APPLICATION OF MACHINE LEARNING IN PAIR TRADING FOR VN30 AND VNMID

Nguyen Ba Hung^a, Dinh Thi Thu Ha^b*, Nguyen Dang Bac^c, Lu Minh Hiep^d, Nguyen Dac Anh Chinh^e

^a&b&c Thai Binh Duong University, 650000, Vietnam

^d&e Bao Viet Securities Company, 700900, Vietnam

ABSTRACT

This study applies machine learning methods to optimize pair trading strategies for VN30 and VNMID stock groups in Vietnam’s stock market. Using daily trading data from January 2020 to August 2024, the research proposes a two-stage analytical framework: structural break detection and trading signal optimization. Three models—the Simple Statistical Model, Decision Tree (DT), and Long Short-Term Memory (LSTM)—were developed and compared based on annualized returns, Sharpe ratio, and maximum drawdown. Results indicated that the DT model delivered the highest average return (1.33%), whereas the LSTM model provided better risk management, reflected by lower maximum drawdowns. Even after incorporating transaction costs (0.15%), machine learning models still significantly outperformed the Simple Statistical Model, with the DT model showing the highest proportion of profitable trading pairs (>5%). Additionally, a Majority Voting strategy was implemented to optimize pair trading portfolios, achieving an annualized return of up to 9.24% with an attractive Sharpe ratio (1.37) and low risk. The use of the AWS SageMaker platform enhanced model performance and market adaptability. Future research directions include utilizing tick-level data, implementing more advanced machine learning models, and integrating macroeconomic factors to further improve trading strategy effectiveness. Keywords: Pair Trading, Decision Tree, LSTM, Machine Learning, Structural Break Detection, Vietnam Stock Market.

---

* Corresponding author. Tel.: +84928603388.

 Email address: ha1.dtt@tbd.edu.vn

Thông tin:

Thuộc danh mục	Hội thảo khoa học HỘI THẢO KHOA HỌC QUỐC GIA VỀ KẾ TOÁN VÀ KIỂM TOÁN LẦN THỨ 7
Tạp chí
Nhà xuất bản
Trang
Tập
Số
Ngày xuất bản	7 2025
DOI