A. Background

3. Purchasing power parity (PPP) factors allow for comparing the size of economies, adjusting for differences in price levels across countries. PPP factors, also referred to as PPP exchange rates, are price relatives that express the rate of one currency in terms of another for a given basket of goods and services. Like market exchange rates, they are used to convert national nominal GDP data, expressed in local currency, into a common currency (the U.S. dollar in practice). However, as price deflators and currency converters, they differ from market exchange rates in an important way, because they eliminate the effect of the differences in price levels between economies, thereby allowing volume-comparisons of GDP. Such price level differences between different countries can be important, as items that are only domestically produced and consumed (“non-tradables”), particularly services, may have persistently differing prices from country to country when measured in a common currency.¹ In particular, price levels of non-tradables tend to be higher in advanced economies (AEs) than in EMDCs, largely reflecting higher wages and higher productivity in AEs (known as the Balassa Samuelson effect). PPP GDP data comparisons correct for such price level differences and reflect merely the differences in volumes of goods and services as a measure of the relative size of economies.²

4. PPP data are widely used, including in the Fund’s current quota formula. WEO PPP GDP data on a country level are used as one input in the Fund’s quota formula to arrive at Fund members’ calculated quota shares. PPP data are also used to analyze welfare (PPP GDP per capita), the relative size of economies (PPP GDP), material well-being (per capita consumption in PPP terms), labor productivity (PPP GDP per hour worked), energy efficiency (energy used to produce one dollar of PPP-adjusted GDP) among others. The Fund’s WEO uses PPP GDP weights to calculate a measure of global GDP growth. PPP data also play an important role in the monitoring of several of the Sustainable Development Goals. They are widely used by other international organizations as well, including for poverty assessments (World Bank), allocation of structural and cohesion funds (European Commission), in the construction of the Human Development Index (United Nations Development Program), and assessment of market size in global competitiveness (The Global Forum).

5. PPP factors are periodically collected through the International Comparison Program (ICP), with the latest collection effort (ICP 2017 round) released this year. The ICP, a very elaborate global statistical initiative led by the World Bank under the umbrella of the United Nations Statistical Commission, has the objective of providing comparable price and volume measures of the expenditure-side GDP and its components, around the world. Price and GDP expenditure data are collected under a partnership with international, regional, and national agencies through periodic surveys, currently conducted every six years. The periodic surveys aim to produce regional comparisons that can be aggregated into a single global comparison for a given reference year. The previous survey dated back to 2011 and used 2011 as a reference year. In May 2020, the results from the latest update were published using 2017 as the new reference year and covering 176 economies that participated in the program. Also, the revised results for 2011 and PPP estimates for 2012–16 were subsequently released.³

6. The ICP 2017 round builds upon an effort that has been conducted for several decades. The first PPP estimates date back to the work in the 1940s and 50s of the British economist Colin Clarke. In 1968, the International Comparison Project was launched as a modest research project jointly conducted by the United Nations Statistical Division and the International Comparisons Unit of the University of Pennsylvania. It has since become a permanent global statistical program. Nine ICP rounds have been conducted to date (Text Table), with coverage increasing from 10 to 176 economies, making the ICP the largest international statistical initiative, covering seven geographic regions.

History of the ICP

Source: World Bank, 2020, FIN staff calculations

History of the ICP

Year	Economy coverage
1970	10
1973	16
1975	34
1980	60
1985	64
1993	115
2005	146
2011	199
2017	176

Source: World Bank, 2020, FIN staff calculations

View Table

History of the ICP

Year	Economy coverage
1970	10
1973	16
1975	34
1980	60
1985	64
1993	115
2005	146
2011	199
2017	176

Source: World Bank, 2020, FIN staff calculations

7. The World Bank has been the global coordinator since 1993. A stronger governance structure was put in place with the 2005 round, establishing the ICP Global Office at the World Bank, which oversees regional agencies which in turn coordinate data collection by national agencies. This allowed greater consistency across regions and a means to link regional comparisons in one global comparison (the “ring comparison technique”). The IMF participates on the Executive Board, Technical Advisory Group, and Inter-Agency Coordination Group of the ICP, as well as on several methodological Task Forces. In addition, the IMF has been a financial contributor to the program since the 2005 round—contributing nearly $1 million over three years. The next ICP comparison is expected to be conducted for the reference year 2021.⁴

8. The compilation of ICP PPP data follows a series of steps.⁵ The conceptual framework for the ICP comparison is guided by the System of National Accounts (SNA) definition of expenditure-based GDP, where the ICP collects prices for the expenditure components of GDP. First, expenditures are broken up into 155 basic headings, which are the building blocks for the ICP comparison. Second, within each basic heading, representative items are selected, priced, and compared. Surveys to collect price information include a global core list of items, and in addition, each region also has its own list of region-specific representative items (e.g., regionally consumed items). The comparison of prices at each basic heading is the lowest level at which PPPs are calculated. The third step involves aggregation, first within regions for regional comparisons and, ultimately, across regions for the global comparison using factors that allow regions to be linked (“interregional linking factors”), which are calculated based on the global core list of items.

9. ICP factors play a key role in the calculation of PPP GDP, but since they are not available for all years, they are complemented by estimates. To calculate PPP GDP for quota purposes, the PPP price indices are typically obtained from the WEO database, which uses the results from the ICP survey, along with estimates. The WEO PPP-based GDP is derived by dividing a country’s nominal GDP expressed in local currency by its PPP factor relative to the numeraire country, which is the United States. PPP factors from the ICP survey are available for the reference year (see also ¶4); ⁶ for other years where no ICP data are available, the WEO estimates PPP GDP data by applying the relative growth of the GDP deflator of a given member country vis-à-vis the U.S. GDP deflator to the previous year’s PPP factor.

B. The 2017 ICP Round and PPP Data Quality

10. Coverage in the 2017 ICP round declined somewhat compared with the 2011 survey. A total of 176 economies (of which 163 are Fund member countries) participated in the 2017 round compared to 199 economies (of which 175 are Fund member countries) in 2011 (text table).⁷ In addition, some economies changed regional grouping, for instance, countries that recently obtained OECD membership joined the Eurostat-OECD regional group. For a number of non-participants, the ICP estimated PPPs using regression methods.⁸

11. The 2017 round maintained the innovations introduced in the 2011 cycle. The 2011 cycle had introduced major methodological updates, including a new way for aggregating regional results and linking them with global results, building on the strong progress made during the 2005 ICP round. The United Nations Statistical Commission had recommended that no methodological changes be introduced for the ICP 2017 cycle―in order to maintain comparability over time and allow for producing reliable PPP timeseries. Consequently, there was a particular focus on ensuring the consistency of the ICP 2017 methodology with the ICP 2011 methodology, and to limit any possible methodological change.

12. The 2017 round did, however, introduce data quality improvements, and established a fully documented and more transparent process for producing ICP results. Enhanced procedures were introduced for price surveys; data editing and validation; and some computation methods, including for dealing with comparison-resistant sectors such as housing, the government expenditure on health and education, machinery and equipment, and construction. In addition, the 2017 cycle includes revised estimates of GDP which align it with the 2008 System of National Accounts (SNA). The GDP estimates included in the 2005 and 2011 cycle were based on the 1993 SNA conceptual framework.

13. PPP data are broadly comparable in quality to the national GDP and price statistics from which they are constructed, and to the other data used in the quota formula. As noted in earlier quota data update reports, all economic statistics, unless based on a perfectly measured, fully covered population of units, are estimators.⁹ The statistics produced by the ICP do not differ in this regard from other economic statistics and, indeed, their properties depend on the properties of the GDP and price estimators underlying them and the methodology employed to link the country estimators. Therefore, overall, and given the further quality improvements introduced in the 2017 ICP round, staff views that PPP data are as reliable as the price and national GDP statistics from which they are constructed.

C. Comparing the 2017 ICP results with extrapolated estimates based on 2011 ICP survey

14. This section compares the results of the newly published PPP factors with earlier estimates. As noted (see ¶9), for the years where no ICP survey data are available, PPP data are estimated using extrapolation from the most recent ICP survey, using the relative growth in GDP deflators of a given country vis-à-vis the numeraire country, as is done under the WEO methodology. While such extrapolation methodology is robust, the estimates based on extrapolation, however, should not be expected to produce PPPs that match those from a new ICP survey (McCarthy, 2013).¹⁰ Revisions to extrapolated series inevitably occur with each ICP cycle, as discussed in some detail below. This section compares the differences between PPP factors obtained from the 2017 ICP survey results and the extrapolated estimates for the same year based on the 2011 ICP survey (henceforth “revisions”) and considers a number or reasons that can help explain the differences.

Conceptual Framework

15. It is important to note that periodic updates of the ICP benchmark, and the revisions vis-à-vis the extrapolated PPP estimates they imply, are both expected and common, while improving upon the quality of PPP data. The compilation of the ICP benchmarks involves international coordinated data collection activities and an intensive data editing and imputation process that is absent in the non-benchmark years. The quality of the ICP benchmarks is higher than the extrapolated values and therefore significant revisions vis-à-vis the extrapolated PPP estimates are both expected and common, with the latest ICP round being no exception. Figure 1 illustrates the revisions resulting from updates in ICP’s benchmarks in the last three rounds (2005, 2011, and 2017). While revisions have become smaller, and more closely centered around one (Figure 1), they remain substantial, as expected.

View Full Size

Ratio of the updated ICP Factor to the Extrapolated Factor, Results from the Last Three ICP Rounds (2005, 2011, and 2017)

Citation: Policy Papers 2021, 016; 10.5089/9781513571492.007.A001

Source: ICP, WEO, and Fund staff calculations.Note: For each round, excludes extreme values, defined as those below (above) the second (98^th) percentile.

• Download Figure
• Download figure as PowerPoint slide

16. The literature has attributed revisions to a variety of factors. A number of studies have sought to explain the sizeable revisions associated with new ICP rounds, pointing to the role of dissimilar economic structures and patterns of spending being compared and the role of large changes in underling relative prices (Deaton and Aten, 2017, see also Box 1),¹¹ while many other practical considerations also play a role (McCarthy, 2013 see also Box 1). The analysis below draws on these earlier findings.

^{Box 1.}

Sources of Divergences Between Extrapolated and Updated ICP Factors – Highlights from the Literature

This box summarizes key findings of Deaton and Aten (2017) and McCarthy (2013) that help explain why updated ICP factors can result in revisions vis-à-vis extrapolated estimates.

Deaton and Aten (2017) provide a simple canonical two-country framework to describe main sources of discrepancies between the ICP results and the extrapolation. They assume that PPP₂ represents the PPP index of country 2, relative to a numeraire country 1, Pi stands for the price index of country i, which is calculated as the weighted sum of prices and shares that are contained in vectors p_i and si respectively. Then, using the Törnqvist index to express the PPPs, in natural logarithms as:

$$\begin{array}{cc}\ln PP{P}_2=0.5(s_2+s_1)\text{'}(\ln p_2-\ln p_1)& \left(1\right)\end{array}$$

And, for simplicity, assuming no changes in share over time, the following can be shown to hold:

$$\begin{array}{cc}d\ln PP{P}_2=(d\ln P_2-d\ln P_1)-0.5(s_2-s_1)\text{'}(d\ln p_2+d\ln p_1)& \left(2\right)\end{array}$$

where the left-hand side variable represents the change in PPP of country 2, relative to a numeraire; the first term on the right-hand side represents the extrapolation, indicating the relation with the relative change in price indices Pi of the two countries; the second term on the right-hand side is the discrepancy: it’s a function of two things: (1) “changes in underlying relative prices” and (2) differences in consumption patterns (or “structure of spending”) between the two countries.

Using equation (2), they show that extrapolated and updated PPP factors can be different for a number of reasons. In particular, they show that the discrepancy factor is zero if (i) there are no relative price changes in either country; (ii) if the structures of spending are the same; and (iii) if the differences in spending structures are uncorrelated with average changes in relative prices. Hence, for countries with very different patterns of consumption, or with different trends in domestic inflation, the discrepancy could be large.

In addition, McCarthy (2013) argued that a number of other practical considerations can play a role as well. McCarthy notes that, although some reasonable results may be obtained provided that some very restrictive assumptions are met, in practice, these assumptions are unlikely to apply, and a number of other practical considerations will result in discrepancies between the ICP results and the extrapolation. These include:

• GDP revisions. Significant GDP revisions occur when a country undertakes a complete reassessment of the data in the national accounts and the assumptions involved in combining various data sets. As a result, large inconsistencies could arise between the GDP estimates in a time series compared with those provided by the ICP.

• Product choice and quality change. The products to be priced in the ICP are defined to ensure comparability between countries, while the products used in estimating the volumes in a country’s national accounts, and in a country’s time series price indices, are selected on the basis that they are the most representative products available in a country. In addition, a broader set of prices (products) could be used in a country’s time series price indices than the set that can be included in the ICP. Further, the prices in a country’s time series price index could be adjusted for quality changes over time, with different countries using different methods, with the result that the quality-adjusted time series are not consistent across countries;

• Economic structure. An assumption underlying the technique of extrapolating PPPs at the level of GDP is that the structure of each country’s economy is similar to that of the numeraire country (e.g., the United States) and is changing in a similar way over time. In practice, the structures of different countries’ economies differ significantly, particularly when developing economies are being compared with a developed economy. Economies may also develop differently to one another over time;

Other. (i) Changes in the terms of trade are treated as a volume effect in the ICP, but they are reflected in the GDP deflators, which are used to extrapolate PPPs, as a price effect. (ii) In the national accounts, very few countries adjust their volumes of non-market services for productivity changes. Therefore, differences in productivity over time in different countries will be reflected in the GDP deflators as part of the price changes, while annual productivity changes are less likely to be reflected in price data from ICP surveys; and often “an assumption that productivity is identical in all the countries in a comparison generally has to be made between countries in calculating PPPs” (McCarty 2011, p.9). The differences in how productivity changes are captured could further contribute to discrepancies between extrapolated series and new ICP benchmark estimates. See: McCarthy, Paul, 2011, Extrapolating PPPs and comparing ICP benchmark results. 6th Technical Advisory Group Meeting October 3–4, 2011 Washington DC. Mimeo.