Date of Graduation
Bachelor of Science in Data Science
Committee Member/Second Reader
Committee Member/Third Reader
The rapid democratization of computing resources and advancements in data science have enabled the development of sophisticated automated trading systems. This experimental research paper investigates the efficacy of a novel autonomous portfolio management system that integrates deep learning techniques for asset selection, optimization, and trade execution. The investment universe includes all stocks in the S&P500. Data acquisition was conducted using the EOD Historical Data API for all historical and fundamental data. Every week, on the last trading day, the system employs a Long Short-Term Memory (LSTM) network, implemented in TensorFlow, trained for weekly forward return prediction. These predictions are combined with engineered alpha-factors to generate a favorability rating for each stock. Using a combination of the return prediction, favorability rating, and risk metrics, the portfolio optimization process combines Post-Modern Portfolio Theory (PMPT) and the Global Minimum Variance Portfolio (GMVP) to generate the best portfolio for the next week. The system underwent rigorous testing and was subsequently backtested over the period from February 4, 2022, to March 19, 2023. Over this time, the system returned 28.47%, generating an alpha of 40.50% when compared to the S&P500, which lost 12.97% over the same period. Unfortunately, live-trading implementation was not possible due to regulatory constraints and brokerage conditions. However, this paper still demonstrates the system's efficacy, reporting substantial improvements in key performance metrics compared when compared to traditional methods. Highlighting its potential as a scalable, accessible framework for future research in the field of autonomous portfolio management.
Data Science, Finance, Portfolio Management, Deep Learning, Portfolio Optimization
Kincannon, J. (2023). Autonomous Portfolio Management Using Deep Learning. Data Science Undergraduate Honors Theses Retrieved from https://scholarworks.uark.edu/dtscuht/2
Available for download on Thursday, May 11, 2028