Abstract
<jats:p>This study aims to compare the accuracy and efficiency of linear regression with nonlinear Logit and Probit models in short-term stock price forecasting. The study uses daily stock data of Isfahan Steel Company from January 1, 2024, to November 19, 2025. After eliminating multicollinear variables, the closing price was considered the dependent variable, while the number of trades, price-to-earnings ratio (P/E), net percentage of retail trading, and per capita buy and sell values of individual investors were selected as independent variables. An out-of-sample forecasting approach was applied, reserving the last 30 trading days as the test set. The results indicate that although linear regression shows acceptable explanatory power, it underperforms in short-term forecasting. The Logit model achieved the lowest MSE, RMSE, and MAE values, followed by the Probit model. The findings confirm the nonlinear nature of stock price dynamics and highlight the superior predictive performance of probability-based nonlinear models, particularly Logit.</jats:p>