90722 International Comparison Program [01.02] Should the Regions and the Global Office use Weighted GEKS? Bettina Aten, Alan Heston and Ehraz Refayet 7th Technical Advisory Group Meeting September 17-18, 2012 Washington DC Contents Background ............................................................................................................................................... 2 Least Squares Estimation of EKS (GEKS) .................................................................................. 3 A. The Linking Schema .......................................................................................................................... 3 B. (1) The 2011 ICP ............................................................................................................................... 4 B. (2) Empirical Results for all 2005 Benchmark Countries ................................................. 6 The CAR Estimates ........................................................................................................................ 6 Weighted GEKS Based on Paasche-Laspeyres Ratios ...................................................... 7 Comparing the Volumes .............................................................................................................. 8 C. A Closer Look at the Regional GEKS ........................................................................................... 9 All regions ......................................................................................................................................... 9 The OECD and EU ........................................................................................................................ 10 Conclusions ........................................................................................................................................... 12 References .............................................................................................................................................. 13 1 Should the Regions and Global Office Use Weighted GEKS Bettina Aten, Alan Heston and Ehraz Refayet Background TAG discussions have often focused on the wide economic differences between countries and regions and how best to put together countries within regions and the world. In the discussions it is often suggested that regions are more homogeneous than the world. This note examines one method that has been mentioned for dealing with this issue, namely weighing the various Fisher indexes that go into the GEKS estimation depending on how disparate are the two components, the Paasche and Laspeyres price indexes. At this point we need to distinguish different types of GEKS weighting. This paper deals with weighting with measures of the economic similarity between countries, SIM-weights. Another criteria for introducing country weights into aggregation estimates is by economic importance of a country, for example by person using population, or by economic size using total GDP as in the usual Geary application. Or each country may receive equal weight as in the way EKS is used in the European Union. However, EKS can also be estimated assigning each country total GDP weight.1 Using the researchers data set for the 2005 ICP we look at SIM-weights estimation of GEKS both within regions and for the ICP world as a whole. We also examine SIM-weighting in the context of the method of putting together the world as recommended by TAG, namely the CAR approach. We take the RMSE from the ordinary least squares EKS estimation as a measure of the variability that one would like to reduce. However, the RMSE declines the more spread there is to the weighting system and in the limit, as one gives more weight to some Fishers and less to others, one moves towards chaining through single links as in Hill’s spanning tree approach. We propose a balance that employs modest weighting in the GEKS estimation both within and across regions. The illustrations of this paper employ 1This is further illustrated in Aten, Heston and Refayet (2012b). The term PEI-weights is used in that paper to stand for Political-Economic-Importance size measure of economic size assigned to each country. 2 only one similarity measure: the ratio of the Paasche to the Laspeyres price indexes. There are other symmetrical similarity measures that might be preferable that are discussed in Diewert (2009). Least Squares Estimation of EKS (GEKS) For convenience we build up EKS estimates from the price side using the price level form, namely the PPP/Exchange Rate as a ratio, with the US as the reference country. In estimating the GEKS we follow an approach of an earlier paper (Aten and Heston, 2009) where the GEKS estimation used a regression approach along the lines originally put forward by Gini (1931). This approach has also been used by Rao, Shankar, and Hajarghast (2010). The least squares estimation of EKS is convenient in that it allows for the addition of other variables and also provides a measure of the variance of the estimates. The only similarity or distance variable between countries that is reported here is the Paasche-Laspeyres ratio, and if the approach seems promising other similarity measures should be examined.2 The form of the estimating equation is in (1) where the indexing now refers to individual Fishers between each pair of countries.. As noted Fij, the Fisher price index between country i and j, is expressed in price level form (PL) where the expenditures and PPPs have been divided by the exchange rate to the reference country or the geo-mean of exchange rates of all countries. ij is the error term assumed log normally distributed; ij =1 if i = j, 0 otherwise. Essentially the GEKS estimate for the PLj is the geometric mean of the direct and indirect Fisher price indexes of each country. A. The Linking Schema 2Cuthbert (2003) examined an approximation to a GLS solution of the variance-covariance matrix structure of Fishers that he judged to support GEKS as used for the 32 relatively homogeneous OECD countries in 1996. 3 It will be convenient for the discussion below to summarize the 2005 ICP Fisher matrix of 146 countries as in Table 1, grouped by the 6 ICP regions. Each cell represents a set of Fishers with the principal diagonal being Fishers among countries in their own region. The off-diagonal elements are Fishers between countries outside their region. Table 1 Summary of Matrix of Fishers, F Region Africa Asia/Pacific CIS OECD 45 South Western 146 48 23 10 America Asia 10 10 AFR F11 F12 F13 F14 F15 F16 ASP F21 F22 F23 F24 F25 F26 CIS F31 F32 F33 F34 F35 F36 OECD F41 F42 F43 F44 F45 F46 LAC F51 F52 F53 F54 F55 F56 WAS F61 F62 F63 F64 F56 F66 For example, the cell F11 would contain the 2304 Fishers between the 48 countries in Africa, while cell F16 would contain the 480 Fishers between African and West Asian countries. Underlying the 21,316 (146 x 146) elements represented in Table 1 are the corresponding Paasche and Laspeyres indexes. For convenience the Fishers are from the price side and are expressed in price level form, namely the PPP/Exchange Rate as a ratio, with the US as the reference country. B. (1) The 2011 ICP The method adopted by the Technical Advisory Group on the 2011 ICP is the CAR method, standing for Country Aggregation and Redistribution. The CAR approach may be described as follows: 1. Run one GEKS on the full matrix F above, estimating PPPs and Domestic Absorption (DA) for each country 2. Calculate the DA for each of the 6 Regions 3. Run 6 separate GEKS on F11, F22, F33, F44, F55 and F66, estimating PPPs and DA for each country 4. Calculate the shares of each country within region from c) 5. Apply the shares from d) to the regional totals in b) for each region 4 The CAR approach maintains the regional shares from a global GEKS but controls the distribution of that total according to the regional GEKS. This leads to the first question: whether there is much difference between taking the country DA estimates from a) or e)? We begin by looking at the Equation (1) statistics for the full matrix and the separate regional matrices, shown in Table 2. The first row includes all countries at once. We call this the global GEKS. Rows 2-7 provides the regional estimates. Row 8 is labeled Off Diagonal and is estimated over all countries but excluding the Fishers within regions.3 Row 9 is the weighted regression discussed in more detail shortly. The number of observations is given by n. The RMSE provides the extent to which price levels vary after taking account of country effects with each group.4 Clearly the RMSE within regions is smaller than across all countries in the ICP 2005 benchmark, with the important exception of the weighted regression. Table 2. Inputs into Equation (1) GROUP N Variance RMSE 1 F: All Countries 21316 8712 0.455 2 F11: Africa 2304 372 0.288 3 F22: Asia 529 116 0.339 4 F33: CIS 100 6.3 0.188 5 F44: OECD 2025 596 0.388 6 F55: S.America 100 12 0.258 7 F66: W. Asia 100 16 0.302 3In an evolving paper, Robert Hill has proposed linking regions in a way that finds a chain of all countries the meets his minimum spanning tree criteria using only the Fishers and underlying Paasche and Laspeyere indexes in the off diagonals of Table 1. The advantage of this approach is that the linking will not be inconsistent with the results in each region using the diagonal elements of Table 1, whereas CAR will be inconsistent. Our experiment with making the off-diagonal elements of Table in the estimated GEKs underlying Table 2 derives from Hill’s idea. 4 Because the Fisher matrix is symmetrical with essentially 2 observations per country pair, equation statistics are somewhat messy to interpret. However, the average unexplained variance when you use all 146 countries compared to say the same measure for each of the 6 regions, it is possible to make meaningful inferences. This is a measure that Cuthbert (2003) also used. And as we discuss below a comparison of the unexplained variance between all 146 in the weighted and un-weighted versions of equation (1) should be a guide to the importance of weighting. 5 8 Fi≠j: Off Diagonal 16158 7592 0.441 9 F: Weighted 21316 6044 0.380 At first glance, the low RMSE for CIS compared to the OECD is surprising. It might be thought this was due to the way in which the CIS was linked to the other regions, namely Russia was included in OECD, and then binary links were made to the other CIS countries. However, the Fishers are built up from basic heading PPPs of the individual CIS countries so the special linking of CIS should not really affect the results reported here. However, if we look at the ratio of the Laspeyres to Paasche indexes (L/P ratio), we see a reason for the difference. There are 42 unique country pairings with L/P ratios greater than five, and Tajikistan belongs to 35 of them. The only other CIS country in the 35 pairings is Kyrgyzstan and only one other CIS country is among the 42 pairings, namely Armenia, paired with Portugal. Tajikistan is noisy across regions but without great effect within the CIS. On the other hand, in the remaining 7 pairings, there are four OECD countries: Portugal, Lithuania, Montenegro and Korea, and 3 of the 7 are within OECD pairings (Korea-Lithuania, Korea-Portugal and Montenegro-Portugal). The other four non-Tajikistan pairings of very high L/P ratios are: Korea-Venezuela, Ghana-Lithuania, Cote D’Ivoire- Lithuania, Armenia-Portugal. B. (2) Empirical Results for all 2005 Benchmark Countries The CAR Estimates In Table A1 we present the country estimates derived from the GEKS and other linking approaches, where the countries are ordered alphabetically within each Region by their 3 digit ISO-code. Column A is the DA (domestic absorption with US reference) of each country obtained by converting the exchange rate 6 converted total by the estimated price level from the global GEKS.5 The country DAs in column A are summed to obtain a total for each region6. Column B expresses the CAR country estimates as a ratio to the total DA for the country in Column A. To obtain B, the share of the DA of each country within its region is derived from the 6 separate regional GEKS. The country share of say, Gambia in Africa, is then applied to the regional total of Africa from column A (fn6) to obtain the CAR DA of Gambia. Column B is the ratio of the CAR DA to the global DA for Gambia from Column A. The OD estimate of total DA for each country is expressed as a ratio to column A in column C.7 Weighted GEKS Based on Paasche-Laspeyres Ratios In Table 2 row 9, the ratio of the Paasche to Laspeyres indexes (P/L) was used as the weight in the GEKS regression (equation 1). The expectation is that a Laspeyres price index will tend to be larger the more different the two countries are in structure, and the opposite for a Paasche price index. So the P/L will be smaller when there is more uncertainty around the estimated Fisher, and less weight will be given to those Fishers. The results in Table 2 support such an interpretation. The 5 Domestic Absorption is GDP-exports + imports, or C + G + Domestic Investment. 6 The regional DA totals from the global GEKS are: Global GEKS DA (US$) % Africa 2,073,832 3.6% Asia Pacific 12,839,757 22.6% CIS 2,360,105 4.1% OECD 36,176,864 63.6% South America 2,637,734 4.6% Western Asia 828,583 1.5% All 56,916,875 100.0% 7 The sum of the global GEKS is provided in footnote 5, and is equal to the total that would be obtained summing all CAR country DAs. Because the total of the DA of all countries underlying columns C and D can differ from the column A total, the country estimates have been normalized to the common total. 7 weighted regression has the lowest variance of the three regressions involving all countries, and even a lower RMSE than the OECD. In column D of Table A1 the GEKS estimates of country DAs from the weighted regression are expressed as a ratio to column A. Comparing the Volumes A summary set of data by regions and all countries is provided in Table 3. The geo means and standard deviations of the ratios (CAR, OD, P/L wgt) are presented in columns B-G. The standard deviations of columns E-G respectively derived from Table A1 are all expressed as percentages. A strong regional pattern is clear for the OD approach, suggesting that the off-diagonal approach is the least attractive option. The weighted version appears to vary the least and the OD approach the most when the base is a global GEKS, so the focus will be on the CAR and weighted GEKS results. Table 3. Regional Differences and Variation from global GEKS Regions Geo Means Stand Ard Dev iations CAR OD P/L Wgt CAR OD P/L Wgt A B C D E F G Africa 1.003 1.200 0.993 0.036 0.043 0.011 Asia Pacific 1.008 1.177 0.994 0.019 0.023 0.006 CIS 0.949 1.092 0.983 0.068 0.078 0.013 OECD 0.991 0.900 0.998 0.015 0.014 0.010 South America 0.999 1.117 0.998 0.009 0.010 0.010 Western Asia 0.996 1.064 0.995 0.027 0.030 0.008 All Regions 0.996 1.077 0.995 0.327 0.133 0.010 The fact that the weighted GEKS is close to the un-weighted version is not surprising. Averaging over 146 direct and indirect Fishers that include many similar countries reduces the influence of outliers like Tajikistan, even without weighting. So weighting is not going to show large departures from un-weighted results. The geo-means in Table 3 however, do suggest that the CAR method can have significant regional results, over 5% for CIS, and near to 1% for Asia and the OECD. The variation as measured by the standard deviation generally follows the pattern the same pattern 8 in the regions as for all countries, namely the weighted is less than the CAR or the off- diagonal. We have referred several times to the RMSE as a good indicator of whether weighting is an improvement in comparison to treating all Fishers the same. We interpret a lower RMSE as meaning that the estimated PLs for countries have less associated error than otherwise. However, given the very special character of the matrix of Fishers used in the weighted GEKS, our claim is limited. Returning to a general point in the introduction, if we use weights with more dispersion like the (P/L)2, the RMSE will continue to fall. In Table 2, the global GEKS had a RMSE of .455, and the weighted version, using the P/L has a RMSE of .380. If we use (P/L)2, the RMSE drops to .327. If the principle of using weighted GEKS is accepted, it is not clear how much dispersion should be sought. For example, raising (P/L) to higher and higher powers is analogous to the binary linking in the chaining of Hill. Our view is that a reasonable compromise would be to use a weight or its square, the main argument being that some weighting on the Fishers should be on the table for discussion in most applications. C. A Closer Look at the Regional GEKS All Regions Following up the findings in Part B, we discuss in this Part application of the SIM-weighted GEKS within the regions with special emphasis on the OECD. Weighting in the regions produces similar effects to that for all countries. The results are given in Table 4 for no weights, a (P/L) weight and (P/L)2 weight. If it makes sense to look at the relative reduction in RMSE, then the largest effect is in the CIS, and the smallest effect is in South America when moving from no weighting to (P/L)2. 9 Table 4. Root Mean Square Errors All Countries and Regions P/L raised to power 0 1 2 All 0.455 0.380 0.327 Africa 0.288 0.236 0.199 Asia Pacific 0.339 0.285 0.245 CIS 0.188 0.144 0.118 OECD 0.388 0.336 0.296 South America 0.258 0.229 0.205 Western Asia 0.302 0.256 0.219 We have also addressed the question of what would happen if you used CAR with SIM-weights, where we have used the simple SIM-weight. Six Sim- weighted GEKS were estimated, one for each region. Then CAR was applied using the both the un-weighted and weighted GEKS over all countries to obtain the regional GDPs. The standard deviation using the un-weighted GEKS over all countries was 3.8%, higher than without weighting the regions. When CAR was applied using SIM-weights for regions and all countries the standard deviation was 2.2%, a modest reduction. Our conclusion is if weighting is to be applied then the CAR method would be moderately better weighting both the global and regional GEKS. The OECD and EU Let us look more closely at the OECD where subgroups have been linked internally since the first participation of the OECD in the 1980 benchmark. The EU distinguishes a core group of 15 countries that became members prior to May 2004 and the 10 countries that joined post May 2004. In addition there are the EFTA (European Free Trade Association) countries, Iceland, Norway and Switzerland, and the 3 candidate and 6 other European countries that are all combined into one group, other European countries. Finally, there are 7 other OECD members and 10 Israel. The groupings are given in Table A2. Table 5 provides a parallel weighting scheme to Table 4 for these OECD countries. In interpreting Table 5 it is worth noting that operationally the countries are divided in geographically homogenous areas, for example, a Northern Group including Norway and Iceland. The geographical groups hold workshops that focus on choice of specifications, each with a group leader from the core members. So whatever heterogeneity is observed in Table 5, it is not due to different methods of obtaining basic heading parities. Also, it should be made clear that the measure of heterogeneity we have been using between regions can arise because of data quality and because countries have very heterogeneous economic structures. Either of these effects can contribute to high P/L spreads. This paper would be sharpened if we knew the relative contribution of each of these factors to heterogeneity within and between regions. Table 5. Root Mean Square Errors for OECD and Subgroups OECD Region P/L raised to power Subgroup 0 1 2 n All Euopean and OECD Countries 0.388 0.336 0.296 45 EU Member States 0.321 0.284 0.256 25 Pre May 2004 0.136 0.114 0.100 15 Post May 2004 0.175 0.148 0.128 10 Other European Countries* 0.505 0.435 0.382 12 Other non-European OECD Countries 0.205 0.176 0.157 8 *Combines EFTA, the 3 candidate countries and the other European Countries What is striking is Table 5 is the heterogeneity of the Other European group compared to say the Other non European OECD group. The latter is spread from Australia to Mexico including some fairly diverse countries, so this was surprising to us. It is also clear that the earlier members and later members are each fairly homogeneous, but some of this is lost when they are combined. At present, the EU imposes fixity on their core 25 countriess (the Pre and Post May, 2004 members), and the OECD fits that into a GEKS over all of the 45 countries. Given the relatively low RMSEs for the 25 core Members, this certainly seems justified. We have also 11 estimated not shown a weighted GEKS over the weighted GEKS estimates for each region. We have not shown these results here but they are in the right direction but suggest that there i something to be gained in precision by using SIM-weighting within the OECD. Conclusions We reported a number of linking results at the level of Domestic Absorption using the 2005 ICP research data-base. The linking was based upon the GEKS approach and used Fishers and associated Paasche and Laspeyres price and derived volume measures. The CAR approach that will be used in the 2011 ICP was described, along with an alternative attempt that uses only those Fishers between countries outside their own region.8 Use of SIM-weighted GEKS on the Fisher binaries is a transparent and operational method, and provides measurable differences for comparisons. We have worked with one commonly used weight, but there are other similarity measures that can be considered. In looking at the OECD, we have illustrated how one large region accords priority to its core countries, in this case the 25 members of the EU, while including the others within a multilateral framework. Other regions might also have core or a number of sub-regions that could be handled in a similar manner. Our findings suggest that SIM-weighting of GEKS can readily be applied within the OECD, and to other regions. Should a SIM-weighted GEKS be used? We would argue that it moves the estimates in the right direction and in general this is a good thing. The effect of Sim- weighting is generally less than 1% on the GEKS estimate for a country compared to a base GEKS with a few countries over 2%. Since the effect of SIM-weighting is not large is it worth another to the estimation? We are suggesting yes both because it moves in the right direction but also because of the relatively large differences 8 Weighting of the OD elements does improve the RMSE as would be expected. From a value of 0.441 for an un-weighted GEKS, the RMSE declines with simple weighting to 0.365, and to 0.311 for (P/L)^2. 12 introduced in the CAR approach, in some countries more than 5% compared to a global GEKS. It is our guess that using SIM-weighted GEKS in conjunction with the CAR approach will reduce some of the larger differences associated with the CAR approach at least for 2005. At a minimum this seems like a direction that would justify further study. References Aten, Bettina and A. Heston, “Are All Fishers’ Equal”?, Draft of 4/30/2009, PWT site under Research papers. Cuthbert, James R. (2003), “On the Variance/Covariance Structure of The Log Fisher Index, and Implications for Aggregation Techniques,” Review of Income and Wealth, March, 69-88. Deaton, Angus and Alan Heston (2008), “Understanding PPPs and PPP -based national accounts” American Economic Journal-Macro. Diewert, W. Erwin, (1999), “Axiomatic and economic approaches to international comparisons,” in Alan Heston and Robert Lipsey, eds., International and interarea comparisons of income, output, and prices, University of Chicago Press, 13–52. _____ (2009)”Similarity Indexes and Criteria for Spatial Linking”,in Purchasing Power Parities of Currencies: Recent Developments in Methods and Application, D.S. Prasada Rao editor, Edward Elgar Publishing Ltd. _____ (2011) “Method of Aggregation Above the Basic Heading Level,” Discussion Paper 11/5, Dept of Economics, University of British Columbia Eltetö, O., and P. Köves, 1964, “On a problem of index number computation relating to international comparison,” Statisztikai Szemle 42, 507–18. Gini, Corrado (1924), “Quelques considerations au sujet de la construction des nombres indices des prix et des questions analogues,” Metron, 4, 3–162. ______ (1931), « On the Circular Test of Index Numbers, » Metron, 2, 3–24. Hill, Robert (1999), “Comparing Price Levels Across Countries Using Minimum Spanning Trees,” Review of Economics and Statistics, February, 135-42. 13 Hill, Robert (2012), Linking the Regions in the International Comparisons Program: Some Ways of Imposing Within-Region Fixity, latest version Rao, D.S. Prasada, Sriram Shankar, and Gholamreza Hajarghasht (2010), A Minimum Distance and the Generalized EKS Approahces to Multilateral Comparisons Of Prices and Real Incomes. Szulc, B., 1964, “Indices for multiregional comparisons,” Przeglad Statystyczny, 3, 239 54. Table A1: Comparison of Linking Methods for 2005 ICP (Volume in Millions of US $ for Total Domestic Absorption) Region ISOCode Country GEKS All CAR OD P/L Wgt A B C D AFR AGO Angola 34276 0.865 1.035 0.989 AFR BDI Burundi 3341 0.999 1.195 0.993 AFR BEN Benin 11456 1.027 1.229 0.988 AFR BFA Burkina 16749 1.022 1.222 1.013 Faso AFR BWA Botswana 17664 0.965 1.154 0.98 AFR CAF Central A.R. 2966 1.027 1.229 0.989 AFR CIV Cote 26237 1.044 1.249 0.976 d`Ivoire AFR CMR Cameroon 35594 1.042 1.247 1.001 AFR COM Colombia 236887 0.974 1.166 0.997 AFR CPG Comoros 720 0.975 1.166 0.979 AFR CPV Cape Verde 1955 1.017 1.216 0.988 AFR DJI Djibouti 1521 0.963 1.152 1.007 AFR EGY Egypt 366710 0.998 1.193 0.974 AFR ETH Ethiopia 43352 1.033 1.235 0.988 AFR GAB Gabon 10597 0.98 1.172 1.013 AFR GHA Ghana 31171 1.015 1.214 0.982 AFR GIN Guinea 9457 1.008 1.206 0.987 AFR GMB Gambia 1476 0.929 1.111 1.022 AFR GNB Guinea- 974 0.973 1.164 0.994 Bissau AFR GNQ Equat 6353 1.016 1.216 1.002 Guinea AFR KEN Kenya 52915 1.025 1.226 0.997 AFR LBR Liberia 1394 0.984 1.177 0.992 AFR LSO Lesotho 4495 1.013 1.211 0.997 AFR MAR Morocco 112909 1.007 1.204 0.989 AFR MDG Madagascar 19349 1.025 1.226 0.988 14 Region ISOCode Country GEKS All CAR OD P/L Wgt AFR MLI Mali 13925 0.996 1.192 0.987 AFR MOZ Mozambique 16120 1.033 1.236 0.982 AFR MRT Mauritania 7234 1.021 1.221 0.995 AFR MUS Mauritius 13321 0.985 1.178 0.992 AFR MWI Malawi 9581 1.042 1.246 0.991 AFR NAM Namibia 10018 0.992 1.186 1.012 AFR NER Niger 8617 1.043 1.248 1.008 AFR NGA Nigeria 227147 0.988 1.182 0.993 AFR RWA Rwanda 8186 1.007 1.204 0.998 AFR SDN Sudan 86302 1.016 1.216 0.99 AFR SEN Senegal 20585 1.029 1.231 0.989 AFR SLE Sierra Leone 5143 0.981 1.173 0.996 AFR STP Sao Tome 306 1.015 1.214 0.994 AFR SWZ Swaziland 5079 1.001 1.197 0.992 AFR TCD Chad 13755 0.97 1.16 1.023 AFR TGO Togo 6226 1.037 1.24 0.992 AFR TUN Tunisia 65332 0.98 1.172 1 AFR TZA Tanzania 36791 1.096 1.31 0.987 AFR UGA Uganda 30112 1.016 1.215 0.995 AFR ZAF South Africa 400606 1.002 1.199 0.99 AFR ZAR Congo, D. 16068 1.013 1.212 0.99 AFR ZMB Zambia 15925 1.037 1.24 0.987 AFR ZWE Zimbabwe 6935 0.961 1.149 0.977 Total AFR 2073832 1.003 1.200 0.993 ASP BGD Bangladesh 206261 0.987 1.153 0.991 ASP BRN Brunei 10130 0.997 1.165 0.993 ASP BTN Bhutan 3070 0.971 1.134 0.987 ASP CHN China 5682517 0.984 1.149 0.988 ASP FJI Fiji 4759 1.036 1.21 0.993 ASP HKG Hong Kong 219327 1.028 1.201 0.983 ASP IDN Indonesia 739535 1.018 1.189 1.007 ASP IND India 2624073 1.014 1.185 0.997 ASP IRN Iran 788292 0.994 1.161 0.992 ASP KHM Cambodia 22714 0.999 1.167 0.993 ASP LAO Laos 12281 0.996 1.163 0.993 ASP LKA Sri Lanka 81441 1.007 1.176 0.999 ASP MAC Macao 11234 0.982 1.148 0.993 ASP MDV Maldives 1610 1.009 1.179 0.984 ASP MNG Mongolia 7591 1.017 1.188 0.987 ASP MYS Malaysia 248377 1.035 1.209 0.997 15 Region ISOCode Country GEKS All CAR OD P/L Wgt ASP NPL Nepal 34008 1.016 1.187 0.992 ASP PAK Pakistan 428219 1.005 1.174 0.999 ASP PHL Philippines 286697 1.005 1.173 0.995 ASP SGP Singapore 126600 1.041 1.217 1.003 ASP THA Thailand 481226 1.037 1.212 0.999 ASP TWN Taiwan 613367 1.012 1.182 0.992 ASP VNM Vietnam 206428 0.989 1.155 0.997 Total ASP 12839757 1.008 1.177 0.994 CIS ARM Armenia 17262 0.973 1.121 0.982 CIS AZE Azerbaijan 39560 1.021 1.176 0.985 CIS BLR Belarus 99010 0.98 1.128 0.982 CIS GEO Georgia 22071 0.974 1.121 0.984 CIS KAZ Kazakhstan 140998 0.936 1.077 0.986 CIS KGZ Kyrgyzstan 13188 0.911 1.049 1.002 CIS MKD Macedonia 24447 0.941 1.083 0.99 CIS RUS Russia 1674104 1.014 1.168 0.984 CIS TJK Tajikistan 15737 0.784 0.902 0.952 CIS UKR Ukraine 313728 0.978 1.126 0.987 Total CIS 2360105 0.949 1.092 0.983 OECD ALB Albania 20659 0.977 0.887 0.988 OECD AUS Australia 719806 0.982 0.891 1.012 OECD AUT Austria 273738 1.001 0.909 1.012 OECD BEL Belgium 330449 1 0.908 0.997 OECD BGR Bulgaria 83596 0.984 0.893 0.991 OECD BIH Bosnia 33845 0.98 0.89 0.992 OECD CAN Canada 1124659 0.995 0.903 0.995 OECD CHE Switzerland 253692 0.993 0.902 1.012 OECD CYP Cyprus 19422 1.012 0.919 0.986 OECD CZE Czech R. 213120 0.983 0.893 0.995 OECD DEU Germany 2477522 0.989 0.898 0.996 OECD DNK Denmark 176837 0.993 0.902 1 OECD ESP Spain 1286553 0.996 0.904 0.996 OECD EST Estonia 24896 0.984 0.893 0.99 OECD FIN Finland 154743 0.992 0.901 0.995 OECD FRA France 1922325 1.003 0.911 0.995 OECD GBR United K. 2101222 0.973 0.884 1.012 OECD GRC Greece 310329 1.005 0.912 0.989 OECD HRV Croatia 65039 1.006 0.914 1.008 OECD HUN Hungary 182572 0.982 0.892 0.997 16 Region ISOCode Country GEKS All CAR OD P/L Wgt OECD IRL Ireland 138491 0.996 0.905 1 OECD ISL Iceland 13242 0.97 0.881 1.009 OECD ISR Israel 163121 0.988 0.897 1.005 OECD ITA Italy 1668849 1.002 0.91 1.013 OECD JPN Japan 3895319 1.01 0.917 0.994 OECD KOR Korea 1034090 1.01 0.917 0.974 OECD LTU Lithuania 54735 0.975 0.885 1.017 OECD LUX Luxembourg 25021 1.005 0.913 1.001 OECD LVA Latvia 36100 0.973 0.884 0.99 OECD MDA Moldova 10889 0.982 0.892 0.996 OECD MEX Mexico 1283346 0.959 0.871 0.999 OECD MLT Malta 8892 1.001 0.909 0.999 OECD MNE Montenegro 5499 1.038 0.943 0.983 OECD NLD Netherlands 530045 1.003 0.91 0.988 OECD NOR Norway 182417 0.994 0.903 1.012 OECD NZL New 108461 0.978 0.888 0.997 Zealand OECD POL Poland 566433 0.967 0.878 1.006 OECD PRT Portugal 237087 0.998 0.906 0.988 OECD ROM Romania 227891 0.98 0.89 0.982 OECD SRB Serbia 77304 0.984 0.894 1.003 OECD SVK Slovak R. 96444 0.965 0.876 1.005 OECD SVN Slovenia 47948 0.996 0.905 0.996 OECD SWE Sweden 274087 0.989 0.898 0.999 OECD TUR Turkey 625429 0.979 0.889 1.002 OECD USA United 13090700 1.013 0.92 1.015 States Total 36176864 0.991 0.900 0.998 OECD S.AMER ARG Argentina 401165 0.988 1.105 0.997 S.AMER BOL Bolivia 33088 1.012 1.131 0.997 S.AMER BRA Brazil 1504811 1.003 1.121 0.992 S.AMER CHL Chile 179035 0.998 1.116 0.997 S.AMER COL Columbia 7559 1.002 1.12 0.993 S.AMER ECU Ecuador 86620 1.01 1.13 1.012 S.AMER PER Peru 165528 1.005 1.124 0.998 S.AMER PRY Paraguay 24385 0.989 1.105 0.991 S.AMER URY Uruguay 29884 0.987 1.103 0.985 S.AMER VEN Venezuela 205659 0.998 1.116 1.017 Total 2637734 0.999 1.117 0.998 S.Amer 17 Region ISOCode Country GEKS All CAR OD P/L Wgt W.Asia BHR Bahrain 16062 1.033 1.146 0.988 W.Asia IRQ Iraq 96954 1.000 1.109 0.994 W.Asia JOR Jordan 36720 0.999 1.108 1.008 W.Asia KWT Kuwait 72895 0.998 1.107 0.987 W.Asia LBN Lebanon 52984 0.96 1.065 0.996 W.Asia OMN Oman 39852 0.993 1.102 0.996 W.Asia QAT Qatar 35501 1.056 1.172 0.991 W.Asia SAU Saudi Arabia 349239 0.998 1.107 0.998 W.Asia SYR Syria 79445 1.019 1.13 1.011 W.Asia YEM Yemen 48931 0.981 1.087 0.995 Total W.Asia 828,583 1.073 1.064 0.971 Global 56916878 0.996 1.077 1.145 Table 2a All European and OECD Countries EU Member Prior or Post May States ,2004 1 BEL Belgium Prior 1 BGR Bulgaria Cand 2 DNK Denmark Prior 2 Romania Cand ROM 3 DEU Germany Prior 3 TUR Turkey Cand Other European 4 ESP Spain Prior 4 ISL Iceland EFTA 5 FRA France Prior 5 NOR Norway EFTA 6 IRL Ireland Prior 6 CHE EFTA Switzerland 7 ITA Italy Prior 7 HRV Croatia OEC D 8 LUX Prior 8 Macedonia, OEC Luxembour MKD FYR D g 9 NLD Prior 9 ALB Albania OEC Netherlands D 1 AUT Austria Prior 1 BIH Bosnia and OEC 0 0 Herzegovin D a 1 PRT Portugal Prior 1 MNE OEC 1 1 Montenegro D 1 FIN Finland Prior 1 SRB Serbia OEC 2 2 D 18 1 Sweden Prior Other non- 3 SWE European OECD Members and Israel 1 GBR United Prior 1 AUS Australia Othe 4 Kingdom r 1 GRC Greece Prior 2 NZL New Othe 5 Zealand r 1 CYP Cyprus Post 3 JPN Japan Othe r 2 CZE Czech Post 4 KOR Korea, Rep. Othe Republic r 3 EST Estonia Post 5 CAN Canada Othe r 4 Hungary Post 6 MEX Mexico Othe HUN r 5 LVA Latvia Post 7 USA United Othe States r 6 LTU Lithuania Post 8 ISR Israel Othe r 7 MLT Malta Post 8 POL Poland Post 9 SVK Slovak Post Republic 1 SVN Slovenia Post 0 19