Mitigating the Effects of Multicollinearity Using Exact and Stochastic Restrictions: The Case of an Aggregate Agricultural Production Function in Thailand

Abstract Ordinary least squares, exactly restricted OLS, stochastically restricted OLS (mixed estimation), and principal components regression each were used to estimate an aggregate agricultural production function for Thailand for which data were highly multicollinear. Pretest considerations, incorporating alternative risk measures, were addressed in detail for purposes of model evaluation. The final mixed and principal components models generally outperformed OLS in terms of risk and overall reasonableness, mitigating a serious multicollinearity problem and permitting a direct examination of the rate and composition of Thai agricultural output growth.