Using Machine Learning Algorithms to Estimate the Functional Form of Optimal Trading Strategies

The Laws of Information for Traders are developed and a framework for the solution of stochastic program- ming problems relevant to trading with an alpha are presented. The solution is expressed in terms of a holding function which represents the optimal policy for a trader to follow at any given decision point based on their alpha, their current position, and any pertinent control parameters. Traditional analysis of this problem uses analytic or numerical integration and analytic or numerical optimization. Machine learning provides an alternative tool to discover the functional form of optimal trading strategies as expressed through the holding function. An analysis of trading strategies shows that it is feasible to train elements of a machine learning system to reproduce these functions. Such a technique provides the potential to “shortcut” the computation- ally complex task of solving stochastic programming problems via neural networks, or other machine learning algorithms.