漫步於隨機森林——輔以多數決學習的台股指數期貨交易策略

A RANDOM WALK DOWN RANDOM FOREST: ENSEMBLE-LEARNING-ASSISTED TRADING STRATEGIES FOR TAIEX FUTURES

應用隨機森林演算法來進行未來期貨價格漲跌的分類預測，本文以技術面與籌碼面指標作為模型訓練的特徵，進而建構輔以多數決學習的台股指數期貨交易策略。藉由參數的重要性衡量，我們辨識出爭議變數，並探究參數配置的屬性擾動之於演算法預測能力及策略績效的影響。利用2007年至2018年的台股指數期貨資料，本文以多重角度測試策略之績效與穩健性。實證結果顯示，在考量交易成本之下，本文所建構之多數決學習台股指數期貨交易策略，要能於其訓練區間及測試區間皆呈現穩定勝出大盤的績效，其隨機森林模型所共同具備的參數配置必須包含3－14日MA與RSI指標、遠月期貨交易量、現貨交易量、期貨外資未平倉量與買賣權未平倉比率。

With the ensemble learning of specific TAIEX market characteristics drawn from technical analysis data, in this paper we construct futures trading strategies where price directional forecasts are generated by Random Forest classification models. By quantifying the model attributes' extent of contribution to the overall prediction outcomes, we identify attributes-in-dispute and explore their perturbative effects on the predictive ability of Random Forest and thus the risk-reward performance of the proposed strategies. Using 2007-2018 TAIEX futures data, our in-sample and out-of-sample test results show that, after transaction costs, risk-adjusted outperformance over the market is consistently observable when the Random Forest models adapt the 3-14 days MA and RSI indicators, far-month futures trading volume, spot transaction volume, foreign capital open interest in futures, and open interest ratio in options.