Stochastic specification and the international GDP series

Abstract

This paper investigates the stochastic properties of GDP series based on purchasing power comparisons for 125 countries from the Penn World Table (PWT) and a GDP series based on exchange rate conversions in 1987 constant dollars for 107 countries from the World Development Indicators (WDI) in the period 1965-90. Because many health and demographic variables are compiled at irregular intervals, models for economic growth are often estimated using data that are separated by 5-year intervals. Panel data on the GDP and growth rate series were analyzed using alternative econometric methodologies. First, the stochastic properties of the series were analyzed by applying classical tests for unit roots in a fixed effects framework. A new ratio was developed for the case where heterogeneous drift parameters are present under the null hypothesis. The 5% critical values of the most powerful invariant tests for unit roots are tabulated for different numbers of countries and time periods. Second, a dynamic random effects framework was used for testing the stochastic specification for the GDP and growth rate series. A sequence of Wald statistics was applied to test various structures for the variance covariance matrix of the GDP and growth rate series. Overall, the GDP series from PWT and WDI showed various forms of non-stationarity. Moreover, GDP growth rates at 5-year intervals possessed simple stochastic properties making them amenable to econometric modeling.