LGBM on Stock Returns Prediction and Portfolio Construction

Hongyu Zhu

2025

Abstract

Investors need accurate stock price forecasts because they raise the chances of successful investments. This research assesses how machine learning methods predict the next day's daily returns of S&P 500 companies based on the past 10 days' data. The study also focuses on building portfolios based on forecasted returns. The study's dataset contains stock prices for S&P500 index companies and S&P500 index prices from 2014 to 2017, while including 497,472 total data points. The extended dataset includes 18 features which cover stock market relationship indicators as well as technical analysis indicators. This study assessed predictive models which consist of Random Forest, eXtreme Gradient Boosting (XGBoost) as well as Long Short-Term Memory (LSTM) and Light Gradient Boosting Machine (LGBM). The analysis results show that the LGBM model shows superior accuracy in forecasting the next day’s daily returns since it achieves an R-square value of 0.7377. Concurrently, this study utilized the predicted daily returns to construct a portfolio comprising 20 companies from the S&P500 index companies. Based on the objective of maximizing the portfolio alpha, the optimal portfolio result contains 20 stocks of S&P500 companies, achieving an alpha over 10% and a total return of by 20%.

Paper Citation

in Harvard Style

Zhu H. (2025). LGBM on Stock Returns Prediction and Portfolio Construction. In Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE; ISBN 978-989-758-765-8, SciTePress, pages 521-526. DOI: 10.5220/0013701300004670

in Bibtex Style

@conference{icdse25,
author={Hongyu Zhu},
title={LGBM on Stock Returns Prediction and Portfolio Construction},
booktitle={Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE},
year={2025},
pages={521-526},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013701300004670},
isbn={978-989-758-765-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Data Science and Engineering - Volume 1: ICDSE
TI - LGBM on Stock Returns Prediction and Portfolio Construction
SN - 978-989-758-765-8
AU - Zhu H.
PY - 2025
SP - 521
EP - 526
DO - 10.5220/0013701300004670
PB - SciTePress