本文應用卡門演算法(Kalman Filter)探討我國遠期外匯市場重新開放後，投資者面對條件性非齊一變異數時，如何評估外匯投資之風險溢酬;以及外匯市場是否具效率性等問題。實證結果顯示，新台幣對美金之遠期匯率呈現ARCH效果乃一普遍現象，致使迴歸模式結構不穩定。故若援引國外研究方法，應用迴歸分析以估計Level 或Percentage Change Specifications都將產生誤導結果。本研究將上述問題轉為State Space之模式，並利用卡門演算法估計投資人接獲市場訊息時，如何逐步調整其風險溢酬。實證結果發現: (1)利用卡門演算法於Percentage Change Specifications比Level Specifications更適合於探討投資人如何逐步調整風險溢酬;(2) 我國外匯市場不具效率性。 This research employs both level and percentage change specifications to test the efficiency of the Taiwanese foreign exchange market and to investigate how the market revises its assessment of risk premium in the presence of conditional variances. The weekly data of the one-month and the three-month forward exchange rates of NT dollar against the US dollar were collected from November 1991 to May 1993 for empirical purposes. The empirical results indicate that the ARCH effect is pervasive. This implies that (1) both the level and the percentage change specifications are misspecified; and (2) the structure of the regression models is unstable. These hypotheses were supported by the empirical results of the Chow test and the Goldfeld-Quandt test. To capture the ARCH effect, the time -varying coefficients regression models were cast into the state space model and the Kalman filter was employed to investigate how the market continuously adjusts its assessment of risk premium following the reopening of the Taiwanese foreign exchange market. Empirical results suggest that the percentage change specification with the application of the Kalman filter to estimate the time-varying parameters may be more appropriate than the level specification. Nevertheless, there is no indication that the Taiwanese foreign exchange market is efficient.