Skip to main content
Open Access Publications from the University of California


UCLA Electronic Theses and Dissertations bannerUCLA

Building an Options Portfolio with Deep Learning


Recent applications of powerful machine learning models within the field of portfolio op- timization have shown promising results. This paper explores the application of similar methods to create an actively managed options portfolio, rather than a traditional portfolio containing stocks and ETFs. Although financial options are popular assets amongst both retail and institutional investors, they are almost exclusively traded for one of two purposes: hedging or intra-asset speculation. Little, if any, literature exists on the topic of inter-asset options strategies. Using percent change in implied volatility data as a proxy for single-day straddle returns, various machine learning models are trained by directly optimizing the Sharpe Ratio–a risk-adjusted return metric. The models output daily volatility positions in 315 underlying assets thereby creating the Options Portfolio. Results show significant potential in such a strategy, with the Attention Transformer model yielding a before-cost average annual Sharpe Ratio of 10.76 compared to the GRU model of 6.41, the LSTM model of 5.07, and the equal-weighted baseline of 0.53.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View