高頻率股市報酬波動性之ANN-GARCH Model

Abstract

當投資大眾產生投資行為時，其所關心的就是投資報酬風險值，也就是這一籃的投資組合將可能會有多少的損失。而會影響投資風險值的便是報酬波動性（Volatility），報酬波動性並非為一固定的常數，它是會隨著時間的改變而改變。由於以往學者再研究此項議題時，大都採用日資料或週資料的報酬率，而本研究則是採用每一交易日內的每分鐘報酬率為研究樣本，以四種波動性模型為研究工具：GARCH、EGARCH、GJR、ANN-GARCH模型，進而比較這四種模型間，何者對於交易期間訊息的捕捉能力較強。經由實證結果顯示，ANN-GARCH模型確實比其他三種傳統的波動性模型，更能捕捉台灣證券市場中所流竄的交易訊息；且若將交易量納入整各類神經網路的架構中，其ANN-GARCH模型的解釋能力就更佳了。

Many of the researchers in finance had used the daily returns to find its volatility model. In recently some researchers also care about the high frequency returns. This study also uses three of the traditional ARCH-family models to discuss the conditional variance in high frequency data. Donaldson and Kamstra（1996,1997）constructed a seminonparametric nonlinear GARCH model, based on the Artificial Neural Network literature, called ANN-GARCH model. We also use it to capture the risk value. In Taiwan Stock Exchange we want to know which volatility model is optimal. In our empirical evidence found that ANN-GARCH can explain what other traditional volatility models can’t explain.