隨機森林演算法於量化選股策略中的應用：以台股為例

Abstract

本研究為探討機器學習方法應用於量化投資選股策略上之成效，並採用隨機森林演算法進行量化投資選股策略，以台股2013年1月1日至2017年12月31日 1600多支上市櫃股票之月横截面因子資料與報酬為樣本，進行模型之訓練與測試，再以2018年1月1日至2019年3月1日資料為樣本外驗證集，進行回測驗證模型之實效性。經實測後，特定轉換過之因子組合下，並以模型預測結果相對正報酬機率高的股票進行每月選股與換倉，能取得較投資元大台灣50更高的年化報酬。

This study explores the effectiveness of machine learning methods in quantitative investment stock selection strategies and uses a random forest algorithm to quantify investment stock selection strategies. The data range is more than 1700 Taiwan stocks from January 1, 2013 to December 31, 2017. The monthly cross-section factor data and the stock price rise and fall are taken as samples to train and test the model, and the data from January 1, 2018 to May 1, 2019 is used as the sample external verification set and the effectiveness of the backtest verification model is carried out. After the actual measurement, a number of specific converted factors are used, and the stocks with a high probability of relative positive returns are used for backtesting, and good investment results can be obtained.