بررسیهای حسابداری و حسابرسی (Sep 2024)
Comparative Analysis of Artificial Neural Networks and Linear Regression in Predicting the Continuation of Shareholders' Overreaction Trends
Abstract
ObjectiveOverreaction is a noticeable anomaly in financial markets that leads to various consequences, including market inefficiency. As such, one of the prominent topics investigated in major global stock exchanges is shareholders' overreaction. This phenomenon is particularly prevalent in emerging and less developed markets, where investors tend to overreact to financial events. Overreaction, as a behavioral bias, distorts investors' decision-making in uncertain conditions, pulling the market away from its efficient state. Predicting and identifying the persistence of these reactions can assist investors in making more rational decisions regarding the purchase or sale of shares and other securities. Therefore, predicting and identifying the continuation of shareholders' overreaction trends can serve as a valuable tool for investors, financial analysts, and investment managers to make decisions based on intuition and precise analysis. To create an effective predictive model, various methods, such as regression analysis, can be employed to analyze the relationship between different variables and the continuation of shareholders' overreaction trends. Additionally, artificial neural networks (ANNs), as an advanced method, can be used to model the non-linear complexities and intricate connections between variables. This topic is directly related to predicting shareholders' behavior and investment decision-making, ultimately helping improve investment strategies and risk management. Hence, the main objective of this research is to conduct a comparative analysis of artificial neural networks and linear regression in predicting the continuation of shareholders' overreaction trends.MethodsThe present study is descriptive-causal, utilizing ex post facto research. To test the research hypotheses, multivariate linear regression based on panel data and a combination of time series was employed. Data was collected using the library research method, and necessary information was gathered by studying the financial statements of companies within the statistical population. The statistical population includes all companies listed on the Tehran Stock Exchange between 2011 and 2021, with 110 companies selected through systematic elimination sampling. In data analysis, regression methods were used to examine the relationships between variables, and the results were compared with those obtained from artificial neural networks.ResultsThe results indicate the superiority of the artificial neural network model in terms of the coefficient of determination and the MSE (Mean Squared Error) index. Specifically, the highest coefficient of determination for the artificial neural network (with 1 hidden layer and 9 neurons) for test data is 0.3880, compared to 0.349 for the linear regression model. Moreover, the results show that the MSE for the artificial neural network (1 hidden layer and 9 neurons) for test data is 0.003266, compared to 0.004 for the linear regression model. Thus, similar to the coefficient of determination, the MSE index is also better in the case of the artificial neural network.ConclusionThe artificial neural network model is capable of uncovering complex and non-linear patterns, providing the most accurate predictions. By using this model, stock return trends can be predicted more precisely and reliably.
Keywords
