Using alternative methods to estimate the determinants of cross-border trips

The relationship between quarterly same day automobile trips by Canadians to the United States and the real exchange rate over the period 1979 - 95 is estimated using the parametric technique Ordinary Least Squares (OLS) and the nonparametric technique Locally Weighted Scatterplot Smoothing (LOWESS). The results suggest that there are important differences in the estimates generated across the two techniques. The nonparametric approach generates an inverse J-curve relationship between the two variables which implies that linear OLS estimates can yield substantial under-and overestimates of the dependent variable as the real exchange rate varies. This can have implications for applied researchers attempting to estimate models of tourism demand for forecasting purposes. At the very least, the nonparametric approach can help suggest a more flexible parametric functional form for multivariate regression that better reflects the data.