PPPs and thus real GDP based on benchmark and interpolated observations and PPPs based on extrapolated observations This is reflected in the sharp drop in correlation to 0 58 in column 2 Comparing five year or ten year growth rates rather than annual growth rates leads to higher correlations but clear differences remain between GDP growth and the change in real GDP from PWT8 0 Table 4 The correlation between annual GDP growth and the annual change in real GDP PWT7 1 versus PWT8 0 Benchmark amp All observations interpolated Extrapolat
11. growth rates as the 90 percent range of revisions shrinks considerable from 4 8 to 1 3 percentage points for growth rates up to 2009 and from 0 9 to 0 6 percentages for growth rates up to 1995 This also confirms that more recent data are more likely to change due to NA revisions This is unsurprising as the most up to date GDP growth numbers tend to be based on incomplete source data The cross country correlations at the bottom of the table indicate that rapidly growing countries in one vintage also tend to grow fast in the other vintage but a correlation of 0 93 for annual growth in 2009 indicates notable variation To help assess the sensitivity of any research results to the NA vintage used we provide the 2009 and 2010 NA data vintages 32 In addition we include the statistical capacity indicator of the World Bank in PWT8 0 This indicator is constructed since 1999 for developing economies and it is based on the quality 32 Old vintages are not made available by the UN 40 of the statistical methodology frequency with which source data is collected and the timeliness with which data is provided Not all of these aspects refer directly to NA data but this indicator can be useful to assess the reliability of a country s data 33 China China also deserves some attention in regards to its NA As discussed above the 2005 ICP results underestimated China s GDP level which we adjust for in PWT8 0 as in PWT7 0 and 7 1 In a
12. median absolute difference is 3 2 percent and in 5 percent of the countries the difference is even larger than 20 percent Figure 4 shows the difference between GDP and GDP per capita in 2005 set against log of GDP per capita in 2005 The larger differences are typically observed in smaller open economies So the choice between GDP and GDP clearly matters For many analyses it is now possible to use the conceptually appropriate real GDP measure For instance for analyzing productivity differences across countries e g Hall and Jones 1999 Caselli 2005 real GDP would be the appropriate measure while for comparing cross country wellbeing e g Jones and Klenow 2011 real GDP would be better suited In comparisons of cross country wellbeing the effect from favorable or unfavorable terms of trade can and should now also be taken into account Moreover to emphasize the breadth of new information we provide not only provides PPPs for total exports and imports but also distinguished by broad economic categories BEC This breakdown by BEC means that a distinction is made of the prices paid for for example imports of capital goods versus imports of industrial materials This could for instance shed new light on the role of imported technology as highlighted in Caselli and Wilson 2004 by accounting for price differences of imported capital goods Implications The newly developed international trade prices in
13. more details on these methods The GK method has traditionally been used in PWT while the GEKS method has gained ground in the statistical community PWT8 0 uses a combination of these methods see below for details 14 In PWT8 0 we provide the correlation between expenditure shares in each country and the US for all benchmark observations In addition we provide a separate file with all the bilateral correlations as those will be more useful when comparing say China and India rather than India and the US While the correlations are not a structural measure of the reliability of PPP estimates they can be used as a warning signal that whenever the correlation is low the real GDP per capita level should not be interpreted with too much precision Furthermore we provide the software so that alternative PWT datasets can be constructed using different PPP methods so that for any set of empirical results the sensitivity to this choice can be assessed The discussion so far has focused on PPP data from the global ICP comparisons In addition to these Eurostat and the OECD also collect and estimate PPP data see Eurostat OECD 2012 These comparisons are done more frequently than the ICP comparisons annually since 1995 for countries covered by Eurostat current and potential future EU members and once every three years since 1996 for other OECD countries Assuming that a benchmark PPP observation leads to a more reliable estimate
14. of assets only covers the so called fixed reproducible assets recognized in the System of 29 The Total Economy Database of The Conference Board does provide TFP growth measures based on growth in capital services rather than growth in the capital stock as in PWT 35 National Accounts Differences in the availability of land inventories subsoil assets e g World Bank 2006 or intangible assets e g Corrado Hulten and Sichel 2009 are not taken into account This will understate capital input in oil producing and other resource intensive countries in countries with large arable land areas and in richer economies that increasingly rely on investment in intangible assets Hours worked Data on the number of persons engaged could be constructed for 164 out of 167 countries in PWT but data on average annual hours worked is only available for 52 countries from the Total Economy Database of The Conference Board Hours worked vary between 1380 and 2800 hours per year with richer countries working relatively fewer hours Labor input of the poorer countries is thus underestimated by using the number of workers Human capital To account for differences in human capital we use data on the average years of schooling from Barro and Lee 2010 and use rates of return for completing different sets of years of education Psacharopoulos 1994 This ignores any variation in these returns over time or across countries Likewise it ignores diffe
15. of real GDP than a non benchmark observation 16 One alternative approach would be to estimate PPPs in country product dummy CPD regression as in Rao 2005 and use the standard errors of the coefficients as a reliability measure However such a measure ignores the variation in expenditure shares and only accounts for the variation in relative prices across products That variation is unrelated to GDP per capita or the correlation of expenditure shares measure 15 on which more below it is important to incorporate such additional benchmark data so that is what we do in PWT8 0 27 Non benchmark countries Combining PPP data from the six global ICP comparisons and the OECD Eurostat comparisons only provides data for a modest number of countries and years 436 observations from ICP and 438 from OECD Eurostat Furthermore only 167 countries have at some point participated in an ICP comparison Compare this to the 209 countries or areas for which the UN National Accounts NA compiles GDP data and there is a clear shortfall This shortfall is even larger when comparing the number of country year observations from UN NA and historical NA data we have a dataset of 10063 observations spanning the period from 1950 to 2011 This means that PPP benchmark data are directly available for only 8 6 percent of all country year observations Furthermore many of these benchmark observations are for a limited number of countries Table 3 shows that 4
16. 1 percent and human capital 69 percent per worker As a result its relative total factor productivity level is higher than its relative labor productivity level 4 Given this broad overview of the type of measures available in PWTS8 0 we now turn to a more in depth discussion on the choices we make 4 Note that both the amount of human capital per worker and total factor productivity have no natural units The amount of human capital per worker is related the average years of schooling and the return to education while total factor productivity is defined as the relative level of PPPs for consumption and investment Comparing GDP levels across countries requires correcting for price differences across countries This challenge is analogous to measuring GDP growth over time knowing the change in the quantity of products produced in an economy requires correcting nominal values for changes in prices But while measuring price changes over time is a well understood and mostly routine part of the work of statistical offices around the world measuring price differences across countries is much more of a challenge This is because spending patterns tend to be very different especially when comparing rich and poor countries see e g Deaton and Heston 2010 For instance people in poor countries tend to spend much more of their income on food Almas 2012 so that food prices are much more important for living standards than in rich count
17. 19 Furthermore the observations for countries with estimated PPPs would subsequently be less useful in testing some cross country relationships This is most obviously the 18 Or to be precise the World Bank s Atlas method which smoothens out large exchange rate fluctuations 19The most common explanation for the Penn effect is the Balassa Samuelson hypothesis according to which prices in the non traded sector rise more quickly than in the traded sector due to lower productivity growth 17 case for testing the Penn effect which was assumed when imputing the PPPs for missing countries but could be a broader problem as well The second approach taken in earlier versions of PWT is to use post adjustment indices These are indices used by for instance the UN or the US State Department to determine the cost of living when expats are posted in a foreign capital city This approach does not suffer from the problem of the first approach namely that the estimated PPPs would already reflect some of the cross country patterns one may wish to test However it is unclear to which degree these indices reflect the same information as PPPs or even the same information as each other The three sets of post adjustment indices used in PWT6 1 show very extensive differences for numerous countries For example according to the UN index prices in Myanmar are 15 percent lower than in the US while the index from Employment Conditions
18. 9 countries only participated in a single benchmark comparisons while only 37 European and OECD countries participated in more than 6 benchmarks How then to deal with the 42 countries that have never participated in an ICP comparison and how to deal with the many years for which there are no benchmark PPP observations for the other 167 countries 17 The Asian Development Bank also constructed updated PPPs for 2009 the results of which may be included in a future update of PWT 16 Table 3 Number of benchmarks per country Benchmarks Countries Observations 1 49 49 2 30 60 3 20 60 4 15 60 5 7 35 6 9 54 gt 6 37 555 There are broadly three approaches to dealing with the 42 non benchmark countries The first approach taken by the World Bank 2008 is to impute PPP converted GDP per capita by showing for other countries how the ratio of this variable to gross national income per capita converted using nominal exchange rates 8 varies systematically with the secondary school gross enrollment rate This approach exploits the common finding that the exchange rate converted GDP per capita level underestimates the PPP converted GDP per capita level for poorer countries the so called Penn effect Samuelson 1994 Inklaar and Timmer 2013a But while this relationship is found in every PPP benchmark comparison it is not a structural relationship as indicated by the changing coefficients over time in World Bank 2008 p 164
19. Feenstra Robert C Alan Heston Marcel P Timmer and Hayan Deng 2009 Estimating Real Production and Expenditures Across Nations A Proposal for Improving the Penn World Tables Review of Economics and Statistics 91 1 201 12 Feenstra Robert C Hong Ma J Peter Neary and D S Prasada Rao 2013 Who Shrunk China Puzzles in the Measurement of Real GDP Economic Journal 45 forthcoming Feenstra Robert C Robert Inklaar and Marcel Timmer 2013 PWT The Next Generation University of California Davis and University of Groningen draft 2012 Hall Robert E and Charles I Jones 1999 Why do some countries produce so much more output per worker than others Quarterly Journal of Economics 114 1 83 116 Hallak Juan Carlos and Peter K Schott 2011 Estimating Cross Country Differences in Product Quality Quarterly Journal of Economics 126 1 417 74 Hanushek Eric A and Ludger Woessmann 2012 Do better schools lead to more growth Cognitive skills economic outcomes and causation Journal of Economic Growth 17 4 267 321 Heston Alan 2013 Government Services Productivity Adjustments in World Bank ed Measuring the Real Size of the World Economy World Bank Washington DC Chapter 16 Holz Carsten A 2006 China s Reform Period Economic Growth How Reliable Are Angus Maddison s Estimates Review of Income and Wealth 52 1 85 119 Hsieh Chang Tai and Peter J
20. Klenow 2007 Relative Prices and Relative Prosperity American Economic Review 97 3 562 585 Inklaar Robert and Marcel P Timmer 2013a The Relative Price of Services Review of Income and Wealth forthcoming Inklaar Robert and Marcel P Timmer 2013b A Note on Extrapolating PPPs in World Bank ed Measuring the Real Size of the World Economy World Bank Washington DC Chapter 18 Annex D Inklaar Robert and Marcel P Timmer 2013c Capital labor and TFP in PWT8 0 mimeo see www ggdc net pwt Jerven Morten 2013 Comparability of GDP estimates in Sub Saharan Africa the effect of revisions in sources and methods since structural adjustment Review of Income and Wealth forthcoming Johnson Simon William Larson Chris Papageorgiou and Arvind Subramanian 2013 Is newer better Penn World Table Revisions and their impact on growth estimates Journal of Monetary Economics 60 2 255 274 Jones Charles I and Peter J Klenow 2011 Beyond GDP Welfare across Countries and Time NBER Working Paper no 16352 Kravis Irving B Aalan Heston and Robert Summers 1982 World Product and Income Johns Hopkins University Press Baltimore MD Maddison Angus 2006 Do Official Statistics Exaggerate China s GDP Growth A Reply to Carsten Holz Review of Income and Wealth 51 1 121 126 Maddison Angus and Harry X Wu 2008 Measuring China s Economic Performance World Economics
21. P benchmark information overturns the finding of Bergin Glick and Taylor 2006 that the Penn effect disappeared when going back further in time Feenstra et al 2013 shows that this finding is an artifact of the extrapolation methodology used in earlier versions of PWT and without that extrapolation the Penn effect is preserved Figure 3 Schematic illustration of a hypothetical PPP computation Extrapolation Interpolation Extrapolation BM BM gt Si 1996 2005 Note BM stands for benchmark This hypothetical country participated only in the 1996 and 2005 benchmarks Figure 3 summarizes the approach to PPP computation that we use in PWT8 0 Say that a hypothetical country participated in the 1996 and 2005 ICP benchmarks so those benchmark data are used In between these two benchmarks PPPs are interpolated using the national accounts price movements for each country relative to the US thereby ensuring that for those intervening years the estimated PPPs are consistent with the benchmarks on either side 23 Again for sensitivity analysis the programs and data we provide would allow one to construct a PWT dataset based on only the 2005 ICP benchmark which is methodologically comparable to PWT7 0 and PWT7 1 or relying on any of the other benchmarks 21 Before 1996 and after 2005 there are no benchmarks for this particular country so PPPs are extrapolated using relative inflation rates The second l
22. PWT 8 0 a user guide by Robert C Feenstra University of California Davis and NBER Robert Inklaar University of Groningen Marcel Timmer University of Groningen June 2013 This research received support through NSF Grant No 27 3457 00 0 79 195 for the project entitled Integrating Expenditure and Production Estimates in International Comparisons Financial support from the Sloan Foundation is also gratefully acknowledged We also thank Alan Heston Charles Jones and participants at the PWT workshops in Groningen 2011 Philadelphia 2012 and Princeton 2013 and a seminar at the Norwegian School of Economics in Bergen for useful comments and suggestions Introduction The introduction of a substantially revised version of the Penn World Table PWT is a useful moment to introduce or reintroduce this dataset to its users The aim of this user guide is to provide a non technical overview of PWT version 8 0 what are the new concepts how is the dataset constructed how can it be best used in research and what are some of the main limitations The central element of PWT has always been real GDP per capita a measure of relative living standards across countries at different points in time This measure requires two main pieces of information namely GDP per capita in national currency and purchasing power parities PPPs to correct for differences in prices across countries Many of the choices necessary for constructing PWT ar
23. To facilitate such research we have included a measure of real GDP in PWT8 0 whose growth rate equals the National Accounts measure of real GDP growth i e at constant national prices and whose level in the benchmark year 2005 equals real GDP on the output side as discussed next International trade PPPs As a second major change PWTS8 0 introduces a new measure of real GDP to the dataset The traditional measure is based only on prices of consumption and investment i e domestic final expenditure also known as domestic absorption while the new measure also accounts for differences in the prices of exports and imports Put differently the new measure of real GDP accounts for differences in the terms of trade A real GDP measure that ignores terms of trade differences is certainly relevant as it can be seen as a measure of real income for consumers it does not matter if income is high because export prices are relatively favorable or because productivity is high But for comparing the productive capacity of 25 economies we do want to make such a distinction and account for favorable or unfavorable terms of trade in comparing GDP across countries 25 PWT8 0 therefore includes two distinct real GDP measures one from the expenditure side GDP and one from the output side GDP This is in addition to the real GDP measure RGDP 4 discussed just above which equals GDP in 2005 but whose growth rate is taken from the Nationa
24. a Gopinath 2013 International Prices and Exchange Rates prepared for the Handbook of International Economics vol IV Caselli Francesco 2005 Accounting for Cross Country Income Differences in Philippe Aghion and Steven N Durlauf eds Handbook of Economic Growth volume 1A Elsevier Amsterdam 679 741 Caselli Francesco and Daniel J Wilson 2004 Importing technology Journal of Monetary Economics 51 1 32 Corrado Carol Charles Hulten and Daniel Sichel 2009 Intangible capital and U S economic growth Review of Income and Wealth 55 3 661 685 Davarajan Shantayanan 2013 Africa s Statistical Tragedy Review of Income and Wealth forthcoming Deaton Angus 2012 Consumer price indexes purchasing power parity exchange rates and updating paper presented at the Penn World Table Workshop May 2012 Deaton Angus and Alan Heston 2010 Understanding PPPs and PPP based National Accounts American Economic Journal Macroeconomics 2 4 1 35 Diewert W Erwin 2002 Similarity and Dissimilarity Indexes An Axiomatic Approach UBC Discussion Paper no 02 10 Diewert W Erwin 2013 Methods of Aggregation above the Basic Heading Level within Regions in World Bank ed Measuring the Real Size of the World Economy World Bank Washington DC Chapter 5 Feenstra Robert C and John Romalis 2012 International Prices and Endogenous Quality NBER Working Paper no 18314
25. ach to productivity measurement it follows that the relative price of output is equal to the relative price of inputs minus relative productivity 10 based on the 1996 PPP For this reason PWT8 0 omits PPP estimates for Zimbabwe in 2005 Figure 1 Comparing 2005 GDP PPPs PWT8 0 versus World Bank difference between PWT and World Bank OBWA GNB OcoD ocIv ongo gMS O SGPO LUX o o oo o o oO 2 ga os o po o o 8Q fe fe O o Qo 8 o o GP O o A 80 o Bo o_ eo oO oR Ake o o 8 o S 2 5 a a Q o 7 OKGZ OOCHN Q O MDA OQAT OTJK OMDV A a O a i a 4 6 8 10 12 log of CGDP per capita Note the PWT8 0 PPP refers to the GDP PPP the World Bank GDP PPP is taken from World Bank 2008 The next step is to compute a weighted average of basic heading PPPs to arrive at an overall GDP PPP There are a large number of index number methods for doing so and each of these corresponds to a different approach to weighting individual basic heading PPPs A detailed comparison of these methods is beyond 9 Using relative inflation rates to extrapolate 1996 PPPs to 2005 shows an estimated price level o 0 14 with USA 1 while the 2005 ICP data imply a price level of over 1 6 The 0 14 is also more in line with the expected price level for a country at Zimbabwe s level of development 11 the scope of this paper but all aim to give larger weight to those products on which an economy spends relati
26. crease the number of GDP concepts that are included in PWTS8 0 In addition we distinguish between the traditional real GDP measures and current price measures Table 5 summarizes 29 the resulting five measures The first RGDP 4 is primarily suited for measuring economic growth of a particular country over time Its level in 2005 is the same as RGDP and CGDP but its growth rate is equal to that in the National Accounts The next two CGDP and CGDP give the best estimate of the level of GDP in a country relative to another at a single point in time where CGDP is a measure of comparative living standards and CGDP is a measure of comparative productive capacity To make the magnitudes comparable over time we account for US inflation but changes in CGDP levels should not be seen as measure of economic growth Finally there are two real GDP measures RGDP and RGDP These aim to comparable across countries and over time see Feenstra et al 2013 for details These measures are primarily useful to compare trends in comparative living standards RGDP or in productive capacity RGDP This can give insight on such questions as how rich is China compared with the US in 1950 In 2005 RGDP equals CGDP and RGDP equals CGDP Table 5 GDP concepts in PWTS8 0 and their use Series Best use Example RGDP Studies only requiring output Dependent variable in cross based growth rates over time and country growth
27. d Bank 2013 which provides health warnings as well as suggestions for the next global comparison of prices in 2011 across 200 countries The 2005 comparison was a great improvement over the five earlier global comparisons Most notably it covered the largest number of countries ever as shown in Table 2 and the data collection processing and PPP computation were also more rigorous The table also illustrates that before 1996 European countries and OECD countries elsewhere made up the lion s share of countries Country coverage in most other regions has steadily risen with the exception of Latin America where coverage has fluctuated over the years Table 2 ICP global benchmark comparisons and country participation by region ICP benchmark 1970 1975 1980 1985 1996 2005 Europe amp OECD 10 18 22 25 31 44 Asia 3 6 6 8 12 22 Latin America 1 4 16 7 21 10 Middle East amp North Africa 1 2 3 4 12 15 Sub Saharan Africa 1 3 13 19 19 44 Former Soviet Union CIS 12 10 Total 16 33 60 63 107 145 Notes Only includes countries currently included in PWT8 0 so omits Yugoslavia which participated in 1975 1980 and 1985 OECD refers to current membership The 1996 benchmark was constructed for PWT6 based on the 1996 survey for OECD and EU countries and the 1993 survey for other regions in the world This excludes Zimbabwe which did participate in the 2005 benchmark but is not included see for more discussion the sectio
28. ddition there have long been doubts about the accuracy of China s growth figures In the academic literature the debate has been between those arguing that the official statistics get it broadly right Holz 2006 and others arguing that official statistics systematically overstate growth Maddison 2006 Maddison and Wu 2008 We find the overstatement argument convincing and use alternative NA data based on data from Wu 2011 Table 8 shows average annual GDP growth for each decade since 1952 comparing the official NA data to the adjusted GDP data we use in PWT8 0 It show that the degree to which growth is overstated varies considerably over time but is present in every period As a result the GDP level in 1952 is more than twice as high according to the adjusted growth figures than according to the official growth figures Since China participated in an ICP comparison for the first time only in 2005 there is no readily available independent information for a possible cross check of this result While we present data based on the adjusted 33 The work by Jerven 2013 Devarajan 2013 and Young 2012 are useful in this regard as well 41 2005 PPP and adjusted growth rates in PWT8 0 we also provide the data to construct an alternative dataset using official PPPs and growth rates for China Table 8 Average annual GDP growth in China 1952 2010 adjusted versus official Official Adjusted Difference 1952 1960 6 2 5 4 0
29. e related to estimating PPPs and we will use this guide to motivate these choices and discuss their consequences GDP data are readily available from National Accounts NA statistics and so require fewer choices However revisions of NA data by statistical offices are often substantial with GDP increasing by half or even doubling in some cases Such revisions are the 1For a more technical discussion of the methodological innovations in PWT 8 0 see Feenstra Inklaar and Timmer 2013 For discussions of earlier versions see e g Summers and Heston 1991 and Kravis Heston and Summers 1982 2 A country s PPP gives the number of local currency units e g euro s that are needed to buy a bundle of products worth one dollar in the US Dividing the PPP by the nominal exchange rate also in local currency units per dollar then gives the price level of that country relative to the US A price level of 0 5 for example indicates that local prices converted to US dollars with the nominal exchange rate are 1 2 as high on average as in the United States as might be the case fora developing country dominant reason for differences between subsequent versions of PWT that were emphasized by Johnson et al 2013 We illustrate this using recent vintages of NA data 3 In version 8 0 we make three major changes to PWT two of which are related to the calculation of PPPs The first change is that we now also measure relative prices of
30. ed PWT7 1 0 89 0 85 0 89 PWT8 0 0 70 0 52 0 76 Number of observations 7776 2664 5112 Note The change in real GDP from PWT7 1 is measured as the growth of variable rgdpch plus population growth GDP growth is computed from variable rgdpna in PWT8 0 and change in real GDP from PWT8 0 is computed from variable rgdpe Benchmark amp interpolated refers to those observations that in PWT8 0 are based on benchmark PPPs or PPP interpolated between benchmarks variable i_cig The new method of estimating PPPs has arguably led to a measure of real GDP that is more reliable than before since older benchmark information is no longer discarded This has substantially changed PWT data as benchmark and 24 interpolated observations now cover one third of all observations in PWT As a consequence though real GDP has become less suitable to measure changes over time in a single country Real GDP has always been less than ideal for this purpose as it is estimated using information on spending patterns across all countries Since a country s spending pattern is the result of its own preferences and relative prices other countries spending patterns are irrelevant when measuring the economic performance of a single economy over time So if an analysis aims to explain cross country differences in GDP growth rates we would strongly recommend using data on the growth of GDP at constant national prices directly based on a country s National Accounts
31. esson is that any PPP extrapolation should be done at a detailed level so not for GDP as a whole but for different components of GDP A guiding principle should be that if expenditure shares and relative price trends differ considerably across countries and over time the relative change of an aggregate price index will not adequately capture the change in PPPs This is in accordance with earlier PWT practice whereby PPPs for household consumption investment and government consumption are separately extrapolated using price trends for each of these components from the National Accounts In each year the PPPs for the three components are then weighted using expenditure shares for all countries to arrive ata GDP PPP The results of Feenstra et al 2013 on the Penn effect indicate though that even this extrapolation below the GDP level can lead to qualitatively different patterns in the data than benchmark or interpolated observations This suggests that an even more detailed breakdown of GDP would be preferable but this is not readily feasible given available National Accounts data We therefore indicate for each observation whether it is based on a PPP benchmark interpolated between PPP benchmarks or extrapolated using relative inflation This way the robustness of any findings to for instance including real GDP observations based on extrapolated PPPs can be established In addition we compared the extrapolated PPPs to benchmark and inter
32. itude are likely to be seen in more countries for instance as the accounting rules of the System of National Accounts SNA 1993 are replaced by those of SNA 2008 In contrast Ghana revised its GDP upwards by 60 percent in 2010 Jerven 2013 Although such a large revision is not typical it is also not as unusual as may be hoped Figure 5 compares GDP per capita in 2005 as published originally in 30 For interpolation between benchmarks we also use national accounts price indexes to determine the precise year to year pattern instead of doing a linear interpolation This is a second order impact of these data 37 World Bank 2008 with GDP per capita using the most recent NA data It shows that Gambia s GDP per capita has more than doubled after revisions in recent years and even Turkey s GDP per capita has increased by over 40 percent In contrast the median absolute revision is a more modest 3 4 percent As in Figure 4 the broad cross country pattern is not materially affected The correlation between the two sets of GDP per capita numbers in Figure 5 is 0 996 though that is scant comfort if your main interest is the level of living standards in Ghana Figure 5 GDP per capita in 2005 NA revisions since 2008 1 5 2 2 5 current NA data original ICP 1 10 11 8 9 log of GDP per capita original ICP Note original ICP denotes the GDP per capita level as reported in World Bank 2008 current NA data
33. l Accounts of each country Computing GDP requires developing new information on the relative price of exports and of imports as these prices are not part of the World Bank s ICP program Instead the World Bank makes the simplifying assumption that the law of one price holds for traded products so that the exchange rate can be used to express the trade balance in real terms Yet there is much evidence that is at odds with this assumption The review by Burstein and Gopinath 2013 concludes that even in the long run exchange rate movements do not fully pass through to export and import prices and that imperfect competition and pricing to market seem to play an important role in explaining these patterns Yet quantifying these price differences has been a challenging undertaking The only readily available data from which price differences of exports and imports can be inferred is data on the value and quantity of traded products Dividing the value by the quantity gives a unit value but this is only an average price of a potentially very heterogeneous bundle of products 2 Recent research by Hallak 25 This argument is made more formally in Feenstra et al 2009 26 For example one product in the 6 digit Harmonized System list is Color television receivers and that is the most detailed level available in a wide cross country setting On Amazon com television prices vary between 100 and 50 000 26 and Schott 2011
34. lative price of CGDP increases at a much faster rate with income than the relative price of exports or imports So in poor countries prices of exports and imports are relatively high reducing the real share relative to the nominal share Table 6 shows that prices of government consumption increase at the fastest rate with income This can be explained by the Balassa Samuelson hypothesis whereby the non traded sector shows little or no productivity improvement as countries grow richer and therefore faster increases in prices Government consumption consists of spending on government administration education and health activities which are typically among the least exposed to international trade However these are also the activities where prices are hardest to measure As discussed above the relative prices for these categories of expenditures are approximated by productivity adjusted relative wages and it is unknown to what extent these are an accurate reflection of true prices 28 Note that Alcala and Ciccone 2004 define their measure or real openness as nominal exports plus imports over real GDP With PWT8 0 real openness can now be measured as real exports plus imports over real GDP due to the introduction of export and import PPPs 33 Finally it is useful to point out that PWT8 0 also includes a residual real expenditure share so the shares of consumption investment exports and imports do not add up to one A
35. m Eurostat and OECD 13 The World Bank uses the exchange rate to convert the trade balance while PWT measures specific PPPs for exports and imports see the next section for details 12 Figure 2 Correlation between expenditure shares of each country and the US across benchmark comparisons compared with income levels and PPP precision 4 hee 2 4 2 4 4 v 4 N 4 N 4 o o oe o o4 Oo Oo T T T T T T T T T T 4 6 8 10 12 6 4 2 0 2 log of CGDP per capita Difference between alternative PPP indexes Note difference between alternative PPP indexes is the percent difference between a GDP PPP computed using the GEKS index number and the Geary Khamis GK index number As discussed above some of the differences seen in Figure 1 are due to differences in index number methods The choice of index number method will matter more when comparing two countries with very different spending patterns Since the literature has advanced many plausible alternative methods and has proven that none of these will be perfectly suited to the job Van Veelen 2002 there will be a margin of uncertainty in every cross country comparison of prices and real income Furthermore that uncertainty will be larger when comparing countries for which spending patterns differ more 13 Since information on those spending patterns is available we compute the correlation between each country s expenditure shares and those in the US
36. n Benchmark comparisons below This discussion illustrates the main features and limitations of the PPP information used in PWT data is available for relatively few benchmark years for an incomplete set of countries and data quality varies across countries and years As the aim of PWT is to provide a broad and complete panel of real GDP estimates the PPP source material requires further choices and estimation The first set of choices is on how to use the basic benchmark material the second on dealing with non benchmark countries and years Benchmark comparisons In each of the six global comparisons prices are collected for many consumption and investment products Together these cover all of domestic absorption i e GDP excluding the trade balance comparing the trade balance i e exports minus imports across countries will be discussed in more detail in the next section Price quotes would be collected on for example rice of different types and sold in different package sizes and the resulting relative prices are then averaged to arrive at a relative price of rice Overall a list of roughly 1000 products is priced in every country to cover all of consumption and investment These prices are combined into 100 or more so called basic headings for which information is available about the expenditure on these products We mostly take these basic 6 The exception is the 1996 global comparison for which only about 30 basic headi
37. nd over time 34 2 Rather than assuming a single share of labor compensation in GDP to weight the importance of human versus physical capital we have constructed new measures from basic National Accounts data These improvements and in particular the use of a country specific and year specific labor share help to reduce the role of TFP differences in explaining cross country income differences as we show in Feenstra et al 2013 Similar to the distinction between different GDP measures in Table 5 PWT8 0 includes a TFP measure that allows for comparisons across countries at a point in time variable CTFP and a measure that allows comparisons within countries across the years RTFP 4 Despite the improvements over earlier work there are still shortcomings in the TFP measures in PWT8 0 due to a lack of data 1 Capital services Jorgenson and Griliches 1967 argued that not every dollar s worth of capital generates the same return Shorter lived assets such as computers would be expected to earn a greater productive return than long lived assets such as buildings Practical difficulties in determining a required rate of return across countries and over time have stopped us from implementing such an approach This is likely to underestimate capital input mostly in the richer economies where investment in information and communication technologies is highest 2 2 Land inventories subsoil assets and intangibles Our current set
38. nflation differential between poorer and richer countries More in general as long as spending patterns and product inflation rates differ across countries there will be systematic differences between changes in economy wide PPPs and differences in overall inflation We draw two lessons from this observation First that it is preferable to use information from benchmark PPPs whenever possible Although PPP benchmarks are by no means perfect observations of relative prices there is less indication that they are systematically biased than PPPs that are extrapolated from another benchmark using relative inflation rates 2 Put differently PPPs benchmarks were the best estimates of comparative price levels at the time so it seems sensible to use the original source material An alternative estimate would only be preferable if it is of demonstrably higher quality than the original In benchmark years PPPs can be used directly while for years in between benchmarks the trend can be interpolated This approach is a departure from earlier versions of PWT which constructed a set of PPPs for a single benchmark 20 In a two country case this gives a T rnqvist PPP 21 See also McCarthy 2013 for an extensive discussion of this topic 22 See also Inklaar and Timmer 2013b 20 year and extrapolated these using relative inflation rates to the full set of years 3 As we show in Feenstra et al 2013 this new approach using all possible PP
39. ng for differences in initial real GDP per capita levels the recommendations for when to use real GDP per capita and when to use GDP at constant national prices will be most relevant GDP per capita a numerical example To illustrate the main concepts fromPWT8 0 Table 1 compares GDP per capita and productivity in China and the United States in 2005 The first row shows GDP per capita in national currency so in renminbi RMB for China and US dollars USD for the US and these data are directly from National Accounts Since these values are in different currencies a comparison between the Chinese and US values makes little sense The second row converts the Chinese value in US dollars using the market exchange rate at the time of 8 2 RMB USD Comparing the Chinese and US values implies that China s GDP per capita level is only 4 percent of that in the US However the market exchange rate will not reflect relative prices of non traded products such as housing and many other services while a PPP is designed to compare prices for all products in the economy The third row shows real GDP per capita based on PWT8 0 and this makes a considerable difference with China s relative income level at 12 rather than 4 percent of that in the US As will be discussed in more detail below real GDP per capita is a measure of comparative living standards as it covers prices for consumption and investment but not for exports or imports As a result
40. ngs are available heading prices and expenditures as given though for the 2005 benchmark two adjustments were made First the relative consumption prices for China were deemed 20 percent too high by Deaton and Heston 2010 so all consumption basic headings were adjusted downwards by this proportion Second for much of government consumption health education collective services no relative output prices are available so instead relative input prices mostly relative wages are used Since there are large productivity differences across countries these relative input prices are a poor predictor of relative output prices 8 The World Bank 2008 made an adjustment for some countries but Heston 2013 discusses a method to make an adjustment for all countries and this method is applied in both PWT7 x and 8 0 The case of Zimbabwe also warrants discussion Although prices were collected the country was suffering through hyperinflation in this period As a consequence the ICP 2005 report World Bank 2008 does report a PPP estimate but not an exchange rate Conversely the World Bank s databank now omits a PPP estimate while showing an exchange rate The PPP estimate we get using ICP 2005 data and the UN National Accounts exchange rate also imply a relative price level that far exceeds what might be expected 7 See also Feenstra Ma Neary and Rao 2013 for evidence that this adjustment is conservative 8 From the dual appro
41. of a dataset with information on economic performance for most countries in the world over the past 60 years is obvious it is of little use if used without due regard of the choices and limitations that underlie it We hope that this guide has given a better understanding of the PWT8 0 dataset so that it is used to its greatest potential 44 References Aghion Phillipe and Peter Howitt 2006 Appropriate Growth Policy A Unifying Framework Journal of the European Economic Association 4 2 3 269 314 Alcala Francisco and Antonio Ciccone 2004 Trade and Productivity Quarterly Journal of Economics 119 2 613 646 Almas Ingvild 2012 International Income Inequality Measuring PPP Bias by Estimating Engel Curves for Food American Economic Review 102 1 1093 1117 Ashraf Quamrul and Oded Galor 2012 The Out of Africa Hypothesis Human Genetic Diversity and Comparative Economic Development American Economic Review 103 1 1 46 Balk Bert M 2008 Price and Quantity Index Numbers Models for Measuring Aggregate Change and Difference Cambridge University Press Cambridge Barro Robert J and Jong Wha Lee 2010 A new data set of educational attainment in the world 1950 2010 NBER Working Paper no 15902 Bergin Paul R Reuven Glick and Alan M Taylor 2006 Productivity tradability and the long run price puzzle Journal of Monetary Economics 53 2041 2066 Burstein Ariel and Git
42. ountries is quite different This is illustrated in Table 6 which shows the relationship between relative price levels for GDP and each of the expenditure categories and the level of CGDP per capita Each cell in the table is based on a separate regression where the price level of that expenditure category is explained by the log of CGDP per capita The bottom row of the table shows that if CGDP per capita the relative price level of GDP increases significantly which is known as the Penn effect Samuelson 1994 The table shows that relative prices of consumption increase most rapidly with income levels relative prices of exports and imports increase only very modestly and investment prices show no systematic relationship with income These findings are in line with those of Hsieh and 31 Klenow 2007 who find that investment goods are relatively expensive in poorer countries because of the low price of consumption in those countries Table 6 Price levels and CGDP per capita levels pl x Household consumption 0 128 Government consumption 0 217 Investment 0 0207 Exports 0 0166 Imports 0 0230 GDP 0 136 Notes each cell in the table is the coefficient on the log of CGDP per capita from a regression explaining the relative price level of an expenditure category denoted pl x denotes a coefficient significantly different from zero at the 1 level based on robust standard errors clustered by countr
43. polated PPPs and to predicted PPPs based on Penn effect regressions This led us a to replace some market exchange rates by estimated rates whenever price levels spiked due to misaligned 22 exchange rates and b to label some observations as outliers whenever price levels would be systematically outside a range we consider plausible based on observed benchmark and interpolated price levels and predicted price levels from Penn effect regressions These choices are motived and discussed in detail in the documentation on the PWT website Implications The choice to use the historical PPP survey data has implications for the use of PWT in research Until now there has always been a clear connection between growth of GDP at constant national prices and the change in real GDP over time Starting from a single benchmark year PPPs in earlier years were estimated using national price trends Since those price trends are the same as those underlying growth of GDP at constant national prices the only differences in growth rates were due to differences in the weights of consumption investment and government expenditures used to aggregate these components with the trade balance to total GDP However the use of multiple PPP benchmarks in constructing real GDP in PWT8 0 means that changes in real GDP will now show less resemblance to growth of GDP at constant national prices This is confirmed in Table 1 which shows the correlation between annual GDP
44. regressions comparing growth rates across countries CGDPe Expenditure based to compare Initial level in growth regressions relative living standards across requiring relative living countries at a single point in time standards CGDP Output based to compare relative Initial level in growth regressions productive capacity across requiring productive capacity or countries at a single point in time productivity RGDPe Expenditure based to compare Living standards of China today relative living standards across 30 compared to the US at some point countries and over time in the past RGDP Output based to compare relative Productive capacity of China productive capacity across today compared to the US at countries and over time some point in the past Expenditure shares and price levels While data on real GDP per capita is most often used in research the data on price levels and shares by GDP expenditure category are frequently used as well Think of studies looking at the effect on growth of openness e g Alcala and Ciccone 2004 the comparison of investment rates and prices Hsieh and Klenow 2007 or the analysis of a real consumption measure rather than a real GDP measure Jones and Klenow 2011 For such purposes PWT8 0 includes expenditure share variables and relative prices of those expenditure categories These relative prices are the constituent parts of the overall GDP price level but the variation across c
45. rences in the cognitive skills that students obtain which may be more important than the simply the number of years in school Hanushek and Woessman 2012 Ignoring cognitive skills likely underestimates labor input in richer countries as richer countries have higher cognitive skills given the average years of schooling However data do not allow a cognitive skills measure to be implemented for a broad enough sample that also includes variation over time 36 Despite these shortcomings we believe the current data on relative TFP levels and on TFP growth in PWT8 0 represent a useful improvement over earlier work and the list of shortcomings is an open invitation to realize further improvements National Accounts Besides the PPP benchmark data the other main data input of PWT is National Accounts NA data These data are used first to estimate PPPs where benchmark or interpolated data is not available using national price indices Second PWT relies on NA for data on GDP at national prices which is converted to real GDP using the GDP and GDP PPPs Comparative GDP figures are thus subject to change if the underlying NA data are revised In advanced economies such revisions are typically quantitatively small For example the 2009 Comprehensive Revision by the US Bureau of Economic Analysis revised US GDP in 2008 upwards by 1 2 percent Seskin and Smith 2009 neither a negligible nor a substantial change Changes of similar magn
46. ries An estimate of relative living standards needs to take both sets of spending patterns into account which is an inherently imperfect endeavor A challenge of a more practical nature is how to compare say the cost of housing in a Nairobi slum to that in a Washington DC suburb Even if one can measure how much is spent on housing in the two places determining how much of the difference in spending is due to price differences and how much due to the difference in the quantity of output divided by the relative level of inputs In both cases there is no natural interpretation of the absolute values only of the relative values 5 Deaton and Heston 2010 discuss this problem and provide an introduction to the broader index number literature Note that these differences in spending patterns can be the result of differences in prices but also of differences in preferences and that taking such differences into account is still challenging Neary 2004 housing is even harder As a result the PPP comparing prices in Canada to the US will be much more precisely estimated than the PPP for Kenya relative to the US In 2005 the World Bank s International Comparisons Program ICP made the most recent benchmark comparison of consumption and investment prices based on a detailed cross country price survey covering 146 countries around the world see World Bank 2008 More recently a broad review of this comparison appeared Worl
47. rison of Volumes Econometrica 70 1 369 375 World Bank 2006 Where is the Wealth of Nations Measuring Capital for the 21st Century World Bank Washington DC World Bank 2008 Global Purchasing Power Parities and Real Expenditures 2005 International Comparison Program World Bank Washington DC World Bank ed 2013 Measuring the Real Size of the World Economy World Bank Washington DC Wu Harry X 2011 The Real Growth of Chinese Industry Debate Revisited Reconstructing China s Industrial GDP in 1949 2008 The Economic Review Institute of Economic Research Hitotsubashi University 62 3 209 224 Young Alwin 2012 The African Growth Miracle mimeo London School of Economics 47
48. s and the country has a real balance of trade surplus then real GDPe will tend to be higher than real GDPo So in addition to the terms of trade the comparison of trade prices to domestic prices also determine the gap between real GDP and real GDP which has a straightforward interpretation countries with a real GDP level that 27 See footnote 2 27 exceeds their real GDP level can consume in excess of their economy s productive capacity and vice versa We caution that the gap between real GDP and real GDP is not a measure of the gains from trade for countries or at least not the gains from trade as compared to autarky i e no trade We have not built anything into the calculations in PWTS8 0 that would allow the gains as compared to autarky to be estimated Rather this gap reflect the ability of countries to trade as prices that better than the average world prices i e higher for exports or lower for imports By construction then the sum over countries of real GDP and real GDP should be close to zero as occurs in the dataset for all years Figure 4 Real GDP and GDP per capita in 2005 OSGP om OLUX OCY ISL O MDA OIRL OGMB O COG OMDV o OAGO ONOR nd OGNQ T T T T T 4 6 10 12 8 log of GDP per capita Note only the 145 countries included from the 2005 ICP benchmark are included 28 For individual countries the gap between GDP and GDP can be considerable The
49. s is set to shrink even further PWT8 0 will thus only include the 167 countries that participated in a global ICP price comparison at some point in the past Non benchmark years That still leaves the majority of country year observations that are not covered by PPP benchmarks Estimating PPPs for non benchmark years will typically rely on data on national price changes If a PPP is the price of goods in country A relative to those in country B then the change in this PPP may be well approximated by the change in prices in country A relative to the change in prices in country B This will hold by definition when comparing prices of individual products but comparing the relative price of a bundle of goods is more complicated This problem arises again because spending patterns differ across countries and over time but also because prices for different product change at different rates over time and across countries Statisticians compiling the consumer price index CPI only have to take into account price changes and the national 19 spending pattern But when compiling a PPP all sets of budget shares have to be taken into account for instance by using the average share to weight the price difference for each product 2 As shown by Deaton 2012 this is likely to lead to systematic differences between national inflation rates and the change in PPPs with the PPPs of poorer countries increasing at a faster rate than indicated by the i
50. s explained in the National Accounts in PWT8 0 document this residual category can include net exports of services since the export and import PPPs only refer to merchandise trade This explains why for example Panama has a large positive residual expenditure the Panama Canal revenues count as exports of transport services This category also includes any statistical discrepancy For example GDP in India is measured from the income side and the statistical discrepancy is the difference between GDP and the sum of expenditures Capital and productivity The construction of the data on capital and productivity in PWT8 0 are discussed in detail in Inklaar and Timmer 2013c but it is helpful to highlight some of the main features and limitations In the past PWT data has often been used to construct measures of total factor productivity TFP such as by Hall and Jones 1999 and Caselli 2005 These would typically use GDP per worker as the measure of labor productivity and correct for differences in tangible capital per worker and human capital per worker as in Table 1 PWT8 0 improves upon those earlier approaches in two important ways 1 Rather than assuming a single depreciation rate that is constant across countries and over time we allow this to be different By distinguishing investment in up to six types of assets including at least machinery transport equipment and buildings our depreciation rate will vary across countries a
51. the value for the United States is also affected 43212 vs 42330 Row 4 shows real GDP per capita which does reflect relative prices of exports and imports and is thereby a measure of comparative productive capacity China has the same comparative living standards and productive capacity relative to the US but this is not true in general as we will demonstrate below Table 1 Comparing income and productivity in China and the United States for 2005 China United States China US A GDP per capita in national currency 14565 42330 in US dollars converted with exchange rate 1777 42330 4 in US dollars GDP 5342 43212 12 in US dollars GDP 5270 42330 12 B productivity GDP per worker US 8967 87483 10 Tangible capital stock per worker US 29221 261588 11 Human capital per worker index 2 46 3 57 69 Total factor productivity index 0 34 1 00 34 The second part of Table 1 looks at productivity of China relative to the US The first row of panel B shows real GDP per worker i e labor productivity which is somewhat lower in China than the relative GDP per capita level as the Chinese employment population ratio of 59 percent is higher than the 48 percent of the US If the amount of tangible and human capital per worker were the same in the two countries the relative total factor productivity level would be the same as the relative labor productivity level However China has considerably less tangible capital 1
52. vely more Here too PWT version 8 0 follows somewhat different procedures than the World Bank including different index number methods different treatment of regional data and a different PPP conversion of the trade balance see World Bank 2008 for details on their approach and Feenstra et al 2013 for details on PWT These factors together with the changes to the basic headings for China and for government services explain why PWT GDP PPPs in 2005 are different than the World Bank 2008 PPPs Figure 1 plots these differences against the real GDP per capita level from PWTS8 0 This shows how differences are often substantial and these occur at all levels of income This will obviously be important for users interested in the income level of specific countries though for users interested only in the broad cross country pattern the cross country correlation of 0 97 between the two sets of PPPs should be reassuring 10 For those interested see e g Diewert 2013 or Balk 2008 11 The World Bank 2008 uses the GEKS procedure in most regions and the IDB procedure in Africa while PWT uses the GEKS procedure to go from the basic heading level to consumption and investment and the GK procedure to combine these to total GDP 12 The World Bank s methods maintains fixed parities within regions also when computing PPPs across regions while in PWT this is only the case for EU OECD countries whose benchmark PPP data come directly fro
53. versions The table shows the 5 and 95 percentile of the revision to growth rates across the two vintages as well as the correlation between the growth rates in the two vintages 31 Earlier versions of PWT up to v6 constructed real GDP using a weighted average of the national accounts growth rates of the components of GDP i e C I and G That approach was criticized by Johnson et al 2013 because the weights would change with the ICP benchmark in different versions of PWT To address this criticism PWT7 used the national accounts growth rate of total GDP instead of the components as we also do in PWT8 0 when constructing the growth rate of RGDP4 39 Table 7 Revisions to UN NA GDP growth 2009 vs 2011 vintage 2009 1995 Annual 5 year 10 year Annual 5 year 10 year Median growth 0 2 4 1 4 1 4 1 2 3 3 2 Revisions 5th percentile 1 6 1 1 0 6 0 8 0 2 0 4 95th percentile 3 2 1 3 0 7 0 1 0 2 0 2 Correlation 0 93 0 96 0 98 0 97 0 99 0 99 Notes Revisions compare GDP growth across the 2009 and 2011 UN NA vintage for each of the 167 countries in PWT8 0 The annual growth rate refers to growth from 2008 to 2009 in the left hand panel and 1994 1995 in the right hand panel The five year growth rate refers to 2004 2009 and 1990 1995 and the ten year growth rate refers to 1999 2009 and 1985 1995 The results in Table 7 confirm the Johnson et al 2013 finding that long run growth rates are less affected by NA revisions than annual
54. y Each regression includes only the 2706 benchmark and interpolated observations in PWT8 0 and includes year dummies This provides a useful link to the real expenditure shares in PWT8 0 which are shares in real CGDP The starting point of Hsieh and Klenow 2007 is the observation that the real investment share is lower in poor countries than in rich countries This could be because poor countries devote less of their nominal GDP i e at national prices to investment than rich countries or it could be because the price of investment is relatively high in poor countries To see this more precisely we express the real share of expenditure category x as a A E EE A where csh is the real share sh is the nominal share i e at national prices pl x is the relative price level of expenditure category x and pl gdp is the relative price level of GDP Hsieh and Klenow 2007 show that real investment shares in poor 32 countries are low purely because the investment prices are high not because nominal investment shares are low The findings in Table 6 also relate to Alcala and Ciccone 2004 who argue that openness i e exports plus imports over GDP should be measured in real terms rather than in nominal terms i e at national prices They find that poorer countries are less open in real terms than in nominal terms and this is also the case in PWT8 0 28 This can be explained using the results in Table 6 namely that the re
Download Pdf Manuals
Related Search