A Comparative Study on the Application of Machine Learning to Dynamic Rebalancing (DSI-SRP)
This DSI-SRP fellowship funded Ziyao Zhang to work in the laboratory of Dr. Mike Neamtu, Ph.D. in the Department of Mathematics during the summer of 2023. Ziyao is a rising junior with majors in Computer Science and Mathematics and a minor in Economics.
In this DSI research, Ziyao and his mentor studied the financial market employing a quantitative approach using data science methodologies. As applications of machine learning algorithms become more common in the field of finance, quantitative strategies were proven to yield excess earnings when compared to traditional value investment. In this research, they used historical financial data and portfolio theories to discover the most effective portfolio rebalancing approach under constraints and the use of multi-dimensional fundamental and technical inputs in the estimations of portfolio returns. They tested the usefulness of reinforcement learning theories in forecasting stock prices and analyze the performance of data fitting. They specifically applied SVM and the combination of autoencoders and LSTM neural networks to forecast the percentage change of individual equities and portfolios. They also implemented the Hidden Markov model (supervised machine learning algorithm for classification) to determine market regimes and use the expected market regime as inputs to the general deep learning neural network. Through the research, they disproved the effectiveness of using technical indicators to predict stock price changes. Even though articles exist demonstrating models that can converge and able to yield a high win rate, they think that this is due to the misuse of futuristic data and the lack of standardized inputs and outputs. They have also achieved 98% accuracy in bull and bear market classification. In their future studies, they will not only apply multi-dimensional input to the Hidden Markov Model to generate more statistically significant classifications of market regimes, but they will also implement this updated result with spatio-temporal analysis regarding the performance of factor-based portfolios.
In addition to receiving support through a DSI-SRP fellowship, this project was supported and facilitated by the DSI Data Science Team through their regular summer workshops and demo sessions.