Modeling Stock Data Using Multiple Linear Regression and LASSO Regression Analysis

Abstract

This study evaluates and compares the model fitting and predictive performance of Multiple Linear Regression (MLR) and Least Absolute Shrinkage and Selection Operator (LASSO) regression in the context of stock price prediction for four leading Nigerian companies. A dataset comprising 1,300 observations from 2019 to 2025 was obtained from Yahoo Finance and Investing.com. Multicollinearity assessment using the Variance Inflation Factor (VIF) revealed substantial collinearity among certain predictors, particularly for the variables "Open" (Honeywell: 55.45; Zenith: 920.30) and "Low" (Oando: 621.81), indicating the need for variable selection or dimensionality reduction. Comparative analysis based on model selection criteria, including Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) demonstrated the superior performance of LASSO over MLR across all companies. For example, Honeywell's LASSO model recorded an AIC of –12,112.64 and an MSE of 0.000021, compared to MLR's AIC of –2,690.54 and MSE of 0.00998. LASSO regression also identified key predictors such as "High" price, which exhibited strong statistical significance for Oando (z = 18.991, p < 0.001) and Zenith (z = 7.066, p < 0.001), whereas trading volume generally showed weak predictive power. The study concludes that LASSO provides a more parsimonious and accurate predictive model for financial time-series data. It is recommended for use in financial forecasting and investment analysis, particularly when dealing with multicollinear datasets and high-dimensional predictor variables.