Have a personal or library account? Click to login
Toward responsible investment: A deep learning frame-work for ESG-differentiated portfolio optimization Cover

Toward responsible investment: A deep learning frame-work for ESG-differentiated portfolio optimization

By: Minh Duc Nguyen  
Open Access
|Jan 2026

Abstract

Aim/purpose – This paper proposes a framework that integrates deep learning-based return forecasting with environmental, social, and governance (ESG)-differentiated optimization to align portfolio performance with financial and sustainability goals, enabling data-driven responsible investment decisions. The study hypothesizes that ESG risk dimensions influence portfolio performance differently: mitigating environmental risk imposes higher financial costs due to regulatory and operational pressures, whereas social and governance risks yield more balanced return-sustainability trade-offs.

Design/methodology/approach – This study employs the N-BEATS deep learning model to forecast one-day-ahead returns for S&P 100 stocks. The predicted returns serve as inputs to an enhanced Mean-Variance with Forecasting (MVF) model that integrates ESG risk as a penalty term. ESG factors are analyzed both collectively and across individual dimensions using a tunable risk-aversion parameter that reflects investor preferences. The dataset includes 99 S&P 100 stocks from January 2017 to December 2024, with distinct training, validation, and test sets for model development and evaluation.

Findings – The study reveals that incorporating ESG risk into portfolio optimization with forecasted returns produces distinct trade-offs across ESG dimensions. Mitigating environmental risk entails the greatest return cost, whereas social and governance risks allow more favorable balances between return and sustainability. The N-BEATS model achieves sufficient forecasting accuracy to inform investment decisions. Moreover, the elbow point method offers a practical means for selecting optimal ESG sensitivity levels, enabling investors to effectively balance performance and sustainability objectives.

Research implications/limitations – This research demonstrates that combining deep learning-based forecasting with ESG-differentiated optimization enables more nuanced and responsible investment strategies, offering practical tools for aligning financial and sustainability goals. However, the study is limited by its focus on a single market (S&P 100) and does not account for real-world factors such as transaction costs or dynamic re-balancing, which could affect practical applicability.

Originality/value/contribution – This study is among the first to integrate N-BEATS-based return forecasting with ESG-differentiated portfolio optimization in a unified framework. It offers a novel approach that enables investors to explicitly manage trade-offs between financial returns and individual ESG risk dimensions, providing both methodological innovation and practical guidance for responsible investing.

DOI: https://doi.org/10.22367/jem.2026.48.01 | Journal eISSN: 2719-9975 | Journal ISSN: 1732-1948
Language: English
Page range: 1 - 16
Submitted on: Mar 3, 2025
|
Accepted on: Dec 4, 2025
|
Published on: Jan 22, 2026
In partnership with: Paradigm Publishing Services
Publication frequency: 1 issue per year

© 2026 Minh Duc Nguyen, published by University of Economics in Katowice
This work is licensed under the Creative Commons Attribution-NonCommercial 4.0 License.