Pairs-Trading Strategies with Recurrent Neural Networks Market Predictions

We propose a deep learning approach to complement investor practices for identifying pairs trading opportunities. We refer to the reversal effect, empirically observed in many pairs of financial assets, consisting in the fact that temporarily market deviations are likely to correct and finally converge again, thereby generating profits. Our study proposes the use of Long Short-term Memory Networks (LSTM) to generate predictions on market performances of a large sample of stocks. We note that pairs trading strategies including such predictions can contribute to improve the performances of portfolios created according to gaps in either prices or returns.