Ranking the Factors Affecting Financial Flexibility Using Artificial Neural Networks and Machine Learning

Abstract

This study aimed to identify and rank the factors affecting the financial flexibility of companies listed on the Tehran Stock Exchange and Iran Fara Bourse using artificial neural network and XGBoost machine-learning models. This applied, quantitative, retrospective, and correlational study used longitudinal firm-level data from companies listed on the Tehran Stock Exchange and Iran Fara Bourse during 2020–2024. Following the application of the inclusion and exclusion criteria, 143 companies were selected through a census approach. Financial flexibility was assessed based on cash holdings and debt capacity. The predictors included default risk, political connections, market power, managerial ability, environmental uncertainty, managerial overconfidence, chief executive officer power, management turnover, and board financial expertise. After data screening and preprocessing, an artificial neural network and the XGBoost algorithm were trained and evaluated. Model performance was assessed using the coefficient of determination and root mean squared error, while variable importance and the direction of predictor effects were interpreted through normalized importance scores and SHAP values. Both algorithms successfully ranked the determinants of financial flexibility, although the artificial neural network demonstrated superior predictive performance. The neural network achieved an (R^2) of 0.499 and an RMSE of 0.179, whereas XGBoost produced an (R^2) of 0.413 and an RMSE of 0.194. Default risk was the most influential predictor in both models, and lower default risk was associated with greater predicted financial flexibility. In the neural network, managerial ability, market power, political connections, and environmental uncertainty followed default risk in importance. In XGBoost, political connections, environmental uncertainty, managerial ability, and market power received the next highest rankings. Managerial overconfidence had moderate importance, whereas chief executive officer power, management turnover, and board financial expertise made comparatively limited predictive contributions. Training and validation errors converged satisfactorily, indicating acceptable model stability and no severe overfitting. Financial flexibility is primarily determined by credit quality, managerial competence, competitive position, institutional access, and environmental conditions, and artificial neural networks provide a more accurate framework than XGBoost for predicting and ranking these nonlinear determinants.