PPP estimation at the basic-heading level
PPP aggregation above the basic headings
PPPs for nonparticipating economies
Overview
In general, price relatives are first computed at the individual item level within each basic heading for each pair of economies being compared. Suppose three economies—A, B, and C—price two kinds of rice under the rice basic heading. For each kind of rice, there are three price relatives: PB/PA; PC/PA; and PC/PB. The basic-heading PPP for each pair of economies can be computed directly by taking the geometric mean of the price relatives between them for the two kinds of rice. This is a bilateral comparison. The PPP between economies B and A can be computed indirectly: PPP C/A × PPP B/C = PPP B/A. The use of both direct and indirect PPPs is a multilateral comparison. This means that the PPPs between any two economies are affected by their respective PPPs with other economies in the comparison. As a result, a change in the mix of economies included in the comparison will also change the PPPs between any two economies.
Different methods can be used to compute multilateral PPPs. The choice of method is based on two basic properties: transitivity and base country invariance. PPPs are transitive when the PPP between any two economies is the same whether it is computed directly or indirectly through a third economy. PPPs are base country invariant if the PPP between any two economies is the same regardless of the choice of base economy. These properties apply for every computational step: computing basic heading PPPs between economies, aggregating basic-heading PPPs to the within-region GDP PPPs, linking PPPs across regions, and finally computing global PPPs.
Another property underlying the computational steps to obtain PPPs is that economies are treated equally regardless of the size of their GDP. Weights based on basic-heading expenditures are used in the methodology to weight a group of basic headings to an aggregate level. Therefore, PPPs are first weighted using economy A’s weights (Laspeyres index) and then weighted again using economy B’s weights (Paasche index). Each index provides a weighted average of the PPP between economy A and economy B. To maintain symmetry, the geometric mean is taken of the two aggregated PPPs for every pair of economies in the comparison. The result is the Fisher index. For each pair of economies, the multilateral PPP is the geometric mean of the direct and indirect Fisher indexes.
This method, however, does not satisfy the additivity requirement. Additivity occurs when the sum of the PPP-based expenditures of the basic headings constituting an aggregate equals the PPP-based expenditures based on the PPPs for the aggregate. Additive methods have the disadvantage of giving more weight to the relative prices of the larger, more developed economies. As a result, the PPP-based expenditures of smaller, less developed economies become artificially larger and move closer to the PPP-based expenditures of rich economies. This is known as the Gerschenkron effect. For the uses of ICP PPPs, such as for poverty analysis, nonadditive methods that avoid the Gerschenkron effect are preferred.
Fixity is yet another concept that determines the methods used. The fixity concept means that the relative volume—the ratio of PPP-based expenditures—between any pair of economies in a region remains the same after the regional results have been combined into a set of global results, including all economies.
PPP estimation at the basic-heading level
The PPP estimation process begins with the participating economies collecting prices for items chosen from a common list of precisely defined items. These common lists include both regional items, priced in a specific region, as well as global core items as set out in the GCL, priced in all ICP regions. These two sets of prices cover the whole range of final goods and services included in GDP: household consumption expenditures, government consumption expenditures, and gross capital formation expenditures. In the current ICP methodology, basic-heading PPPs are first estimated at the regional level by the regional implementing agencies, and subsequently linked at the global level through the global linking procedure outlined in the following sub-section.
Two basic methods are used in the ICP to calculate basic-heading PPPs. The first approach is based on the Jevons index made transitive by the Gini-Éltető-Köves-Szulc (GEKS) method, which transforms bilateral PPPs into multilateral PPPs. The other method uses a regression model known as the country product dummy (CPD) method, which directly estimates PPPs that are transitive and base-country invariant in one step. The results obtained by both methods are the same if every economy prices every item, and no item level weighting is applied.
Both methods can be modified to include weighting at the item level. As there are no expenditure weights below the basic-heading level, participating economies are asked to use their expert judgment to determine which items would have the larger expenditure shares within each basic heading. For instance, if basmati rice is deemed to be important within the rice basic heading, then basmati rice will have a higher weight in the PPP calculation for that economy. This weighting at the item level is only applied in the household consumption survey. The modified methods are simply known as GEKS* and weighted CPD (CPD-W). However, the results provided by the two methods differ. This is because these methods provide different results in their unweighted form, and in addition the GEKS* method assigns a weight of 1 to the most important items and a weight of zero to the least important items, while the current CPD-W method assigns weights of 3 and 1, respectively.
Reference PPPs
For some basic headings, expenditure data exist, but price collection is considered too expensive or time-consuming or the price data are unreliable. For these basic headings, reference PPPs are used, and they can be categorized as follows:
• Price-based reference PPPs, specific or neutral
• Market exchange rate reference PPPs.
Price-based reference PPPs form the majority of all reference PPPs used. They are based on the PPPs of other basic headings for which prices were collected. These PPPs are referred to as specific reference PPPs. They may refer to the PPP for a single basic heading or an average of the PPPs for several basic headings. In the latter case, they will be GEKS averages of the selected PPPs weighted by expenditure shares. In other cases, reference PPPs are the PPPs of a large group of basic headings, such as all the basic headings under gross capital formation for which prices have been collected. In this case, the purpose is to ensure that the use of a reference PPP will not change the PPP for that larger group. These are referred to as neutral reference PPPs because the intention is for them to have no impact on the PPPs of the larger group of basic headings.
Market exchange rate reference PPPs are used for the following four basic headings: net purchases abroad, acquisitions less disposals of valuables, exports of goods and services, and imports of goods and services. For the calculation of PPPs for exports and imports, it would be prohibitively expensive to collect prices in the same manner as for other items of final expenditure, and the use of market exchange rates can be justified on practical grounds.
PPP aggregation above the basic headings
Once PPPs are computed for each basic heading for all participating economies within a region, they are used as inputs for the higher levels of aggregation using the GEKS method. This method consists of two steps:
· Step 1. Aggregate the basic-heading PPPs using the national accounts expenditure structures to obtain the bilateral PPPs for each pair of economies. Usually Fisher-type PPPs will be used, which require calculating both Paasche-type and Laspeyres-type PPPs. The Fisher-type binary PPPs will simply be the geometric mean of the Laspeyres-type and Paasche-type PPPs.
· Step 2. Average the Fisher-type PPPs obtained to arrive at the final vector of GEKS PPPs. The GEKS calculations are performed separately for each aggregation level and for each category within a given aggregation level.
As previously discussed, the GEKS method is not additive. While additivity may be desirable for some uses, this property has the disadvantage of rendering the PPP-based expenditure of low-income economies artificially larger. In applications of PPPs such as global poverty estimates, this might lead to unwanted biases, and therefore a non-additive method is preferred. The resulting linked global PPPs maintain the fixity of the regional results.
Calculation of global PPPs
Standard linking approach
At the global level, regional PPPs are linked to form a global set of PPPs and measures of price and volume relatives. In order to link the regional basic-heading PPPs for each participating economy, the so-called interregional linking factors are calculated based on the prices of global items from the GCL collected in all ICP regions.
The GEKS aggregation method, with further redistribution of regional volumes in accordance with an economy’s regional volume shares (known as the country aggregation with redistribution [CAR] procedure), is used to obtain PPP-based expenditures (hereafter referred to as volumes) and aggregated PPPs with regional fixity. All economies in the standard ICP regions participated simultaneously and equally in the global aggregation using the GEKS method.
Linking at the basic-heading level involves the following four steps:
· Step A1. Calculate the regional basic-heading PPPs based on both the regional and GCL items. The regional basic-heading PPPs must follow the ICP expenditure classification.
· Step A2. Convert all GCL item prices in local currency to a common regional currency by using the regional basic-heading PPPs from step A1.
· Step A3. Calculate the interregional linking factors by applying the weighted Region-Product-Dummy regression method (a modified version of the CPD-W) to the GCL item prices from step A2.[i]
· Step A4. Multiply each economy’s regional basic-heading PPP from step 1 by the interregional linking factor resulting from step A3. The PPPs derived from this step are the global basic-heading PPPs with regional fixity.
The estimation of PPPs above the basic-heading level involves the following six steps:
· Step B1. Calculate the regional PPPs by applying the GEKS aggregation to the regional basic-heading PPPs from step A1 and the national accounts basic-heading expenditures in local currency units for each level of aggregation up to GDP.
· Step B2. Obtain an economy’s volume shares in the regional results for each level of aggregation up to GDP using data from step B1.
· Step B3. Calculate an economy’s aggregated PPPs in the global comparison by applying the unrestricted GEKS aggregation to the global basic-heading PPPs derived from step A4 and the national accounts basic-heading expenditures in local currency for each level of aggregation up to GDP.
· Step B4. Obtain the regional volume totals in the global comparison by summing up the total volumes for individual economies for each region derived from step B3 for each level of aggregation up to GDP.
· Step B5. Distribute the regional volume totals from step B4 among the economies in the regions according to the economy shares in the regional results derived from step B2 in order to uphold regional fixity for each level of aggregation up to GDP.
· Step B6. Calculate the aggregated global PPPs indirectly by dividing the economies’ nominal expenditures by the volumes derived from step B5 for each level of aggregation up to GDP.
Nonstandard linking approaches
While the standard linking approach applies to most household consumption basic headings for all regions, other basic headings require different approaches due to the specific nature of their surveys.
Housing
All economies participating in ICP 2021 were asked to collect annual average rents for a global list of dwelling types, as well as dwelling stock data: for example, number of dwellings, usable surface area in square meters, and information on three quality indicators—availability of electricity, water supply, and an in-house toilet. National accounts expenditure data on actual and imputed rentals were also collected. However, not all economies were able to report rents and dwelling stock data, and some economies were only able to provide rents for a limited subset of dwelling types or limited dwelling stock data. Each regional coordinating agency decided on the best way to use the collected data for its region.
Rental data were used to link the Africa, Latin America and the Caribbean, and Western Asia regions. The linking factors for these three regions were calculated using the CPD method. The Asia and the Pacific, the Commonwealth of Independent States, and the Eurostat-OECD regions were linked to each other and to the rest of the world using the dwelling stock approach. This involved calculating housing volumes and housing quality indexes to adjust total per capita expenditure on housing and regional housing PPPs in order to render them comparable globally.
Government compensation
The ICP approach for estimating PPPs for government services is based on an input approach in which compensation data for selected government occupations are collected across economies. Given the differences in productivity, adjustment factors are applied to account for differences in capital per worker. These adjustment factors are based on differences in countrywide levels of capital per worker and their estimated contribution to output using the aggregate share of capital income in GDP from the Penn World Tables.[ii] Adjustments were made to the PPPs for government in the Africa, Asia and the Pacific, Latin America and the Caribbean, and Western Asia regions. No productivity adjustments were applied to the Eurostat-OECD and Commonwealth of Independent States regions because differences in labor productivity within each of those regions were considered to be relatively low. However, productivity adjustments were made to all regions when the interregional linking factors were estimated to maintain consistency in the global comparison.
Education and health
Eurostat-OECD economies did not report prices nor regional PPPs for private education. As a result, the global linking factors for this basic heading were adjusted on the basis of the global linking factors for the “Production of Education Services” aggregate under Individual Consumption Expenditure by Government.
For all other education and health components, Eurostat-OECD followed an output approach to calculating their PPPs. Thus, it was necessary to link their PPPs to those of other ICP regions that follow the input approach to calculating their PPPs for health and education.
For education, since Eurostat-OECD does not collect expenditure weights at the basic-heading level, a simplified weighting system was used, based on information on education expenditure structures in the OECD–United Nations Educational, Scientific, and Cultural Organization (UNESCO) database. Data on average compensation of employees in education in Eurostat-OECD economies were used to bridge their output-based PPPs to those of other ICP regions that follow the input approach.
For health, since Eurostat-OECD does not collect expenditure weights at the basic-heading level, the System of Health Accounts (SHA) was used, as it offers a breakdown with a significant overlap with basic headings in the ICP. For comparison purposes, it was necessary to combine the basic headings for household, NPISHs, and government consumption for other ICP regions because the SHA does not distinguish between these different types of expenditures.
Construction and civil engineering
Due to differences in the regional approaches for estimating the construction and civil engineering PPPs, the linking of these PPPs at the global level required a special linking approach. Regional item prices in local currency were used to calculate the nine construction sub-heading PPPs using the CPD method for the economies participating in the global linking, which were then aggregated in the respective three construction basic-headings PPPs (residential buildings, nonresidential buildings, and civil engineering works). Linking factors for the three construction basic headings were calculated as geometric means of the aggregated PPPs for the economies in a region.
The regional Eurostat-OECD approach to estimating construction and civil engineering PPPs differs from the ICP approach. In order to link construction and civil engineering PPPs between the Eurostat-OECD economies and the rest of the world, eleven economies in the Eurostat-OECD comparison conducted the ICP survey.
Similarly, the regional CIS approach to estimating construction and civil engineering PPPs differs from the ICP approach. A procedure called the Resource-Technology Model (RTM) method was used to compare the construction output of all participating CIS economies. The RTM method consists of defining a standard facility and the notional cost of its construction. It is based on the accounting of prices for resources and corresponds to the information capabilities and the resource-based method of determining the cost of construction used in construction and design organizations in CIS economies. In order to link construction and civil engineering PPPs between the CIS economies and the rest of the world, all economies in the CIS comparison conducted the ICP survey.
Special linking cases
Special linking calculations were conducted in tandem with the calculation of global results. This included the linking of Caribbean economies with Latin American economies, the treatment of dual-participation economies in the Africa and Western Asia regions, and the inclusion of the special participation economies, Georgia and Ukraine.
Linking the Caribbean with Latin America
The method used to link the Caribbean economies with Latin America included three steps. First, PPPs were produced for the full set of Latin American and Caribbean economies. Second, separate subregional PPP aggregations were carried out, one for the Latin American economies and another for the Caribbean economies. As a third and final step, the PPPs in the first step were reindexed in accordance with results from the second step in order to maintain the fixity of both Latin American and Caribbean PPPs.
Linking dual-participation and single economies
The Arab Republic of Egypt, Mauritania, Morocco, Sudan, and Tunisia participated in both the Africa and Western Asia comparisons. Published global PPPs for these economies are geometric means of their respective global PPPs in the Africa and Western Asia comparisons. The single economies Georgia and Ukraine were included as guest participants in the Eurostat-OECD comparison and mainly followed the Eurostat-OECD methodology.
[i] CPD is used for selected basic headings under household consumption (housing and education) and for all nonhousehold consumption expenditure components.
[ii] The Penn World Tables are a data set of National Accounts developed and maintained by the University of California, Davis, and the University of Groningen to measure GDP across economies.
PPPs for nonparticipating economies
In the 2021 cycle, 176 economies participated in the ICP. Other economies did not participate in the comparison for various reasons, including civil unrest, lack of resources, or lack of capacity. Although these nonparticipating economies account for a small share of the world economy and population, it is important to estimate the size of these economies for various purposes.
The method used for imputing PPPs for nonparticipating economies uses three regression models, one based on the price level index (PLI) for GDP, one based on the PLI for individual consumption expenditure by households, including NPISHs and the third, introduced for the first time in the ICP 2021 cycle, based on the PLI for actual individual consumption. The three regressions are estimated jointly using the “seemingly unrelated regression” method. The required explanatory variables are as follows: GDP per capita in US dollars based on market exchange rates; imports as a share of GDP; exports as a share of GDP; and the age dependency ratio. Dummy variables are required for the Sub-Saharan African economies, the Eurostat-OECD PPP Programme participants, island economies, and landlocked economies. Interaction terms of GDP per capita in US dollars based on market exchange rates and the dummy variables are also required.
Interpolated time series PPPs
For the years between reference years 2017 and 2021—namely, 2018, 2019, and 2020—PPPs were calculated based on an approach in which basic-heading PPPs were first interpolated between the reference years and subsequently aggregated using the standard GEKS method. In addition, regional PPPs between the reference years, where available, were incorporated using the country approach with redistribution (CAR) procedure. The resulting annual PPPs uphold the same properties of base-country invariance and fixity as the PPPs from reference year comparisons.
The data required to construct time series PPPs included global PPPs for the two reference years; regional PPPs between the reference years, where available; national accounts deflators and consumer price indexes; national accounts expenditures at current prices in local currency units; market exchange rates; and population. Where participating economies could not provide any of the above information required for time series PPPs estimates, data were complemented with internal estimates by the ICP global implementing agency.