Beyond Traditional Risk–Return: A Hybrid SEM and Neural Network Approach to Climate Risk, Mispricing, and Investment Decision Efficiency

Abstract

This study explores the transformation of the risk-return paradigm in the contemporary era through the integration of artificial intelligence (AI), climate risk, and the dynamics of market efficiency. Adopting a Systematic Literature Review (SLR) methodology following the PRISMA protocol, 47 selected articles from the Scopus database were comprehensively analyzed to ensure the quality of the findings. The study's results reveal a fundamental shift from conventional linear models to a multidimensional framework that can accommodate market complexity and environmental uncertainty. Econometric analysis of CAT bonds confirms the existence of mispricing driven by information asymmetry, while multi-criteria portfolio optimization techniques demonstrate that the inclusion of climate risk variables significantly alters the investment decision architecture. Furthermore, the implementation of generative neural networks has been shown to outperform traditional models in predicting volatility in volatile market conditions through a non-linear approach. Bibliometric mapping via VOSviewer identifies asset pricing, geopolitical risk, ESG, and investor sentiment as dominant and promising research clusters for the future. This study contributes theoretical novelty by synergizing econometric, optimization, and AI approaches, and provides practical guidance for data-driven investment management.