Comparative Analysis of Regression Models for Tesla Closing-Price Prediction

This study addresses the challenge of short-term stock-price forecasting by comparing six regression techniques for predicting the same-day closing price of Tesla, Inc. (TSLA). A ten-year dataset (September 2014–September 2024) of daily open, high, low, close, and volume data was enriched with technical indicators—simple and exponential moving averages, relative strength index, and on-balance volume—and split chronologically into 80% training and 20% testing sets. Models evaluated include ordinary least squares, ridge regression (L2 regularization), lasso regression (L1 regularization), k-nearest neighbors, random forest, and gradient boosting. Hyperparameters were selected via nested five-fold, time-series cross-validation, and out-of-sample performance was measured by root mean squared error, mean absolute error, mean absolute percentage error, and coefficient of determination. Results indicate that ridge regression with a tuned penalty coefficient (α = 0.1) achieved the lowest test RMSE of $2.60, closely followed by ordinary least squares with RMSE of $2.53, MAE near $2.00, MAPE under 1%, and R2 above 0.99. In contrast, k-nearest neighbors and ensemble methods exhibited significant overfitting. These findings demonstrate that carefully engineered technical features combined with regularized linear models yield robust forecasts for highly volatile equities.