本研究分別利用傳統 OLS、誤差修正模型以及 Bivariate GARCH 模型研究以摩根台股指數期貨規避台灣股價指數的避險效果，現貨部分除了摩根台股指數現貨之外，亦考慮了台灣加權股價指數，目的在於瞭解摩根指數期貨的避險效果，並提出未來台灣加權股價指數上市後一套研究指數期約避險績效的研究架構。

本研究實證結果發現：

(1)將台灣加權股價指數、摩根台股指數現貨以及摩根台股指數期貨的每日收盤價取對數值，並且依照避險期間分為三種情況，利用 ADF(Augmented Dicky and Fuller)進行單根檢定，結果顯示三個時間數列皆非定態(stationary)。

(2)時間數列取一階差分之後，視為指數的報酬率，同樣利用 ADF 進行單根檢定，結果顯示三個時間數列呈現定態(stationary)，亦即時間列服從 I (1)。此時，報酬的迴歸式存在具有實質意義。進行供整合檢定之結果顯示，無論是台灣加權指數與摩根台股指數期貨市場間，或是摩根台股指數之現貨與期貨市場間存在長期穩定之均衡關係。因此欲研究現貨與期貨市場的避險比率，應考慮誤差修正項。

(3)在加權股價指數與摩根指數期貨間避險效果方面：

1.在樣本內實證中，傳統 OLS 除了在避險期間為每日的情況之外，所造成投資組合變異數降低幅度較大，有較好的樣本內避險效果表現。

2.在樣本外實證中，傳統 OLS 無論在何避險期間，所造成投資組合變異數降低幅度較小，其避險效果皆較差。

3.避險誤差均方根比較方面，傳統 OLS 表現較差。

(4)在摩根台股指數現貨與摩根指數期貨間避險效果方面

1.在樣本內實證中，傳統 OLS 在各避險期間，所造成投資組合變異數降低幅度較大，有較好的樣本內避險效果表現。

2.在樣本內實證中，傳統 OLS 無論在何避險期間，其避險效果差皆較差。

3.避險誤差均方根比較方面，同樣以傳統 OLS 表現較差。

Investors of Taiwan Stock Market have been long lack of hedging tools. SIMEX has provided a new merchant, MSCI Taiwan Index Future on January 9,1997. In addition, Taiwan Futures Exchange is going to run on July, 1998. Though investors are still not familiar with the new derivatives. Futures will be the new markets in Taiwan and it is the right time for us to analyze it. This research use different econometrics methods to check if it is a good hedge tool for the investors. The results are as followed.

1.The time series of MSCI Taiwan Index futures, MSCI Index Spots and Taiwan Weighted Index are not stationary. They are integrated of order 1.

2.There exist cointegrations between MSCI Taiwan Index futures and MSCI Index Spots, in addition to MSCI Taiwan Index futures and Taiwan Weighted Index.

3.OLS Regression, Error Correction Model and Bivariate GARCH Model are applied to find the optimal hedge retio. Among them, the hedge ratios of Bivariate GARCH Model are dynamic while the other two are constant.

4.According to the in-sample hedging effects results, the OLS are outstanding. The low variance of hedging portfolios and the reduction percentage compared to the no-hedged portfolios prove that.

5.Investors may care more about the out-sample results. From the table we know that Error Correction Model and Bivariate GARCH Model perform better than OLS, especially when the time period is longer.

6.When we check the RMSE, we get the same conclusion that OLS is the worst one among the three methods.