FORECASTING TOURIST ARRIVALS IN GUILIN CITY USING STATISTICAL TECHNIQUES

Abstract

            This study forecasts tourist arrivals in Guilin using quarterly data from the Guilin Bureau of Statistics spanning 1999–2024 (n=104). Six univariate time-series techniques were benchmarked—Trend Analysis, Classical Decomposition (additive/multiplicative), Moving Average, Single and Double Exponential Smoothing, and the Holt-Winters method—with model selection based on Mean Absolute Deviation (MAD) and Mean Absolute Percentage Error (MAPE). The series displays a sustained upward trend with pronounced quarterly seasonality. Consistent with the research objective of identifying the most accurate forecasting approach, the Holt-Winters model with multiplicative seasonality achieved the best fit and predictive accuracy (MAPE=17; MAD=216), capturing peak–off-peak patterns and recent dynamics robustly. These findings indicate that incorporating seasonality multiplicatively is crucial for destinations where demand growth amplifies seasonal fluctuations. The selected model provides reliable short-term forecasts to support capacity planning, resource allocation, and targeted marketing in Guilin’s tourism sector, and offers a practical baseline for method selection in destination forecasting.