Increasing FOREX trading profitability with LSTM candlestick pattern recognition and tick volume indicator

The paper introduces the research of the effectiveness of using LSTM (Long–Short Term Memory) for candlestick data and a technical analysis indicator for a large number of the most common currency pairs (27 in total) over a long period in order to build automatic trading strategies. The achieved average total and annual return for 8 years of a model trading were 286% and 15.4%, respectively. It is more than 50 times higher than the values for the classic Buy & Hold trading strategy for the same period. In addition, the paper introduces a new technical indicator based on tick volumes which is an alternative trading strategy (the total and annual returns of LSTM models exceed it by an average of 7.2 and 2.3 times) as well as an additional feature to increase the profitability of the neural network strategy through the use of ensemble learning. For 37% of the analyzed currency pairs, the use of an ensemble of LSTMs allows one to increase further the total return by an average of 17.2%.