When hidden risks surface: Overconfident bank CEOs and portfolio reallocation under policy uncertainty

In October 2025, US regional bank stocks experienced one of their sharpest declines since the spring 2025 tariff-related market shock. Zions Bancorporation and Western Alliance disclosed fraud-tainted commercial loans; Fifth Third recorded a charge-off linked to a failed subprime auto lender; and JPMorgan, after writing off $170 million from the same borrower, publicly warned that isolated credit problems can often signal wider vulnerabilities.

Analysts largely treated these losses as idiosyncratic underwriting failures against a backdrop of elevated economic policy uncertainty (EPU). Aggregate credit growth remained subdued and loan-loss reserves looked adequate. Yet the losses concentrated in specialised, opaque commercial exposures that conventional supervisory metrics had not flagged.

Our recent paper in the Journal of Banking and Finance provides a behavioural explanation for such episodes. We examine whether all bank CEOs respond similarly to rises in EPU, or whether managerial overconfidence amplifies risk.

Data and research design

We study a panel of US listed bank holding companies from 2000 to 2021. CEO overconfidence is measured using the revealed-beliefs approach of Malmendier and Tate (2005, 2008): CEOs are classified as overconfident if they persistently fail to exercise deep-in-the-money vested stock options (the Holder67 measure). EPU is measured by the Baker et al (2016) index, orthogonalised following Xu (2020) to isolate policy-driven uncertainty from broader macroeconomic conditions.

Bank risk is captured by a battery of market- and accounting-based measures (idiosyncratic volatility from Fama–French five-factor residuals, equity return volatility, expected shortfall, log Z-score, and the non-interest income ratio).

Main findings

A one-standard-deviation increase in exogenous EPU raises bank risk by approximately 8–20% more at banks led by overconfident CEOs than at otherwise similar banks. The amplification is robust to difference-in-differences designs that exploit plausibly exogenous CEO turnovers (illness, death, or pre-announced departures) and to instrumental-variables estimation using CEO age. It is not explained by CEO power, gender, founder status, or compensation incentives (delta and vega).

The transmission channel: portfolio composition, not volume or provisioning

Contrary to crisis-period evidence (Ho et al 2016), the risk does not arise from more aggressive aggregate lending or weaker loan-loss provisioning. Overconfident CEOs neither expand total originations nor under-reserve relative to their peers during high-EPU periods.

Instead, they reallocate the loan book. Originating activity shifts away from safer, collateralised, loss-buffering categories (residential mortgages and consumer credit) toward commercial and industrial (C&I) lending, which is harder to collateralise and more sensitive to information asymmetries.

Vintage-loss analysis that traces originations forward into subsequent charge-offs and non-performing loans shows that loans made by overconfident CEOs in high-EPU periods generate significantly higher realised losses over the following five years. The damage stems primarily from the erosion of the safer portion of the portfolio rather than from the expansion of the riskier portion. Bank performance (ROA, ROE, and Tobin’s Q) deteriorates accordingly.

The underlying mechanism is psychological rather than strategic. Overconfident managers exhibit overprecision: they place excessive weight on their private assessments and under-adjust their beliefs when the information environment becomes noisier—precisely the condition created by elevated policy uncertainty (Moore and Healy 2008; Ben-David et al 2013). The result is a growing gap between perceived and actual risk that is invisible in aggregate volume and provisioning data because it operates through composition.

Why recent episodes fit the pattern

The banks affected in October 2025 were not characterised by runaway loan growth; several were described as conservative institutions with solid reserves. The costly exposures were concentrated in specialised commercial niches—the very segments toward which our results show overconfident CEOs reallocate credit under uncertainty.

Policy implications

Supervisory frameworks centred on aggregate loan growth, capital ratios, and provisioning coverage can systematically miss risk that accumulates through shifts in loan mix. Stress tests and examination programmes would benefit from systematically decomposing portfolio changes by loan category during spikes in policy uncertainty.

Boards possess an underused governance lever. Executive risk-perception traits are, in principle, observable before losses materialise—through persistent failure to exercise deep-in-the-money vested options and other revealed-belief indicators. Integrating such behavioural signals into succession planning and risk oversight could improve resilience.

Policy uncertainty itself does not render bank CEOs reckless. It does, however, widen the scope for error among those already too confident in their own judgement. As recent market events illustrate, that error tends to surface precisely where conventional monitoring is weakest.

The authors’ article is available here.

Kwabena Aboah Addo is an Assistant Professor of Finance at the Open Universiteit, the Netherlands.

Shams Pathan is a Professor of Finance at the German International University.

Steven Ongena is a Professor of Banking at the University of Zurich, and holds positions at the Swiss Finance Institute, KU Leuven, NTNU Business School, and CEPR.