本文利用傳統時間序列模型GARCH與EGARCH，引入最佳化演算法PSO建構新模型PSOGARCH與PSOEGARCH。繼而透過追蹤誤差對匯率預測進行比較，期望建構預測能力較高的匯率預測模型。本文首先引用Chang and Tzeng（2009）挑選對匯率影響的27個變數，並以傳統時間序列模型GARCH與EGARCH進行預測，進而利用最佳化技術PSO模型篩選變數，再投入GARCH與EGARCH進行預測，最後將上述模型進行追蹤誤差比較，結果顯示PSOEGARCH模型具有最小的預測誤差，亦即顯示其預測能力最佳;其次為PSOGARCH模型；傳統時間序列GARCH模型之預測能力則為最差。

This research utilized traditional time series models GARCH and EGARCH, then introducing PSO optimization algorithm for composing new models PSOGARCH and PSOEGARCH. The tracking error methods are compared among the models for expecting to construct a higher performance forecasting model. First of all, this article refers to Chang and Tzeng (2009) selection effect on the exchange rate of 27 variables. Furthermore, traditional time series models GARCH and EGARCH are used to the foreign exchange forecasting, and then PSO model filter variables for re-entering models GARCH and EGARCH prediction. Finally, comparing for the above models, the results showed of PSOEGARCH model with the smallest error that display the best forecasting ability, second PSOGARCH model, and the traditional times series forecasting of GARCH models are the worst.