Improving Portfolio Optimization Using Option Skewness Regimes

Abstract

Portfolio optimization methods have had many different approaches and additional developments since the introduction of modern portfolio theory and the mean-variance optima. Research has shown that the financial derivative of options contains valuable future-looking implied information regarding the markets that can drastically improve the Sharpe ratios of portfolios. While the most common use case has been using option implied volatility to improve underlying variance estimations, higher moments such as skewness has been found to also provide portfolio improvements. However, these higher moments are very difficult to predict correctly, and thus their impact has been largely neglected. The purpose of this thesis will be to see if applying the machine learning method of Hidden Markov Models will be effective in predicting option skewness regimes rather than explicit values, evaluating effectiveness as the ability to improve portfolio performance. By predicting regimes of option skewness, the goal will be to gain greater accuracy in evaluating these higher moments rather than predicting explicit values, and thus derive more accurate information to feed into portfolio optimization. The method of optimization will be held constant to control for studying the effect of implied information, and effectiveness of combining higher moments with regime prediction will be how much the portfolio optimization process is improved.