Decomposition Techniques on Forecasting Tourist Arrivals from Western European Countries to Sri Lanka

Abstract

The forecasting tourist arrival is an essential discipline in planning, resource management, and other decision-making processes in micro and macro level. It facilitates to ensure the sustainable development by minimizing the risk to all aspects. There is a growth of arrivals from all the regions to Sri Lanka. Increasing of tourist arrivals could cause a positive or negative effect on the country. To get the maximum benefits from the positive impacts and to overcome the negative impacts, it is vital to forecast arrivals. This study focuses on identifying the highest tourist producing Western European countries and to forecast the arrivals from them. Monthly tourist arrival data from the UK, Germany, France, Netherland, and Italy for the period of, January 2008 to December 2014 was obtained from Sri Lanka Tourism Development Authority (SLTDA). Time Series plots and AutoCorrelation Functions (ACF) were used for pattern recognition of arrivals. It revealed that the arrivals have both trend and seasonal patterns. As such, the Decomposition Techniques were tested for forecasting arrivals. The Residual plots and the Anderson-Darling test were used as the goodness of fit tests in model validation. The residuals of both; additive and multiplicative models for all the countries were found normally distributed and independent. The best fitting model was selected by comparing the relative measurements of errors and the absolute measurements of errors. Measurement errors of all the fitted models were satisfactorily small. Among them, the models with least errors were selected for forecasting. It is concluded that the Additive Decomposition models are the most suitable models for forecasting arrivals from Western European countries. However, arrivals show wave-like patterns with the trend. The Circular Model is a newly introduced technique for modeling wave-like patterns. It is recommended to test the Circular Model on de-trended data to see whether the forecasting accuracy increases.