Comparative Evaluation of Estimation Techniques for Purchasing Power Parity in African Countries Using the Country-Product-Dummy Regression Framework

Abstract

Purchasing Power Parity (PPP) is a popular macroeconomic analysis metric used to compare economic productivity and standards of living between countries. This study examines the estimation of PPP within the International Comparison Program (ICP) at Basic Heading (BH) level stage and leverages on the data from the 2011 ICP round. Focusing on five BHs out of 12 BHs across 50 Africa countries, to empirically evaluate the validity of the classical Ordinary Least Square (OLS) assumptions in the estimation of Country Product Dummy (CPD) regressions. Given the widespread use of OLS for BH level PPP computation, a rigorous examination of these assumptions is essential to ensuring the statistical reliability of PPP estimates. For the five BHs considered, it was observed that the problem of Heteroscedasticity and Non-Normality is common to all the BHs while one of the BH also has the problem of Autocorrelation inclusive. The Generalized Least Square (GLS), Robust standard error, Quantile Regression, and Box-Cox was used to address the problem detected. To evaluate the precision of estimate and model fit performance, standard error and residual standard error was used as criteria respectively. For the estimate precision, Quantile regression has the least value of standard error across all the BHs and GLS also has the least value for residual standard error across all the BHs for model fitting