Machine Learning Feature Importance vs. Fixed-Effects Regression: Which Variables Actually Drive CSR–Firm Value Relationships in Africa?

Domain:

socioeconomic

Record type:

paper

Creator:

MutProOsm

Publisher:

Spr

Host:

Abstract Background Empirical research on the relationship between CSR disclosure and firm value in African financial markets shows methodological consistency and uses mostly fixed-effects regressions together with carefully chosen theoretical controls. Such an approach implies linearity and may suffer from variable selection bias, omitting important nonlinearities and thresholds. Methods This study employs feature importance rankings obtained through the Random Forest and Gradient Boosting machine learning algorithms combined with Shapley Additive Explanations for a balanced panel with 800 firm-years of data, consisting of 80 firms in South Africa, Nigeria, Kenya, and Ghana for the years 2015–2024. Results are compared against two-way fixed-effects regression coefficient estimates based on 83 predictors. Results There is considerable congruence between the results obtained via the three methods used in this paper with regard to the most important predictors: the first four places are taken by the CSR score, return on assets, institutional quality, and firm size. In turn, the fixed-effects approach fails to give proper weight to board independence, political risk, CEO duality, and environmental disclosure due to the presence of thresholds or nonlinearity. The SHAP algorithm reveals a threshold value of 40% for board independence, an interaction effect between CSR disclosure and firm size limited to the small-cap category, and an effect of CEO duality depending on institutional quality. Conclusions Machine learning feature importance rankings serve as valuable means of variable discovery that help improve specification of fixed-effects regressions. The suggested two-step procedure involves variable discovery through machine learning followed by causal inference with the use of fixed-effects regression. The proposed congruence framework is replicable to other CSR panels. JEL Classification: C55; G15; M14; O55

Visit

doi.org

Licenses

https://creativecommons.org/licenses/by/4.0/

Does CSR Improve Firm Value?

The relationship between Corporate Social Responsibility (CSR) and firm value has garnered significa

Machine Learning Models Evaluation and Feature Importance Analysis on NPL Dataset

Predicting the probability of non-performing loans for individuals has a vital and beneficial role f

Feature importance score.

There is a substantial increase in sexually transmitted infections (STIs) among men who have

The feature importance of the climate variables in predicting maize yield in each region of Ghana.

The feature importance of the climate variables in predicting maize yield in each region of Ghana

Effects of Some Monetary Variables on Fixed Investment in Selected Sub-Saharan African Countries

Monetary variables are not only important for the attainment of stable inflation but also for exerci

Weighted variable importance of predictor variables given by the three machine-learning models included in the model ensemble.

(A) West Africa; (B) East Africa. Stacked bars show the relative variable importance given by XGB