Forecasting U.S. Tourist arrivals using optimal Singular Spectrum Analysis

Highlights:
• This study applies Singular Spectrum Analysis (SSA) to forecasting tourist arrivals to the US.
• The Vector SSA model outperforms ARIMA, Exponential Smoothing and Neural Networks.
• SSA outperforms ARIMA in forecasting U.S. Tourist arrivals by country of origin.
• The improvement in performance refers to both short- and long-run.
• The results are statistically significant.

Abstract:
This study examines the potential advantages of using Singular Spectrum Analysis (SSA) for forecasting tourism demand. To do this it examines the performance of SSA forecasts using monthly data for tourist arrivals into the Unites States over the period 1996 to 2012. The SSA forecasts are compared to those from a range of other forecasting approaches previously used to forecast tourism demand. These include ARIMA, exponential smoothing and neural networks. The results presented show that the SSA approach produces forecasts which perform (statistically) significantly better than the alternative methods in forecasting total tourist arrivals into the U.S. Forecasts using the SSA approach are also shown to offer a significantly better forecasting performance for arrivals into the U.S. from individual source countries. Of the alternative forecasting approaches exponential smoothing and feed-forward neural networks in particular were found to perform poorly. The key conclusion is that Singular Spectrum Analysis (SSA) offers significant advantages in forecasting tourist arrivals into the US and is worthy of consideration for other forecasting studies of tourism demand.

Please contact UAL Research Online to request access to this article.