A Study On Stock Price Trend Forecasting Based On GBDT Model

While doing investment analysis,it may be one-sided if you just use one or several technical indicators,however,to use too many indicators will lead to operational difficulties.This paper is designed to examine the applicability of the Gredient Boost Decision Tree model in predicting the stock price’s trend,which combines a set of technical indicators.First,discretizing technical indicator values as an input matrix of the model.Second,using the future price of the stock to classify the trend as category labels of the model.Third,the daily trading data of SH300 index,000623.sz,600115.sh,600999.sh from 2010 to 2014,2011 to 2015,2012 to 2016 years are used as sample data to train GBDT model.At last,doing strategy back testing with the buy signal and sell signal generated by GBDT model.The empirical results show that the trend signal generated by the GBDT model is slightly behind the real trend.However,the profitability of the GDBT model is significantly higher than the buy-hold strategy.It has been shown that the GDBT model has caught the trends of ups and downs from 2014 to 2016,which indicates that the trend signal generated by the GBDT model have a certain reference value.The innovations of this paper are as following: firstly,it is the first time to apply GBDT model in forecasting the stock trend.Secondly,by discretizing the technical indicators,it reduces the training difficulty and improves the training speed of the model.Thirdly,by using the stock’s future price to calculate the current trend label,the training model would be much closer to the reality.