Pairs trading on different portfolios based on machine learning

Victor Chang, Xiaowen Man, Qianwen Xu, Ching Hsien Hsu

Research output: Contribution to journalArticlepeer-review

485 Downloads (Pure)


This article presents an advanced visualization and analytics approach for financial research. Statistical arbitrage, particularly pairs trading strategy, has gained ground in the financial market and machine learning techniques are applied to the finance field. The cointegration approach and long short-term memory (LSTM) were utilized to achieve stock pairs identification and price prediction purposes, respectively, in this project. This article focused on the US stock market, investigating the performance of pairs trading on different types of portfolios (aggressive and defensive portfolio) and compare the accuracy of price prediction based on LSTM. It can be briefly concluded that LSTM offers higher prediction precision on aggressive stocks and implementing pairs trading on the defensive portfolio would gain higher profitability during a specific period between 2016 and 2017. However, predicting tools like LSTM only offer limited advice on stock movement and should be cautiously utilized. We conclude that analytics and visualization can be effective for financial analysis, forecasting and investment strategy.

Original languageEnglish
Article numbere12649
Pages (from-to)1
Number of pages25
JournalExpert Systems
Issue number3
Early online date18 Nov 2020
Publication statusE-pub ahead of print - 18 Nov 2020
Externally publishedYes

Bibliographical note

Funding Information:
This work is supported by VC Research (VCR 0000052) and the National Natural Science Foundation of China (Grant No. 61872084).

Funding Information:
National Natural Science Foundation of China, Grant/Award Number: 61872084; VC Research, Grant/Award Number: VCR 0000052 Funding information

Publisher Copyright:
© 2020 John Wiley & Sons Ltd

Copyright 2020 Elsevier B.V., All rights reserved.


Dive into the research topics of 'Pairs trading on different portfolios based on machine learning'. Together they form a unique fingerprint.

Cite this