| Summary: | 碩士 === 逢甲大學 === 應用統計研究所 === 105 === This study evaluates the Taiwan government’s initiative “New Southbound Policy” based on the performances of eight prediction methods for the country’s inbound tourism arrivals from ASEAN 5. We examine out-of-sample forecast accuracy by using a holdout set and consider three main types of a hierarchical time series forecasting approach: (1) bottom-up approach,(2) top-down approach, and (3) optimal combination approach; we then employ an (4) aggregation of forecasts from each ARIMA model; (5) Holt-Winters exponential smoothing method for aggregated tourism arrivals from ASEAN 5; (6) a time series regression for aggregated tourism of ASEAN 5; (7) the ARIMA model for aggregated tourism arrivals from ASEAN 5; and (8) the unobserved components model for aggregated tourism arrivals from ASEAN 5. We evaluate how well the eight models/methods can forecast by computing four criteria: the root mean square error (RMSE), the mean absolute deviation (MAD), the mean percent error (MPE), and the mean absolute percentage error (MAPE). We measure the forecast accuracy of these eight forecasting models/methods with four criteria by a rolling window approach. Based on the forecasts’ performances, we conclude that the “New Southbound Policy” does help boost Taiwan’s tourism market.
Moreover, the multiplicative Holt-Winters method is at the top based on a ranking system, while the unobserved components model exhibits the second-best rank.
|
|---|
