Bayesian hidden Markov models and strategies for buying and selling stocks

Abstract

Predicting stock market behavior is a highly sought-after area of research, offering valuable insights for both individual investors and corporations. This dissertation aims to address this challenge by utilizing Bayesian Hidden Markov Models to determine underlying stock market regimes in order to inform trading decisions. By working under a Bayesian framework, uncertainty in parameters is described through posterior distributions, allowing for a more uncertainty-aware and reliable approach when compared to traditional point estimates. This study utilizes the daily closing prices of three financial assets, varying in their inherent volatility, and uses results to implement regime-based trading strategies. The findings showed that the Bayesian hidden Markov model was able to identify regime shifts successfully, and implement trading strategies that outperformed a passive buy-and-hold trading approach. Moreover, results suggested that higher volatility assets yielded a greater return on investment by allowing the model to fully leverage its regime identification ability in more dynamic markets. Lastly, the implications that transaction costs have on the trading strategies were assessed, further simulating a real life trading scenario.