Dual-CLVSA: A Deep Learning Approach to Financial Market Prediction using Trading Data and Sentiment Measurements

Department INFORMATION SYSTEMS AND OPERATIONS MANAGEMENT (ISOM)

Speaker : Hongwei (Harry) ZHU

from University of Massachusetts Lowell

Abstract:

"It is a challenging task to predict financial markets. The complexity of this task is mainly due to the interaction between financial markets and market participants, who cannot stay rational and are often affected by emotions. Extending a hybrid convolutional LSTM-based variational sequence-to- sequence model with attention (CLVSA) model, we develop dual-CLVSA to predict financial market movement with both trading data and the corresponding social sentiment measurements, each through a separate sequence-to-sequence channel. We evaluate the performance of our approach with backtesting on historical trading data of SPDR SP 500 Trust ETF over eight years. The experiment results show that dual-CLVSA can effectively fuse the two types of data. We further show that sentiment measurements are informative for financial market prediction and using dual-CLVSA we can extract profitable features to boost the performance of our prediction system."