The predictive power of mispricing Stocks based on financial and governance criteria, using linear and nonlinear models (CART, LASSO, PINSVR)

This paper introduces a predictive model for stock mispricing by examining the financial and governance factors that influence it. The model utilizes panel data from 133 companies listed on the Tehran Stock Exchange, selected through systematic elimination, during the years 2013 to 2022. It compares the predictive capabilities of these factors and evaluates the learning and predictive power of linear and nonlinear models using CART, LASSO, and PINSVR algorithms, which are considered artificial intelligence models in the fields of data mining and pattern recognition. The results indicate a significant difference in the error rates between linear and nonlinear models in predicting stock mispricing, suggesting that linear models, especially in times of high volatility, are less effective. Additionally, based on the Mean Absolute Error (MAE), the prediction of stock mispricing using corporate governance metrics generally indicates lower accuracy compared to financial metrics, even in nonlinear algorithms.