Tourism Demand Forecasting

Use a time-series model such as SARIMA when you only have the demand history, when seasonality is strong, and when accurate short-to-medium-horizon forecasts are the goal; these models are robust and need little data. Use an econometric model such as an ADLM or error-correction model when you also want to understand or simulate the effects of drivers like income, prices, or exchange rates, and you have reliable data on them. Song and Li's review shows neither family dominates universally, and combining forecasts from both often beats either alone.

Because a model can fit the historical data almost perfectly yet forecast badly, a phenomenon known as overfitting. The only honest test of a forecasting model is how well it predicts data it did not see during estimation. Standard practice holds out a recent portion of the series, forecasts it, and compares errors using measures like MAPE, RMSE, and Theil's U against rival models and a naive benchmark. Song and Li treat this out-of-sample discipline as the central criterion for model selection in tourism demand forecasting.

Not well on their own. Time-series and econometric models assume the future resembles estimated patterns and relationships, so they tend to fail around structural breaks such as financial crises, terrorism, or pandemics. Analysts respond with intervention and dummy variables, regime-switching or other nonlinear models, scenario analysis, and judgmental adjustment, and they widen uncertainty bands. The honest position, consistent with the review literature, is that quantitative models inform but do not replace human judgment when a major shock breaks historical continuity.