An Evaluation of the Use of Ordinary Least Squares for Estimating Tourism Demand Models

Abstract

Estimating tourism demand models involves a set of related equations with errors that may not satisfy the common assumptions of being independent, with constant variance and normal distribution. In such circumstances, seemingly unrelated regression estimation may be considered a better estimation technique than ordinary least squares. Results from a simulation exercise, however, show that generally there is little difference between ordinary least squares and seemingly unrelated regression. The ordinary least squares technique performs well, and the results give little reason to use more complex estimation techniques. Another feature of tourism data is that strong growth in tourist numbers is often observed. This feature implies that models in which such series are the dependent variable are not consistently estimated by least squares methods. A percentage error loss function is proposed as a more appropriate criterion for estimating tourist data of this type.