Estimating growth rate in the presence of serially correlated errors

Abstract

The aim of this study is to address the difficulties frequently encountered in estimating average growth rates by a log-linear time trend in the presence of serially correlated errors. There are a few studies in the literature that provide some guidance on choosing the appropriate method depending on the degree of first order serial correlation. However, the higher order serial correlation case is generally ignored. This study proposes the Nelder-Mead simplex method as a general solution to estimating linear trend in the presence of serial correlation of any order. The proposed method and the conventional methods are applied to the real GDP per capita series of 27 OECD countries. Twelve series seem to be better modelled by a log-linear trend with AR(2) residuals, and five of them yield remarkably different growth rates.