Chapter 2 - Traditional methodology of tourism demand modelling

Abstract

Most of the published studies on causal tourism demand models before the 1990s were classical regressions with ordinary least squares (OLS) as the main estimation procedure. The functional form of most of these models was single-equation, and either linear or power models. The data used in estimating tourism demand models are mainly time series, and since most of these time series, such as tourist expenditure, tourist arrivals, income (measured by personal disposable income or GDP), tourists' living costs in the destination, transport prices, and substitute prices are trended (non-stationary), the estimated tourism demand models have tended to have high R 2 values due to these common trends in the data. We shall see in Chapters 3 and 4 that statistical tests based on regression models with non-stationary variables are unreliable and misleading, and therefore any inference drawn from these models is suspect. Moreover, tourism demand models with non-stationary variables tend to cause the estimated residuals to be autocorrelated, and this invalidates OLS. The problem of autocorrelation in tourism demand models has normally been dealt with by employing the Cochrane-Orcutt (1949) iterative estimation procedure. However, the use of the Cochrane-Orcutt procedure diverts attention from searching for the correctly specified model (autocorrelation normally indicates model misspecification).