01/07/2021
By Tong Sun

Doctoral Dissertation Defense: Tong Sun, Thursday, Jan. 21, 2021, 10 a.m. EST, Online Zoom Meeting

Title: Predicting Financial Market Movement Using Deep Neural Networks

Ph.D. Candidate: Tong Sun
Time: Thursday, Jan. 21, 2021, 10:00 AM EST
Location: This will be a virtual defense via Zoom. Those interested in attending please contact tong_sun@student.uml.edu at least 24 hours prior to the defense to request access to the meeting.

Committee Members:

  • Chair (Advisor): Benyuan Liu, Professor, Department of Computer Science, University of Massachusetts Lowell
  • Yu Cao, Professor, Department of Computer Science, University of Massachusetts Lowell
  • Wei Ding, Professor, Department of Computer Science, University of Massachusetts Boston
  • Hongwei Zhu, Professor, Department of Operations and Information Systems, University of Massachusetts Lowell

Abstract:
Recently there have been many efforts to study the predictability of financial market trends using various machine learning approaches. Recent advancements of deep recurrent neural networks has brought significant improvements for time series prediction of financial market data. In this thesis we explore the idea of using machine learning and deep neural networks to analyze and predict financial market movements and achieve promising returns. Due to tremendous complexity of the financial market, we choose two different approaches to address this challenge.

We first investigate the relationship between market price movement and social media sentiment. We collect data from various types of social media sites (microblogs, chat rooms, web forums) and find that data from these sites exhibit distinct characteristics in activity level, post length, and correlation with market behavior. We then investigate several machine learning models to classify post sentiment and achieve state-of-the-art prediction results for short posts. We find that there is strong correlation and Granger causality between chat room post sentiment and stock price movement, indicating that post sentiments can be used to improve the prediction of price movement. Based on this finding, we propose a prediction model that uses chat room sentiment to forecast the stock market direction and develop a trading strategy that utilizes the prediction as trading signals. Backtesting using our strategy achieves promising portfolio returns. A total return of 19.54% is obtained at the end of the seven-month period with slippage and commissions taken into account, compared to a loss of -25.26% by a passive buy-and-hold baseline strategy.

We further propose another approach that adopts deep long short term memory (LSTM) as the main model architecture and predicts financial market movement using live data stream augmented with technical indicators from an online broker. Training and testing of our model is performed in a rolling fashion to ensure the validity and reliability of the prediction. We discuss the design trade-off of several variations of our model and evaluate the impact of various parameter choices on the model performance using backtesting. We further design and implement an end-to-end deep learning based trading platform to evaluate our approach. This platform integrates all the core components of a trading system and can effectively evaluate quantitative models and perform backtesting, paper trading and real trading using real-time market data. The platform is designed to enable automated learning and tuning capabilities for both prediction model and trading strategy in a rapidly changing market environment. Backtesting and live paper trading of our model on this platform achieves promising returns. Moreover, a total return of 58.69% is obtained with live paper trading for a twelve-month period, which demonstrates the effectiveness of our proposed approach.