Estimating the stochastic formulation of purchasing power parities: Allowing for spatial autocorrelation

Abstract

When price levels of goods and services display spatial interdependence across regions or countries, the problem of spatial autocorrelation arises. In the context of multilateral index estimation, this phenomenon makes the disturbance term of the CCD stochastic formulation proposed by Prasada Rao and Selvanathan (1992a) spatially autocorrelated, leading to inefficient estimates of the multilateral indices and the purchasing power parities associated with these indices. In this study, a more generalised form of the stochastic formulation of the CCD multilateral index that accounts for spatial autocorrelation (GCCD(SA)) is explored. The presence of significant spatial autocorrelation among the estimated residuals of the CCD multilateral index model is also analysed and tested using the Moran's 'I' statistic with the aid of three alternative proximity measures namely, contiguity, distance and trade. All these three measures have shown their usefulness in the simultaneous estimation of the GCCD(SA) formulation. The results obtained indicate that with the existence of significant positive spatial autocorrelation among the price relatives of goods and services across countries, the GCCD(SA) specification is able to provide plausible, unbiased and efficient estimates of the multilateral indices for spatial comparisons, specifically the purchasing power parities.