Expenditure Weights for PPP Aggregations

Prepared by

Barbro Hexeberg, Chandrakant Patel and Thilak Ranaweera

1. Background

Lack of sufficiently detailed expenditure weights of good quality is one major problem in the estimation of purchasing power parities (PPP) under the International Comparison Programme (ICP). So far, most work relating to ICP have been on the development of aggregation methods and the collection of prices, as the expenditure weights in principle should be readily available from the national accounts (NA). However, for many countries expenditures on gross domestic product (GDP) are simply not available at a sufficiently detailed level, if available at all. Thus, the aim of this paper is to look into alternative sources/methods for the derivation of expenditure weights, as the quality of the aggregated PPP will depend on the quality of the weights as well as on the quality of the price data and the aggregation methods applied. The paper describes past practice when expenditure data were not submitted by the countries, discusses problems and suggests solutions, and illustrates these problems and solutions by portraying six countries.

For the ICP, GDP is compared across countries by expenditure category breakdown, that is actual consumption by households (consumption expenditures by households and non-profit institutions serving households, plus individual consumption expenditures by government), collective consumption by government, gross fixed capital formation, changes in inventories, acquisitions less disposals of valuables, and net exports.

NA expenditures need to be reported at the level of 150 to 250 basic headings or expenditure categories (the number of required basic headings varies between regions), as the weights are used to combine basic heading parities in order to arrive at PPPs for aggregated measures such as GDP.

Table 1 Expenditure breakdown as required for the ICP[1]

GDP

/ Consumption exp. by households
(incl. NPISH) / Consumption exp. by government / Gross fixed capital formation / Changes in inventories / Acquisitions of valuables, net / Net exports
Level of required details / Breakdown by purpose / product-groups (CPC or similar)
117 or more ‘basic headings’ / Breakdown by purpose (COFOG) – to separate individual and collective consumption.
In addition, collective consumption should be broken down by cost: Compensation of employees; Intermediate consumption; and Consumption of fixed capital / Breakdown by type of asset
15-20 ‘basic headings’ / No breakdown required / No breakdown required
(Not included in the 1968 SNA nor the list of expenditures needed for ICP) / No breakdown required

Out of the 190 (+/-) countries reporting NA statistics to the United Nations (UN), approximately 130[2] countries report GDP by expenditure, while only 40+ (of which 22 are OECD countries) private final consumption expenditure by purpose (COICOP). Furthermore, the detailed breakdown in the UN database refers to two digit COICOP level (12 purposes), while the required reporting level for the ICP (basic heading level) is four digit level (app. 120 purposes) supplemented with even more detailed product-information for certain groups.[3] Therefore, national statistical offices will only be able to extract this information directly from their NA if detailed supply-use tables are part of their national accounts. Much of the same story holds for government consumption expenditures. While government expenditures by purpose (COFOG) is needed at a two digit COFOG level, no more than 35 countries (of which 10 are OECD countries) provide a breakdown by one digit COFOG.

International guidelines for compiling national accounts are designed to yield internationally comparable measures by including production and of all goods and services (with a few exceptions) whether or not they are produced for the market, for own use, or provided to others free of charge. In particular, informal, illegal and subsistence should all be included. In practice, however, international comparability is hampered by the use of different international standards (1968SNA versus 1993SNA) and by differences in compilation practices, in coverage and in quality of the estimates. The latter is the most severe problem, an may result in systematic biases in the estimates; it is generally thought that GDP in developing countries is underestimated compared to that of developing countries. Nevertheless, official GDP will need to be the starting point for the ICP.

2. Past Practice in the Absence of National Accounts Data

As a number of countries in the past have been unable to submit the required information, guesstimates of expenditures at basic headings level as well as aggregates have been made by World Bank staff for a number of countries.[4]

The estimation method started with the expenditure data available in the NA of a given country. When data were available on the broad expenditure categories of GDP, these data were used as control totals, to be distributed over their respective more detailed categories.

In those cases where no expenditure data on GDP were submitted, the first option was to consult national statistical offices, and statistical publications of other international organizations for actual data, however, if no data were available, the option used was to apply the percentage distribution of a comparable country that participated in the ICP phase.

The detailed components of the control totals were then obtained by first examining national and international statistical sources for actual data, such as household expenditure surveys, or if the data from the household survey was not published, use the CPI weights (based on household expenditure surveys) which could either be taken directly from the national statistical offices or from the International Labour Organizations database. In those cases where no data were available, guesstimates would be based on the expenditure structure of a comparable ICP country.

3. ICP Requirements, Statistical Sources, and Recommended Practice

3.1. Consumption

The consumption concept utilized in the ICP is actual consumption, rather than final consumption expenditures which is the more commonly used concept of consumption.

Actual consumption by households is defined as households final consumption expenditure plus social transfers in kind from government and non profit institutions serving households (NPISH). Examples of transfers in kind are health services and free (or sponsored) admission to cultural events. It follows that actual consumption by government (also called collective consumption) is defined as final consumption expenditure less individual consumption (social transfers in kind). Per definition, no actual consumption is taking place in NPISH, as all their final consumption expenditures are on individual goods and services.

3.1.1. Requirements

The ICP not only requires a detailed breakdown of households final consumption expenditures, but for the estimation of actual consumption by household an estimate of individual consumption by government is needed. This requires that government final consumption expenditures are classified by purpose, following the Classification of the Functions of Government (COFOG) on a two digit level.[5] For collective as well as individual consumption expenditure by government, breakdown by cost is required.

The list of required details can be found in Annex I.

3.1.2. Statistical sources

(i) Households final consumption expenditure

The estimate of households final consumption expenditure could be either independently estimated, or be the residual in the equation with GDP given from the production side. In those cases where household consumption expenditures are compiled directly, one of the three methods/sources, or a combination is usually applied: (i) household expenditure surveys; (ii) data on retail trade sales or turnover; and/or (iii) commodity flow approach.

Household expenditure surveys are more commonly used in developing countries, while developed countries tend to rely more on retail trade statistics among other sources and commodity flow techniques. This difference in approach can in part be explained by the importance own produced goods play in the consumption estimates in developing countries, for which (household) surveys are the source. Furthermore, the statistical infrastructure might not be in place to produce retail trade statistics of good quality.

Some other frequently used sources are VAT records, statistics on sales of alcohol and tobacco, energy statistics, special housing/rent surveys, government accounts (for fees paid etc), surveys of private health services, records on new registration of motor vehicles, and transport statistics.

(ii) Government consumption

The sources for government consumption expenditure are the different government units’ accounts, sometimes presented in the shape of the Government Finance Statistics (GFS) tables. The degree of detail made available to the statistical office will differ considerably between countries, and there is a fact that the availability of accounts for local government units can be scarce/uncurrent.

For more detailed data, which might be needed in particular areas such as for intermediate consumption in the production of health services, administrative records from the Ministry of Health is a potential source, or if the country primarily imports medical equipment and medicines, import statistics/trade data would be another potential one. Likewise, for a breakdown of compensation of employees by level of education services provided, administrative records from the Ministry of Education will most likely include this information.

3.1.3. Recommended practice

The optimal way of deriving expenditure weights for the ICP is to extract the weights from the national accounts, as this will ensure consistency when the PPP data are used to convert national accounts aggregates. As this will not be an alternative for many countries, details from other sources more readily available, such as household surveys, should be utilized.[6]

(i) Weights for Households Final Consumption Expenditures

As reporting on basic heading level requires even a higher degree of details than four-digit COICOP[7] provides, integrated supply-use framework on a relatively detailed level is necessary if all components are to be extracted from the SNA. Even if detailed supply-use tables are available, there will not always be full match between purposes/products reported in the use tables and basic headings needed for the ICP weights, thus, there might be a need to go back to the sources used for the national accounts compilation.

A more common situation will be that the NA includes an aggregated estimate or an aggregated breakdown by purpose of household consumption. In this situation, the NA data will need to be supplemented with detailed information from the household survey and/or other suitable sources available.

An exact link between the household survey result and the NA consumption expenditure groups might not exist. Therefore, while estimating weights at the more detailed level based on the household survey, NA estimate(s) should be used as constraint(s). A simple example is given below.

If no breakdown of GDP by expenditure categories is available, total household consumption expenditure as well as the detailed weights should be based on the household expenditure survey (supplemented with additional sources available).

There are limitations to the direct use of household survey data for the estimation of expenditure weights. The coverage of the household survey data (even with nationwide coverage) is usually somewhat more limited than the coverage of household consumption expenditure in the NA. This is due to the fact that most household surveys do not cover any consumption by institutional households, e.g., orphanages and prisons. Furthermore, survey data often underestimate household spending on alcohol, tobacco and the like, as most people deliberately forget to report any (or the true) spending on such items.

Table 2 Derivation of ICP weights based on aggregated expenditure data from the NA supplemented with details from the household expenditure survey.

Data from the system of national accounts
Current price / Implicit weights
GDP / 559149 / 1.000
Household consumption expenditure / 278346 / 0.498
01 / Food and non-alcoholic beverages / 50190 / 0.090
02 / Alcoholic beverages, tobacco and narcotics / 10470 / 0.019
03 / Clothing and footwear / 21210 / 0.038
….
Data from the household expenditure survey
Amounts
… / …
03 / Clothing and footwear / 19940
Clothing / 13150
Men's clothing / 3500
Ladies' clothing / 3900
Children's and infants' clothing / 4700
Clothing materials and accessories / 900
Repair and maintenance of clothing / 150
… / …
Derivation of basic heading weights
ICP weights
… / …
03 / Clothing and footwear / 0.038
Clothing
Men's clothing / =(3500/19940)*(21210/559149) / 0.007
Ladies' clothing / =(3900/19940)*(21210/559149) / 0.007
Children's and infants' clothing / =(4700/19940)*(21210/559149) / 0.009
… / …

On the other hand, household surveys are better suited to include own account consumption, as information on this is missing from other statistical sources frequently used to estimate household consumption expenditure in the SNA, such as retail trade or turnover data.

Other common problems with survey data might be linked to sampling design – stratification and selection of units, and to grossing up survey results. Therefore, surveys might not always be representative. Nevertheless, for the derivation of weights on the most detailed level of the ICP – basic heading level, this should not pose a significant problem. It could, however, in some cases be more of a problem when aggregated data from the household survey are needed. Thus, coverage adjustments to survey data should be done both on the overall level and on details, where other sources are deemed more reliable.

Recent household surveys are not always available. Because the weights in an old household survey can be non-representative due to relative price changes, the results of the most recent survey should be extrapolated with the move in component indices in the consumer price index (CPI). The weights based on the price adjusted survey results will of course not reflect changes in demand due to relative price changes or changes in taste etc. Hence, it will not represent an ideal source of data, and the need to use additional source statistics is even greater. This will be particularly important for goods and services where demand is price- and income sensitive such as restaurant visits, wine, furniture, services of hair dressers and beauty shops, and photographic equipment. [8]

Albeit the above, household survey data should be considered a good source for the derivation of weights when detailed expenditure data for the households are lacking in the NA.

(ii) Weights for individual and collective consumption expenditures by government

For many countries, their NA data lack the details required to distinguish between individual and collective consumption expenditures by government, as they do not classify government expenditures by COFOG at such a detailed level (70 functions).

However, a number of countries use a more aggregated version with a breakdown on 12 divisions, thus, previous practice has been to allocate COFOG divisions 07 Health and 09 Education to households, in order to obtain an estimate of actual consumption by households. As most of COFOG divisions 08 Recreation, Culture and Religion and 10 Social Protection are also individual, the coverage of social transfers in kind should be estimated as more than only COFOG divisions 07 and 09 even when detailed data is not available.