配對交易與機器學習在台灣股票市場之應用

Abstract

本研究根據Vidyamurthy (2004)以及後續相關文獻所提出的統計套利配對交易方法對台灣股票市場進行實證研究。本文使用的模型為Engle and Granger (1987)提出的二階段共整合檢定。我們利用上述模型檢定台灣股票，找出具共整合性質之股票配對，利用技術指標－布林通道找出價格異常的時間點進行交易，建構配對交易投資組合；本研究進一步將類神經網路模型加入，用於預測共整合殘差走勢，建構類神經網路結合布林通道之配對交易策略並建構投資組合。實證結果顯示和Avellaneda and Lee (2010)結果相同，市場上確實存在市場中立性的報酬，且兩個策略的投資組合皆有優於大盤的績效和穩健性；此外類神經網路確實有幫助我們減少進場次數提高勝率，並且使投資組合的最大虧損下降，但也因此降低了投資組合的總報酬。

This paper used the statistic arbitrage pairs trading method according to Vidyamurthy (2004) and other papers based on this book. This paper followed papers to conduct empirical research on Taiwan stock market. The models used in this paper is two-steps cointegration test that proposed by Engle and Granger (1987). We tested Taiwan stocks through the above models to test cointegration, and find the investable pairs. After finding out investable pairs, we used Bollinger Band to find out abnormal stock price to trade. Then we constructed the portfolio to study its performance. This study further adds the neural network model to predict cointegral residual and constructs a strategy with Bollinger Band and neural network model. The result shows that the strategy helping us find market neutral return, which is the same as the result of Avellaneda and Lee (2010). Furthermore, our portfolio is also better than investing in benchmark. Neural network model truly helps us reduce trading frequency and decrease drawdown, but it also decreases return at the same time.