Estimate stock returns using CAPM models and CAPM integration with neural networks (comparative study): An analytical study in the Iraq Stock Exchange

Abstract

The current research presents the idea of ​​using deep learning tools and employing them in financial aspects due to their significant role and ability to explore unobservable aspects in light of financial models governed by a set of restrictions, conditions and linear relationships. On the other hand, the nature of financial data that tends to be non-linear and suffers from the missing of monthly closing prices, which imposes a state of data loss. All of this provides preference for deep learning models, including the neural network tool. The research aims to estimate financial returns in light of the capital asset pricing model CAPM as a financial model and neural networks as a deep learning tool in addition to the mask & padding tool to address the problem of missing data. The knowledge gap was determined by the inability of the capital asset pricing model to explore hidden and invisible aspects and overcome non-linear relationships. The research sample consisted of 42 organizations listed on the Iraq Stock Exchange for the period from 1/1/2021 to 31/12/2024 with 60 observations. The research concluded that the neural network tool is able to overcome the determinants in light of financial models and provide accurate estimates of returns are close to estimates under the capital asset pricing model.