美國聯邦基金利率預測模型之實證探討

本研究分別以 ARIMA 及 GARCH 模型針對 1991 年 1 月至 2005 年 12 月間美國聯邦基金利率(The U.S. Federal Funds Rate)之月資料進行分析，並以 MAE，MSE，RMSE以及 MAPE 等方法比較兩模型之預測績效。而在研究之過程中得出下列幾點實證結果: (1) 利用 ARIMA 模型與 GARCH 模型在預測美國聯邦基金利率時，在預測期為 24 期的預測績效，以 GARCH 模型較優。(2) ARIMA 模型與 GARCH 模型在預測美國聯邦基金利率時，不同的預測期間會有不同的預測績效，在預測期為 6 個月時兩模型之績效均優於 12 個月、18 個月及 24 個月。(3) ARIMA 模型與 GARCH 模型受限於其自身的參數假設，其預測值往往較受到其預測期前幾期之走勢而變動。(4)在利率有明顯變動時，ARIMA 模型與 GARCH 模型並無法準確預測其變動方向，僅能依照預測期之前幾期利率的走勢勾勒出預測值。

This study employs uni-variable time-series models to forecast the trends of the U.S. Federal Funds Rate (FFR). The empirical data covers FFR monthly data from January 1991 to December 2005. Two models, ARIMA and GARCH, are used to analyze the data respectively. In order to compare the accuracies of these two models, the methods of MAE, MSE, RMSE and MAPE are the main tools used to measure their prediction performances.Major conclusions of this study are stated as follows. First of all, the forecasting performance of GARCH model is better than that of ARIMA. Secondly, the setting of time points and length of forecasting periods lead to different performances of these two models. For instance, performances of both models for six months are better than those of twelve months, eighteen months and twenty-four months. The prediction performances of both models are depends heavily on the length of the in-sample data. Finally, these two models cannot predict precisely the change of the Federal Funds Rate especially when the interesting rate comes across with a sudden change. They can only outline roughly the trend of the interest rates based upon the previous in-sample period.