本研究使用Smart Beta因子與技術指標作為輸入特徵(features),並搭配深度學習中的多層感知器(multilayer perceptron, MLP)和卷積類神經網路(convolutional neural networks, CNN)來建構Smart Beta交易策略,驗證深度學習在投資交易策略上之效果。運用2007年至2017年的台灣股票資料,透過每季進行模型的更新及訓練,結果發現深度學習Smart Beta交易策略績效表現優於台灣加權指數和運用Asness, Frazzini, and Pedersen(2017)的Smart Beta特徵分數方法建構之投資組合。此外,透過特徵篩選法保留重要輸入特徵,可使深度學習Smart Beta交易策略績效更進一步提升。
Using Smart Beta factors and technical indicators as input features, this paper employs multilayer perceptron (MLP) and convolutional neural networks (CNN) to construct deep learning Smart Beta trading strategies. The frequency of updating and training deep learning model is quarter base from 2007 to 2017. The empirical results demonstrate that the performance of Deep Learning Smart Beta trading strategies is better than the Taiwan Capitalization Weighted Stock Index (TAIEX) and benchmark Smart Beta portfolio based on Asness, Frazzini, and Pedersen (2017). In addition, the refined feature through feature selection processes can further enhance the performance of deep learning Smart Beta trading strategies.