配對交易與機器學習在台灣股票市場之應用

Applications of Pairs Trading and Machine Learning in Taiwan Stock Market

徐瑀暄(Yu-Hsuan, Hsu) ； 林士貴(Shih-Kuei, Lin) ； 蔡瑞煌(Rua-Huan, Tsai) ； 林朝陽(Chao-Yang Lin)

《統計與資訊評論》 19卷 (2019/12) Pp. 1-37

摘要

本研究根據統計套利的配對交易方法對台灣股票市場進行實證研究，其模型為二階段共整合檢定。將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，類神經網路確實可以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。

關鍵字

共整合 ； 配對交易 ； 布林通道 ； 類神經網路 ； 投資組合

並列摘要

This study conducts empirical research on the Taiwan stock market based on the statistical arbitrage pairing transaction method. The model is a two-stage cointegration test. The above model is used to verify Taiwan stocks, to find stocks with a common integration nature, to use the technical indicators - Bollinger Bands to find out the time points of price anomalies to conduct transactions, and to construct a matching trading portfolio; this study further adds a neural network model. It is used to predict the trend of cointegration residuals, construct a paired trading strategy based on the neural network and the Bollinger Band, and construct a portfolio. The empirical results show that there is indeed market neutrality in the market, and the investment portfolio of both strategies has better performance and robustness than the broader market; in addition, the neural network can indeed reduce the number of entries to improve the winning rate and make the portfolio The biggest loss has fallen, but it has also reduced the total compensation for the portfolio.

並列關鍵字

Cointegration ； Pairs Trading ； Bollinger Band ； Neural Network ； Portfolio