Auto-Regressive Distributed Lag Model on Forecasting Tourist Arrivals from Asian Region to Sri Lanka

Abstract

Sri Lankan tourism industry shows the historical development of tourist arrivals every year. The Asian region is the highest tourist producer to Sri Lankan tourism market. With the increase in tourist arrivals from the Asian region, the government needs the correct method of forecasting tourist arrivals to cope with uncertain situations and resource management. Therefore, the study was focused on identifying suitable to test the ADLM on forecasting tourist arrivals to Sri Lanka from the Asian region.Monthly tourist arrival data from 2009 to 2014 were obtained from statistical reports of 2009 and 2014 of Sri Lanka Tourism Development Authority (SLTDA). Autoregressive Distributed Lag Model (ADLM) was tested on forecasting tourist arrivals. One way Analysis of Variance (ANOVA) technique was used for overall model testing and t-test was used for individual parameter testing.The residual plots, Anderson-Darling and Durbin- Watson tests for residuals were used as a model validation criterion. Stationary of the series tested by Augmented Dickey-Fuller (ADF) test and Auto Correlation Function (ACF).The forecasting ability of the models was assessed by considering both relative and absolute measurements of errors.The results revealed that model with variable lag 1 confirmed the significance of the model. It is also confirmed the normality and uncorrelated of residuals. ADF and ACF confirmed the non-stationary of the series. The P value of Anderson-Darling test was (P= 0.534) and Durbin-Watson statistic was 1.93. Adjusted R2 of the model is 76.8%. The Mean Absolute Percentage Error (MAPE) values of fitting and verification of model with lag 1 were 1.36 % and 1.03% respectively. Mean Absolute Deviation (MAD) is 0.133 and 0.110 of fitting and verification and Mean Square Error (MSE) is 0.028 and 0.012. The deviations of the model also very small. It was concluded that the ADLM with lag one is suitable for forecasting tourist arrivals from Asian region to Sri Lanka. It is recommended to try other time series techniques namely decomposition techniques, Circular model, Seasonal Auto Regressive Integrated Moving Average (SARIMA) models etc, to capture the seasonal behavior of the series.