Random Forest Algorithm vs. Linear Regression for Financial Performance Determinants: Evidence from Iraqi Mixed-Sector Firms

Abstract

This paper examines the predictive factors of financial performance among Iraqi mixed-sector firms, comparing the predictive power of traditional Ordinary Least Squares (OLS) regression with the Random Forest (RF) machine learning algorithm. Utilizing annual data from 11 firms over the period 2015–2020, the study measures financial performance using Return on Assets (ROA), Return on Equity (ROE), and Return on Sales (ROS). OLS regression identified the directional relationships and p-values, whereas Random Forest evaluated predictive accuracy and modeled nonlinear relationships. The OLS results indicate that firm size, liquidity, productivity, and cash-to-asset ratio have significant positive effects on financial performance, while the debt ratio shows a significant negative effect, particularly on ROA. The impact of fixed asset investment is mixed; it is statistically insignificant for ROA but negatively associated with ROE. The explanatory power R2 of the OLS models varies widely, accounting for 56.9% of the variation in ROA, 72.3% in ROE, and 24.9% in ROS. In comparison, the Random Forest yields an R2 of 0.395 for ROA and demonstrates a lower prediction error than OLS, particularly for ROS which exhibits more complex nonlinear relationships. Finally, both methods yield consistent results for ROE, whereas OLS provides higher explanatory estimates for ROA. Our findings highlight that these financial determinants are central to enhancing corporate financial performance via liquidity management, capital structure optimization, and productivity gains. Crucially, this study demonstrates that combining traditional econometric modeling with machine learning provides complementary insights, thereby maximizing both statistical interpretability and prediction accuracy within the context of emerging markets.