Tourist Arrival Forecasting in Sri Lanka: A Bayesian Spline and Interrupted Regression Approach

Tourism is a major contributor to the Sri Lankan economy and is heavily affected by external factors like the COVID-19 pandemic. This study aims to explore the impact of the pandemic on tourist arrivals using advanced statis-tical methods to model the complex time series in tourism data. Interrupted regression and Bayesian spline regression are used to capture the non-linear, phase specific patterns in tourist arrivals during and after the pandemic. In-terrupted regression models the structural shift caused by the pandemic and gives insights into the sudden changes in tourist flows. Bayesian spline re-gression provides flexibility to model the non-linear trends and gives a more detailed understanding of how tourism patterns evolve over time. The study highlights the importance of addressing over-dispersion in the data which is common in count data like tourist arrivals. A Negative Binomial model is used to account for this over-dispersion and improves the predictions. Unlike the Poisson distribution which assumes the variance is equal to the mean, the Negative Binomial model allows for more variability and hence better model fit. The results show a general recovery in tourist arrivals but the uncertainty in the forecasts raises questions whether the tourism sector will go back to pre-pandemic levels or new patterns will emerge. This uncertainty highlights the need for ongoing monitoring and adaptive strategies to ensure the long term sustainability of the sector. Moreover, the use of posterior predictive distributions gives more insights into future tourist arrivals by accounting for uncertainty and providing a range of possible outcomes rather than single point estimates. This is useful for stakeholders including government bod-ies, travel agencies and businesses to make informed decisions and develop strategies to support the tourism industry’s recovery and growth in the post pandemic era.