Application of Machine Learning With News Sentiment in Stock Trading Strategies

This study empirically tested the feasibility of machine learning in trading strategies using technical indicators and news information as feature variables for learning. Six were adopted this study, including moving average (MA), convergence/divergence (MACD), relative strength index (RSI), stochastic oscillator (KD), on-balance volume (OBV), sentiment ratio (SR) developed via text mining. Selected models, support vector (SVM), eXtreme Gradient Boosting (XGBoost), recurrent neural network (RNN), long short-term memory (LSTM), also employed investigation. backtested daily historical data constituent stocks Taiwan Top 50 ETF from January 1, 2003, to December 31, 2018, three categories along with conventional countertrend operations. The following conclusions drawn after analyzing performance these various means: 1. Technical such MA, MACD, RSI performed poorly most cases. 2. Specific parameters importance several indicators, RSI, OBV. 3. OBV was a indicator positive impact on strategies. 4. learning-based XGBoost models able outperform under specific scenarios. 5. SR, could not significantly improve models. empirical results suggest that machine-learning are capable long-term stock price movements some extent.