Indicator selection for daily equity trading with recurrent reinforcement learning

Abstract

Recurrent reinforcement learning (RRL), a machine learning technique, is very successful in training high frequency trading systems. When trading analysis of RRL is done with lower frequency financial data, e.g. daily stock prices, the decrease of autocorrelation in prices may lead to a decrease in trading profit. In this paper, we propose a RRL trading system which utilizes the price information, jointly with the indicators from technical analysis, fundamental analysis and econometric analysis, to produce long/short signals for daily trading. In the proposed trading system, we use a genetic algorithm as a pre-screening tool to search suitable indicators for RRL trading. Moreover, we modify the original RRL parameter update scheme in the literature for out-of-sample trading. Empirical studies are conducted based on data sets of 238 S&P stocks. It is found that the trading performance concerning the out-of sample daily Sharpe ratios turns better: the number of companies with a positive and significant Sharpe ratio increases after feeding the selected indicators jointly with prices information into the RRL system.