Environmental, Social and Governance Disclosure and the Financial Stability of Quoted Deposit Money Banks in Nigeria: Heterogeneous Effects on Profitability, Capital, Liquidity, Asset Quality and Insolvency Risk

Abstract

This study examines the relationship between environmental, social and governance (ESG) disclosure and the financial stability of quoted deposit money banks in Nigeria, and whether governance structure moderates that relationship. Using panel data hand-collected from the audited annual and sustainability reports of twelve banks listed on the Nigerian Exchange Group over 2015–2025, this study relates an 18-item content-analysis ESG disclosure index and its environmental, social and governance sub-indices to five bank-level stability indicators: capital adequacy, liquidity, asset quality (non-performing loans), profitability (return on assets) and insolvency risk (Z-score). Bank fixed-effects models with Driscoll–Kraay standard errors form the baseline, and fixed-effects two-stage least squares (FE-2SLS) is used as a robustness check for potential endogeneity. The composite ESG index is positively and significantly associated with profitability, a result that persists in the FE-2SLS estimates, but its associations with liquidity, non-performing loans and the Z-score are insignificant and its association with capital adequacy is negative; the associations between disclosure and stability are therefore heterogeneous across its dimensions rather than uniformly positive. Governance structure, entered as a separate regressor, is directly associated with higher capital adequacy and profitability and with lower non-performing loans, whereas its interaction with disclosure is statistically significant only for capital adequacy and liquidity and is sensitive to the exclusion of individual banks, so that the moderation evidence is treated as fragile. Among the disclosure dimensions, governance disclosure is most consistently associated with stability, environmental disclosure is associated with lower capital adequacy and social disclosure is associated with higher non-performing loans. The near-saturation of the disclosure index among these large banks limits within-bank variation and counsels caution in interpretation. The findings suggest that regulators should prioritise the credibility and quality of disclosure, and the governance mechanisms that underpin it, rather than its quantity.

1. Introduction

The global financial crisis of 2007–2008 exposed the fragility of banking systems across the world and renewed attention to the factors that determine the resilience of banks (Berger & Bouwman, 2013; Demirgüç-Kunt et al., 2013). Nowhere is that resilience of greater consequence than in emerging economies, whose capital markets are shallow and whose growth depends heavily upon bank intermediation (Bashir et al., 2025; Olowofela et al., 2025). In Nigeria, the banking sector has experienced recurrent episodes of distress, from the systemic crisis and public bailout that followed the consolidation of the mid-2000s to the more recent strains imposed by macroeconomic volatility, currency depreciation and the fluctuating fortunes of an oil-dependent economy (Aroghene, 2022; Ozili, 2019). Against this background, the factors capable of strengthening the stability of banks have attracted sustained scholarly and regulatory attention, and among the factors more recently canvassed is the disclosure by banks of their environmental, social and governance (ESG) conduct. Banks with sound sustainability practices, it is argued, may advance the goals of sustainable development through responsible lending even as they strengthen their own resilience (Bashir et al., 2025; Nwobu et al., 2017).

Environmental, social and governance disclosure denotes the communication by a bank of information concerning its management of environmental and climate-related risk, its relationships with employees, customers and communities, and the structures through which it is governed (Velte, 2022; Arvidsson & Dumay, 2022). The theoretical case for a relationship between such disclosure and stability rests upon several complementary mechanisms. Disclosure reduces the information asymmetry between banks and their stakeholders, signals the quality of a bank’s management of non-financial risk, sustains the confidence upon which funding depends, and, through its governance dimension, disciplines the conduct of management (Spence, 1973; Jensen & Meckling, 1976; Bilyay-Erdogan et al., 2023). Yet despite the intuitive appeal of these mechanisms, the empirical evidence on the disclosure–stability relationship remains inconclusive (Ahmed, 2026; Chiaramonte et al., 2022; Musa et al., 2026), and it is especially thin for the Nigerian banking sector, where studies have concentrated upon financial performance and market value rather than upon the several dimensions of stability, and have seldom examined the conditioning role of governance (Iyawe & Chijuka, 2025; Ogboi et al., 2024; Owolabi et al., 2026).

The regulatory context lends the question particular salience. The Central Bank of Nigeria issued the Nigerian Sustainable Banking Principles in 2012, enjoining banks to integrate environmental and social considerations into their operations and lending (Central Bank of Nigeria, 2012), and the Nigerian Exchange followed with sustainability-disclosure guidelines in 2019, while the anticipated adoption of the sustainability-reporting standards of the International Sustainability Standards Board portends a further tightening of disclosure obligations (Erin et al., 2022). These developments have coincided with a far-reaching recapitalisation programme under which, from 2024, commercial banks operating at the international, national and regional levels were required to raise their minimum capital bases substantially, with a compliance deadline of 31 March 2026. The confluence of a maturing sustainability-disclosure regime and a demanding recapitalisation exercise, undertaken in alignment with the Basel III framework (Basel Committee on Banking Supervision, 2021; Oyetade et al., 2022), makes the present moment an apposite one at which to reassess whether disclosure, and the governance that underpins it, contributes to the resilience of Nigerian banks.

Recent studies have provided useful but conflicting evidence on the broader question of what determines bank stability in Nigeria. Some report positive and significant relationships between prudential and governance factors and performance (Kafidipe et al., 2021; Edeh & Iwedi, 2024), while others report insignificant or even negative associations (Samson & Tukur, 2024), a divergence attributable in part to differences in variable measurement, in estimation methodology and in the treatment of the cross-sectional dependence and endogeneity that characterise banking panels (Ozili, 2019; Pham et al., 2021). Bank-level observations in Nigeria are subject to common shocks arising from monetary policy, exchange-rate fluctuations, inflation and regulatory change (Omankhanlen et al., 2021), so that conventional standard errors may yield misleading inference where heteroskedasticity, cross-sectional dependence or serial correlation is present (Driscoll & Kraay, 1998; Pesaran, 2015). The present study addresses these concerns directly by estimating the relationship between ESG disclosure and five indicators of stability, controlling simultaneously for bank-specific and macroeconomic factors, using a bank fixed-effects estimator with Driscoll–Kraay standard errors to accommodate cross-sectional dependence, and employing fixed-effects–two-stage least squares (FE-2SLS) as a robustness strategy for the potential endogeneity arising from reverse causality between disclosure and stability (Staiger & Stock, 1997; Wooldridge, 2010).

Several research questions guide this study. Does ESG disclosure improve the capital adequacy, liquidity, asset quality, profitability and overall stability of quoted Nigerian deposit money banks? Which dimension of disclosure environmental, social or governance is the most consequential? Does governance structure strengthen or weaken the relationship between disclosure and stability? And do these relationships survive controls for bank-specific and macroeconomic factors and corrections for endogeneity? To answer these questions, this study employs a panel of twelve listed banks over the period 2015 to 2025.

This study makes several contributions to the literature. First, it provides new evidence on the disclosure–stability relationship within the specific and under-researched setting of quoted Nigerian deposit money banks, employing a comprehensive set of five stability indicators, drawn largely from the CAMELS framework and complemented by the Z-score, rather than the single measure that characterises much of the existing work, and it shows that the relationship between disclosure and stability is heterogeneous across these indicators. Second, it decomposes ESG disclosure into its environmental, social and governance dimensions and relates each to the indicators it is theorised most directly to affect, thereby avoiding the aggregation that obscures the offsetting effects of the several dimensions. Third, it models governance structure as a moderator of the disclosure–stability relationship, a role seldom examined for Nigerian banks. Fourth, it employs an econometric strategy that combines Driscoll–Kraay standard errors with FE-2SLS, thereby accommodating the cross-sectional dependence and mitigating the potential endogeneity characteristic of banking panels and providing evidence from non-GMM estimates, and it does so at a moment of heightened recapitalisation requirements and rising pressure for sustainable financial intermediation. The findings are expected to inform the policies of regulators, the decisions of bank managers, and the assessments of investors concerned with the resilience of the Nigerian banking system.

The remainder of this paper proceeds as follows. Section 2 reviews the theoretical and empirical literature linking ESG disclosure and governance to bank stability and develops the hypotheses. Section 3 discusses the data, variable construction and econometric methodology. Section 4 presents the results and discusses their implications. Section 5 concludes with recommendations for regulators, bank managers and other stakeholders.

2. Literature Review

The banking industry is a key driver of economic activity through its role in financial intermediation, and it is consequently among the most heavily regulated sectors of the economy, with considerable attention paid to its conduct in order to preserve public confidence and to support the wider economy (Omankhanlen et al., 2021; Bashir et al., 2025). The global financial crisis of 2007–2008 continues to resonate with stakeholders and to raise questions about the influence of prudential and non-prudential factors upon bank stability and performance. Increasingly, the disclosure by banks of their environmental, social and governance conduct has been added to the traditional prudential concerns of capital, liquidity and asset quality as a potential determinant of resilience.

2.1. The Evolution of Sustainability and Governance Regulation in Nigeria

The regulatory framework within which Nigerian banks disclose has evolved considerably over the past two decades. Following the consolidation exercise of 2004–2005, which raised the minimum paid-up capital of banks and reduced a fragmented multitude of institutions to a smaller number of larger ones, the Central Bank of Nigeria progressively strengthened its supervisory architecture, introducing risk-based supervision in alignment with the Basel II framework and, subsequently, the capital buffers, countercyclical requirements and liquidity standards associated with Basel III. Upon this prudential foundation, a sustainability architecture was erected: the Nigerian Sustainable Banking Principles of 2012 enjoined banks to integrate environmental and social risk into their lending and operations, and the sustainability-disclosure guidelines issued by the Nigerian Exchange in 2019 extended the expectation of disclosure to listed companies. Most recently, the recapitalisation directive of 2024 required commercial banks to raise their minimum capital substantially by 31 March 2026, an exercise that has returned the adequacy of bank capital and the resilience of the sector to the centre of regulatory attention. This evolving framework has progressively interwoven environmental, social and governance considerations with the prudential concerns that have traditionally defined banking supervision in Nigeria.

2.2. Theoretical Framework

The relationship between ESG disclosure and bank stability is underpinned by several complementary theories. Stakeholder theory (Freeman, 1984) holds that a bank is accountable not only to its shareholders but to the wider constituency of parties who affect or are affected by its conduct, and that responsiveness to their interests, communicated through disclosure, sustains the confidence upon which the bank depends. Legitimacy theory (Suchman, 1995) conceives disclosure as a legitimation strategy through which a bank secures the social licence upon which its continued access to resources is conditional. Signalling theory (Spence, 1973) frames disclosure as a costly and credible signal through which a bank of superior quality communicates its sound management of non-financial risk, thereby reducing information asymmetry and the risk premium demanded by the providers of capital. Agency theory (Jensen & Meckling, 1976) locates disclosure among the mechanisms that reduce the agency costs arising from the separation of ownership and control, and it identifies governance as the structure that disciplines managerial conduct and lends disclosure its credibility.

Taken together, these theories imply that ESG disclosure should enhance bank stability through the reduction of information asymmetry, the maintenance of confidence and the disciplining of risk-taking, and that governance should condition the strength of this relationship by determining whether disclosure is substantive or merely symbolic. This integrated framework, which complements the capital-buffer perspective that dominates the prudential literature, motivates both the direct hypotheses relating disclosure to the several indicators of stability and the moderation hypothesis relating governance structure to the disclosure–stability relationship.

2.3. Empirical Literature

A growing body of empirical scholarship has examined the relationship between ESG engagement and bank outcomes, though the evidence remains mixed and is concentrated outside Nigeria. On capital adequacy, Otitolaiye et al. (2023) found that sustainability reporting exerted a positive and significant effect on the capital adequacy of West and East African banks, and Lamanda and Tamásné Vőneki (2024) documented a strong association between capitalisation and disclosure among Central European banks. Ahmed (2026), however, found capital adequacy to be the least responsive of the risk dimensions to disclosure quality, and evidence from broad cross-country samples points to heterogeneous effects of the several ESG indicators upon capitalisation. These mixed findings motivate the examination of the capital-adequacy relationship in the Nigerian context.

On liquidity, the evidence points to a confidence channel that is partly offset by the liquidity costs of particular sustainability activities. Xu et al. (2024) found that stronger ESG performance enhanced customer stability and thereby the stability of funding, and Gangwani and Kashiramka (2024), employing the system generalised method of moments across emerging markets, reported that higher disclosure was associated with reduced liquidity risk. Yet studies that decompose sustainability practice have found that environmental activities may tighten liquidity even as they improve loan quality, so that the net effect of disclosure on liquidity is an empirical question. On asset quality, Olowofela et al. (2025) found that environmental innovation reduced non-performing loans among Nigerian banks, and studies employing propensity-score matching in other emerging economies have found the environmental and governance dimensions to lower non-performing loans, while the social dimension, through the extension of credit to riskier borrowers under a financial-inclusion agenda, may raise them.

On profitability, the Nigerian evidence of Iyawe and Chijuka (2025) and Ogboi et al. (2024) is broadly favourable, with the governance dimension emerging as the strongest, though Cantero-Saiz et al. (2025) caution that in developing economies of low or moderate sustainability the profitability effect of disclosure may be muted or even negative in the near term. On overall stability, measured by the Z-score, the most rigorous global studies find the composite ESG measure positively but often weakly associated with stability: Chiaramonte et al. (2022) found ESG engagement associated with greater stability among European banks, especially during crises, whereas Musa et al. (2026), in a global sample of 688 banks, found the effect of the composite ESG score economically negligible, with the governance pillar positively and the social pillar negatively associated with stability. On the moderating role of governance, Del Sarto et al. (2025) found that board diversity and independence mitigated the risks associated with ESG controversies, and Kimunei (2025) found that the association between corporate governance mechanisms and the financial stability of Kenyan microfinance banks was itself conditioned by firm age, illustrating that governance–stability relationships are context-dependent; yet the moderating role of governance in the disclosure–stability relationship has seldom been examined for Nigerian deposit money banks.

Complementing the ESG literature, a substantial body of scholarship within the capital-buffer tradition has examined the determinants of bank stability more broadly, and it furnishes the control structure and the estimation approach that the present study adopts. This literature establishes that stability is shaped jointly by capital adequacy, asset quality, liquidity and bank size, and by the macroeconomic environment, and that its estimation must contend with unobserved heterogeneity, cross-sectional dependence and endogeneity. Studies employing the fixed-effects estimator with Driscoll–Kraay standard errors, and instrumental-variable methods for endogeneity, have become the methodological benchmark for single-country banking panels of the kind analysed here, precisely because banks within one jurisdiction are exposed to common monetary, exchange-rate and regulatory shocks. The present study situates the ESG–stability question within this established framework, treating disclosure and governance as determinants of stability alongside the conventional bank-specific and macroeconomic factors, and adopting the estimation strategy that the literature has shown to be appropriate for such data.

The foregoing literature reveals that, while some studies report a positive and significant relationship between ESG disclosure and bank outcomes, others report insignificant or negative associations, and that the evidence for the specific setting of Nigerian deposit money banks, examined across a comprehensive set of stability indicators and with attention to the moderating role of governance, remains scant. Given the ongoing recapitalisation of the Nigerian banking sector and the significant role of banks in economic growth, it is essential to reassess the effect of disclosure, and of the governance that underpins it, upon bank stability. On the strength of the theoretical framework and the balance of the empirical evidence, this study advances the following hypotheses, stated in alternative form:

Hypothesis 1 (H1).

ESG disclosure has a positive and significant effect on the capital adequacy of quoted deposit money banks in Nigeria.

Hypothesis 2 (H2).

ESG disclosure has a positive and significant effect on the liquidity of quoted deposit money banks in Nigeria.

Hypothesis 3 (H3).

ESG disclosure has a negative and significant effect on the non-performing loans of quoted deposit money banks in Nigeria.

Hypothesis 4 (H4).

ESG disclosure has a positive and significant effect on the profitability of quoted deposit money banks in Nigeria.

Hypothesis 5 (H5).

ESG disclosure has a positive and significant effect on the overall stability (Z-score) of quoted deposit money banks in Nigeria.

Hypothesis 6 (H6).

Governance structure positively moderates the relationship between ESG disclosure and the financial stability of quoted deposit money banks in Nigeria. Because financial stability is measured by five distinct indicators, H6 is decomposed into five sub-hypotheses that correspond one-to-one to the estimated models:

Hypothesis 6a (H6a).

Governance structure strengthens the positive effect of ESG disclosure on the capital adequacy of quoted deposit money banks in Nigeria.

Hypothesis 6b (H6b).

Governance structure strengthens the positive effect of ESG disclosure on the liquidity of quoted deposit money banks in Nigeria.

Hypothesis 6c (H6c).

Governance structure strengthens the negative effect of ESG disclosure on the non-performing loans of quoted deposit money banks in Nigeria.

Hypothesis 6d (H6d).

Governance structure strengthens the positive effect of ESG disclosure on the profitability of quoted deposit money banks in Nigeria.

Hypothesis 6e (H6e).

Governance structure strengthens the positive effect of ESG disclosure on the overall stability (Z-score) of quoted deposit money banks in Nigeria.

Each hypothesis is evaluated separately in the model for the indicator to which it refers, and no overall verdict on financial stability as a whole is inferred from any single model. A hypothesis is regarded as supported when the relevant coefficient (β₁ for H1–H5 and the interaction coefficient β₃ for H6a–H6e) carries the hypothesised sign and is statistically significant at the five per cent level, as weakly supported when it is significant only at the ten per cent level, and as not supported otherwise; a significant coefficient of the opposite sign is reported as evidence contrary to the hypothesis. Throughout this paper, financial stability refers to the financial soundness of individual banks rather than to the stability of the financial system as a whole.

3. Methodology

3.1. Data Sources and Variable Selection

This study utilises panel data to examine the relationship between ESG disclosure and the financial stability of Nigerian deposit money banks (DMBs). The use of panel data is essential to control for the unobserved heterogeneity that may be associated with individual banks and to permit the analysis of both cross-sectional and time-series variation (Hsiao, 2014; Baltagi, 2021). This study uses data from the audited annual reports and sustainability reports of the sample banks to construct the bank-level variables, namely the ESG disclosure index and its sub-indices, the capital adequacy ratio, the liquidity ratio, the non-performing-loan ratio, the return on assets, the Z-score, governance structure and bank size. The data for the macroeconomic variables, namely the monetary policy rate and the inflation rate, were sourced from the Central Bank of Nigeria Statistical Bulletin. The sample was derived through a sequential filtering procedure. Unlisted banks were excluded first, because they are not subject to the sustainability-disclosure guidelines of the Nigerian Exchange and their reports are not consistently publicly available, which would compromise comparability. From the remaining deposit money banks, this study retained those for which audited annual reports and sustainability reports or sustainability sections within the annual report were available for the whole period 2015–2025, so that each bank contributes a continuous institutional record. Banks with incomplete disclosure series over the period were excluded. These filters yielded twelve banks: five tier-one banks (Access Holdings, FBN Holdings, Guaranty Trust Holding Company, United Bank for Africa and Zenith Bank) and seven tier-two banks (Fidelity Bank, Stanbic IBTC Holdings, Sterling Bank, Union Bank of Nigeria, Wema Bank, Polaris Bank and Jaiz Bank). Polaris Bank is retained as the successor institution to the previously quoted Skye Bank, and Jaiz Bank, a non-interest bank, is retained because it is subject to the same disclosure guidelines and prudential reporting as the other deposit money banks. The five tier-one banks, all of which are in the sample, are the domestic systemically important banks and alone held 57.2 per cent of total banking-industry assets at the end of 2025 (Central Bank of Nigeria, 2026), so that the sample accounts for well over half of the assets of the sector. The sample therefore covers the systemically important core of the sector, but it under-represents smaller and unlisted banks, and the findings should be generalised to the sector as a whole with this qualification in mind. The study period spans from 2015 to 2025, giving a maximum of one hundred and thirty-two bank-year observations across the twelve banks and eleven years. Because a small number of banks did not report every variable in every year, the panel is unbalanced, and the number of usable observations therefore varies across the variables and across the models, as reflected in the descriptive statistics and the regression tables; this is a common feature of hand-collected banking panels and is accommodated by the estimators employed (Baltagi, 2021). Because the macroeconomic series are available only up to 2024, the regression samples cover 2015–2024. The study period encompasses the implementation of the Nigerian Sustainable Banking Principles, the introduction of the sustainability-disclosure guidelines of the Nigerian Exchange, the recessions of 2016 and 2020, and the commencement of the current recapitalisation programme.

ESG disclosure, the principal independent variable, is measured through a content-analysis index constructed from the annual and sustainability reports of the sample banks. The index comprises eighteen items, six in each of the environmental, social and governance dimensions, each scored one where the item is disclosed and zero otherwise, following the disclosure-index tradition established by Velte (2022) and applied to banks by Ahmed (2026). The items were selected from the Nigerian Sustainable Banking Principles (Central Bank of Nigeria, 2012) and the sustainability-disclosure guidelines of the Nigerian Exchange, and the complete checklist, together with the decision rule applied to each item, is reported in Appendix A (Table A1). The composite ESG index is the sum of the eighteen items divided by eighteen, so that it lies in the zero-to-one interval, whereas the environmental (ENV), social (SOC) and governance (GOV) sub-indices are the unscaled counts of the disclosed items in each dimension and therefore range from zero to six:

ESG_it = (1/18) Σ_{j = 1}¹⁸ d_ijt, ENV_it = Σ_j∈E d_ijt, SOC_it = Σ_j∈S d_ijt, GOV_it = Σ_j∈G d_ijt

where d_ijt equals one if item j is disclosed by bank i in year t and zero otherwise, and E, S and G denote the six items of each dimension. The content analysis was conducted manually using a written scoring guide that specifies, for each item, the evidence required for a score of one (Appendix A, Table A1); an item was scored one only where the annual report or sustainability report contained a specific, identifiable disclosure on the matter, generic statements of intent were scored zero, and the source document was recorded for every bank-year, so that each score can be traced to its source. The coding was conducted by a single primary coder (the first author). As a reliability check, a sample of the reports was independently re-coded by a second coder (a co-author) without reference to the original scores, and any discrepancies were resolved by discussion before the final scores were adopted. As a consistency check, the coding of the three items that have quantitative counterparts in the dataset was compared with the availability of the corresponding data: board-composition disclosure (G1) agrees with the availability of board-size and board-independence data in 96.8 per cent of bank-years, gender-diversity disclosure (S2) with the availability of board-gender data in 96.9 per cent, and audit-committee disclosure (G3) with the availability of audit-committee data in 96.8 per cent.

Financial stability, the explained construct, is measured by five indicators: the capital adequacy ratio (CAR), the ratio of total regulatory capital to risk-weighted assets; liquidity (LIQ), the ratio of liquid assets to short-term liabilities; asset quality, proxied by the ratio of non-performing loans to total loans (NPL); profitability, measured by the return on assets (ROA); and overall stability, captured by the Z-score, entered in natural-logarithm form (lnZ). Financial stability is understood here as the financial soundness of the individual bank rather than the stability of the financial system as a whole. The five-indicator set is not presented as a standard or universally accepted measure; it is a study-specific, multidimensional combination adapted to the objectives of the research. Four of the indicators correspond to components of the CAMELS supervisory framework, namely capital adequacy (C), asset quality (A), earnings (E) and liquidity (L), which are widely used to monitor the condition of banks (Sahajwala & Van den Bergh, 2000; Betz et al., 2014), and each corresponds to a distinct channel through which ESG disclosure is theorised to operate: the regulatory-capital channel (capital adequacy), the credit-screening channel (asset quality), the cost-of-capital and reputational channel (earnings) and the funding-confidence channel (liquidity). The Z-score is not a component of CAMELS; it is an accounting-based measure of the distance to insolvency that combines profitability, capitalisation and the volatility of earnings (Boyd & Runkle, 1993; Laeven & Levine, 2009), and it is included to capture overall default risk, which none of the four ratio-based components measures on its own. Two CAMELS dimensions are deliberately excluded. Management quality (M) is not observable from public data and is ordinarily proxied by governance attributes; because board and audit characteristics enter the present study through the governance-structure moderator and the governance disclosure sub-index, including them in the dependent construct would introduce circularity between the explanatory and the explained variables. Sensitivity to market risk (S) requires repricing-gap, foreign-exchange-position and trading-book data that the sample banks do not disclose on a consistent basis over the study period; exposure to common market conditions is instead controlled for through the macroeconomic variables and the Driscoll–Kraay correction for cross-sectional dependence. Because the five indicators capture different and potentially conflicting aspects of stability, they are analysed as separate dependent variables rather than combined into a single composite, and the results are interpreted dimension by dimension.

The Z-score of bank i in year t is computed as

$$Z_{it}=\left(RO{A}_{it}+E{Q}_{it}/T{A}_{it}\right)/\sigma {(ROA)}_{it},\ln Z_{it}=\ln \left(Z_{it}\right),\sigma {(ROA)}_{it}=\surd [(1/2){\Sigma }_{k=0^2}(RO{A}_{i,t-k}-{{\mu \left(ROA\right)}_{it})}^2]$$

where ROA_it is net income after tax divided by total assets; EQ_it is the book value of total shareholders’ equity reported in the audited statement of financial position, used in preference to regulatory capital so that the measure is computed consistently across banks and years irrespective of changes in the regulatory definition of capital; TA_it is total assets; and σ(ROA)_it is the sample standard deviation of ROA over a three-year rolling window comprising years t, t − 1 and t − 2, with μ(ROA)_it the mean of ROA over the same window. The rolling window follows Beck et al. (2013) and Lepetit and Strobel (2013), allowing the volatility of returns, and hence the Z-score, to vary over time rather than remaining constant within each bank. Because the sample begins in 2015, the window for 2016 contains two observations (2015 and 2016), and for 2015, for which no rolling window is available, σ(ROA) is the standard deviation of the bank’s ROA over the full sample period. A higher Z-score indicates a greater distance from insolvency, and because the Z-score is highly skewed, it is entered in natural-logarithm form (Laeven & Levine, 2009).

Governance structure (GOVSTR), the moderating variable, is constructed from four components, each coded on the zero-to-one interval: board independence (BIND), the proportion of independent non-executive directors on the board; board gender diversity (BGD), the proportion of women on the board; audit-committee independence (ACI), the proportion of non-executive members of the statutory audit committee; and external-audit quality (BIG4), a binary indicator equal to one where the financial statements of the bank are audited by one of the Big Four audit firms and zero otherwise. The index is the unweighted mean of the four components:

GOVSTR_it = (BIND_it + BGD_it + ACI_it + BIG4_it)/4

Equal weighting is adopted for three reasons. First, it is transparent and replicable, in keeping with the additive construction of governance indices (Gompers et al., 2003). Second, data-driven weights from principal component analysis would be unstable with only twelve cross-sectional units and would vary with the composition of the sample. Third, the literature provides no theoretical basis for privileging any one of the four mechanisms. The components are those that the Nigerian governance literature identifies as most relevant to monitoring, to the credibility of reporting and to the containment of credit risk (Adegboye et al., 2020; Kafidipe et al., 2021; Edeh & Iwedi, 2024); those studies inform the choice of components, whereas the construction and aggregation set out here are specific to the present study.

The control variables comprise bank size (BSIZE), the natural logarithm of total assets, and the macroeconomic conditions captured by the monetary policy rate (MPR) and the inflation rate (INFL), which are standard determinants of bank performance and risk (Demirgüç-Kunt & Huizinga, 1999; Athanasoglou et al., 2008). Table 1 presents the description and measurement of the variables.

Table 1. Description and measurement of variables.

Variable	Symbol	Measurement	Category
Capital adequacy ratio	CAR	Total regulatory capital divided by risk-weighted assets	Dependent variable
Liquidity ratio	LIQ	Liquid assets divided by short-term liabilities	Dependent variable
Non-performing loans	NPL	Non-performing loans divided by total loans	Dependent variable
Return on assets	ROA	Net income divided by total assets	Dependent variable
Z-score	lnZ	ln[(ROA + book equity/total assets)/σ(ROA)], with σ(ROA) computed over a three-year rolling window (t, t − 1, t − 2)	Dependent variable
ESG disclosure index	ESG	ESG = disclosed items/18 (0–1); ENV, SOC and GOV = counts of six items each (0–6); see Appendix A	Independent variable
Governance structure	GOVSTR	Unweighted mean of board independence, board gender diversity, audit-committee independence and Big Four auditor (0–1)	Moderating variable
Bank size	BSIZE	Natural logarithm of total assets	Control variable
Monetary policy rate	MPR	Central Bank of Nigeria monetary policy rate	Control variable
Inflation rate	INFL	Annual inflation rate	Control variable

Note. Source: Authors’ compilation (2026).

3.2. Model Specification

This study examines the relationship between ESG disclosure and bank stability, controlling for governance structure and for bank-specific and macroeconomic factors. The baseline relationship is stated in functional form as

STAB_it = f (ESG_it, GOVSTR_it, BSIZE_it, MPR_t, INFL_t)

(1)

where STAB denotes one of the five indicators of stability (CAR, LIQ, NPL, ROA, lnZ). The baseline econometric model with bank fixed effects is specified as

STAB_it = α₀ + β₁ESG_it + β₂GOVSTR_it + ϕX_it + μ_i + ε_it

(2)

where X is the vector of control variables (bank size, the monetary policy rate and inflation); μ denotes the unobserved bank-specific effect; ε is the idiosyncratic error; i indexes banks and t indexes years. Year fixed effects are not included because the monetary policy rate and inflation vary over time but not across banks and would be perfectly collinear with a full set of year dummies, so that their coefficients could not be identified. Common macroeconomic conditions are instead captured explicitly through MPR and INFL, and any remaining common shocks are accommodated in inference through the Driscoll–Kraay standard errors, which are robust to cross-sectional dependence (Driscoll & Kraay, 1998). MPR and INFL capture important macroeconomic conditions, but they may not fully account for all common shocks during the study period, such as the exchange-rate adjustments of 2016 and 2023–2024, the disruption associated with the COVID-19 pandemic in 2020, and changes in the regulatory environment. The Driscoll–Kraay correction makes inference robust to such residual common shocks, but it does not remove any bias that could arise if those shocks were correlated with disclosure, and the estimates should be read with this qualification. Because the environmental, social and governance dimensions of disclosure are theorised to operate through distinct channels, this study also estimates a disaggregated model in which the composite index is replaced by its three sub-indices:

STAB_it = α₀ + β₁ENV_it + β₂SOC_it + β₃GOV_it + ϕX_it + μ_i + ε_it

(3)

To examine the moderating role of governance structure, this study augments the baseline model with the interaction between ESG disclosure and governance structure:

STAB_it = α₀ + β₁ESG_it + β₂GOVSTR_it + β₃(ESG × GOVSTR)_it + ϕX_it + μ_i + ε_it

(4)

where a positive and significant coefficient β₃ indicates that governance structure strengthens the effect of ESG disclosure on stability (for the non-performing-loan model, a negative β₃ indicates strengthening). Because GOVSTR also enters Equation (4) as a separate regressor, β₂ captures the direct, conditional association between governance structure and stability, whereas β₃ alone captures moderation; the two are distinct and are interpreted separately (Brambor et al., 2006). In the presence of the interaction term, β₁ measures the effect of disclosure when GOVSTR equals zero, and the marginal effect of disclosure is β₁ + β₃GOVSTR_it. To address the potential endogeneity between disclosure and stability, arising because more stable and profitable banks may have both the resources and the incentive to disclose more fully, the study instruments current disclosure with its first and second lags. The first-stage model is

ESG_it = α₀ + β₁ESG_i,t−1 + β₂ESG_i,t−2 + ϕX_it + μ_i + ε_it

(5)

and the second-stage model relates the predicted disclosure to the stability indicator:

STAB_it = α₀ + β₁ESĜ_it + β₂GOVSTR_it + ϕX_it + μ_i + ε_it

(6)

3.3. Estimation Techniques

This study employs the bank fixed-effects (FE) model with Driscoll–Kraay standard errors as the baseline estimator, following the necessary specification tests. This estimator is well suited to a panel with a small cross-sectional dimension (N = 12) and a moderate time dimension (T = 11), and the Driscoll–Kraay standard errors are essential for simultaneously correcting cross-sectional dependence, serial correlation and heteroskedasticity. The choice between the fixed-effects and random-effects estimators is examined through the Hausman test but, as explained in Section 4.7, is ultimately made on conceptual grounds. The presence of cross-sectional dependence, which justifies the Driscoll–Kraay correction, is assessed through the Pesaran cross-sectional dependence test. This study also considered the generalised method of moments (GMM) as an alternative for the treatment of endogeneity (Arellano & Bover, 1995; Blundell & Bond, 1998; Roodman, 2009); however, given the small cross-sectional sample, the GMM estimator would be prone to instrument proliferation, and this study therefore adopts the fixed-effects–two-stage least squares (FE-2SLS) estimator for the robustness check.

The FE-2SLS estimator combines the fixed-effects transformation with instrumental-variable estimation, instrumenting current ESG disclosure with its first and second lags. The instruments are theoretically motivated by the persistence of disclosure practice, which renders lagged disclosure a plausible predictor of current disclosure while reducing the likelihood of contemporaneous reverse causality. The relevance of the instruments is evaluated through the first-stage F-statistic, with an F-statistic above ten being the conventional diagnostic for the absence of weak instruments (Staiger & Stock, 1997), and the endogeneity of disclosure is assessed through the Durbin–Wu–Hausman test. The model is kept deliberately parsimonious to avoid instrument proliferation. The FE-2SLS estimates are interpreted as a robustness assessment of the baseline FE–Driscoll–Kraay results rather than as definitive causal estimates. Prior to estimation, this study examines the stationarity of the variables through the Fisher-type and the cross-sectionally augmented (CIPS) panel unit-root tests, assesses multicollinearity through the variance inflation factor, and winsorises the five dependent variables at the first and ninety-ninth percentiles to limit the influence of outliers (Wooldridge, 2010).

The validity of the instruments additionally requires that lagged disclosure affect current stability only through its effect on current disclosure (the exclusion restriction). Three considerations support this restriction. First, lagged disclosure is predetermined with respect to current-year shocks to profitability and risk, since the reports for years t − 1 and t − 2 are prepared and published before year-t outcomes are realised, so that current shocks cannot feed back into them. Second, the bank fixed effects absorb time-invariant factors, such as corporate culture or franchise value, that could otherwise drive both past disclosure and current performance. Third, disclosure practice in the sample is strongly persistent, being anchored in reporting templates and regulatory guidelines that change slowly, so that the economically relevant variation in lagged disclosure operates through its continuation into current disclosure. The restriction cannot be tested directly, however, and it would be violated if past disclosure were correlated with persistent, time-varying unobservables, such as evolving management quality, that also affect current stability (Reed, 2015; Bellemare et al., 2017). Because the model has two instruments for one endogenous regressor, it is over-identified, and this study reports the Hansen J test of over-identifying restrictions, which provides a partial check on validity conditional on at least one instrument being valid (Hansen, 1982), together with the Kleibergen–Paap rk Wald F-statistic for weak identification under non-i.i.d. errors (Kleibergen & Paap, 2006).

4. Results and Discussion

4.1. Preliminary Insights

4.1.1. Descriptive Statistics

Table 2 reports the descriptive statistics of the variables. The number of observations varies across the variables, from one hundred and thirteen for the capital adequacy ratio to one hundred and twenty-eight for the disclosure indices (one hundred and twenty for the macroeconomic series, which end in 2024), because the panel is unbalanced: a small number of banks did not report every item in every year, so that the available observations differ across the variables while all are drawn from the same twelve banks over the same eleven years. The average ESG disclosure index is 0.94, indicating that the sample banks, which are the largest and most visible in the sector, disclose the majority of the index items; the near-saturation of the composite index among leading banks is consistent with the pattern documented for large emerging-market banks by Ahmed (2026), who notes that disclosure quantity converges among visible institutions while quality continues to vary. The environmental sub-index (a mean of 5.19 out of 6) displays greater variation than the social and governance sub-indices, reflecting the more uneven adoption of environmental disclosure that Owolabi et al. (2026) report for Nigerian banks. The mean capital adequacy ratio of 0.29 lies comfortably above the regulatory minimum, indicating that the sample banks maintained adequate capital buffers during the study period, in line with the buffer-holding behaviour predicted by Jokipii and Milne (2008). The mean non-performing-loan ratio of 0.07 reflects the credit-risk pressures of the period, and its considerable dispersion highlights differences in credit-risk exposure across banks, attributable in part to macroeconomic shocks and to the volatility of oil output (Ozili, 2019). The mean return on assets of 0.03 is typical of the sector (Iyawe & Chijuka, 2025). The macroeconomic variables reflect the volatility of the period, with inflation averaging seventeen per cent, underscoring the need to control for macroeconomic conditions (Omankhanlen et al., 2021).

Table 2. Summary statistics of the variables.

Variable	Obs.	Mean	SD	Min	Median	Max
ROA	124	0.032	0.030	0.000	0.021	0.139
lnZ (Z-score)	124	3.467	1.078	−0.224	3.306	7.521
CAR	113	0.286	0.144	0.130	0.248	1.061
LIQ	124	0.427	0.191	0.013	0.430	1.017
NPL	114	0.072	0.087	0.000	0.047	0.552
ESG index	128	0.943	0.110	0.500	0.944	1.000
ENV	128	5.188	1.362	0.000	5.000	6.000
SOC	128	5.883	0.541	3.000	6.000	6.000
GOV	128	5.906	0.747	0.000	6.000	6.000
GOVSTR	124	0.456	0.089	0.125	0.463	0.625
BSIZE	124	21.829	1.292	18.029	21.906	24.666
MPR	120	0.152	0.047	0.110	0.140	0.275
INFL	120	0.172	0.068	0.090	0.161	0.332
ER	120	513.421	388.047	197.000	344.000	1535.820

Note. SD = standard deviation; Obs. = number of observations. The panel is unbalanced, so the number of observations varies across variables; the macroeconomic series cover 2015–2024. Statistics are computed before winsorisation. CAR, LIQ and NPL are ratios; ROA is a decimal; ESG ranges 0–1; ENV, SOC and GOV range 0–6; BSIZE is ln(total assets); MPR and INFL are decimals; ER is the USD/NGN exchange rate. Source: Authors’ computation (2026).

The near-saturation of the disclosure measures has implications for identification that warrant explicit discussion. The social and governance sub-indices average 5.883 and 5.906 out of a maximum of six and are at their maximum in 95.3 and 98.4 per cent of bank-years, respectively, and the composite index equals one in 48.4 per cent of bank-years. Because the fixed-effects estimator identifies the coefficients from variation within banks over time, the relevant variation is the within-bank variation. Its standard deviation for the composite index (0.095) is not much smaller than the overall standard deviation (0.110), but it is highly concentrated: five of the twelve banks record no change in the composite index over the period, and four banks (Jaiz Bank, Wema Bank, United Bank for Africa and Stanbic IBTC Holdings) account for 97.6 per cent of the within-bank variation. Three consequences follow. First, the coefficients on the composite index, and especially those on the social and governance sub-indices, are identified from a small number of banks and bank-years, so that statistically insignificant results for liquidity, non-performing loans and the Z-score may reflect limited power rather than the absence of an underlying relationship. Second, the estimated effects may be sensitive to the inclusion of individual banks and may not extrapolate to levels of disclosure below the observed range. Third, a ceiling implies that an extent-based binary index cannot discriminate between banks that disclose all items but differ in the depth and quality of their disclosure (Arvidsson & Dumay, 2022; Ahmed, 2026). The environmental sub-index, which is at its maximum in only 48.4 per cent of bank-years, is correspondingly the dimension for which within-bank variation is most informative. These considerations inform the interpretation of the results below and are acknowledged as a limitation in Section 5.

4.1.2. Panel Unit-Root Tests

To ensure the reliability of the estimates and to guard against spurious regression, this study examines the stationarity of the variables through a Fisher-type panel unit-root test, having first established through the Pesaran (2015) cross-sectional dependence test that the variables exhibit cross-sectional dependence. Because the Fisher-type test assumes cross-sectional independence, the cross-sectionally augmented IPS (CIPS) test of Pesaran (2007), which is robust to such dependence, is also reported for the bank-level variables; given the short time dimension, its p-values are obtained by simulating the null distribution for the dimensions of the present panel. Banks whose series do not vary over the period cannot enter the test, so the CIPS statistic for the ESG index is based on the seven banks whose disclosure changed. Table 3 reports the results. Under the Fisher-type test, profitability and non-performing loans are stationary at their levels, and the Z-score, liquidity, the ESG index, governance structure and bank size are stationary after first differencing. The CIPS test rejects the unit-root null at level for the Z-score (p = 0.032), non-performing loans (p = 0.049) and bank size (p < 0.001), and in first differences for every bank-level variable (p ≤ 0.036). For the capital adequacy ratio, the Fisher-type test does not reject the unit-root null either at level (p = 0.480) or in first differences (p = 0.876), but the CIPS test, which accounts for cross-sectional dependence, rejects the null in first differences (p = 0.010), so that the ratio is classified as integrated of order one. For profitability, the two tests disagree at level (Fisher p = 0.044; CIPS p = 0.857), and both reject the null in first differences, so that the order of integration of ROA is zero or one. For the monetary policy rate and inflation, the reported p-values (1.000 at level, and 1.000 and 0.999 in first differences) do not support stationarity at either level. These two series vary only over time and are identical across banks, so that a panel unit-root test is in effect applied to twelve replications of a single ten-year time series and is uninformative about their order of integration. They enter the models only as controls for common macroeconomic conditions; they are policy-determined rates whose long-run behaviour is bounded, and none of the substantive conclusions of this study depend on their coefficients; the possibility that their persistence affects the estimated control coefficients is nevertheless acknowledged. This mixed order of integration mirrors the pattern reported for African banking panels by Ozili (2019) and Yitayaw et al. (2023). Because the bank-level variables are stationary at level or become stationary after first differencing, and none is integrated of order two, a panel cointegration framework is not essential (Baltagi, 2021), and it is appropriate to proceed with the bank fixed-effects–Driscoll–Kraay and fixed-effects–2SLS estimations.

Table 3. Panel unit-root test results.

Variable	Level (p)	1st Diff. (p)	CIPS p (Level/1st Diff.)	Order	Decision
ROA	0.044	0.002	0.857/0.000	I(0)/I(1)	Stationary at level (Fisher); at first difference (CIPS)
lnZ	0.205	0.000	0.032/0.036	I(1)	Stationary at first difference
CAR	0.480	0.876	0.503/0.010	I(1)	Stationary at first difference (CIPS)
LIQ	0.584	0.000	0.386/0.004	I(1)	Stationary at first difference
NPL	0.000	0.000	0.049/0.000	I(0)	Stationary at level
ESG	0.248	0.000	0.165/0.001	I(1)	Stationary at first difference
GOVSTR	0.905	0.001	0.658/0.008	I(1)	Stationary at first difference
BSIZE	1.000	0.000	0.000/0.000	I(1)	Stationary at first difference
MPR	1.000	1.000	n.a.	—	Not stationary at level or first difference; common time series (see text)
INFL	1.000	0.999	n.a.	—	Not stationary at level or first difference; common time series (see text)

Note. Fisher-type panel unit-root test based on augmented Dickey–Fuller regressions; CIPS = cross-sectionally augmented IPS test of Pesaran (2007), robust to cross-sectional dependence, with p-values from 2000 simulations of the null distribution for N = 12 and T = 11; the ESG statistic is based on the seven banks whose index varies; n.a. = not applicable, because MPR and INFL do not vary across banks. Source: Authors’ computation (2026).

4.1.3. Correlation and Multicollinearity

Table 4 presents the correlation matrix. ESG disclosure is positively correlated with the return on assets (r = 0.24), and governance structure is positively correlated with capital adequacy (r = 0.53) and liquidity (r = 0.40) and negatively correlated with non-performing loans (r = −0.17), foreshadowing the regression findings and echoing the associations reported by Kafidipe et al. (2021) and Edeh and Iwedi (2024) for Nigerian banks. The positive correlation between non-performing loans and profitability in this sample of large banks reflects the coexistence of higher gross earnings and elevated credit exposure among the more aggressive lenders, a configuration also observed by Ogboi et al. (2024). With the exception of the monetary policy rate and inflation, which are closely related macroeconomic series, the correlations among the explanatory variables lie well below the 0.80 threshold at which multicollinearity would be a concern (Wooldridge, 2010). The variance inflation factors reported in Table 5 confirm this assessment: every factor lies below the conventional threshold of ten, and the mean factor of 3.14 indicates that multicollinearity does not threaten the reliability of the estimates. The elevated factors for the monetary policy rate and inflation concern two control variables whose individual coefficients are not interpreted, and collinearity among control variables does not bias the coefficients on the variables of interest (Wooldridge, 2010). The factors are computed for the augmented specification reported in Section 4.5, which also includes NPL and LIQ as regressors.

Table 4. Correlation matrix.

	ROA	lnZ	CAR	LIQ	NPL	ESG	GOVSTR	BSIZE	MPR	INFL
ROA	1.00
lnZ	−0.41	1.00
CAR	0.08	−0.09	1.00
LIQ	−0.35	−0.03	0.47	1.00
NPL	0.48	−0.13	−0.21	−0.34	1.00
ESG	0.24	−0.06	−0.02	0.11	0.05	1.00
GOVSTR	−0.01	−0.07	0.53	0.40	−0.17	0.36	1.00
BSIZE	0.04	−0.22	0.38	0.29	−0.23	0.31	0.58	1.00
MPR	0.12	−0.17	0.16	0.16	−0.06	0.12	0.28	0.37	1.00
INFL	0.11	−0.16	0.12	0.14	−0.03	0.16	0.31	0.41	0.93	1.00

Note. Source: Authors’ computation (2026).

Table 5. Variance inflation factors.

Variable	VIF	Tolerance
ESG	1.141	0.876
GOVSTR	1.560	0.641
BSIZE	1.578	0.634
NPL	1.194	0.837
LIQ	1.536	0.651
MPR	7.161	0.140
INFL	7.840	0.128
Mean VIF	3.144

Note. VIF = variance inflation factor. Source: Authors’ computation (2026).

4.2. Baseline Regression Results

Table 6 reports the baseline bank fixed-effects estimates with Driscoll and Kraay (1998) standard errors for the five indicators of stability. The results show that ESG disclosure is positively associated with bank profitability: the coefficient on the ESG index in the return-on-assets model is positive and significant at the one per cent level (β = 0.011). In economic terms, a one-standard-deviation increase in the composite index (0.110) is associated with an increase in the return on assets of about 0.12 percentage points, approximately four per cent of the sample mean of 3.2 per cent, and a move from the lowest (0.50) to the highest (1.00) observed disclosure score is associated with an increase of about 0.54 percentage points; given the limited within-bank variation in the index, the latter comparison lies at the edge of the identifying variation and should be read as indicative. This is consistent with the signalling account of Spence (1973), under which credible disclosure lowers information asymmetry and the cost of funding, and with the Nigerian evidence of Iyawe and Chijuka (2025) and Ogboi et al. (2024). Hypothesis H4 is therefore supported. Governance structure, which enters the baseline model as a separate regressor, displays strong direct associations with stability: its coefficient is positive and significant in the capital-adequacy (β = 0.875) and profitability (β = 0.026) models and negative and significant in the non-performing-loan model (β = −0.347), while it is insignificant in the liquidity and Z-score models. These are conditional associations estimated within banks over time; they indicate that, holding disclosure and the controls constant, a stronger governance structure is associated with higher capital adequacy and profitability and with lower non-performing loans. They describe the direct role of governance, which must be distinguished from its moderating role—that is, whether governance alters the disclosure–stability relationship—examined separately in Section 4.3. These direct associations are consistent with the agency-theoretic proposition of Jensen and Meckling (1976) that governance disciplines risk-taking, and with the evidence of Del Sarto et al. (2025) and Kafidipe et al. (2021). The coefficient on the composite ESG index in the capital-adequacy model is negative and significant, a finding that may reflect the limited variation in the near-saturated composite index among these large banks and the dominance of the governance channel, and that is better understood through the disaggregated analysis reported below (Arvidsson & Dumay, 2022); it constitutes evidence contrary to H1. The effects of the composite index on liquidity, non-performing loans and the Z-score are not statistically significant, so that hypotheses H2, H3 and H5 are not supported, although the signs of these coefficients are as expected. Taken together, the baseline results show that the composite ESG index is significantly associated with only one of the five dimensions of stability, namely, profitability, and that the positive association with profitability cannot be generalised to financial stability as a whole: profitability captures current operating performance, whereas capital adequacy, liquidity, asset quality and the Z-score capture resilience and risk exposure, and a bank may be more profitable without being better capitalised, more liquid or further from insolvency. Bank size carries a negative and significant coefficient in the profitability and Z-score models, and the monetary policy rate is positively associated with profitability, consistent with the macroeconomic sensitivities documented by Omankhanlen et al. (2021).

Table 6. Baseline bank fixed-effects estimates with Driscoll–Kraay standard errors.

Variable	CAR	LIQ	NPL	ROA	lnZ
ESG	−0.1571 ***	0.0232	−0.0214	0.0108 ***	0.8214
	(0.0486)	(0.0640)	(0.0500)	(0.0027)	(0.5421)
GOVSTR	0.8750 ***	−0.0275	−0.3474 ***	0.0256 ***	2.3901
	(0.2255)	(0.1198)	(0.0878)	(0.0089)	(2.0126)
BSIZE	−0.0389 **	0.0135	−0.0047	−0.0030 ***	−0.8489 ***
	(0.0155)	(0.0135)	(0.0173)	(0.0010)	(0.1361)
MPR	0.6091 **	0.6390 **	−0.1871 *	0.1859 ***	−3.2673
	(0.2406)	(0.3099)	(0.1050)	(0.0331)	(4.6957)
INFL	−0.0788	−0.1739	0.0972	−0.0553 *	4.3813
	(0.1158)	(0.2069)	(0.1832)	(0.0306)	(5.0763)
Observations	104	114	105	114	114
Within R2	0.266	0.078	0.089	0.258	0.284

Note. Driscoll–Kraay standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. All models include bank fixed effects; year fixed effects are not included because they would be perfectly collinear with MPR and INFL, which vary only over time. Source: Authors’ computation (2026).

4.3. Moderating Role of Governance Structure

Table 7 reports the estimates of Equation (4), in which the interaction between ESG disclosure and governance structure is added to the baseline specification for each of the five indicators, so that each of the sub-hypotheses H6a to H6e is evaluated in the model to which it refers. The interaction coefficient is positive and significant at the one per cent level in the capital-adequacy (β₃ = 8.257) and liquidity (β₃ = 1.495) models, and insignificant in the non-performing-loan (p = 0.120), profitability (p = 0.970) and Z-score (p = 0.773) models. Under the decision rule set out in Section 2.3, H6c, H6d and H6e are not supported, and H6a and H6b would be supported on the full sample alone; because the two significant interactions are not robust (see below), H6a and H6b are regarded as receiving only tentative, sample-specific support rather than as evidence of a robust moderating effect. Taken at face value, the two significant interactions would imply that the association between disclosure and capital adequacy, which is negative for the average bank, becomes positive where governance structure exceeds about 0.50, and that the association between disclosure and liquidity becomes positive where governance structure exceeds about 0.35.

Table 7. Moderating effect of governance structure on the ESG disclosure–stability relationship.

Variable	CAR	LIQ	NPL	ROA	lnZ
ESG	−4.1265 ***	−0.5160 ***	−2.2951	0.0104	1.4048
	(0.5846)	(0.1453)	(1.3715)	(0.0128)	(1.6083)
GOVSTR	−7.0498 ***	−1.2774 **	−4.8866	0.0245	3.7422
	(1.0983)	(0.3983)	(2.6784)	(0.0301)	(5.2051)
ESG × GOVSTR	8.2571 ***	1.4947 ***	4.7296	0.0013	−1.6170
	(1.2625)	(0.4541)	(2.7571)	(0.0329)	(5.4384)
BSIZE	−0.0400 **	0.0052	−0.0053	−0.0030 **	−0.8400 ***
	(0.0153)	(0.0097)	(0.0193)	(0.0010)	(0.1157)
MPR	0.4584 *	0.6816 *	−0.2735 **	0.1860 ***	−3.3134
	(0.2196)	(0.3393)	(0.0962)	(0.0326)	(4.7236)
INFL	0.0635	−0.1979	0.1788	−0.0554	4.4072
	(0.1144)	(0.2276)	(0.1900)	(0.0305)	(5.1282)
Observations	104	114	105	114	114
Within R2	0.315	0.122	0.121	0.258	0.285

Note. Each column reports the estimates of Equation (4) with bank fixed effects; Driscoll–Kraay standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Source: Authors’ computation (2026).

These two results are, however, fragile, and they should not be read as robust evidence of moderation. When the models are re-estimated twelve times, each time excluding one bank, the capital-adequacy interaction falls to 0.95 (p = 0.79) when Guaranty Trust Holding Company is excluded, and the liquidity interaction becomes −0.95 (p = 0.87) when Jaiz Bank is excluded, while the remaining exclusions leave both coefficients significant. Each result therefore rests on a single bank. The fragility reflects the features of the data discussed in Section 4.1.1: five banks record no change in disclosure, so that for these banks the interaction term moves one-for-one with governance structure, and the near-collinearity of ESG × GOVSTR with GOVSTR produces the large, offsetting main-effect coefficients reported in Table 7 (McClelland & Judd, 1993). The overall conclusion is therefore that the evidence of moderation is limited to two dimensions of stability and is not robust. Importantly, the significant direct associations of governance structure reported in Section 4.2 do not in themselves constitute evidence of moderation, since a variable may be directly associated with an outcome without altering the effect of another variable on that outcome (Brambor et al., 2006).

4.4. Disaggregated ESG Dimensions

Table 8 reports the estimates when ESG disclosure is decomposed into its environmental, social and governance dimensions, following the recommendation of Arvidsson and Dumay (2022) that the pillars be examined separately rather than aggregated. The results lend support to the theoretical expectation, grounded in agency theory (Jensen & Meckling, 1976), that the governance dimension is the most consequential for bank stability. The governance dimension carries a positive and significant coefficient in the profitability, liquidity and Z-score models, and a negative and significant coefficient in the non-performing-loan model, indicating that governance disclosure is the dimension most consistently associated with stability, a pattern that accords with the pre-eminence of the governance pillar documented by Ahmed (2026) and Gyimah and Owusu-Afriyie (2025). However, it also carries a small negative and significant coefficient in the capital-adequacy model (β = −0.018), so governance disclosure is not associated with higher capital. The social dimension exhibits the tension anticipated in the literature, being associated with higher non-performing loans and lower liquidity (both significant only at the ten per cent level), consistent with the extension of credit to riskier borrowers under a financial-inclusion agenda that Samson and Tukur (2024) identify. The environmental dimension carries a negative and significant coefficient in the capital-adequacy model but is otherwise insignificant, which may reflect the near-term costs of environmental integration noted by Olowofela et al. (2025). These dimensional findings, which the composite index obscures, suggest the value of decomposition (Liu, 2022). Given the near-saturation of the social and governance sub-indices (Section 4.1.1), however, these dimension-level coefficients are identified from a small number of within-bank changes and should be interpreted with caution.

Table 8. Fixed-effects estimates by ESG dimension.

Dimension	CAR	LIQ	NPL	ROA	lnZ
ENV	−0.0141 ***	0.0076	−0.0106	−0.0007	0.1444
	(0.0035)	(0.0082)	(0.0094)	(0.0008)	(0.1122)
SOC	0.0252	−0.0236 *	0.0441 *	0.0035	−0.2286
	(0.0163)	(0.0139)	(0.0225)	(0.0022)	(0.1901)
GOV	−0.0184 ***	0.0194 **	−0.0074 ***	0.0017 ***	0.0919 *
	(0.0038)	(0.0079)	(0.0024)	(0.0005)	(0.0498)

Note. Each column is a separate regression of the stability indicator on the three ESG dimensions with governance structure, bank size, the monetary policy rate and inflation as controls, and bank fixed effects. Driscoll–Kraay standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.10. Source: Authors’ computation (2026).

4.5. Macroeconomic Conditioning of the Disclosure Effect

To examine whether the relationships between disclosure and profitability and between disclosure and the Z-score depend on macroeconomic conditions, this study estimates an augmented specification that adds non-performing loans and liquidity as further bank-level controls and interacts ESG disclosure with the monetary policy rate and inflation, following the practice of single-country Nigerian panel studies (Ozili, 2019; Pham et al., 2021). As in the baseline, the specification includes bank fixed effects without year effects, which would absorb the macroeconomic main effects. Table 9 reports the results. Because the ESG coefficient in this specification measures the effect of disclosure when both the monetary policy rate and inflation equal zero, a value far outside the sample range, it is not directly comparable with the baseline coefficient (Brambor et al., 2006). Evaluated at the sample means of the monetary policy rate (0.152) and inflation (0.172), the implied marginal effect of ESG disclosure on the return on assets is approximately 0.008 (−0.0255 + 0.5171 × 0.152 − 0.2637 × 0.172), close to the baseline estimate of 0.011, and the interactions with the macroeconomic variables are individually insignificant in the profitability model, indicating that the relationship between disclosure and profitability is not systematically altered by the level of the policy rate or of inflation. Governance structure retains a positive coefficient, significant at the ten per cent level, and bank size retains its negative and significant coefficient in the profitability model. In the Z-score model, the macroeconomic terms and their interactions carry very large coefficients of offsetting sign, and the implied marginal effect of disclosure at the sample means is approximately −0.74. These magnitudes reflect the near-collinearity of the monetary policy rate and inflation (r = 0.93) and of their interactions with a near-saturated disclosure index; they should not be given a substantive economic interpretation, and the Z-score results of this specification are reported for completeness only.

Table 9. Results with macroeconomic specification.

Variable	ROA	lnZ
ESG	−0.0255	−6.4987 *
	(0.0207)	(3.5906)
GOVSTR	0.0371 *	−0.7748
	(0.0204)	(1.7172)
BSIZE	−0.0051 ***	−0.7806 ***
	(0.0017)	(0.1870)
NPL	−0.0093	−0.9744
	(0.0194)	(0.8533)
LIQ	0.0018	0.2045
	(0.0053)	(0.6278)
MPR	−0.3072	−237.5246 ***
	(0.5266)	(85.9597)
INFL	0.2116	178.8889 ***
	(0.3865)	(61.0742)
ESG × MPR	0.5171	242.9657 ***
	(0.5632)	(84.2845)
ESG × INFL	−0.2637	−181.2593 ***
	(0.4160)	(58.6491)
Observations	105	105

Note. Bank fixed effects without year effects, as in the baseline; the ESG main effect is evaluated at MPR = INFL = 0. Driscoll–Kraay standard errors in parentheses. *** p < 0.01, * p < 0.10. Source: Authors’ computation (2026).

4.6. Robustness Checks with Fixed-Effects–2SLS

To check the robustness of the findings and to address the potential endogeneity between disclosure and stability, arising because more profitable banks may have both the resources and the incentive to disclose more fully (Bilyay-Erdogan et al., 2023), this study employs the fixed-effects–two-stage least squares estimator, instrumenting ESG disclosure with its first and second lags. The Durbin–Wu–Hausman test (χ² = 6.27, p = 0.012 in the profitability model; χ² = 3.99, p = 0.046 in the Z-score model) supports the treatment of disclosure as endogenous, and this study accordingly proceeds to the instrumental-variable estimation (Hausman, 1978). The first-stage diagnostics are reported in Table 10: the first lag of disclosure is a strong and significant predictor of current disclosure, whereas the second lag is individually insignificant (p = 0.895), so that the first lag carries the identifying variation. The conventional joint first-stage F-statistic of 21.85 exceeds the threshold of ten suggested by Staiger and Stock (1997), but the heteroskedasticity-robust Kleibergen–Paap rk Wald F-statistic, which is the appropriate diagnostic under non-i.i.d. errors, is only 4.93, so that the two-instrument model is at risk of weak-instrument bias. Two further checks address this concern. First, the Anderson–Rubin test, whose validity does not depend on instrument strength, rejects the null hypothesis that disclosure has no effect on the return on assets (χ²(2) = 14.56, p = 0.001; Anderson & Rubin, 1949). Second, a just-identified model that uses only the first lag, which has a Kleibergen–Paap F-statistic of 16.52, yields a coefficient on disclosure in the profitability model of 0.042 (p = 0.001). Instrument relevance does not, however, establish instrument validity. The Hansen J test of over-identifying restrictions does not reject the joint null that the instruments are uncorrelated with the error term (J = 0.045, p = 0.832 for profitability; J = 0.299, p = 0.584 for the Z-score), but with one weak instrument its power is limited, and the exclusion restriction, discussed in Section 3.3, is maintained by assumption. Table 11 reports the second-stage estimates. The coefficient on ESG disclosure in the return-on-assets equation remains positive and significant (β = 0.051, p < 0.01) and is larger than the baseline coefficient, which would be consistent with the attenuation that measurement error induces in fixed-effects estimates (Wooldridge, 2010), although the larger standard error of the FE-2SLS estimate means that the two coefficients are not sharply distinguished, and the just-identified estimate (0.042) lies between them. The FE-2SLS results therefore indicate that the positive association between disclosure and profitability is robust to an estimation strategy that addresses potential reverse causality and to weak-instrument-robust inference, under the maintained assumption that lagged disclosure is a valid instrument; they are presented as robustness evidence rather than as proof of a causal effect. The effect of disclosure on the Z-score remains insignificant, governance structure is positive in both models and significant in the Z-score model, and the control variables retain their expected signs, so that the estimation using fixed-effects–2SLS does not alter the substantive conclusions.

Table 10. First-stage instrument diagnostics.

Instrument/Test	Coefficient/F	p
ESG (t − 1)	0.588 ***	0.000
ESG (t − 2)	−0.013	0.895
Joint first-stage F (conventional)	21.847 ***	0.001
Kleibergen–Paap rk Wald F (robust)	4.933	—
Hansen J, ROA/lnZ model	0.045/0.299	0.832/0.584
Anderson–Rubin χ2(2), ROA model	14.564	0.001
Just-identified (ESG t − 1 only): KP F; ROA coefficient	16.521; 0.0417 ***	0.001
Durbin–Wu–Hausman χ2, ROA/lnZ model	6.270/3.991	0.012/0.046

Note. The Hansen J test is reported because the model is over-identified (two instruments, one endogenous regressor). The Kleibergen–Paap, Hansen and Anderson–Rubin statistics are heteroskedasticity-robust; the just-identified model is estimated on 102 observations. *** p < 0.01. Source: Authors’ computation (2026).

Table 11. Second-stage fixed-effects–2SLS estimates.

Variable	ROA	lnZ
ESG	0.0514 ***	−2.1963
	(0.0191)	(1.6540)
GOVSTR	0.0140	4.7003 ***
	(0.0169)	(1.3831)
BSIZE	−0.0099 ***	−0.6941 ***
	(0.0037)	(0.2592)
MPR	0.1429 ***	1.0076
	(0.0400)	(3.2530)
INFL	0.0411	−2.1111
	(0.0410)	(3.7031)
Observations	90	90

Note. ESG disclosure instrumented by its first and second lags. Robust standard errors in parentheses. *** p < 0.01. Source: Authors’ computation (2026).

4.7. Diagnostic Tests

This study conducts a battery of post-estimation diagnostic tests to validate the reliability of the estimates. The results, reported in Table 12, support the chosen estimation strategy. The Wooldridge (2010) test does not reject the null of no first-order serial correlation, and the Breusch–Pagan test does not reject the null of homoskedasticity, so that the residuals are well behaved in these respects. The Pesaran (2015) cross-sectional dependence test rejects the null of cross-sectional independence, confirming that the banks are subject to common shocks arising from the shared macroeconomic and regulatory environment and thereby justifying the use of Driscoll and Kraay (1998) standard errors in the baseline specification. The Jarque–Bera test rejects the null of normality, a common feature of financial-ratio panels driven by a small number of extreme observations; the winsorisation of the dependent variables mitigates this concern, and the fixed-effects estimator remains consistent under non-normality (Wooldridge, 2010). The Hausman (1978) test does not reject the null hypothesis that the random-effects estimator is consistent (χ² = 1.053, p = 0.958). This result does not establish a statistical preference for fixed effects; rather, it indicates that the bank effects are not detectably correlated with the regressors, in which case the random-effects estimator would be consistent and more efficient. The fixed-effects estimator is nevertheless retained for three reasons. First, the sample effectively comprises the whole population of deposit money banks with continuous disclosure records rather than a random draw from a larger population, which is the setting in which a fixed-effects treatment of bank heterogeneity is conceptually appropriate (Baltagi, 2021). Second, the conventional Hausman test assumes that the random-effects estimator is fully efficient, an assumption that is violated in the presence of the cross-sectional dependence detected by the Pesaran test, so that its outcome is not reliable here (Wooldridge, 2010). Third, with twelve banks the test has low power to detect correlation between the bank effects and the regressors, and the fixed-effects estimator remains consistent whether or not such correlation exists, so that the efficiency forgone is the price of robustness. Taken together, the diagnostics indicate that the inference reported above rests upon an appropriate econometric foundation.

Table 12. Post-estimation diagnostic test results.

Test	Statistic	p	Remark
Serial correlation (Wooldridge)	0.154	0.103	No serial correlation
Heteroskedasticity (Breusch–Pagan)	4.769	0.445	No heteroskedasticity
Cross-sectional dependence (Pesaran)	3.232	0.001	Dependence present
Normality (Jarque–Bera)	61.130	0.000	Mitigated by winsorisation
Estimator (Hausman)	1.053	0.958	RE not rejected; FE retained (see text)

Note. FE = fixed effects; RE = random effects. Source: Authors’ computation (2026).

4.8. Discussion of Findings

Because the estimates are conditional associations from observational data, and the FE-2SLS results serve as robustness evidence rather than causal estimates, the findings are interpreted throughout as associations, and the theoretical mechanisms discussed below are offered as plausible explanations rather than as demonstrated causal channels. The results yield a nuanced rather than uniform account of the relationship between ESG disclosure and the stability of Nigerian deposit money banks: disclosure is positively associated with one dimension of stability, profitability, is not significantly associated with liquidity, asset quality or the Z-score, and is negatively associated with capital adequacy. The most robust finding is that disclosure is positively associated with profitability, a result that persists in the FE-2SLS estimates and that accords with the signalling account of Spence (1973) and the stakeholder account of Freeman (1984), under which credible disclosure reduces information asymmetry and the cost of funding and sustains the confidence upon which earnings depend. The finding is consistent with the Nigerian evidence of Iyawe and Chijuka (2025) and Ogboi et al. (2024), and it qualifies the more cautionary cross-country evidence of Cantero-Saiz et al. (2025) by suggesting that, for the large and visible banks that dominate the Nigerian sector, the positive association between disclosure and profitability is observed rather than suppressed. This divergence from the developing-country average that Cantero-Saiz et al. (2025) report is itself informative: it suggests that where banks are large, visible and subject to close regulatory scrutiny, the reputational and funding-cost channels through which disclosure operates are stronger, which would be consistent with the near-term costs of disclosure being more readily offset by its benefits.

The second finding concerns governance, whose direct and moderating roles must be distinguished. As a direct correlate, governance is the factor most consistently associated with stability: governance structure is associated with higher capital adequacy and profitability and lower non-performing loans, and governance disclosure with higher liquidity, profitability and Z-scores and lower non-performing loans, though with marginally lower capital adequacy. As a moderator, by contrast, the interaction between governance structure and disclosure is statistically significant only for capital adequacy and liquidity, and each of these interactions depends on a single bank; the moderation evidence is therefore fragile and is not interpreted as a robust moderating effect, and the strength of the direct associations cannot be taken as support for the moderation argument. The direct associations are consistent with the agency-theoretic proposition of Jensen and Meckling (1976) that governance disciplines risk-taking and lends disclosure its credibility, and they are consistent with the international evidence of Del Sarto et al. (2025) and with the pre-eminence of the governance dimension documented by Ahmed (2026). That the governance dimension dominates is intelligible in light of the special features of banking emphasised by Kafidipe et al. (2021): high leverage and opacity magnify the consequences of managerial misconduct, so that the monitoring function of governance is of first-order importance for stability. The dimensional analysis further reveals a tension in the social dimension, which is associated with higher non-performing loans, whereas governance disclosure is associated with lower ones; this tension may be attributable to the extension of credit to riskier borrowers under a financial-inclusion agenda and is consistent with the evidence of Samson and Tukur (2024) from the Nigerian setting. The environmental dimension, by contrast, is associated with lower capital adequacy, which may reflect the investment required to integrate environmental risk into lending before its benefits accrue (Olowofela et al., 2025).

It is worth dwelling on the economic magnitude of the central result. The baseline coefficient implies that a one-standard-deviation increase in the disclosure index is associated with a rise in the return on assets of about 0.12 percentage points, roughly four per cent of the sample mean, which, applied to the large asset bases of the sample banks, is a non-trivial absolute increment in earnings, though a modest one relative to the dispersion of profitability (Berger & Bouwman, 2013). The FE-2SLS estimate is larger, which would be consistent with attenuation of the baseline estimate by measurement error in the disclosure index (Wooldridge, 2010), but the estimates are imprecise, and the difference should not be over-interpreted. The magnitude of the governance coefficients tells a similar story: a one-standard-deviation increase in governance structure (0.089) is associated with a capital adequacy ratio higher by about 7.8 percentage points and a non-performing-loan ratio lower by about 3.1 percentage points, equivalent to roughly one-half and one-third of a standard deviation of those indicators respectively, underscoring that governance is a first-order correlate of bank stability in this setting (Edeh & Iwedi, 2024).

The mechanisms that may underlie these associations merit closer articulation; they are offered as theoretically grounded interpretations rather than as channels tested in this study. On the signalling account of Spence (1973), credible disclosure lowers the information asymmetry between a bank and the providers of its funding, so that better-disclosing banks face a lower cost of deposits and of wholesale funding; the resulting improvement in the net interest margin is one plausible channel through which disclosure may raise the return on assets (Bilyay-Erdogan et al., 2023). On the stakeholder account of Freeman (1984), disclosure sustains the loyalty of customers and the confidence of regulators, reducing both the volatility of funding and the likelihood of costly regulatory intervention (Xu et al., 2024). On the agency account of Jensen and Meckling (1976), governance disclosure and governance structure may together discipline the conduct of management, curbing the imprudent lending that generates non-performing loans and the excessive risk-taking that erodes capital (Del Sarto et al., 2025). That the governance dimension dominates the empirical results is consistent with the special importance of the agency channel in banking, where the interests of shareholders, depositors and regulators are imperfectly aligned (Kafidipe et al., 2021).

The absence of a robust moderating effect of governance structure, and the insignificance of the composite index in several of the models, are themselves instructive. They may reflect in part the limited variation in disclosure and governance among the large banks that constitute the sample, and in part the offsetting of the dimensional channels that the composite index aggregates, a phenomenon that the disaggregated analysis brings to light and that Liu (2022) and Arvidsson and Dumay (2022) anticipate. In particular, the absence of significant associations with liquidity, asset quality and the Z-score means that the evidence does not support the broader claim that ESG disclosure enhances the financial stability of Nigerian banks; it supports the narrower claim that disclosure is associated with higher profitability, while its relationships with the resilience dimensions of stability are heterogeneous and, for capital adequacy, negative. These findings caution against reliance upon an undifferentiated composite measure and upon the assumption of a uniform moderating effect, and they point to the governance dimension as the locus of the disclosure–stability relationship in the Nigerian setting. The findings as a whole are consistent with a regulatory approach in which sound corporate governance, effective credit-risk management and a favourable macroeconomic environment complement the strengthening of disclosure in support of stability (Erin et al., 2022).

5. Conclusions and Policy Implications

This study examined the relationship between environmental, social and governance disclosure and the financial stability of quoted deposit money banks in Nigeria, and the moderating role of governance structure, using panel data on twelve listed banks over the period 2015 to 2025. This study employed bank fixed effects with Driscoll–Kraay standard errors as the baseline estimator and fixed-effects–two-stage least squares as a robustness check for endogeneity, together with a battery of specification and diagnostic tests. The findings reveal that ESG disclosure is positively and significantly associated with bank profitability, a result that persists when potential endogeneity is addressed through FE-2SLS, whereas its associations with liquidity, asset quality and the Z-score are not significant and its association with capital adequacy is negative. The associations between disclosure and bank stability are therefore heterogeneous across dimensions, and the profitability result should not be generalised to financial stability as a whole. Governance structure has strong direct associations with stability, being associated with higher capital adequacy and profitability and lower non-performing loans, and governance disclosure is the ESG dimension most consistently associated with stability. The environmental dimension is associated with lower capital adequacy (Olowofela et al., 2025), the social dimension exhibits a tension between financial inclusion and loan quality (Samson & Tukur, 2024), and the interaction between governance structure and disclosure is statistically significant only for capital adequacy and liquidity but is sensitive to the exclusion of individual banks, so it is not interpreted as a robust moderating effect.

Overall, the results show that banks that disclose more fully tend to earn higher returns and that more strongly governed banks tend to have stronger balance sheets; because these are associations, they do not establish that fuller disclosure or stronger governance causes higher returns or greater resilience. The findings are of particular relevance to the ongoing recapitalisation of the Nigerian banking sector, for they suggest that the strengthening of capital may usefully be accompanied by the strengthening of governance and disclosure. This is consonant with the objectives of the Basel III framework (Basel Committee on Banking Supervision, 2021), which seeks to enhance the capacity of banks to absorb losses while sustaining their support for the real economy, and with the sustainable-banking agenda that the Central Bank of Nigeria (2012) and the Nigerian Exchange have progressively advanced.

The findings carry several implications for policy and practice; because they rest on associations rather than causal estimates, these implications are indicative rather than prescriptive. For regulators, the evidence that governance is closely associated with stability, and that disclosure is associated with higher profitability, strengthens the case for the continued development of the sustainability-disclosure requirements of the Central Bank of Nigeria and the Nigerian Exchange, and for the reinforcement of the governance mechanisms—board independence, board diversity and audit quality—that render disclosure credible, particularly as the country advances towards the adoption of the international sustainability-reporting standards. For bank managers, the evidence that disclosure is associated with higher profitability and that governance is closely associated with stability supports genuine investment in substantive disclosure and in effective governance, and in the strengthening of loan-screening, monitoring and recovery so as to contain non-performing loans. For investors, the findings furnish a clearer understanding of how disclosure, and its governance dimension in particular, may signal the resilience of banks (Spence, 1973; Bilyay-Erdogan et al., 2023). For policymakers, the results support the integration of sustainability and governance considerations into the framework of financial-stability supervision and into the design of the recapitalisation programme, and they underscore the importance of containing inflation, which the analysis associates with weaker bank performance.

Beyond its immediate findings, this study carries a broader message for the design of the sustainability-disclosure regime in emerging economies. The evidence that the governance dimension dominates the disclosure–stability relationship suggests that disclosure frameworks which treat the environmental, social and governance pillars as interchangeable, or which reward the quantity of disclosure without regard to its governance underpinnings, may misallocate regulatory attention. A framework that privileges the credibility of disclosure, through assurance requirements, board-level accountability and audit oversight, may be more likely to translate reporting into resilience than one that multiplies disclosure items (Arvidsson & Dumay, 2022; Gyimah & Owusu-Afriyie, 2025). For a banking system in the midst of recapitalisation, this suggests that the strengthening of capital and the strengthening of governance are complements rather than substitutes, and that the returns to recapitalisation may be greater where it is accompanied by genuine improvements in board independence, diversity and audit quality.

This study is subject to limitations that qualify its conclusions and point to directions for further research. The sample is confined to twelve deposit money banks, so that the findings may not generalise to unlisted banks, microfinance institutions or fintech firms, and the modest cross-sectional dimension of the panel limits the power to identify moderating effects. The disclosure index, though carefully constructed, measures the extent of disclosure and captures its quality only imperfectly; moreover, it is close to its ceiling, particularly for the social and governance dimensions, which concentrates within-bank variation in a few banks, limits the power to detect effects on the resilience dimensions of stability, and renders the moderation results sensitive to individual banks. The FE-2SLS estimates rest on the untestable exclusion restriction that lagged disclosure affects current stability only through current disclosure, and they are therefore best regarded as robustness evidence rather than as causal estimates. Because year effects cannot be included alongside the macroeconomic controls, common shocks that are not captured by the monetary policy rate and inflation may remain in the error term. The short time dimension also limits the power of the panel unit-root tests, so that the order of integration of the macroeconomic controls could not be established. Future research might extend the sample across the West African sub-region, employ quality-weighted disclosure measures, examine non-linearities in the disclosure–stability relationship, test for an optimal level of disclosure beyond which additional disclosure yields diminishing returns, and exploit the regulatory changes of the recapitalisation programme as sources of identification (Staiger & Stock, 1997). Notwithstanding these limitations, this study provides timely evidence that ESG disclosure is associated with higher profitability among Nigerian deposit money banks, that its relationships with the other dimensions of stability are heterogeneous, and that governance is closely associated with bank stability, and it thereby advances both the scholarly and the policy conversation on sustainable banking in emerging economies.