Predictive intraday correlations in stable and volatile market environments: Evidence from deep learning

Standard methods and theories in finance are ill equipped to capture complex data interactions presented in financial prediction problems. Deep learning approaches however offer more useful insights into these complex big data interactions. In this paper, using deep-layered feedforward neural networks, which applies econometrically constructed gradients, we learn and exploit time-shifted correlations among S&P 500 stocks to predict intraday and daily stock price movements for target stocks with only other stocks' lagged prices as inputs. Our findings show that time-shifted correlations can be exploited to predict stock prices; our model is also consistent in volatile markets.