Document Type : Original Article
Authors
1
Ph.D. in Financial Management, Razi University, Kermanshah, Iran
2
M.Sc. in Financial Management, Razi University, Kermanshah, Iran
3
Ph.D. Candidate in Financial Management, Razi University, Kermanshah,
Abstract
Accurate asset pricing remains one of the fundamental challenges in finance and investment analysis. Traditional asset pricing models, including the Capital Asset Pricing Model (CAPM) and multifactor models, are primarily based on linear assumptions and static relationships among variables, which limits their ability to capture the complex and interconnected nature of financial markets. Recent advances in deep learning, particularly Graph Neural Networks (GNNs), have provided powerful tools for modeling nonlinear dependencies and structural interactions within complex systems. This study aims to develop a novel asset pricing framework inspired by Graph Neural Networks. In the proposed approach, financial assets and their interrelationships are represented as a graph structure, enabling the model to incorporate information propagation and interaction effects among assets when estimating expected returns. The proposed model is trained using capital market data, and its performance is evaluated against conventional asset pricing models and alternative machine learning techniques using standard prediction and pricing metrics.
The primary contribution of this research lies in exploiting the network structure of financial markets and uncovering hidden relationships among assets to enhance pricing accuracy. Furthermore, the proposed framework enables the integration of fundamental, trading, and inter-firm relationship data within a unified graph-based architecture. To evaluate the effectiveness of the model, several performance measures, including forecasting error, explanatory power, and risk assessment indicators, will be employed. It is expected that the graph-based framework will improve return forecasting accuracy, enhance risk factor identification, and increase the explanatory power of asset pricing models. In addition, the proposed approach can support portfolio management, optimal asset allocation, and intelligent investment strategy design. The findings of this study may contribute significantly to the advancement of artificial intelligence-driven financial modeling and the emerging field of network-based asset pricing.
Keywords
