Developing a Hybrid Model to Estimate Expected Return Based on Genetic Algorithm

• Main Authors: Mehdi Asima, Amir Ali Abbaszadeh Asl
• Format: Article
• Language: fas
• Published: University of Tehran 2019-05-01
• Series: تحقیقات مالی
• Subjects: : genetic algorithm; hybrid model; local kernel regression; standard capital asset pricing model; threshold regression
• Online Access: https://jfr.ut.ac.ir/article_71567_56283bcda7539e9b1affc7c4ae0f7d8e.pdf
• ISSN: 1024-8153; 2423-5377

<strong>Objective:</strong> Capital asset pricing model (CAPM) has been among the most common models to estimate the expected return. In the standard CAPM model, a) the beta coefficient is fixed and b) the relationship between stock returns and market returns is assumed to be linear. While in financial markets, it is possible that the beta coefficient varies over time by changing the cost-benefit analysis on returns and risks, and also in a nonlinear environment, the beta coefficient estimate will be linearly inappropriate and oblique. Therefore, it seems necessary to use other models in estimating expected return. <br /><strong>Methods:</strong> In this study, in addition to the standard CAPM model, the threshold regression and kernel regression models were used to estimate the CAPM model. Considering that the basis of each of these models is based on different assumptions; therefore, this research has tried to use a genetic algorithm in the time period from 2008 to 2017 to propose a hybrid model in order to estimate the expected return. <br /><strong>Results:</strong> Expected return was calculated using standard CAPM, threshold regression, kernel regression and the hybrid model of these three models, and the results were compared with the realized returns. The mean square error (MSE) index was used to measure the predictive power of research models. Using the paired t-test on the mean square error, the research models were compared with each other. <br /><strong>Conclusion:</strong> The results show that applying the hybrid model increases the predictive power of realized return compared to other research models.