隨機森林及深度強化學習在台指期交易策略之應用

Abstract

近年來有許多論文嘗試以不同的機器學習演算法建立模型應用在交易上，並且取得不錯的成效。本次研究採用了機器學習二大分支中的兩種演算法，分別為監督式學習中的隨機森林及深度強化學習來建構台灣加權股價指數期貨交易策略。\n本次研究將2007年7月2日至2019年3月31日的期貨價格、技術面及籌碼面等相關資料分為訓練、驗證及測試區間，以訓練區間資料作為訓練樣本，訓練隨機森林及深度強化學習模型，並以樣本外的驗證及測試區間驗證其績效。\n在隨機森林模型之中我們輸入了62項特徵進行訓練，結果在驗證區間年化報酬率為13.06%、在測試區間年化報酬率則為12.32%。此外，我們透過隨機森林模型篩選特徵重要性，發現外資籌碼面指標占重要性前10名當中的4名，自營籌碼面指標則占前10名當中的5名，其中外資台指期貨多空交易淨額、自營金融期貨多空交易淨額及自營台指期貨多空交易淨額特徵重要性排名分別為第1、第2及第3。在深度強化學習模型之中，我們額外輸入了隨機森林篩選過的特徵進行訓練。結果在驗證區間年化報酬率為0.50%、在測試區間年化報酬率則為5.03%。

In recent years, there are many researches have tried to build trading strategies by different machine learning algorithms and achieved good results. In this paper, we use random forest and deep reinforcement learning which belong two branch of machine learning to construct our models for TAIEX future trading strategies.\nIn this paper, we use the relevant data of TAIEX future including price, technical indicators and chip factors from July 2, 2007 to March 31, 2019 and divide the data into training, validation and test data set. The training data set is used to train the models. The validation and test data set are used to verify the performance.\nIn this paper, we input 62 features to train the random forest model, as a result, the annualized rate of return is in validation data set is 13.06%, and the annualized return rate in the test data set is 12.32%. In addition, we use the random forest model to rank the importance of features and find that factors of foreign investors chip account for 4 of the top 10 and factors of dealer chip account for 5 of the top 10. The feature importance of foreign investors TAIEX future volume net, dealer finance future volume net and dealer TAIEX future volume net is ranked first, second and third respectively. Beside, we additionally input top 10 important feature to train the deep reinforcement learning model, as a result, the annualized rate of return is in validation data set is 0.5%, and the annualized return rate in the test data set is 5.03%.