The concept of sustainability has gained significant importance due to environmental degradation. Enterprise Risk Management (ERM), a holistic approach to risk management, is increasingly adopted by companies aiming for sustainable operations, improved performance, and controlled risks. Unlike traditional silo-based risk management, ERM considers interconnected risks across all organizational levels. While some argue that the value proposition of ERM remains unclear (Lundqvist, 2015), others suggest it enhances decision-making and firm performance (Meidell and Kaarbøe, 2017). This study addresses the gap in research by analyzing the relationship between ERM adoption, firm performance, firm value, and various risk indicators for banking companies listed on the Borsa Istanbul XBANK index. The study uses panel data regression models and PLSR to determine the influence of ERM implementation level and organizational structure sophistication on the dependent variables.

Existing literature extensively explores ERM implementation and its relationship with firm performance and value. Gordon et al. (2009) found a positive association between ERM and firm performance across various industries. McShane et al. (2011) showed a positive correlation between ERM ratings and firm value. Hoyt and Liebenberg (2011) also reported a favorable relationship between ERM utilization and firm value in the insurance sector. Other studies have investigated the impact of ERM on accounting performance (Florio and Leoni, 2017), operational performance (Callahan and Soileau, 2017), and firm characteristics (Farrell and Gallagher, 2019). However, some research indicates that ERM adoption doesn't always improve performance in non-financial sectors (Otero González et al., 2020). This study aims to contribute to this complex discussion by focusing on the banking sector and employing multiple regression techniques, including data mining methods, to analyze the impact of ERM.

This study uses panel data (PD) from ten banking companies listed on the Borsa Istanbul XBANK index, covering the period from 2019 to 2022. The data includes financial ratios (ROA, ROE, Q, Sz_score, Vol, Beta), ERM implementation level (ordinal variable with four levels: basic, standard, modern, advanced), and ERM organizational structure sophistication (ordinal variable with three levels: low, moderate, high). Three panel data regression models (pooled regression, fixed effects model, random effects model) are employed to analyze the relationship between independent variables (IVs) and dependent variables (DVs). Specification tests (Hausman test, modified Wald test, Levene's test, Pesaran's CS dependence test, Bhargava et al. Durbin-Watson test, Baltagi-Wu LBI test) were performed to select the most appropriate model for each DV. To address heteroscedasticity, robust regression models were used for some DVs. Partial Least Squares Regression (PLSR) is used as an alternative prediction model to compare its performance with panel data models. The prediction accuracy of the models is evaluated using R-squared (R²). The SIMPLS algorithm is used for PLSR, and cross-validation is implemented to determine the optimal number of components for each PLSR model.

The study's key findings are based on panel data regression models and Partial Least Squares Regression (PLSR) analyses. Using robust regression models due to heteroscedasticity in the data, the results reveal statistically significant relationships between ERM adoption and various financial indicators. Specifically, a higher ERM implementation level significantly and positively impacts the return on assets (ROA) at the 99% confidence level. The sophistication level of the ERM organizational structure significantly and positively impacts return on equity (ROE) at the 90% confidence level, the Tobin's Q (firm value) at the 90% confidence level, and the standard z-score (insolvency risk) at the 95% confidence level. However, no significant relationship was found between ERM adoption and volatility (Vol) or systematic risk (Beta). When comparing the prediction accuracy of panel data models and PLSR, PLSR models (PLSR10 for ROA, ROE, Q; PLSR6 for Sz_score, Vol) generally demonstrated higher R² values compared to the robust regression and random effect models; however, Beta was better predicted using the robust regression model. The specific R² values for each model are detailed in Table 14 in the paper.

The findings of this study support the notion that implementing an effective ERM system positively influences the financial performance, firm value, and insolvency risk of banking companies. The positive relationship between advanced ERM implementation and higher ROA aligns with the idea that better risk management leads to improved operational efficiency and profitability. The positive impact of a sophisticated ERM organizational structure on ROE, Q, and Sz_score suggests that a well-structured ERM framework enhances strategic decision-making, leading to greater firm value and reduced insolvency risk. The insignificant impact of ERM on volatility and systematic risk might be due to other macroeconomic factors and market dynamics affecting these variables. The higher R² values obtained from PLSR models for some indicators compared to the panel data models suggest that data mining techniques may offer improved predictive capabilities in this domain. The overall results highlight the importance of considering both the implementation level and the organizational structure of ERM for maximizing its benefits.

This study provides empirical evidence supporting the positive impact of ERM adoption on the financial performance, firm value, and insolvency risk of banks. The findings highlight the importance of both a high ERM implementation level and a well-structured, sophisticated ERM organizational framework for achieving superior financial outcomes. The study also demonstrates the potential of PLSR as a robust prediction model for certain financial indicators. Future research could focus on extending this analysis to other sectors, comparing ERM effectiveness across different governance systems, and incorporating data from periods beyond the COVID-19 pandemic to better understand the influence of extraordinary events on ERM's impact. Exploring the application of metaheuristic algorithms for optimizing PLSR components to improve prediction accuracy is also suggested.

The study's limitations include its focus on a specific sector (banking) and a relatively small sample size (ten companies) from a single country (Turkey). The time period covered includes the COVID-19 pandemic, which may have influenced the results. The use of self-reported data from annual reports to measure ERM implementation and organizational structure might introduce bias. Future research should address these limitations by expanding the sample size, including companies from various sectors and countries, and considering a longer time horizon that excludes periods of extraordinary economic events. Additionally, employing more objective measures of ERM implementation would strengthen the study's conclusions.