當自我迴歸型式的不齊一條件變異數（autoregressive conditional heteroskedasticity；ARCH）發生結構變動時，本文以馬可夫鏈（Markov chain）描述不同狀態之間的隨機轉換，對條件變異數進行非線性 Markov-switching ARCH（SWARCH）模型之估計與預測。在估計 SWARCH 模型之前，本文先以變異數結構變動檢定統計量檢驗資料，以避免發生過度配適的問題。而檢驗結果拒絕變異數未發生結構變動的虛無假說，因此本文進一步估計Markov-switching ARCH模型，並根據參數估計之結果，預測條件變異數。 於數種不同形式的ARCH模型、GARCH模型、與SWARCH之估計與預測結果中，以Gaussian SWARCH（3,2）模型的log likelihood估計值為最大，而且在樣本內預測表現上，其∣LE∣與LE�斑怳p。此外，該模型更能表現出 ARCH 過程在各種狀態間的轉移情況，符合條件變異數發生結構變動的可能性。至於樣本外預測的表現上，雖在短期預測方面Gaussian SWARCH（3,2）模型比Threshold GARCH（1,1）模型稍差，但是在長期預測上，其表現則較優異。

　　　　　This paper uses the Markov regime-switching (SWARCH) model to estimate and forecast the volatility while there exist structural changes in the conditional variance. The Lagrange multiplier (LM) test, proposed in Lin and Chang (1997), is adopted to avoid the problem of overfitting. The data we use are the returns from the Taiwan weighted stock index (TWI) during the priod from January 5, 1990 through January 24, 1998. We find that the null hypothesis of no structural change in volatility is not accepted. The results of the likelihood ratio (LR) test suggest that the Gaussian SWARCH (3,2) model fits better than the other models studied in this paper. In addition, the Gaussian SWARCH(3,2) model outperforms the other competing models in terms of the in-sample and long-term out-of-sample forecasting ability.