The fifth (2005) round of the World Bank's International Comparison Program, which produces estimates of the gross domestic product at purchasing power parity prices, has been the most extensive and carefully monitored project so far. The preliminary estimates were noticeable for the large downward revisions of gdp estimates of China and India. This article suggests that there were five reasons for the revisions in the fifth round. The single most important factor was the adjustment to productivity estimates of government services; a correction was required but the icp perhaps produced too large an adjustment for China and India.
SPECIAL ARTICLEEconomic & Political Weekly EPW march 15, 200865The 2005 Global Report on Purchasing Power Parity Estimates: A Preliminary ReviewAlan HestonThe fifth (2005) round of the World Bank’s International Comparison Program, which produces estimates of the gross domestic product at purchasing power parity prices, has been the most extensive and carefully monitored project so far. The preliminary estimates were noticeable for the large downward revisions of GDP estimates of China and India. This article suggests that there were five reasons for the revisions in the fifth round. The single most important factor was the adjustment to productivity estimates of government services; a correction was required but the ICP perhaps produced too large an adjustment for China and India. After the World Bank in late 2007 released its preliminary report on the 2005 round of the International Compari-son Program (ICP) the result that generated the most comment was the drop in the size of the economies of China and India, as measured in purchasing power parity (PPP) terms, compared to what had previously been published. In particular, in the Bank’s World Development Indicators (WDI) and in the Penn World Table (PWT) the estimates for 2005 were considerably larger than the new benchmark comparison, the largest ever carried out involving over 145 countries. The comparative numbers are provided in Table 1 (p 66). In general, the position of China and India in the world economy is about 40 per cent less inICP 2005 than in the WDI or PWT, comparing column (1) with columns (2) and (3), or column (5) with column (6). However, all the PPP results on a per capita basis are closer to each other than the exchange rate basis in column (4). The underlying input data will hopefully be available for researchers after the final report is released and the protocol for access is made final. The purpose of this note is to provide some of the likely reasons for these dramatic changes.The 2005ICP covered 145 countries broken into six regions: Africa, Asia, Commonwealth of Independent States (CIS), the Organisation for Economic Cooperation and Development (OECD), South America, and west Asia. The CIS countries carried out their own comparison including Russia, which participated in theOECD so it provided a single country link of the CIS to the Global Comparison. Egypt participated in both the African and west Asian comparisons. However, the linking between the five regions was carried out through a group of 18 ring countries. The resources devoted to the 2005 ICP were much larger than in previous rounds and it was thus possible to hold many more regional and global meetings to organise and improve data com-parability. Further, the data validation in the 2005 round, which had been quite refined in the OECD in earlier rounds, was much more thorough in the other regions than in any previousICP benchmarks. Given the presumption that the 2005 round has a stronger empirical basis than previous rounds, why the large dif-ferences from our previous perceptions of the global economy?In this note we consider five factors that could figure in the explanation and there may well be more. The first is the basis for our prior view of the PPP global economy, the WDI andPWT. The second is the treatment of government administrative services and other non-priced output; the third is linking of regional com-parisons into the world economy; and the fourth is the treatment of the net foreign balance. The fifth is the nature of urban price Alan Heston (aheston@sas.upenn.edu) is at the University of Pennsylvania, United States.
SPECIAL ARTICLEmarch 15, 2008 EPW Economic & Political Weekly66collection in China, which initially attracted a great deal of atten-tion, but not perhaps for the right reasons.1 The Basis for Our Previous View of the World Economy In its review of the 2005ICP the World Bank appropriately dis-cusses the differences between the formerWDI estimates as illus-trated in Table 1 with the new numbers. The theme of this review is that most of the countries, except theOECD, were based on ex-trapolations of a set of PPP estimates for 1993. The 1993 estimates were rightly considered weak because they were cobbled togeth-er from regional benchmarks and questionable linking across the regions. Those countries not in 1993 estimates, like India, were estimated by short-cut methods; a 1988 estimate for China was the basis for the latter. In addition to the weaknesses of the base estimates, the extrapolations at the GDP level were over a 12-year period when relative prices were changing within most countries. As one closely involved withPWT, I naturally have less interest in distancing myself from our most recent estimates than does the World Bank from their estimates, but many of the above prob-lems also apply toPWT. And in the interests of full disclosure, it should also be mentioned that I served in an advisory capacity to the 2005 round, with some of the advice taken and some not. Accepting that the earlier underlying PPP estimates for 2005 have a much weaker base than the current estimates, is there any reason that they should sys-tematically increase the dispersion of per capita incomes between rich and poor countries? For example, were poorer countries systematically price more lower quality goods and services in earlier ICP benchmarks than in 2005? If so, would removal of this bias in 2005 account for the differences from earlier ICP studies, or might that correction itself have overshot the mark? There are a number of questions of this sort and to gain a better perspective on these issues it will require a very detailed analysis by persons both involved and not involved in the production of the estimates. Put another way, the development of ICP methodology has come a long way and the 2005 Handbook online at the Bank is a quantum step above what has gone before. But the ICP is a work in progress, and like national income accounting and price index construction, it is important to get out work like the 2005 report, both for its content as well as for peer review, so that future rounds can continue the improvement of methods and implementation.2 The Equal Productivity AssumptionHow does one compare the output of civil servants and healthand education workers across countries? These comparative-resistant services also plague constant price estimates in the national ac-counts because these outputs are not typically priced. In national accounts, deflation is necessary to make some assumption about what is happening to the productivity of such workers over time, and the same is true across space for the ICP. In pastICP rounds volumes have been derived by dividing compensation by a PPP that was derived from a detailed comparison of salaries for spe-cific occupations. It had been recognised that this procedure assumed equal productivity across countries in a given occupa-tion, which was unlikely given very different amounts of capital per worker and opportunity costs of labour across countries. There is much less inducement to organise work to improve pro-ductivity of their employees, including in administrative, health and education services in very low-wage economies. In the 2005 benchmark, the range of countries was much greater than in previous rounds, and the consequences of the equal-productivity assumption loomed much larger. In Asia, for example, salaries for the same occupation differ by a factor of 100 between Laos and Hong Kong. Similar differences exist between Yemen and Kuwait in the western Asia comparison. Without some adjustment for productivity, the resulting per capita volumes of comparative-resistant services in Yemen or Vietnam would greatly exceed those of its richer neighbours, an improbable outcome. Such adjustments have been considered earlier by the OECD and the ICP, but the 2005 Asian comparison is the first actual case where the equal productivity assumption has been significantly modified.1Asia, west Asia and Africa have also carried out such adjust-ments based on estimates of capital per worker in the whole economy of each country. In Asia, for example, it means that the volume of GDP of China and India relative to Hong Kong or Singapore will be lower than in previous ICP rounds. This poses a problem of comparability across regions in 2005 because EU-OECD-CIS and South America did not make such adjustments. Further, because capital per worker data were not available for many countries, it was often necessary to apply the same adjustment factor to low- income countries that were at different stages of development. The actual procedure used is well documented in the Preliminary and Final Asian Reports that are on the web sites of the Asian Development Bank and World Bank. Clearly, this adjustment is in the right direction and earlier benchmarks did attribute too large a volume of such services to poorer countries, and consequently imparted an upward bias to their PPP converted GDPs. The particular procedure was based upon limited information applied uniformly over groups of coun-tries within each region, so its suitability for particular countries is certainly something requiring further study. Our concern is what does this mean for comparing the 2005 results to previous bench-marks? In previous benchmarks, the volume of administrative, health and education services for very low wage countries in Africa, Asia, and west Asia would have been substantially lowered if the 2005 procedure had been adopted in those years. Everything else remains the same, the methods adopted for these sectors has the effect of producing a smaller spread in real GDP per capita between rich and poor in 2005 than in previous benchmarks. What is the consequence for the 2005 comparison of the mixed application of an adjustment for productivity in some regions and not in others? Roughly, Asian GDP was reduced by over 12 per cent compared to theOECD countries as a consequence of the productivity adjustment.2 Indian’sGDP was lowered by 15.6 per Table 1: Comparison of Results of 2005 ICP with Prior Estimates GDP Per Capita PPP GDP PC US$ GDP PPP ($ billion) ICP 2005 WDI 2005 PWT 6.3 (ICP 2005 ICP 2005 WDI 2005 Exchange Rate) 1 2 3 4 5 6China 4,091 6,760 6,637 1,721 5,333.28,818.6India 2,126 3,452 3,536 707 2,341.0 3,779.0Japan 30,290 30,736 27,726 35,604 3,870.3 3,927.3United States 41,674 41,890 41,674 41,674 12,376.1 12,416.5
SPECIAL ARTICLEEconomic & Political Weekly EPW march 15, 200867cent and China’s by 14.5 per cent, or by about 38 per cent of the difference between theWDI andGDP estimates in Table 1. This means that comparisons with peer countries like Brazil, Mexico and many eastern European countries would be especially affect-ed by the productivity adjustment because presumably their level of productivity in these services would not be so different than in India and China. This is not an argument against a producti-vity adjustment, though the actual implementation was of a “one size fits all countries” nature; rather it helps us better understand where the new view of the position of China and India in the global economy of 2005 is coming from. 3 Linking of the RegionsIn previous global comparisons, linking of regions has often been through only one or two countries, in which case the results can be quite sensitive to the particular link countries.3 In 2005 a method that was less sensitive to the choice of countries was adopted for linking regions at the basic heading in the ICP.4 These basic heading parities in each region were used to convert the national currency expenditures in each country to a volume in the currency of the numeraire country of a region, like Oman in western Asia. Thenext step is to aggregate these expenditures and parities for each region to a total, like consumption or GDP. In 2005 this was done by the EKS method, which in effect gives equal weight to Africa, Asia, OECD, South America and western Asia. PWT, in contrast, uses a method that weights each country by its GDP converted at exchange rates.5 Without going into the merits of each approach, let it just be said that the PWT approach would tend to increase the share of output of Asia in the global GDP by 13-14 per cent or more com-pared to the 2005 ICP. And within regions this would have the ef-fect of lowering incomes for countries like Japan and raising them for China and India within Asia. China would be about half the size of the US economy and India under one-fourththeUS, but fourth largest just above Germany, but still well below Japan. However, it needs to be pointed out that the way the world is put together, each region produces its own comparison, and by convention (fixity) these results are maintained when the global comparison is put together. So evenif GK were used at the global level, it would not change the position of China and India within Asia, only with respect to the OECD. While the World Bank and PWT are closer to each other than to the 2005 ICP for China the comments above only apply to PWT. The World Bank uses a different method to obtain their WDI numbers. The WDI for 2005 relied on extrapolation of 1993 results at the GDP level. This means that if exports are constant but their prices fall, as in the case of microchips for Singapore, GDP growth will overstate the ability of Singapore to convert cur-rent production into current domestic expenditures in 2005 that underlies their benchmark estimate of GDP. Other changes in the terms of trade will similarly drive a positive or negative wedge between extrapolations and current price PPP conversions.4 Current Exports and ImportsIn the 2005 round the current net foreign balance is converted at exchange rates, a practice common in the OECD countries. In fact it has long been recognised that the proper way to deal with trade is to deflate exports by a PPP based on export prices and the same for imports. However, to do this requires collecting another large set of producers prices, making an already formidable task ap-pear too daunting. PWT does not do it right either as discussed in a new paper [Feenstra, Heston, Timmer and Deng forthcoming].Are there alternative approaches to a fullPPP conversion of ex-ports and imports. What has been done inPWT is to convert the net foreign balance at the PPP for domestic absorption. This has one major advantage for the GK approach used inPWT, namely that it is base-country invariant.6 In Feenstra, et al, it was found that there was a significant difference for a number of high trade countries betweenGDP based on expenditures as inPWT and GDP based on production taking into account the difference in parities and volumes of exports and imports. However, among the 145 countries in the 2005 ICP there are a number in Asia, Africa and western Asia where this difference can be quite large. Because China had a large excess of exports and imports in 2005, PWT methods would have raised their output by about 2 per cent and lowered India by a negligible amount compared to the ICP treatment.5 National Average PricesIt is instructive to begin this section with a discussion of price collection in China because it points out some inherent issues in comparing large countries like India, with small countries. China agreed to participate in the ICP in the 1993 and 2005 com-parisons but on a limited level, namely providing mainly urban prices. In 1993 the plan was to compare Shanghai with Tokyo and Guangdong with Hong Kong; the Shanghai comparison was nev-er made public but the Guangdong comparison was completed and was described in the publication of ESCAP(1999). Of course, that leaves the question of how you go from Guangdong to all of China, and in theESCAP publication, this was not attempted. Interestingly there is a long tradition of such city-to-city compari-sons going back to a Shanghai-Tokyo comparison for 1955.7 The price collection by China in 2005 took place in 11 cities and their surrounding rural hinterland. The expenditures refer to all of China and the prices were moved to an all-China basis to replicate the inputs of fully participating countries. Like the 1993 comparison, the relationship of urban to rural prices is the critical step. There have been no official studies of rural-urban price differences. There are urban and rural expenditure surveys that permit comparisons of unit values of many food items and several non-food items like tobacco, fuels and power, and trans-port. Some research has been carried out with these expenditure surveys but permission to make any results of these studies pub-lic has not been forthcoming. Further, even if these results were available, the consumption items for which price differences appear largest are items like housing, medical, and personal services.8 These are precisely the items that have not been surveyed or measured very well in most studies. Brandt and Holz (2006) have made the most comprehensive set of comparisons of rural-urban and regional price levels in China for 1990 updating the results to 2004. As more and more of the urban housing is market-priced, the rural-urban differen-tials for rented and owner-occupied housing have increased in
SPECIAL ARTICLEmarch 15, 2008 EPW Economic & Political Weekly68China. However, in their work Brandt and Holz only approximate rental differences by the cost of construction taking no account of the scarcity value of land. If there is a direction of error in their estimates, it is to understate the difference between rural and urban prices in China. For a common or joint basket of goods there appears to be much more difference between prices across the provinces than between rural and urban areas within prov-inces. For example, the joint basket in rural Beijing is 84.7 per cent higher than in rural Chongquing. The largest urban-rural difference across provinces is 43.5 per cent in Chongquing. The costs of a common basket in urban Beijing is 50.9 per cent higher than in urban Chongquing. These results make it clear why it is difficult to move from urban to national prices in a large country like China. But is the situation of price collection for the ICP so different in China as other large developing countries?There are other large countries like India, Nigeria, Egypt, Brazil, Bangladesh or Indonesia, where the extent of coverage of price collection, or adjustment to national averages is not as clear. For the ICP, India uses prices in several cities broadly covering the most important regions of the countries. However, in rapidly growing countries like India or China, price collec-tion is not being carried out in an economic environment where spatial differences are close to being in long-term balance. That is, disparities are probably larger than they will be in less in-teresting times, and less well measured than they will be when the framework for price collection catches up with the econo-mies. And the big differences by region are not for commodities, which are easier to price, but for services, where differences can easily be 50 per cent or more. The main reasons for service price differences are site rent which means services provided in large urban areas include housing must cost more than smaller cities, less accessible regions and rural areas; and the lower wage rates of service workers in these areas.This raises a more general question for the ICP as between large, diverse countries and smaller city-states. National average prices for countries like Hong Kong, Singapore and Luxembourg are quite fully covered in the outlets used for their CPIs. However, for large countries like Brazil, China, the US or India this is less likely because prices within a country will move in unison over time so theCPI does require full country coverage to reflect time-to-time price changes. This means the framework for collecting prices for non-tradable services does not adequately reflect inter-nal price differences within large developing countries compared to very small countries or more affluent highly urban econo-mies. This factor tends to bias price levels in large developing countries upward.9A final point relates to the quality of items compared across the wide range of economies in the ICP. On the one hand, the average quality of goods entering into the CPI tends to be lower in poorer countries so it can be argued that they may match lower quality items with higher quality items in richer countries, making their price levels too low. This has been suggested in the past as a rea-son the ICP may have overstatedGDP levels in poorer countries. However, there are also offsetting factors that may have been much stronger in the 2005 round than in previousICPs. Many of the qualities available and in the CPIs of poorer countries are not available in higher income countries, while the qualities in the CPIs of richer countries can also be found in poorer countries. Also, the higher quality items are frequently international brands while regional or brand-less products are more important for lower quality items. The consequence is that higher quality items tend to dominate the actual list of items compared in the ICP. These items will not typically be in the CPIs of poorer countries nor necessarily available in the outlets normally sampled in their CPIs. The consequence is that prices are collected in higher-end outlets with the effect of raising price levels of poorer countries. These two effects are in opposite directions and the net impact on comparisons is likely to raise price levels more for countries that do not have much local production, as in a small country in Africa than in say a large country in Asia. More research is needed to evaluate what is the net effect of the quality factor on the overall comparisons. The conclusion drawn from this discus-sion of national average prices is that there is some upward bias in the price levels of China, and to a lesser extent of India, com-pared to affluent and especially very small countries. And this effect is likely to be larger in the 2005 benchmark than in previous ICP comparisons. In future benchmarks it is planned to treat the type of outlet as a characteristic of each price observation within a country.6 Where Does This Leave the Discussion?What conclusions can be drawn about where China and India fit into the world economy based on the 2005 benchmark? The largest source of difference from past practice is the productivity adjustment for government, which certainly is a change in the right direction. However, my judgment is that these adjustments produced too large a reduction for China and India. In addition because the adjustments were not carried out in the OECD, the most important region for comparison, the overall results for Asia are understated in this round. The difference in aggrega-tion methods betweenPWT and the global report also operated to reduce Asia substantially in the world, and there is legitimate disagreement on whether this was appropriate. The difference in treatment of foreign trade is in any event not a large effect. In understanding where the WDI and PWT may have overstated the size of India and China, the effect of extrapolation is partic-ularly hard to quantify. There are possible errors in the national constant price numbers, which has created a large controversy for China [Maddison and Wu 2007 and Maddison 2007]. However, there is such a variety in treatment in national accounts practice in estimating constant price series both across sectors and countries that it would require much more research to understand where extrapolations my be misleading. Even if we accept the 2005 results, the extrapolation question still remains when backward and forward casting of the 2005 results is to be carried out.Finally, there is the difficult problem of judging whether pric-ing of items of higher quality or in unusual outlets in the 2005 may have led to an understatement of real volumes in a number of developing countries in Africa. In China and India many of the international brands are widely available in the large cities though typically at upper-end outlets that are not usu-ally sampled in the CPI, which tends to impart an upward bias
SPECIAL ARTICLEEconomic & Political Weekly EPW march 15, 200865The 2005 Global Report on Purchasing Power Parity Estimates: A Preliminary ReviewAlan HestonThe fifth (2005) round of the World Bank’s International Comparison Program, which produces estimates of the gross domestic product at purchasing power parity prices, has been the most extensive and carefully monitored project so far. The preliminary estimates were noticeable for the large downward revisions of GDP estimates of China and India. This article suggests that there were five reasons for the revisions in the fifth round. The single most important factor was the adjustment to productivity estimates of government services; a correction was required but the ICP perhaps produced too large an adjustment for China and India. After the World Bank in late 2007 released its preliminary report on the 2005 round of the International Compari-son Program (ICP) the result that generated the most comment was the drop in the size of the economies of China and India, as measured in purchasing power parity (PPP) terms, compared to what had previously been published. In particular, in the Bank’s World Development Indicators (WDI) and in the Penn World Table (PWT) the estimates for 2005 were considerably larger than the new benchmark comparison, the largest ever carried out involving over 145 countries. The comparative numbers are provided in Table 1 (p 66). In general, the position of China and India in the world economy is about 40 per cent less inICP 2005 than in the WDI or PWT, comparing column (1) with columns (2) and (3), or column (5) with column (6). However, all the PPP results on a per capita basis are closer to each other than the exchange rate basis in column (4). The underlying input data will hopefully be available for researchers after the final report is released and the protocol for access is made final. The purpose of this note is to provide some of the likely reasons for these dramatic changes.The 2005ICP covered 145 countries broken into six regions: Africa, Asia, Commonwealth of Independent States (CIS), the Organisation for Economic Cooperation and Development (OECD), South America, and west Asia. The CIS countries carried out their own comparison including Russia, which participated in theOECD so it provided a single country link of the CIS to the Global Comparison. Egypt participated in both the African and west Asian comparisons. However, the linking between the five regions was carried out through a group of 18 ring countries. The resources devoted to the 2005 ICP were much larger than in previous rounds and it was thus possible to hold many more regional and global meetings to organise and improve data com-parability. Further, the data validation in the 2005 round, which had been quite refined in the OECD in earlier rounds, was much more thorough in the other regions than in any previousICP benchmarks. Given the presumption that the 2005 round has a stronger empirical basis than previous rounds, why the large dif-ferences from our previous perceptions of the global economy?In this note we consider five factors that could figure in the explanation and there may well be more. The first is the basis for our prior view of the PPP global economy, the WDI andPWT. The second is the treatment of government administrative services and other non-priced output; the third is linking of regional com-parisons into the world economy; and the fourth is the treatment of the net foreign balance. The fifth is the nature of urban price Alan Heston (aheston@sas.upenn.edu) is at the University of Pennsylvania, United States.
SPECIAL ARTICLEmarch 15, 2008 EPW Economic & Political Weekly66collection in China, which initially attracted a great deal of atten-tion, but not perhaps for the right reasons.1 The Basis for Our Previous View of the World Economy In its review of the 2005ICP the World Bank appropriately dis-cusses the differences between the formerWDI estimates as illus-trated in Table 1 with the new numbers. The theme of this review is that most of the countries, except theOECD, were based on ex-trapolations of a set of PPP estimates for 1993. The 1993 estimates were rightly considered weak because they were cobbled togeth-er from regional benchmarks and questionable linking across the regions. Those countries not in 1993 estimates, like India, were estimated by short-cut methods; a 1988 estimate for China was the basis for the latter. In addition to the weaknesses of the base estimates, the extrapolations at the GDP level were over a 12-year period when relative prices were changing within most countries. As one closely involved withPWT, I naturally have less interest in distancing myself from our most recent estimates than does the World Bank from their estimates, but many of the above prob-lems also apply toPWT. And in the interests of full disclosure, it should also be mentioned that I served in an advisory capacity to the 2005 round, with some of the advice taken and some not. Accepting that the earlier underlying PPP estimates for 2005 have a much weaker base than the current estimates, is there any reason that they should sys-tematically increase the dispersion of per capita incomes between rich and poor countries? For example, were poorer countries systematically price more lower quality goods and services in earlier ICP benchmarks than in 2005? If so, would removal of this bias in 2005 account for the differences from earlier ICP studies, or might that correction itself have overshot the mark? There are a number of questions of this sort and to gain a better perspective on these issues it will require a very detailed analysis by persons both involved and not involved in the production of the estimates. Put another way, the development of ICP methodology has come a long way and the 2005 Handbook online at the Bank is a quantum step above what has gone before. But the ICP is a work in progress, and like national income accounting and price index construction, it is important to get out work like the 2005 report, both for its content as well as for peer review, so that future rounds can continue the improvement of methods and implementation.2 The Equal Productivity AssumptionHow does one compare the output of civil servants and healthand education workers across countries? These comparative-resistant services also plague constant price estimates in the national ac-counts because these outputs are not typically priced. In national accounts, deflation is necessary to make some assumption about what is happening to the productivity of such workers over time, and the same is true across space for the ICP. In pastICP rounds volumes have been derived by dividing compensation by a PPP that was derived from a detailed comparison of salaries for spe-cific occupations. It had been recognised that this procedure assumed equal productivity across countries in a given occupa-tion, which was unlikely given very different amounts of capital per worker and opportunity costs of labour across countries. There is much less inducement to organise work to improve pro-ductivity of their employees, including in administrative, health and education services in very low-wage economies. In the 2005 benchmark, the range of countries was much greater than in previous rounds, and the consequences of the equal-productivity assumption loomed much larger. In Asia, for example, salaries for the same occupation differ by a factor of 100 between Laos and Hong Kong. Similar differences exist between Yemen and Kuwait in the western Asia comparison. Without some adjustment for productivity, the resulting per capita volumes of comparative-resistant services in Yemen or Vietnam would greatly exceed those of its richer neighbours, an improbable outcome. Such adjustments have been considered earlier by the OECD and the ICP, but the 2005 Asian comparison is the first actual case where the equal productivity assumption has been significantly modified.1Asia, west Asia and Africa have also carried out such adjust-ments based on estimates of capital per worker in the whole economy of each country. In Asia, for example, it means that the volume of GDP of China and India relative to Hong Kong or Singapore will be lower than in previous ICP rounds. This poses a problem of comparability across regions in 2005 because EU-OECD-CIS and South America did not make such adjustments. Further, because capital per worker data were not available for many countries, it was often necessary to apply the same adjustment factor to low- income countries that were at different stages of development. The actual procedure used is well documented in the Preliminary and Final Asian Reports that are on the web sites of the Asian Development Bank and World Bank. Clearly, this adjustment is in the right direction and earlier benchmarks did attribute too large a volume of such services to poorer countries, and consequently imparted an upward bias to their PPP converted GDPs. The particular procedure was based upon limited information applied uniformly over groups of coun-tries within each region, so its suitability for particular countries is certainly something requiring further study. Our concern is what does this mean for comparing the 2005 results to previous bench-marks? In previous benchmarks, the volume of administrative, health and education services for very low wage countries in Africa, Asia, and west Asia would have been substantially lowered if the 2005 procedure had been adopted in those years. Everything else remains the same, the methods adopted for these sectors has the effect of producing a smaller spread in real GDP per capita between rich and poor in 2005 than in previous benchmarks. What is the consequence for the 2005 comparison of the mixed application of an adjustment for productivity in some regions and not in others? Roughly, Asian GDP was reduced by over 12 per cent compared to theOECD countries as a consequence of the productivity adjustment.2 Indian’sGDP was lowered by 15.6 per Table 1: Comparison of Results of 2005 ICP with Prior Estimates GDP Per Capita PPP GDP PC US$ GDP PPP ($ billion) ICP 2005 WDI 2005 PWT 6.3 (ICP 2005 ICP 2005 WDI 2005 Exchange Rate) 1 2 3 4 5 6China 4,091 6,760 6,637 1,721 5,333.28,818.6India 2,126 3,452 3,536 707 2,341.0 3,779.0Japan 30,290 30,736 27,726 35,604 3,870.3 3,927.3United States 41,674 41,890 41,674 41,674 12,376.1 12,416.5
SPECIAL ARTICLEEconomic & Political Weekly EPW march 15, 200867cent and China’s by 14.5 per cent, or by about 38 per cent of the difference between theWDI andGDP estimates in Table 1. This means that comparisons with peer countries like Brazil, Mexico and many eastern European countries would be especially affect-ed by the productivity adjustment because presumably their level of productivity in these services would not be so different than in India and China. This is not an argument against a producti-vity adjustment, though the actual implementation was of a “one size fits all countries” nature; rather it helps us better understand where the new view of the position of China and India in the global economy of 2005 is coming from. 3 Linking of the RegionsIn previous global comparisons, linking of regions has often been through only one or two countries, in which case the results can be quite sensitive to the particular link countries.3 In 2005 a method that was less sensitive to the choice of countries was adopted for linking regions at the basic heading in the ICP.4 These basic heading parities in each region were used to convert the national currency expenditures in each country to a volume in the currency of the numeraire country of a region, like Oman in western Asia. Thenext step is to aggregate these expenditures and parities for each region to a total, like consumption or GDP. In 2005 this was done by the EKS method, which in effect gives equal weight to Africa, Asia, OECD, South America and western Asia. PWT, in contrast, uses a method that weights each country by its GDP converted at exchange rates.5 Without going into the merits of each approach, let it just be said that the PWT approach would tend to increase the share of output of Asia in the global GDP by 13-14 per cent or more com-pared to the 2005 ICP. And within regions this would have the ef-fect of lowering incomes for countries like Japan and raising them for China and India within Asia. China would be about half the size of the US economy and India under one-fourththeUS, but fourth largest just above Germany, but still well below Japan. However, it needs to be pointed out that the way the world is put together, each region produces its own comparison, and by convention (fixity) these results are maintained when the global comparison is put together. So evenif GK were used at the global level, it would not change the position of China and India within Asia, only with respect to the OECD. While the World Bank and PWT are closer to each other than to the 2005 ICP for China the comments above only apply to PWT. The World Bank uses a different method to obtain their WDI numbers. The WDI for 2005 relied on extrapolation of 1993 results at the GDP level. This means that if exports are constant but their prices fall, as in the case of microchips for Singapore, GDP growth will overstate the ability of Singapore to convert cur-rent production into current domestic expenditures in 2005 that underlies their benchmark estimate of GDP. Other changes in the terms of trade will similarly drive a positive or negative wedge between extrapolations and current price PPP conversions.4 Current Exports and ImportsIn the 2005 round the current net foreign balance is converted at exchange rates, a practice common in the OECD countries. In fact it has long been recognised that the proper way to deal with trade is to deflate exports by a PPP based on export prices and the same for imports. However, to do this requires collecting another large set of producers prices, making an already formidable task ap-pear too daunting. PWT does not do it right either as discussed in a new paper [Feenstra, Heston, Timmer and Deng forthcoming].Are there alternative approaches to a fullPPP conversion of ex-ports and imports. What has been done inPWT is to convert the net foreign balance at the PPP for domestic absorption. This has one major advantage for the GK approach used inPWT, namely that it is base-country invariant.6 In Feenstra, et al, it was found that there was a significant difference for a number of high trade countries betweenGDP based on expenditures as inPWT and GDP based on production taking into account the difference in parities and volumes of exports and imports. However, among the 145 countries in the 2005 ICP there are a number in Asia, Africa and western Asia where this difference can be quite large. Because China had a large excess of exports and imports in 2005, PWT methods would have raised their output by about 2 per cent and lowered India by a negligible amount compared to the ICP treatment.5 National Average PricesIt is instructive to begin this section with a discussion of price collection in China because it points out some inherent issues in comparing large countries like India, with small countries. China agreed to participate in the ICP in the 1993 and 2005 com-parisons but on a limited level, namely providing mainly urban prices. In 1993 the plan was to compare Shanghai with Tokyo and Guangdong with Hong Kong; the Shanghai comparison was nev-er made public but the Guangdong comparison was completed and was described in the publication of ESCAP(1999). Of course, that leaves the question of how you go from Guangdong to all of China, and in theESCAP publication, this was not attempted. Interestingly there is a long tradition of such city-to-city compari-sons going back to a Shanghai-Tokyo comparison for 1955.7 The price collection by China in 2005 took place in 11 cities and their surrounding rural hinterland. The expenditures refer to all of China and the prices were moved to an all-China basis to replicate the inputs of fully participating countries. Like the 1993 comparison, the relationship of urban to rural prices is the critical step. There have been no official studies of rural-urban price differences. There are urban and rural expenditure surveys that permit comparisons of unit values of many food items and several non-food items like tobacco, fuels and power, and trans-port. Some research has been carried out with these expenditure surveys but permission to make any results of these studies pub-lic has not been forthcoming. Further, even if these results were available, the consumption items for which price differences appear largest are items like housing, medical, and personal services.8 These are precisely the items that have not been surveyed or measured very well in most studies. Brandt and Holz (2006) have made the most comprehensive set of comparisons of rural-urban and regional price levels in China for 1990 updating the results to 2004. As more and more of the urban housing is market-priced, the rural-urban differen-tials for rented and owner-occupied housing have increased in 
SPECIAL ARTICLEmarch 15, 2008 EPW Economic & Political Weekly68China. However, in their work Brandt and Holz only approximate rental differences by the cost of construction taking no account of the scarcity value of land. If there is a direction of error in their estimates, it is to understate the difference between rural and urban prices in China. For a common or joint basket of goods there appears to be much more difference between prices across the provinces than between rural and urban areas within prov-inces. For example, the joint basket in rural Beijing is 84.7 per cent higher than in rural Chongquing. The largest urban-rural difference across provinces is 43.5 per cent in Chongquing. The costs of a common basket in urban Beijing is 50.9 per cent higher than in urban Chongquing. These results make it clear why it is difficult to move from urban to national prices in a large country like China. But is the situation of price collection for the ICP so different in China as other large developing countries?There are other large countries like India, Nigeria, Egypt, Brazil, Bangladesh or Indonesia, where the extent of coverage of price collection, or adjustment to national averages is not as clear. For the ICP, India uses prices in several cities broadly covering the most important regions of the countries. However, in rapidly growing countries like India or China, price collec-tion is not being carried out in an economic environment where spatial differences are close to being in long-term balance. That is, disparities are probably larger than they will be in less in-teresting times, and less well measured than they will be when the framework for price collection catches up with the econo-mies. And the big differences by region are not for commodities, which are easier to price, but for services, where differences can easily be 50 per cent or more. The main reasons for service price differences are site rent which means services provided in large urban areas include housing must cost more than smaller cities, less accessible regions and rural areas; and the lower wage rates of service workers in these areas.This raises a more general question for the ICP as between large, diverse countries and smaller city-states. National average prices for countries like Hong Kong, Singapore and Luxembourg are quite fully covered in the outlets used for their CPIs. However, for large countries like Brazil, China, the US or India this is less likely because prices within a country will move in unison over time so theCPI does require full country coverage to reflect time-to-time price changes. This means the framework for collecting prices for non-tradable services does not adequately reflect inter-nal price differences within large developing countries compared to very small countries or more affluent highly urban econo-mies. This factor tends to bias price levels in large developing countries upward.9A final point relates to the quality of items compared across the wide range of economies in the ICP. On the one hand, the average quality of goods entering into the CPI tends to be lower in poorer countries so it can be argued that they may match lower quality items with higher quality items in richer countries, making their price levels too low. This has been suggested in the past as a rea-son the ICP may have overstatedGDP levels in poorer countries. However, there are also offsetting factors that may have been much stronger in the 2005 round than in previousICPs. Many of the qualities available and in the CPIs of poorer countries are not available in higher income countries, while the qualities in the CPIs of richer countries can also be found in poorer countries. Also, the higher quality items are frequently international brands while regional or brand-less products are more important for lower quality items. The consequence is that higher quality items tend to dominate the actual list of items compared in the ICP. These items will not typically be in the CPIs of poorer countries nor necessarily available in the outlets normally sampled in their CPIs. The consequence is that prices are collected in higher-end outlets with the effect of raising price levels of poorer countries. These two effects are in opposite directions and the net impact on comparisons is likely to raise price levels more for countries that do not have much local production, as in a small country in Africa than in say a large country in Asia. More research is needed to evaluate what is the net effect of the quality factor on the overall comparisons. The conclusion drawn from this discus-sion of national average prices is that there is some upward bias in the price levels of China, and to a lesser extent of India, com-pared to affluent and especially very small countries. And this effect is likely to be larger in the 2005 benchmark than in previous ICP comparisons. In future benchmarks it is planned to treat the type of outlet as a characteristic of each price observation within a country.6 Where Does This Leave the Discussion?What conclusions can be drawn about where China and India fit into the world economy based on the 2005 benchmark? The largest source of difference from past practice is the productivity adjustment for government, which certainly is a change in the right direction. However, my judgment is that these adjustments produced too large a reduction for China and India. In addition because the adjustments were not carried out in the OECD, the most important region for comparison, the overall results for Asia are understated in this round. The difference in aggrega-tion methods betweenPWT and the global report also operated to reduce Asia substantially in the world, and there is legitimate disagreement on whether this was appropriate. The difference in treatment of foreign trade is in any event not a large effect. In understanding where the WDI and PWT may have overstated the size of India and China, the effect of extrapolation is partic-ularly hard to quantify. There are possible errors in the national constant price numbers, which has created a large controversy for China [Maddison and Wu 2007 and Maddison 2007]. However, there is such a variety in treatment in national accounts practice in estimating constant price series both across sectors and countries that it would require much more research to understand where extrapolations my be misleading. Even if we accept the 2005 results, the extrapolation question still remains when backward and forward casting of the 2005 results is to be carried out.Finally, there is the difficult problem of judging whether pric-ing of items of higher quality or in unusual outlets in the 2005 may have led to an understatement of real volumes in a number of developing countries in Africa. In China and India many of the international brands are widely available in the large cities though typically at upper-end outlets that are not usu-ally sampled in the CPI, which tends to impart an upward bias 
