Abstract

Title: Tourism Forecasting: A Comparison of Statistical and Machine Learning Models
Author: Kieva Cadogan, Economic Statistician, Central Bank of Barbados
Strand 1: New sources and tools for effectiveness: AI and Machine Learning Statistics
Abstract:
Tourism plays a central role in the economic performance of Small Island Developing States (SIDS) such as Barbados, yet tourism demand is highly sensitive to seasonality, external shocks, and global uncertainty. The COVID-19 pandemic further exposed the limitations of traditional forecasting approaches in capturing sudden structural breaks. In this context, the growing availability of advanced statistical and machine-learning tools offers new opportunities to enhance forecasting effectiveness for policy and planning.
This paper compares the forecasting performance of traditional time-series, econometric and machine learning models in predicting quarterly tourist arrivals to Barbados from its three main source markets: the United States, the United Kingdom, and Canada. Using quarterly data spanning 196 – 2024, the study evaluates Seasonal Naïve, SARIMA, and SARIMAX models alongside econometric Autoregressive Distributed Lag Models (ADLM) and Support Vector Regression (SVR) models optimised using genetic algorithms. Model performance is assessed using RMSE, MAE, and MAPE across two evaluation periods: a stable pre-COVID period (2016 – 2019) and a volatile post-COVID recovery period (2021-2024).
The results show that while traditional time-series models perform reasonably well under stable conditions, machine learning models – particularly multivariate SVR, exhibit superior forecasting accuracy during periods of heightened volatility and structural change. Econometric models provide valuable theoretical insights but show reduced stability in the presence of large shocks. Overall, the findings demonstrate how machine learning and modern statistical tools can complement conventional methods to improve forecasting robustness in data-constrained environments, while providing evidence to support improved statistical forecasting and policy planning in tourism-dependent economies.