Analysis and Prediction of Patterns in Futures Trading Datasets Using LSTM
One of the most promising tools in recent years for the analysis and prediction of time series data, which includes financial market data has been the use of neural networks. To ensure the accuracy of the outcomes of these systems, it is critical to overcome the vanishing gradient and exploding gradient problems that often occur when recurrent neural networks (RNN) process data. Long Short-Term Memory (LSTM) has been shown to provide good performance when dealing with time series datasets. This paper will explore the feasibility of using an RNN with LSTM as a predictive tool for use with futures trading data. Using a dataset comprised of all futures trading occurring on the Bourse de MontrÃ©al (TMX) during a 9-month period from January to September 2015, we assessed the predictive effectiveness of an RNN in predicting the price of front-end contracts for the futures symbol BAX. We found that while an RNN provided a degree of short-term predictive capability, this capability did not extend beyond a couple of days. Although it failed as a trading instrument to predict futures prices, the RNN could detect, identify, and reflect underlying trends in the data, indicating the tool may hold promise in the detection of trading patterns.Â
Copyright (c) 2018 B. Park, C. Chmelyk, D. Heslop, G. Zhengyu
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Proceedings of Manitoba's Undergraduate Science and Engineering Research by University of Manitoba is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. The authors hold the copyright to published articles without restriction, and retain publishing rights.Â