Comparison of Forecasting Accuracy With Time Series Models

This study attempts to identi the most accurate univariate time series model among contenders when ex post forecast is made. For the empirical analysis, three time series models including decomposition, Winter`s exponential smoothing and Box-Jenkins (ARIMA) are employed in order to forecast the sales revenue of five-star hotels in Gyongju. These models are chosen because they all can deal well with seasonal, cyclical and trended data. 120 observations ranging from January 1989 to December 1998 are used and the actual data of 1999 are left for the comparison with the ex-post forecast. In addition, four criteria including forecast error, MAPE, MAD and RMSE are applied to measure the accuracy. In conclusion, Winter`s exponential smoothing technique turned out to be the most accurate among competing models. This implies that no sophisticated models can always outperform the less sophisticated one. Nevertheless, the forecasts` raison d`etre lies in that it can help minimize the risk of hotel management, and the combination of quantitative and qualitative forecasting would be ideal for forecasting the demand for perishable products such as hotel rooms.