A Bayesian Approach to Stock Trading

Abstract

This project focuses on using Probabilistic Programming and more specifically using the Bayesian approach to devise an effective strategy to trade. This project does so by implementing a novel model on co-integration for pairs-trading using probabilistic programming. As opposed to using the traditional and simpler frequentist approach for pair determination I have implemented a more sophisticated Bayesian approach for pair trading using probabilistic programming. Pair trading is a market neutral strategy that enables traders to profit from virtually any market conditions be it uptrend, downtrend, or sideways movement. It is characterized as statistical arbitrage and convergence trading strategy. Pair Trading combined with co-integration as criteria makes for a successful and reliable trading strategy. Unlike simpler frequentist cointegration tests, the Bayesian approach allows to monitor the relationship between a pair of equities over time, which further allows to follow pairs whose cointegration parameters change steadily or abruptly. Bayesian statistics also accounts for uncertainty in in making predictions. It provides with mathematical tools to update beliefs about random events considering seeing new data or evidence about those events and it can do without having the need for a large dataset. It interprets probability as a measure of believability or confidence that an individual might possess about the occurrence of a particular event while including uncertainty in the equation. Along with a mean reversion trading algorithm, this approach can be effectively used as a viable trading strategy, open for further evaluation and risk management.