Before and After the Inclusion of Intervention Events: An Evaluation of Alternative Forecasting Methods for Tourist Flows

Abstract

This article investigates the intervention impacts on tourist flows and evaluates the accuracy of various forecasting techniques to predict travel demand before and after the inclusion of intervention events. The forecasting methods used in this study include (1) Naïve 1, (2) Naïve 2, (3) Holt-Winter's model, (4) Seasonal Autoregressive Integrated Moving Average (SARIMA) model, and (5) Artificial Neural Networks (ANN). The Holt Winter's and Naïve models are included for comparison purposes to ensure that minimum performance standards are being met. Data on air transport passengers including international arrivals and domestic air transport flows of the US (from January 1990 to June 2003) were obtained from the US Bureau of Transportation Statistics. This study focuses firstly on the importance for forecasting accuracy of allowing for intervention events in the modeling process. SARIMA models are therefore estimated both with and without intervention effects (the September 11th events). These models are used to generate forecasts for 2002 and the first part of 2003, and forecast accuracy is assessed using mean absolute percentage error and root mean square percentage error. The second focus of the study is to examine the impacts on tourism demand of the major crises that occurred during the period 2001–2003.