Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized THE THREE FACES OF THE INTERNATIONAL COMPARISON PROJECT Irving B. Kravis FILhE CUSY he system of international income and purchasing power comparisons produced by the United Nations International Comparison Project (icP)is improvingour perceptionsof the world economy in three major ways.' The most direct and widely known revelations, of course, are the true size of countries' average incomes and the differences between them. The non-icP method commonly used-the conversion of income in own-currency to a numeraire currency by the use of the exchange rate-exaggerates the dispersion of per capita incomes by systematicallyunderstating those of poorer countries. What has attracted less attention is the insight the icp study brings to comparative national price levels. Until now the vast literature on prices and exchange rates has been confined to comparative movementsof the price levels of different countries over time. This approach requires adopting some base period when price levels are assumed to be in equilibrium. There has been no way of measuringthe differencesin the levelsof prices for GDP at a givenperiod. The icP fillsthis gap. It enables us to observe, for example, that Germany's price level was 130 percent of the U.S. level in 1980 and 76 percent of it in 1984, not just that German prices declinedby more than 40 percent relative to U.S.prices. The icp's third and still less noticed contribution is the opportunity to compare the relativequantities and pricesof the goods that make up the Theauthoris gratefulto LaszloDrechsler, AlanHeston,andRobertE. Lipseyfor their helpfulcommentson an earlier draft of this paper,but none of them is responsiblefor the viewsexpressedhere. DavidRobinsondid the statisticalwork and Nancy Bonsallprepared the manuscript. 3 of different countries. Take, for example, the share of spending on producers' durable goods. For most countries, the share expressed in own-currency is available from the national accounts. Without the Icp comparisons, however, one cannot tell to what extent a country's spending share is higher than another's because it absorbs larger quantities of producers' durable goods and to what extent it is merely because their prices are higher. All three of these new insights arise because purchasing power parities (PPPs)replace the exchange rate as the means of converting GDP and its components to a common currency. The exchange rate has great importance and many uses, but it is not a reliable indicator of the purchasing power of a currency. For example, the average German mark (DM)/U.S. dollar exchange rate in 1984 was DM 2.85 but it took only DM 2.17 to buy the bundle of GDP that $1 would buy. The German GDP per capita was DM 24,186; the U.S. was $11,363. Converted at the exchange rate of DM 2.85, the German figure was only 75 percent of the U.S. GDP per capita. The correct ratio was in fact 98 percent, based on the relative purchasing power of the two currencies. One purpose of this paper is to elaborate on these three faces of the icp. Another is to place these results in the context of the history of international income comparisons, the methods used to obtain them, and the problems of maintaining and developing the system of comparisons. GDPS A Little History 4 The story of real income comparisons goes back to Gregory King (1936). He found that in 1688 per capita incomes in England and Holland were about the same, but almost a fourth lower in France. Nearly three centuries later, the United Kingdom and the Netherlands are still very close; France is roughly 10 percent higher. Since King's work, some other estimates have been made. But until the last few decades, major, well-organized efforts were rare. The most ambitious was an investigation of the real wages and welfare of industrial workers in about 200 towns and cities of Belgium, France, Germany, the United Kingdom, and the United States carried out by the British Board of Trade and published in five reports between 1908 and 1911. Data were collected in the field on money wages in certain ubiquitous trades; price comparisons were made for foods, fuel, and rents; and descriptions of workers' welfare (such as the extent of overcrowding) were provided. At the risk of pressing the data beyond what they can support, it may be said that weekly real wages in October 1905 were about 60 percent higher in the United States than in England, while in the other three countries they were some 30 percent below the English level (Kravis 1984). Another landmark study of a very different character was Colin Clark's comparisons of real income in his Conditions of Economic Progress (1940). Using a remarkable range of secondary sources, Clark compared the Research Observer 1, no. 1 Uanuary 1986) purchasing power of thirty currencies for consumption goods with a y 1929 reference date and more than a dozen with a 1946 reference date. Two advances in statistics and technology paved the way for the next stages of international comparisons. The first was the development of national accounts; the other was the computer. The first widened the focus from wages and wage earners to all types of income and income recipients and provided a statistical framework of concepts (for example, GDP) and accounting rules. The computer made it possible to compare many more countries in a completely even-handed and consistent way (KHS 1982, p. 71). In particular, researchers could satisfy the requirement that the index numbers for income should be "base-country invariant"that is, the quantitative relationships among countries should be independent of the choice of the numeraire country. Another important requirement is that the index numbers should be transitive: that is, Ij.b + Ik/b should equal Ij,, when each index is computed directly and where I is an index of quantities or prices, and j, k, and b, are countries. The first study based on national accounts was the Gilbert and Kravis (1954) comparison for four western European countries and the United States, sponsored by the Organisation for European Economic Cooperation (OEEC). In several important respects, this study established the broad approach adopted in later work, including not only the Gilbert and Associates (1958) extension to four more western European countries but also the icp. All worked with a breakdown of GDP in terms of final expenditures for functional categories (such as food and clothing) instead of trying to compare the output of the GDP originating in each industry. All also used the relationship that prevails for a given good for any pair of countries: the price ratio between the countries times the quantity ratio equals the expenditure ratio. These basic principles were applied to work on Latin America (Braithwaite 1968; Salazar-Carrillo 1973) and eastern Europe. However, many other studies were made during the 1950s and 1960s, mainly involving pairs of countries. Almost all were based on secondary data sources and differed in scope, method, timing, and quality. The result was a patchwork that fell far short of a cohesive picture of world incomes (KKHS 1975, pp. 2-3). In 1968 the United Nations Statistical Office (UNSO)felt the time was ripe to meet head on the major source of incomparability that prompted these scattered efforts-the deviation of exchange rates from purchasing power parities. The UNSO had already promoted international comparability of national accounts by sponsoring the development of a standarized system of national accounts (United Nations 1968). With help from the Ford Foundation, the icP was organized. Half of its small central staff came from UNSO,the rest from the University of Pennsylvania. The ICP's first effort, phase 1, covered seven countries with a 1967 reference date and ten countries for 1970. Phase II extended the comparisons Irving B. Kravis to 5 sixteen countries with 1970 and 1973 data. Phase III included thirty-four countries with a 1975 reference date. The basic methodology was established in phase I, but was improved and extended in the later phases. In phase IV, the icp was taken over by the UNSO. This phase included about sixty countries with a 1980 reference date. Much of the work was done by regional groups of countries, with much less coordination in the selection of specifications and in methods of quality matching. How It Is Done The heart of international income comparisons is making price, rather than quantity, comparisons. For most kinds of goods, it is easier to obtain a representative sample of country-to-country price ratios than quantity ratios, and the sampling variability of price ratios tends to be smaller. If aggregation problems are ignored, the approach may be simply described as deriving a quantity (real income) comparison by dividing a price ratio into an expenditure ratio: Qb Eb Pb where j and b are countries; Q's are physical quantities, E's are expenditures (for GDP or its appropriate components), and the P's are prices, the E's and P's being in own-currencies. (P/Pb is the purchasing power parity.) The tasks involved are: * Dividing GDP into categories for which expenditure data and price comparisons can be obtained * Selecting and pricing a sample of specifications for each expenditure category * Aggregating the price relatives at the category level * Aggregating the categories to form price and quantity indexes for GDP and its subaggregates. The methods used by the European Community in its parallel series of comparisons through which its member countries participated in the ICP were broadly similar, but involved greater detail in classifying expenditures and different ways of matching qualities in price comparisons (SOEC 1983, pp. 22f). The way these tasks have been performed in the iCP may be summarized briefly, with particular attention to those elements that affect the quality of the results. Final expenditures on GDP are subdivided into about 150 detailed categories, following closely the United Nations (1968) recommendations. The degree of disaggregation depended primarily on the availability of expenditure data in various countries. The classification employed is mainly a functional one: household consumption, capital formation, and government consumption are differentiated and then subdivided-for example, food and then beef within consump- 6 Research Observer 1, no. 1 (January 1986) tion. Some effort is made to select classifications that have homogeneous international price relationships for their products. *The next step is to make price comparisons for each detailed category, with a few exceptions (such as number of teachers and of government employees) for which quantity comparisons are more feasible. This phase of the work has the greatest influence on the quality of the income comparison. The sample of specifications for each category must be representative of price formation influences in each country, and the items actually priced in the different countries must be equivalent in quality. To insure such equivalence, the icp arranged international exchanges of samples, on-the-spot inspection of goods in shops by visiting price experts, plus the advice of merchants, manufacturers, and engineers. Once specifications were established, the country's statistical authorities provided the national average price for each. In phase III the thirty-four participating countries each priced an average of nearly 400 consumer goods, 38 construction specifications, and nearly 100 producer durable goods. This approach ran into a major aggregation problem because the countries varied widely in their consumption habits. This variety made it impossible to price a standard list of specifications in all countries and still retain the principle that only representative items be compared. Thus, for most detailed categories, few if any countries reported prices for the full list of items. Methods were devised for deriving transitive, base-country invariant index numbers from incomplete sets of prices for the detailed categories. The missing prices in country A were inferred from the price relationships found in other countries between them and other items for which prices were available in country A (Summers 1973). The formula for aggregating across the categories valued the quantities of goods in each country's GDP at world average prices. Added together, these values yield the desired real income comparisons for GDP or its subaggregates. The formula, which was suggested by Robert Geary and amplified by S. Khamis, involves deriving the price comparisons and the average world prices simultaneously in two subsets of equations (KHS 1982, p. 90). Although there has been a lively academic discussion of aggregation methods, it seems unlikely that equally plausible methods would produce significantly different results (KHS 1982, pp. 95f). In the first three phases of the icp, the methods described above were applied uniformly to all countries, regardless of their location or political affiliation. In phase III, for example, the Geary-Khamis formula was used to calculate "world" average prices, taking into account the price structures of all thirty-four countries. Malawi and the United States were treated alike-although each was assigned a "supercountry" weight, based on the importance of that type of country in the world, not just Irving B. Kravis 7 the country per se (KHS 1982, p. 79). A single set of prices-a common measuring rod-was therefore used to value the product of every country or supercountry. The same is not true of the phase IV results, because the exercise has been regionalized. For example, comparisons among fifteen participating African countries were carried out using average African prices to value each country's components of GDP. At a second stage this method was applied to several "core" countries from each region to produce interregional comparisons. The regional GDP totals obtained from the second stage were applied pro rata to the countries within the region in the proportions of real GDP produced by the first stage. The regional groupings included two sets of countries in Europe (the European Communities, or EC, and one led by Austria), Africa, Asia, and Latin America. In addition, the EC insisted that the relative per capita GDPS of its member countries obtained in exclusively EC calculations be kept fixed in the worldwide comparisons of the UNSO (and also in the OECD comparisons for its member countries). As a result, a common measuring rod was no longer used for comparisons between countries in different regions. The EC's average prices were used to value and compare the GDPS of its member countries, and world prices were used to determine the total GDP of the EC. This meant that a different set of prices was used to value the quantities in, say, a Germany/Norway comparison than in a Germany/United States comparison, and both are different from the EC average prices that produced the internal EC comparison of, say, Germany and France. A valid argument for having regional estimates is that the sample of specifications compared in price and the weights used in aggregation will be closer to those of the countries being compared when the comparison is limited to a region. For internal policy purposes within the EC, comparisons involving only EC members are preferable. But the EC's fixity rule requires that the worldwide comparisons may not present alternative results for EC member countries based on the application of world average prices. The aim is to avoid undermining the confidence that policymakers would have in the estimates were there alternative sets of indexes available. However, the fixity rule imposes a high price on worldwide uses and users. Fixed regional results cannot be embodied in the worldwide indexes so as both to preserve the GDP relationships among countries in a region and to have the sum of the region's quantities for each category equal to the sum established by the interregional comparison. It is impossible to produce a table characterized by "matrix consistency" like the one below, that is, with all the rows adding up correctly for each column, each row showing the proper quantitative relationship between the country columns, and the regional and world versions differing by a fixed multiple. 8 Research Observer 1, no. 1 (January 1986) Country A 150 30 20 200 Consumption Investment Government GDP International units Country B 200 15 35 250 Region 350 45 55 450 All in all, the case for taking a world view in phase V, now being planned with a 1985 reference date, is compelling. The UNSO would produce "universal" comparisons, leaving it to regional bodies to produce purely regional estimates where there is a practical need for them. The results of phase IV are available for some of the regional groupings. But the worldwide comparisons being prepared by the UNSO are not completed at the time of writing. In this article, we have linked the European (Ward 1985) and African (SOEC 1985a) comparisons for 1980 and extrapolated Asian and Latin American benchmark comparisons from 1975 phase III results to 1980. These procedures give only rough interim estimates, but they serve to illustrate the icp findings. To prepare table 1, the forty-nine countries for which such estimates could be made were listed in ascending order of real per capita income and divided into seven income classes, each with seven countries. The table shows the simple averages of the real (Ppp-converted) per capita Real Income Comparisons Table 1. Per Capita GDP Converted to Dollars by PPPsand by Exchange Rates, Averages for Countries Grouped by Income Class, 1980 Average GDP per capita converted by Exchange rate PPP Exchange rate deviation index (U.S. = 100) (1) (2) (3) 4.8 8.7 15.6 26.1 46.5 76.5 93.6 2.3 5.4 9.1 16.2 37.3 86.2 109.1 2.1 2.0 1.8 1.7 1.3 0.9 0.9 1 2 3 4 5 6 7 Income class 7.0 or less 7.0-9.9 10.0-21.9 22.0-34.9 35.0-59.9 60.0-84.9 85.0 or more Note: Forty-nine countries were arrayed according to their PPP-converted per capita GDP in 1980 (from low to high) and divided evenly into the seven income classes. Only OECD countries (including Canada, Japan, and the United States) are found in classes 6 and 7, but a few OECD countries are found in classes 4 and 5. Classes 1 to 5 comprise mainly developing countries: fifteen African, ten Asian, and five Latin American and Caribbean countries. Sources: See tables 2-4. Irving B. Kravis 9 Table 2. Real and Nominal 1983 GDP Per 1980 GDP per capita (U.S. = 100) converted by Country Portugal Yugoslavia Greece Ireland Spain Italy United Kingdom Austria Finland Netherlands Belgium France Denmark Germany Luxembourg Norway Capita, European Countries, 1980 and 1983 GDP per capita (U.S. = 100) converted by Exchange rate deviation index 1980 1983 PPPs (1) Exchange rates (2) PPs (3) Exchange rates (4) (1) (2) (5) (3) - (4) (6) 34.4 35.6 43.8 51.1 55.7 69.1 73.1 75.2 77.9 82.2 84.7 86.0 85.8 88.7 93.8 98.9 21.8 27.7 36.4 48.4 49.5 61.3 82.8 89.7 81.5 104.5 106.2 106.5 113.0 115.5 111.1 123.3 35.1 n.a. 40.9 52.2 56.5 67.4 75.8 76.8 82.0 79.2 84.7 87.3 89.1 88.7 88.2 102.5 14.7 n.a. 24.2 36.7 29.7 44.4 57.9 63.5 68.4 65.9 59.6 67.9 78.8 76.2 62.5 95.3 1.58 1.29 1.20 1.06 1.13 1.13 0.88 0.84 0.96 0.79 0.80 0.81 0.76 0.77 0.84 0.80 2.39 n.a. 1.69 1.42 1.90 1.52 1.31 1.21 1.20 1.20 1.42 1.29 1.13 1.16 1.41 1.08 n:a. Not available. Sources: Ward (1985), except Yugoslavia (UN 1984). (column 1) and the nominal (exchange-rate-converted) per capita (column 2). The ratio of the real to the nominal per capita GDPS set out in column 3 has been called the exchange rate deviation index (ERDI); it provides a gauge of the extent to which the exchange rate, valuable though it is for many purposes, fails as a measure of purchasing power. It can be seen that the lower a country's real GDP per capita, the higher its ERDI tends to be. This generalization also ocurred in earlier phases of the ICP. In phase III, for example, the ERDI averaged 2.6 for eight countries with 1975 real per capita GDPS of 15 percent or less of the U.S. level, but it was unity for nine countries with real incomes from 60 to 90 percent of the U.S. level (KHS 1982, p. 22). Thus, the dispersion of real per capita incomes is smaller than that of nominal per capita incomes; the coefficients of variation for the underlying 1980 benchmark observations in columns 1 and 2 are 84 percent and 108 percent respectively. The larger number of countries included in phase IV makes possible a closer examination of the relationships between real and nominal per capita GDP. Column 5 in table 2 shows the 1980 ERDI for sixteen European economies. As in table 1, the lowest-income countries have the highest ERDIs. In 1980, the average ERDI for the five countries with the GDP GDP 10 ResearchObserver1, no. 1 (January1986) lowest real incomes was 1.25; for the next five it was 0.91; for the remaining six it was 0.80. Among the more industrialized countries of western Europe, however, the relationship between the ERDI and real GDP per capita is much more tenuous. For the countries in the lower half of the table (rows 7 to 16), the negative association between per capita income and the ERDIS is very weak despite a wide spread of per capita incomes (Norway's is 35 percent greater than the United Kingdom's). The underlying reason for this weak relationship may be seen by considering the ERDI in terms of its reciprocal-that is, the price level: ERDI Own-currency GDP . PPP Own-currency GDP - Exchange rate Exchange rate ppp In terms of the earlier illustration, a Pppof DM 2.17 to the dollar and an exchange rate of DM 2.85 to the dollar means that German prices are on average 76 percent of the U.S. level. Since the price levels of western Table 3. Real and Nominal GDP Per Capita, African Countries, 1980 1980 GDP per capita (average for 15 countries = 100) converted by Country Ethiopia Mali Tanzania Malawi Madagascar Kenya Senegal Zambia Nigeria Zimbabwe Cameroon Morocco Ivory Coast Botswana Tunisia Africa (15 countries) PPPs (1) Exchange rates (2) 37 43 47 S3 73 82 88 94 115 115 117 154 176 205 256 100 20 29 37 30 55 62 76 94 145 105 121 128 186 167 199 100 Exchange rate deviation index (1)-(2) (3) 1.85 1.48 1.27 1.77 1.33 1.32 1.16 1.00 0.79 1.10 0.97 1.20 0.95 1.23 1.29 1.00 Note: To prepare table 1, the figures in column 1 were converted to a basis of U.S. = 100 by comparing the phase III and SOEC results for the three countries included in both studies (Malawi, Kenya, and Zambia). The average phase Ill/SOEc 1980 ratio for these three countries, 8 percent, was applied to the column 1 figures in this table to shift them to a U.S. base. Source: SOEC(1985a). Irving B. Kravis 1l European countries are closely linked by extensive trading and institutional ties, differences among them in real per capita income are more weakly correlated with the ERDI. A similar analysis of the phase IV results for fifteen African countries may be carried out on the basis of table 3. Here, ERDIs are highest for the six lowest-income countries. An exception is found in that some of the richer countries in the table have higher ERDIS than those in the middle. For the world as a whole, however,the basic relationship still applies: the poorer a country, the higher its ERDI. This receives further support from the data for Asian countries in table 4, the remaining region for which we can assemble more than a handful of benchmark countries. Comparisons Price Levels of There are two possible explanations for the fact that the purchasing power of the currency of low-income countries tends to be greater than is suggested by their exchange rates. In one, the productivity differential model, the productivity gap between rich and poor countries is held to be smaller in nontradable goods (especially services) than in tradables. This is consistent with the widespread observation, confirmed in the ICP, that nontradables tend to be cheap in low-income countries. In 1975, for example, Average price levels (U.S. = 100) Tradables Nontradables GDP Eight countries with per capita incomes less than 15 percent of U.S. per capita income Nine countries with per capita incomes from 60 to 90 percent of U.S. per capita income 60 25 41 118 97 108 All goods were cheaper in the low-income countries, but the nontradables were cheaper by a larger margin (KHS 1982, p. 196). The reason is that wages in a low-income country are set to a national standard, and the relatively higher productivity in nontradables ensures low prices (KHS 1982, p. 333). With the prices of tradables more nearly aligned with world levels, low prices for nontradables mean a low price level for GDP as a whole. However, even tradable prices are pulled down because they are always sold with a mixture of services (retailing, for example). The alternative explanation, cast in terms of factor proportions, assumes that low-income countries are labor-abundant and that services (an important part of nontradables) are labor-intensive. Since the price equalizing tendencies of international trade operate weakly on nontrad12 Research Observer 1, no. 1 (January 1986) Table4. Real and Nominal GDP Per Capita, Asian Countries, 1980 and 1983 Country India Pakistan SriLanka Thailand Philippines Korea Malaysia Syria Israel Japan 1980 GDP per capita (U.S. = 100) converted by PPPs Exchange rates (1) (2) 1983 GDP per capita (U.S. = 100) converted by PPPs Exchange rates (3) (4) 6.4 7.2 10.0 15.0 11.8 25.3 24.9 25.2 59.1 73.5 6.7 7.8 11.1 16.5 13.8 29.4 27.5 27.2 S9.0 80.3 2.2 2.6 2.4 6.3 6.5 14.4 15.1 12.9 47.8 77.1 1.9 2.2 2.4 5.9 4.8 13.8 14.2 14.8 44.2 69.3 Exchange rate deviation index 1980 1983 (1) (2) (3) (4) (5) (6) 2.91 2.77 4.17 2.38 1.82 1.76 1.65 1.95 1.24 0.95 3.59 3.55 4.63 2.80 2.88 2.13 1.94 1.84 1.33 1.16 Sources:PrPconversions:extrapolations from phase III(1975) results (KHS 1982), except for Israel(SOEC1985b) and Japan (Ward 1985)for which 1980 benchmarkswere available; 1983 Israelextrapolated from 1980; Japan from Ward(1985). Exchangerate conversions:based on IMF (1985) data on current GDP,population, and exchangerates, except for Israel and Japan. ables, they will be cheap in low-income countries and again produce a low price level (Kravis and Lipsey 1983; Bhagwati 1984). These and other structural factors affecting prices in different countries are also joined by short-run influences. Obvious examples include changes in the supply of and demand for money in a given country relative to changes in the numeraire country, and changes in expectations that produce a volatility of exchange rates relative to goods prices (Kravis and Lipsey 1983, p. 18). The literature on the determination of exchange rates has paid most attention to these and other short-run influences. However, it has been argued (Kravis and Lipsey forthcoming) that the price levels of different countries tend to conform to a long-run structural relationship. Such a relationship should prove more useful than the assumption that equilibrium exchange rates conform to purchasing power parities, but the issue needs further research. The overall variation in price levels may be inferred from the relationships between nominal and real GDP per capita shown in tables 1 to 4. The ERDI for countries in income class 1 of table 1 indicates, for example, that the 1980 average price level for the seven lowest-income countries was 48 percent of the U.S. level-that is, their price level is the reciprocal of their ERDI. Table 5 examines more closely price level data for four key countries. The first group of figures shows the variability of exchange rates in recent years-a variability muted here by the use of annual averages, but still large. It is far greater, for example, than the tariff changes achieved by arduous trade negotiations. The next two Irving B. Kravis 13 Table5. Comparative GDP PriceLevels, Selected Countries and Dates France (F/US$) Germany (DM/US$) Japan ( 3 1/US$) United Kingdom (f£US$) Exchange rates 1975 1980 1983 1984 4.29 4.23 7.62 8.74 2.46 1.82 2.55 2.85 297 227 238 238 0.452 0.430 0.659 0.752 PPPs 1975 1980 1983 1984 4.62 5.24 5.93 6.18 2.79 2.37 2.19 2.17 277 240 207 202 0.349 0.487 0.503 0.511 Price levels 1975 1980 1983 1984 108 124 78 71 113 130 86 76 93 106 87 85 77 113 76 68 Source: Ward (1985). banks of figures, the Ppps and the price levels, are unique to icp studies. In the absence of icp data, an analyst might choose a base year and then measure relative changes in price levels. If 1975 were picked as the base and the German PPPs correctly reflected the relative movement of German and U.S. prices, what is often called the "real exchange rate" might be calculated as shown in table 6. The index of the real exchange rate (column 4) is the nominal rate adjusted for the relative price changes in Germany and the United States. The same information may be derived from the price-level data by converting it to an index series with 1975 100 and taking the reciprocals. The icp price levels for Germany-113, 130, 86, 76-give information that the real exchange rates do not. To be sure, the latter imply big fluctuations in relative price levels, but they do not show that the German level shifted from 30 percent above that of the United States in 1980 to 24 percent below it in 1984. Comparisons of Economic Structure 14 This article has concentrated so far on an exchange rate's deviation from its purchasing power over the bundle of goods that comprise GDP. However, the deviation of the exchange rate from Ppp is not uniform. It varies from one good to another. This is, of course, a way of saying that the relative price structures of countries differ. The varying deviations of exchange rates from PPPsalso mean that intercountry quantity relationResearch Observer 1, no. 1 (January 1986) Table6. Calculationof the Real ExchangeRate Real exchange rate 1975 1980 1983 1984 Nominal DM/US$ exchange rate (1) Index of German prices to U.S. pricesa (2) Nominal rate adjusted for price change (1).(2) (3) Index of price-adjusted rate (3) with 1975=100 (4) ICP price levels (5) 2.46 1.82 2.55 2.85 100.0 86.0 80.0 79.0 2.46 2.12 3.19 3.61 100 86 130 147 113 130 86 76 a. Ratio of German GDPdeflator to U.S. GDP deflator. Sources: Column 1: table 5; column 2: IMF (1984,1985); and column5: table 5. ships for components of GDP are not what exchange rate conversions make them appear to be. Thus, the ICP results open the path to the study of both price and quantity structure. Here we illustrate the possibilities by showing the contrast for two sets of goods-producer durables and services-seen first in terms of real quantity comparisons and then in exchange rate conversions. For convenience, table 7 groups the thirty-four countries from IcP phase III into six classes according to the size of their 1975 real GDP per capita. The figures in column 1 suggest that the four lowest-income groups, consisting mainly of developing countries, devoted a larger proportion of their GDP to producer durable goods than did the industrial countries. However, when all components of GDP (including the various kinds of producer durables) are valued in every country at a common set of world average prices, a different picture emerges. The real share of GDP spent on producer durables (column 2) tends to rise with income. This difference stems from the fact that producer durables are relatively expensive in the price structure of a typical low-income country and inexpensive in the price structure of a high-income country. The twist in price relationships which produces this result may be seen in columns 3 and 4. Column 3 shows no marked association between the level of producer goods prices per se and real GDP per capita. But most of the other components of GDP do become more expensive as countries grow richer, and the price-level indexes for GDP in column 4 reflect this. When the prices of producer durables are expressed in relative terms-as column 5 does, taking them in relation to the country's GDP price levelthey are indeed seen to be expensive relative to other GDP components in poor countries, but comparatively cheap in rich countries. Developing countries spend a relatively large part of their incomes on producer durables, but do not get as much as appears for what they spend. Irving B. Kravis 15 Table 7. Expendituresand PriceLevels for ProducerDurable Goods, by Income Class, 1975 Income class (U.S.=100)a 1 2 3 4 5 6 0-14.9 15.0-29.9 30.0-44.9 45.0-59.9 60.0-89.9 90.0-100.0 Share of expenditures in GDPb In nominal In real GDP GDP (1) (2) 10.7 11.6 10.5 11.9 8.9 6.9 5.1 8.8 7.5 10.5 11.2 9.9 Price levels (U.S. = 100) Producer durables (3) 130.1 105.6 135.8 116.4 125.8 100.0 Relative producergoods GDP (4) price levels (3) . (4) (5) 40.7 51.7 64.5 73.6 107.4 100.0 3.20 2.04 2.11 1.58 1.17 1.00 a. U.S. per capita GDPof $7,176 is equal to 100. The countries in each real income class are: (1) Malawi, Kenya, India, Pakistan, Sri Lanka, Zambia, Thailand, and Philippines; (2) Korea, Malaysia, Colombia, Jamaica, Syria, and Brazil; (3) Romania, Mexico, Yugoslavia, Iran, Uruguay, and Ireland; (4) Hungary, Poland, Italy, and Spain; (5) United Kingdom, Japan, Australia, Netherlands, Belgium, France, Luxembourg, Denmark, and Germany; and (6) United States. b. Averages of country shares. Source: KHS (1982), tables 6-6, 6-5, and 6-8. This general argument can be strikingly illustrated by examples of individual countries. In the phase III report, the Philippines was found to have spent 13.9 percent of its 1975 GDP on producer durables while Korea spent only 11.8 percent. Translated into dollars via exchange rates, their respective expenditure per capita was $52.40 and $68.97. Such figures are grossly misleading, however. Producer durables were 36 percent more costly in the Philippines than in the United States, while in Korea they were 24 percent cheaper. When world prices are used as a common measuring rod, Koreans bought nearly 2.5 times as many producer durables per capita as did Filippinos ($131.00 compared with $55.60). The Korean share of real Ppp-converted GDP going to producer durables was 8.8 percent; the Philippine share was 5.9 percent. Another illustration of the misleading comparisons that can be produced by exchange rate conversion is the widely held belief that richer countries produce and consume more and more of their output in the form of services. This view seems to be justified by the figures in column 1 of table 8, showing the shares of GDP spent on services by countries in the six income classes. However, when world average prices are used to value the goods and services in each country's GDP, the share of services remains virtually constant (column 2). The contrast can be explained by the behavior of prices (columns 3 and 4). For both goods and services, prices levels rise with income; but service prices are higher in the rich countries by much larger margins than goods prices. For example, the ratio of the price index for class S to class 1 countries is 4.6 for services, 16 Research Observer 1, no. 1 (January 1986) Table 8. Share of Services in GDP and Services Price Levels for Countries Grouped by Income Class, 1975 Share of services in GDP Income class, National prices (1) 1 2 3 4 5 6 22.2 28.4 27.3 25.6 36.8 43.9 World World average prices Goods (2) (3) 33.8 31.7 31.8 30.3 31.2 32.3 57.2 65.9 83.1 94.0 119.0 100.0 ~~~~Price levels Services (4) GDP (5) 20.7 34.1 41.2 46.3 94.6 100.0 40.6 51.7 64.7 73.5 107.5 100.0 a. See table 7. Source: KHS (1982), tables 6-10 and 6-12. only 2.1 for goods. Richer countries do not devote more of their real GDP to services than poorer ones. The cross-country income elasticities of demand for services and for goods are close to 1.0 (KHS 1983). However, if the calculations are confined to consumption rather than all of GDP, the income elasticities of demand are 1.13 for services and 0.93 for goods. The system of international comparisons developed through phase IV will include benchmark estimates for about sixty countries, most of them for only one or two reference dates. Extensions will clearly involve extending the system to all countries and providing annual estimates of at least real GDP per capita and the GDP price level. However, these extensions will be a slow process. The UNSO is finding it difficult to finance an adequate central staff, and some countries will lack the data to participate in future rounds of the icP. For nonbenchmark countries, estimates of real per capita GDP or of the price level can be made by one of several methods. The "reduced information" approach, experimented with at the World Bank, aims to produce close-to-benchmark results with data that meet two criteria: (1) they do not require special collection, but are at hand or easily made available; and (2) the PPPsthey yield are close to the average Pppfor GDP in the cases of the benchmark countries. Another approach is based on relationships found between real GDP per capita in benchmark countries and certain physical indicators such as per capita milk consumption (Beckerman 1966; Economic Commission for Europe 1980). These physical indicators are then used to estimate the real per capita GDP of Irving B. Kravis Filling the Gaps in the System of Comparisons 17 nonbenchmark countries. A third approach, often referred to as the "shortcut" method, is based on estimating for the set of benchmark countries an equation relating real GDP per capita (or the price level) to certain variables that are widely available for all countries. The equation is then used to estimate real per capita GDP (or the price level) for nonbenchmark countries. An example of such an equation based on the Phase III results is as follows (with t-ratios in parentheses): In r= 1.57 + 0.89 In n (7.0) (7.7) - 0.038 (In n) 2 (1.9) - 0.103 In OP R2 = 0.97 (2.1) SEE = 0.157 where r and n are indexes of real (Ppp-converted) and nominal (exchange-rate-converted) GDP per capita, with the United States = 100; OP (openness) is the sum of exports and imports divided by GDP. The rationale for including n among the independent variables rests on the earlier finding that price levels (PL) are positively correlated with r; this implies that n, the product of PL and r, is positively associated with r. The quadratic term picks up curvature in the relationship between n and r. A high degree of openness tends to raise the prices of abundant factors. In a low-income, labor-abundant country the result is likely to raise the price of services, which tend to be labor-intensive, and thus to raise the price level for GDP as a whole. With n held constant, a higher price level means a lower real income. In addition to estimates for nonbenchmark countries for the benchmark years, estimates are needed for nonbenchmark years. Although annual repetition of benchmark surveys would be burdensome, different components of the comparisons might be carried out at different times, completing the cycle over, say, five years. However, benchmark studies once every five years seem the most probable pattern. (Phase III was for 1975, phase IV for 1980, and phase V is being targeted for 1985.) At first sight, the problem of filling in the between-benchmark years seems simple. One has first to alter the benchmark year Ppp estimate by the percentage change in a country's implicit deflator relative to that of the numeraire country. If, for example, country A's Ppp was 10.0 pesos per dollar in the benchmark year of 1980, and its implicit deflator rose by 10 percent in 1981 while U.S. prices rose by 5 percent, the 1981 ppp would be 10.0 x (1.10/1.05) = 10.48. The country's 1981 real per capita income would then be obtained by dividing the new Ppr into the country's own-currency per capita GDP. However, closer inspection reveals some drawbacks in this approach. Country A's increase of 10 percent in prices is based on its own quantity weights, and the U.S. increase of 5 percent on U.S. weights. The relative change is applied to a 1980 Ppp for country A based on yet another set of quantity weights. This disadvantage could be moderated by extending into the nonbenchmark years PPps for individual components of GDP, not just GDP as a whole. Each Ppp would be used to convert a particular component to the numeraire 18 Research Observer 1, no. 1 (January 1986) currency and the results summedto get real per capita GDP. This has the benefit of restricting domestic weights to within-category effects; the aggregationof categoriescan be based on a common set of international weights applied to all countries. Several studies have applied these methods both to nonbenchmark countries and nonbenchmark years. One study set out a basic approach to a shortcut estimation method (KHS 1978b);another offered a refinement of the techniqueand provided annual estimates for 1950-80 (Summers and Heston 1984).Such efforts will have a firmer basis when the phase IV comparisonsfor sixty countriesbecome available. Despite the availabilityof these extrapolations from benchmark countries and years to nonbenchmark countries and years, widespread use has continued to be made of exchange rate conversions. The use of the exchangerate as a convertercan be justifiedonly on the ground that it is the best approximation to the Ppp that is available. The question is then whether exchange rate conversionsdo in fact come closer to benchmark estimatesthan the shortcut estimatesbased on Ppp concepts. Some newly available benchmark data for European and African countries from phase IV (Ward 1985,SOEC 1985a)provide an opportunity to make some preliminary tests of this issue. Specifically,we can match these new 1980benchmark figureswith the previous estimatesfor 1980 made on an exchange rate basis (see World Bank 1983)and with the estimates of Summers and Heston (1984) derived from shortcut methods. The new OECD phase IV estimatesof 1980 real GDP per capita included five countries for which no previous benchmark studies had been done.2 The OECD benchmark, World Bank, and shortcut estimates Table 9. Estimates of Real GDP Per Capita, 1980 Absolute deviation Country OECD benchmark (1) Shortcut' (2) World Bankb (3) Shortcut minus benchmark (4) Portugal Greece Finland Norway Canada 34.4 43.8 77.9 98.9 102.1 38 49 71 90 93 20 36 84 111 88 3.6 5.2 6.9 8.9 9.1 World Bank minus benchmark (5) 14.4 7.8 6.1 12.1 14.1 a. The shortcut method (Summersand Heston 1984) was applied to obtain 1975 estimates for these and other nonbenchmark countries. The estimates were then extrapolated to other years, including1980 as shown here. b. WorldBank (1983)estimates are for GNP. Sources: Column 1: Ward(1985); column 2: Summersand Heston (1984); and column 3: WorldBank(1983). Irving B. Kravis 19 are listed in table 9. In four of the five new benchmark countries, the shortcut method came closer to the benchmark estimate than the World Bank's estimates based on exchange rate conversions. Another comparison between World Bank and shortcut estimates is possible for fifteen African countries for which 1980 benchmark estimates were recently published by the SOEC (1985a). These will eventually be linked to the United States and other countries by the UNSO; at the moment, the only official version is a within-Africa comparison. In column 1 of table 10, the benchmark estimate of each country's real GDP per capita is expressed as an index number based on the unweighted average of the fifteen original indexes as 100. (The original indexes were based on the weighted average of the real GDP of the fifteen countries; the shift to unweighted averages is made here to facilitate comparisons with World Bank and shortcut estimates.) Comparing columns 4 and 5, showing the absolute deviations, it can be seen that in ten of the fifteen cases the shortcut methods come closer to the benchmark results; their Table 10. Comparison of Shortcut and World Bank Per Capita GDPEstimates with Benchmark Estimates, Fifteen Countries, 1980 Absolute deviations Country Ethiopia Mali Tanzania Malawi, Madagascar Kenyaa Senegal Zambiaa Nigeria Zimbabwe Cameroon Morocco Ivory Coast Bostwana Tunisia Unweighted meanb SOEC benchmark (1) Shortcut (2) World Bank (3) Shortcut minus benchmark (4) 33.5 39.0 42.6 48.0 66.2 74.3 79.8 85.2 104.2 104.2 106.0 139.6 159.5 185.8 232.0 100.0 38.1 27.6 54.3 37.5 60.2 49.8 75.7 77.1 171.6 103.6 103.3 137.4 163.5 186.2 214.1 100.0 21.7 30.1 45.2 31.8 58.5 65.2 70.2 97.0 145.5 130.4 122.1 138.8 185.6 157.2 210.7 100.0 4.6 11.4 11.7 10.5 6.0 24.5 4.1 8.1 67.4 0.6 2.7 2.2 4.0 0.4 17.9 - World Bank minus benchmark (5) 11.8 8.9 2.6 16.2 7.7 9.1 9.6 11.8 41.3 26.2 16.1 0.8 26.1 38.6 21.3 - a. 1975 benchmark estimates, which were available for these countries, extrapolated to 1980 by Summers and Heston (1984). b. The unweighted mean of the original estimates has been taken as 100 as a normalizing device for columns 1 to 3. Sources: Column 1: SOEC(1985a); column 2: Summers and Heston (1984); and column 3: World Bank (1983). 20 Research Observer 1, no. 1 (January 1986) J ; average absolute deviation is nearly 30 percent less than that of the World Bank estimates. The superiority of shortcut methods over exchange conversions is the more notable because all the fifteen countries had low incomes; Tunisia, the richest country in the set, has a real GDP per capita roughly one-fifth that of the United States. Even if shortcut methods were used to complete the IcP's coverage, an important gap in our ability to compare economic structures would remain. The system of comparisons that has been developed is based on a final expenditure breakdown of GDP, because that was the easier approach. However, much valuable information about the structure of the world economy-for example, the comparative productivity of given industries in different countries-requires comparisons of GDP by industry of origin. This work, begun by Paige and Bombach (1959), needs to be picked up again. It also requires careful price comparisons-which are more demanding, for an equal degree of disaggregation, than those of the final expenditure approach. There is little justification for the bland assumption that goods with the same name are of equal quality in different countries. Rice, for example, seems like a simple product whose price can easily be compared. But rice comes in numerous varieties which dominate different national markets. And the average bag of rice in a low-income country is likely to contain more impurities and broken kernels than in a rich country. International comparisons of quantities that ignore quality differences can be far off the mark. Although the icP reports have been favorably reviewed in the professional journals (Hill 1976; Isard 1983), they have also provoked doubts that the real-income gap between rich and poor countries is really reduced by so much. One objection has been to the use of world average prices to value the components of the countries' GDPS (Isenman 1980); it is claimed that world average prices are dominated by the large weights of the rich countries. The Gerschenkron effect thus comes into operation; price weights that are very different from those of the low-income countries push up their relative quantity indexes. While this argument is valid in principle, empirical investigation shows that its quantitative importance is small (Kravis 1984). For example, dividing the weights equally between the set of poor countries and the set of rich ones lowers the estimate of real per capita GDP for the eight lowest-income countries (class 1) in phase III by between 9 and 13 percent. This difference is much smaller than the one between the exchange-rate-converted GDP and almost any reasonably devised PPp-type estimate. (The phase III report put the average ERDI for the eight lowest-income countries at 2.6.) As this suggests, the narrowing of the income gap between poor and rich countries depends more on using a common set of average prices to value the components of the different countries' GDPS than on which Irving B. Kravis What Has Been Questioned 21 particular set is chosen. For example, applying Indian prices in a 1975 binary comparison with the United States yields an India/United States index of 4.1, twice the exchange-rate-converted index of 2.03. Another area of icP methods that has been questioned is the treatment of services. Most services, such as haircuts and telephone calls, can be compared in price. But some, like the services of teachers and government employees, are "comparison resistant"; they are in a sense "unpriceable," having no readily identifiable unit of output which can be priced. In the national accounts of individual countries, price and quantity changes of such services are tracked over time in terms of their inputs. The icP comparisons used the same approach over space. Phase III tried to correct for international differences in productivity of the providers of the services-that is, for medical personnel, and teachers (KHS 1982, chap. 5). Some allowances were also made for the use of capital inputs. These corrections to the raw quantity data were based on a careful sifting of the available evidence, but they cannot claim to be solid. Perhaps the most explicit challenge to the icP treatment of services is that of Maddison (1983). His estimates of 1965 real GDP per capita for ten developing countries are on average almost one-third less than the 1965 estimates extrapolated from icP results for 1975. Maddison says that the "most important" source of the difference is the treatment of services, but he does not offer any other explanation. Maddison's statistical procedures raise serious doubts. His estimates, based on a breakdown of GDP by industry of origin, utilized only secondary sources and were made without the benefit of careful price comparisons. Maddison focuses his criticisms on icp treatment of comparison resistant services and on what he calls "disguised services." For the former, he simply assumes that their productivity in different countries varies from one-quarter to three-quarters of the U.S. level, depending on productivity differentials in industry (Maddison 1970, p. 292). He offers no justification for this assumption. Even if Maddison's productivity ratios were correct, however, the change in the phase III results would be small. As Maddison recognizes, comparison resistant services accounted for only 10 percent of the GDP of most low-income countries; so if their true productivity were one-third or one-half of the iCP estimate, the estimate of real per capita GDP would be 5 to 7 percent smaller-tiny compared with the difference between exchange rate and Ppp conversions. Maddison ascribes greater importance to "disguised services," which he defines as the difference between the service industry share in GDP and the share of final product services in GDP. Substantively, Maddison seems to have in mind distributive services. He criticises the icP for failing to adjust for a larger amount of distributive service per unit of consumption in the rich countries, which would therefore lead to an overstate22 Research Observer 1, no. 1 (January 1986) ment of the relative real income of developing countries. Again, the figures suggest that the Maddison criticism is exaggerated. The share of distributive services in total services is probably in the 15 to 18 percent range for developing countries and below 15 percent for the developed countries.3 The true value of distributive services would have to be massively understated in order to swell the 15 percent by enough to raise real per capita incomes in rich countries and thereby significantly widen the spread between poor and rich countries.4 It is in any case doubtful that services alone can account for the bulk of the differences between the Maddison and phase III estimates. Services constitute a little under one-third of the real income of developing countries. For all of the difference between Maddison and icp estimates to be due to an icP exaggeration of service consumption in poor countries, the absorption of services in low-income countries would need to be zero. The icP results show that poor and rich countries divide their income between goods and services in roughly the same proportions. Critics who claim the icp estimates for service consumption in poor countries are too high are suggesting some perverse economic behavior. They imply that people in poor countries where goods are dear consume lots of them, while they buy only small amounts of relatively cheap services. People in rich countries also behave perversely, buying expensive services and economizing on cheap goods. The constant real share of the icP (column 2 in table 8) seems much more plausible. It may be added that the set of IcP outputs-prices, quantities, and real incomesperformed well econometrically, when fitted both to simple demand functions and to a complete demand system (KHS 1982, chap. 9). There is little doubt that purchasing power parities are the right way to convert GDP and its components from domestic currencies to dollars (the usual numeraire). Using standard exchange rates as converters produces biased results, because exchange rates systematically understate the purchasing power of the currencies of low-income countries. These points are now well understood and accepted. Nonetheless, exchange rates continue to be widely used as converters. One reason given for not using PPPs is that the icp has unresolved methodological problems. In fact, methodological improvements can and doubtless will be made, but the indexes of real per capita GDP are little changed by the use of alternative methods. In the European Community Ppp conversions, produced by the methods described above, are being used for policy purposes. Furthermore, it is not enough to claim that there are errors and uncertainties in the Ppp conversions to reject them in favor of exchange rate conversions; one has to ensure that exchange rates are better approximations to PPPs than are produced by the icP. Irving B. Kravis Concluding Comments 23 A second objection, that PPls were available for only a limited number of countries, will lose some of its force when the UNSO publishes benchmark comparisons for sixty countries later this year. More important, shortcut methods could be used to estimate Ppp figures for nonbenchmark countries. They are a better guide to real GDP per capita than figures obtained from exchange rate conversions. In addition, if benchmark studies can be repeated at five-year intervals, estimates for years between the benchmarks can be obtained by extrapolation. The world statistical system could therefore produce an annual set of estimates of real GDP per capita for all the countries of the world. Abstract Purchasing power parities (PPPs),this article confirms, are the correct converters for translating GDP and its components from own-currencies to dollars (the usual numeraire); the alternative measure, exchange rates, obscures the relationship between the quantity aggregates of different countries. Drawing on the reports of the United Nations International Comparison Project (IcP), the article contends that exchange rates systematically understate the purchasing power of the currencies of low-income countries and thus exaggerate the dispersion of national per capita incomes. Where full-scale (benchmark) ppi estimates are not available, estimates based on shortcut methods better approximate what the benchmark estimates would be than do the exchange rate conversions. The icP results also illuminate price and exchange rate relationships among countries by providing a measure of the difference in the levels of prices in different countries. iCP price comparisons for components of GDP make possible the analysis of comparative price and quantity structures of different countries and provide the raw materials for many types of analytical studies. Notes 1. The first three phases of the icP are reported in Kravis, Kenessey, Heston, and Summers 1975; Kravis, Heston, and Summers 1978a; and Kravis, Heston, and Summers 1982 (hereafter, KKHS and KHS). A report on phase IV is expected from the United Nations Statistical Office late in 1985. 2. They included thirteen other countries for which 1975 benchmarks were available from phase III. For these countries Summers and Heston had only to update from 1975 to 1980; they did not have to use shortcut methods. 3. Based on data for Brazil, Colombia, India, Korea, Mexico, Pakistan, Philippines, Spain, and Thailand; and for France, Germany, Italy, Japan, and the United Kingdom (United Nations 1984). 4. It is not clear whether Maddison considers that nondistributive services are also "disguised." In fact, they enter into the final product comparisons in appropriate ways. Some, such as government employee compensation, are also final products and enter directly. Others, such as financial services, enter into the prices of goods and services and are captured in the price comparisons. References Beckerman, Wilfred. 1966. International Comparisons of Real Income. Paris: Organisation for Economic Co-operation and Development (OECD)Development Center. Bhagwati, Jagdish N. 1984. "Why Are Services Cheaper in the Poor Countries?" Economic Journal 94 (June): 279-86. Board of Trade. 1908a. Report on an Enquiry by the Board of Trade into Working Class Rents, Housing and Retail Prices Together with the Rates of Wages in Certain Occupations 24 Research Observer 1, no. 1 (January 1986) in the Principal Industrial Towns of the United Kingdom. Cd. 3864. London: Printed for H.M. Stationery Office by Darling & Son. 1908b. German Towns. Cd. 4032. _.___ __ ._1909.French Towns. Cd. 4512. 1910. Towns of Belgium. Cd. 5065. ____.* . 1911. American Towns. Cd. 5609. Braithwaite, Stanley N. 1968. "Real Income Levels in Latin America." Review of Income and Wealth 14, no. 2 (June): 113-82. Clark, Colin. 1940. The Conditions of Economic Progress. London: Macmillan. Economic Commission for Europe. 1980. Economic Bulletin for Europe 31 (2). Gilbert, Milton, and Irving B. Kravis. 1954. An International Comparison of National Products and the Purchasing Power of Currencies: A Study of the United States, the United Kingdom, France, Germany, and Italy. Paris: Organisation for European Economic Cooperation (OEEC). Gilbert, Milton, and associates. 1958. Comparative National Products and Price Levels. Paris: OEEC. Hill, T. P. 1976. "Review of A System of International Comparisons." Economic Journal (March): 16144. IMF (International Monetary Fund). 1984a. International Financial Statistics 37, no. 7(July). . 1984b. Yearbook of International Financial Statistics. Washington, D.C. 11985.International Financial Statistics 38, no. 7(July). Isard, Peter. 1983. "Review of A System of International Comparisons." Journal of International Economics 15 (August): 177-81. Isenman, Paul. 1980. "Inter-Country Comparison of 'Real' (Ppp)Incomes: Revised Estimates and Unresolved Questions." World Development 8, no. 1 (January): 61-72. King, Gregory. 1936. Two Tracts. George C. Barnett, ed. Baltimore, Md.: Johns Hopkins University Press. Kravis, Irving B. 1984. "Comparative Studies of National Incomes and Prices." Journal of Economic Literature 22 (March): 1-39. Kravis, Irving B., Alan W. Heston, and Robert Summers. 1978a. International Comparisons of Real Product and Purchasing Power. Baltimore, Md.: Johns Hopkins University Press. -____1978b. "Real GDP Per Capita for More than One Hundred Countries-" Economic journal 88, no. 350 (June): 215-42. -_____ 1982. World Product and Income: International Comparisons of Real Gross Product. Baltimore, Md.: Johns Hopkins University Press. Kravis, Irving B., Zoltan Kenessey, Alan W. Heston, and Robert Summers. 1975. A System of International Comparisons of Gross Product and Purchasing Power. Baltimore, Md.: Johns Hopkins University Press. Kravis, Irving B., and Robert E. Lipsey. 1983. Towards an Explanation of National Price Levels. Special Studies in International Finance no. 52. Princeton, N.J.: Princeton University Press. - Forthcoming. "The Assessment of National Price Levels." In S. Arndt and D. Richardson, eds. Real-Financial Linkages in Open Economies. Washington, D.C.: American Enterprise Institute. Maddison, Angus. 1970. Economic Progress and Policy in Developing Countries. New York: Norton. -____1983. "A Comparison of the Levels of GDP Per Capita in Developed and Developing Countries, 1790-1980." Journal of Economic History (March): 27-41. Irving B. Kravis 25 Paige, Deborah, and Gottfried Bombach. 1959. A Comparison of National Output and Productivity in the United Kingdom and the United States. Paris: OEEC. Salazar-Carrillo, Jorge. 1973. "Price, Purchasing Power and Real Product Comparisons in Latin America." Review of Income and Wealth 19, no. 1 (March): 117-32. (Statistical Office of the European Community). 1983. Comparisons in Real Values of the Aggregates of ESA, 1980. Luxembourg: European Economic Communities (EEC). SOEC -_____.1985a. Comparison of Price Levels and Economic African Countries. Luxembourg: Aggregates: The Results for EEC. . 1985b. Comparison of National Accounts Aggregates between Israel and the European Community. Luxembourg: EEC. Summers, Robert. 1973. "International Comparisons Based upon Incomplete Data." Review of Income and Wealth 19, no. 1 (March): 1-16. Summers, Robert, and Alan W. Heston. 1984. "Improved International Comparisons of Real Product and Its Composition: 1950-80." Review of Income and Wealth 30, no. 2 (June): 207-62. United Nations. 1968. A System of National Accounts. Studies in Methods, ser. F, no. 2, rev. 3. New York: United Natons Statistical Office. United Nations, Economic and Social Council, Statistical Commission and Economic Commission for Europe, Conference of European Statisticians. Summary of the Results of the European Comparison Programme CES/S14 (May). New York. Ward, Michael. 1985. Purchasing Power Parities and Real Expenditures in the OECD. Paris: OECD. World Bank. 1983. 1983 World Bank Atlas. Washington, D.C. 26 Research Observer 1, no. 1 (January 1986) 4