Modeling and Constructing Panels of Purchasing Power Parities for GDP and Its Components

Abstract

The International Comparison Program (ICP) of the World Bank carries out price surveys to provide measures of purchasing power parities (PPPs). These surveys have been carried out intermittently since 1970, with the number of countries participating in the ICP exercises increasing from about ten when ICP started in 1970 to 199 in the latest round of 2011. Due to the needs of international comparison, researchers and policy makers, extrapolation of these PPPs is needed to produce figures for the missing countries and years. This thesis contributes to the literature of PPP extrapolation in three di!erent aspects. The first contribution is on how to capture measurement errors when extrapolating ICP PPPs across time and space; the second is on the specification of structural models to explain the price level of the components of GDP (private consumption (C), government expenditures (G) and gross capital formation (I)). The third is on the aggregation of the PPPs of GDP components to form PPPs at GDP and Domestic Absorption level.In the extrapolation of PPPs, different sources of information are used to form the bases for the estimation process. Among these are the PPP observations from the ICP and the GDP deflators from the National Accounts (NAs). The information provided by these sources is subject to measurement errors (MEs). The existing econometric based method to construct afull panel of PPPs across time and space, Rao et al (2010a,b)- RRD, assumes the MEs are fullyheteroskedastic with the variances inversely related to the level of development of each country, and thus the assumption is that richer and more developed countries devote more resources to data collection and therefore provide measurements that are more precise. In the first paper, alternative assumptions for the ME variances are proposed and empirical analyses are carried out. The empirical findings suggest that by incorporating the information of ICP regions and the World Bank’s country groupings by level of income to model the MEs variances as groupheteroskedastic, standard errors of the PPP estimated are significantly reduced, especially for developing countries. We label this the modified RRD method.The second paper deals with the modelling of the price levels for the components of GDP, Consumption, Government Expenditure and Investment; and uses the modified RRD econometric method to produce panels of PPPs for the components. In the existing literature, PPPs for these components are not produced using economic theory of the price level for each component, and point estimates are produced without any indication of the uncertainty level. By using elements from the macroeconomic literature to define structural determinants for each component, we propose economic models for each of the GDP components and integrate them into the modified RRD econometric framework to produce PPP figures for each component together with their corresponding standard errors.In the last paper the estimated series resulting from the second paper are aggregated to form alternative series of PPPs at Domestic Absorption and GDP level at current prices. With positive prices and non-negative quantities as input data, the standard aggregation methods (GEKS, GK and IDB) produce strictly theoretically correct average price indices and quantity indices. However, the application of the standard formulas when headings with negative value have a significant share of the aggregate can lead to meaningless results. This is especially the case in the aggregation of PPPs from component level to GDP level, where we encounter a problem of negative nominal quantities of net exports when exports are smaller than imports for some countries. We apply alternative methods in the literature by either carrying out the aggregation at Domestic Absorption level, which do not involve net exports; or using the augmented GK method, which is currently being used by the Penn World Table. An extension of the existing methods is that a weighted average of the standard errors of the estimated component PPPs are computed to provide an indication of uncertainty for the aggregated PPPs of countries. The di!erence between PPPs at GDP level and Aggregated PPPs at GDP level obtained from alternative methods are computed and evaluated.