Deep Learning Statistical Arbitrage

Statistical arbitrage identifies and exploits temporal price differences between similar assets. We propose a unifying conceptual framework for statistical develop novel deep learning solution, which finds commonality time-series patterns from large panels in data-driven flexible way. First, we construct portfolios of assets as residual conditional latent asset pricing factors. Second, extract the time series signals these with one most powerful machine solutions, convolutional transformer. Last, use to form an optimal trading policy, that maximizes risk-adjusted returns under constraints. conduct comprehensive empirical comparison study daily cap U.S. stocks. Our strategy obtains consistently high out-of-sample Sharpe ratio substantially outperforms all benchmark approaches. It is orthogonal common risk factors, asymmetric local trend reversion patterns. strategies remain profitable after taking into account frictions costs. findings suggest compensation arbitrageurs enforce law price.