Identifying CSE market signals by integrating technical analysis with machine learning techniques

Abstract

Investing in publically listed companies in Colombo Stock Exchange is a trendy option among investors. These investments, though carry extremely higher risks, if decisions are taken correctly, investors can achieve extremely higher returns, potentially multiplying the initial investment. However, due to the market volatility, the investors have to keep their close eye on the market sentiments to get the maximum gain, which is not an option for those who don’t have time for active trading. Technical indicators are mathematical and statistical equations based on historical stock price and volume data visually represented as graphs to help investors understand market sentiments. In this study, we model the problem of stock market investment as a game and try to solve it with a reinforcement learning algorithm. The neural network is a recurrent neural network that understands the market patterns within a 14 day moving window period. The study includes multiple experiments, training the model with different training data sets, different episode counts, and different types of RNN strategies in the neural network, gated recurrent networks (GRU networks ) and long short-term memory networks ( LSTM networks ), and comparing their profits in different periods. The study’s findings suggest that we cannot create a ’train once use forever’ model for stock market predictions. The model performs better when we train it with a recent batch of data, and the model has to be retrained periodically with the previous 200-day price points. Further, it suggests that GRU networks work significantly better than LSTMs and for training, using an episode count of around 500 is sufficient to train.