This thesis investigates whether the quality of bank Environmental, Social and Governance (ESG) disclosures reduces stock return volatility in a global sample of 80 listed commercial banks drawn from the S&P Top 100 Banks ranking and spanning the period 2020-2024. Using computational text analysis on 390 bank-year reports (285 standalone ESG/sustainability documents and 105 integrated annual reports), four theoretically motivated disclosure-quality metrics are constructed: the Gunning Fog Index of readability; the Loughran-McDonald net positive sentiment ratio (tone); ESG vocabulary density; and a three-component SASB material-topic coverage metric capturing depth, balance and breadth across the six SASB Commercial Banking material topics. The dependent variable is annualised stock return volatility, computed from daily log returns. The regression specifications include year-varying market capitalisation, year-varying beta against MSCI ACWI, year-varying return on equity, the year-varying Tier 1 capital ratio, total assets, a report-type dummy, log word count, and year and region fixed effects, with standard errors clustered at the bank level. The headline findings strongly support the linguistic-credibility dimension of disclosure quality: the Loughran-McDonald tone of bank ESG reports is robustly and negatively associated with stock return volatility (beta = -0.122, p = 0.001 in the preferred cross-sectional specification, beta = -0.055, p = 0.030 once year-varying ROE and beta are added, and beta = -0.203, p < 0.001 in the standalone-reports subsample), is significant at the 1% level under Beck-Katz panel-corrected standard errors and feasible GLS with AR(1) errors alike, and survives explicit bank-size control and exclusion of the high-volatility COVID year. The Gunning Fog Index displays a positive association with volatility consistent with the obfuscation channel, significant at 5% in the cross-section (beta = +0.0072, p = 0.022) and in the standalone-reports subsample (beta = +0.0068, p = 0.029), and marginally significant at 10% in the full-control specifications and under all four alternative error-structure estimators. Both stylistic results are robust to adding the Tier 1 capital ratio as a solvency control, and to a battery of identification checks: a Mundlak correlated-random-effects decomposition with a regression-based Hausman test confirms that pooled OLS consistently estimates the disclosure effects, which load on the betweenbank dimension (a test restricted to the disclosure variables does not reject pooled consistency, F = 1.97, p = 0.107); and under a one-year lag the readability effect is essentially unchanged — prior-year obfuscation predicts current-year volatility — whereas the tone effect attenuates, indicating a contemporaneous association. The classical impression management hypothesis is partially supported: banks with lower ROE produce systematically less readable ESG reports (beta = -5.55, p = 0.028), providing direct evidence of strategic obfuscation when financial performance is weak; the companion tone-inflation channel is not supported in either direction (beta = +0.366, p = 0.175). The tone effect is, however, performance-contingent: a Tone x ROE interaction (beta = 0.717, p = 0.039) shows that the volatility-dampening association of positive tone is concentrated among weak performers and is absent for strong performers, locating the tone channel precisely where impression-management incentives are strongest. A novel descriptive contribution emerges from a topic-conditioned analysis of disclosure quality: in 99.5% of bank-year reports the Fog Index is materially higher in sentences referencing SASB material topics than in the rest of the report (mean within-document gap of +9.5 Fog points, Cohen's d = +1.76), and in 91.3% of reports the Loughran-McDonald tone of material content is markedly less positive than that of non-material content (mean gap -0.27, Cohen's d = - 1.43). This pattern of selective obfuscation and de-tonalisation of material content is documented for the first time in a banking context, is highly statistically significant under every test, and is consistent across years, report types and regions; the regression evidence indicates that it operates as a structural feature of the industry rather than a discriminator across banks. The substantive content metrics (ESG vocabulary density, SASB materialtopic coverage) are not significantly associated with volatility in the cross-section, and the materiality premium hypothesis is not supported. The findings have implications for the design of mandatory sustainability disclosure frameworks (CSRD, ESRS, ISSB), suggesting that readability and tone standards should complement the substantive coverage requirements that dominate current regulatory practice, and that targeted constraints on material-topic readability may have particular regulatory leverage.