REVIEW 4 major objections 6 minor 2 references
Hedging Deposit Run Risk Prior to the 2023 Regional Banking Crisis
T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read Banks did not appear to hedge deposit run risk before the 2023 regional banking crisis: interest-rate option holdings and equity show no significant relation to changes in uninsured deposits during the 2022–23 rate-hike cycle.
desk verdict A transparent but identification-limited null result: the paper's strong 'no evidence' conclusion overreaches a sample that drops to derivative-using banks and flips the equity sign. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The engine is a cross-sectional regression of the change in the share of uninsured deposits (total deposits minus insured deposits, divided by total deposits) from Q4 2021 to Q3 2023 on pre-crisis bank characteristics, including the notional amounts of interest-rate options bought and sold, the equity-to-assets ratio and its change, notional futures/forwards and swaps, the brokered-deposit share, the initial uninsured-deposit share, log assets, and liability-repricing maturities. The load-bearing comparisons are the t-tests on the option and equity coefficients: the model predicts negative coefficients if banks hedged, so insignificant coefficients carry the conclusion. The reach of these tests is constrained by sample size: deposit data cover roughly 4,500 banks but derivative data only about 1,123, and the equity coefficient flips sign between the full sample and the derivative-using subsample.
What would settle it
Re-run the regression with confidential supervisory data that report option direction and type, or add controls for digital-banking density and deposit-pricing; if the coefficient on convex option positions or on equity growth becomes significantly negative within the derivative-using sample, the paper's conclusion that banks did not hedge run risk would be contradicted.
Extended reading notes
Core claim
On its own terms, the paper's central discovery is a set of null results: in cross-sectional regressions of the change in the percentage of uninsured deposits between Q4 2021 and Q3 2023, the notional amount of interest-rate options bought or sold in Q4 2021 is statistically insignificant, and the level of equity as well as the change in equity over the hiking cycle are insignificant in the derivative-using sample. Futures and forward positions are associated with larger declines in uninsured deposits and swaps with smaller declines, but the paper reads these instruments as not being effective hedges against run risk. Because the theoretical benchmark says the only balance-sheet hedges are convex-payoff derivative contracts and increased equity, the insignificance of options and equity is treated as evidence against proactive run-risk management. The paper is explicit that the test is limited by data granularity: call reports supply option notional amounts but not option type or direction.
Load-bearing premise
The load-bearing premise is that the regression's controls absorb every other factor that could jointly drive a bank's derivative and equity choices and its deposit outflows; if anything correlated with both is missing, the insignificant coefficients do not establish that hedging had no effect.
Editorial extensions
If this is right
- If correct, observable call-report measures of option notional and equity ratios should not be used to identify banks preparing for deposit runs; prior uninsured and brokered deposit shares are the clearer risk markers.
- The mixed signs on futures/forwards versus swaps imply that using interest-rate derivatives at all did not protect a bank's deposit base during the hiking cycle.
- Because the data lack option direction and type, the conclusion is strictly about reported notional amounts, not about all possible forms of option-based hedging.
- Banks entering a rate-hiking period with high uninsured deposit shares should expect larger outflows regardless of their option or equity positions.
Reading between the lines
- A natural extension is to re-estimate the model with confidential supervisory data identifying option direction, strikes, and maturities; the null could be an artifact of netting bought and written options into a single notional measure.
- The flip in the equity coefficient between the full sample and the derivative-using subsample suggests selection: banks that report interest-rate derivative positions differ systematically, so the restricted sample may not speak to the whole banking sector.
- Because digital-banking density and deposit-pricing strategy were shown elsewhere to drive deposit flows, their omission from these regressions means the null coefficients cannot fully separate 'no hedging' from 'hedging confounded with other flow drivers'.
- A sharper test would examine subsequent liquidity-stress outcomes beyond deposit changes, such as discount-window borrowing or asset sales, to see whether convex option positions predicted resilience.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper tests whether banks proactively hedged deposit run risk ahead of the 2022–2023 rate-hiking cycle, following Drechsler et al. (2023) which predicts that convex-payoff derivatives (options) and equity increases are the main hedging tools. Using FDIC Call Report data, the authors regress the change in the percent of uninsured deposits from Q4 2021 to Q3 2023 on equity levels, equity changes, and interest-rate derivative notional amounts (futures/forwards, swaps, options bought/sold). They find insignificant coefficients on option use and on equity levels/equity changes, but significant effects of futures/forwards and swaps. The conclusion is that there is no evidence of banks managing run risk via their balance sheet.
Significance. If the result held, it would speak to an important policy question: whether regional banks took anticipatory measures before the 2023 crisis. The paper directly tests a theoretical prediction from Drechsler et al. (2023) and uses a comprehensive regulatory dataset, which are notable strengths. However, the central null result is currently identified on a nonrandom, derivative-reporting subsample and is vulnerable to omitted confounders that the paper itself cites. These issues, rather than the theoretical framing, are what prevent the paper from providing a convincing answer. The paper is transparent about data limitations, which is commendable, but the inference issues require substantive additional analysis.
major comments (4)
- [§2, Table 3] The central null result is identified only on the derivative-reporting subsample. Adding derivative variables drops the sample from 4,544 to 1,123 banks, and the equity coefficient changes sign from +0.075*** (column 1) to -0.097* (column 3). Because the derivative data exist only for banks that chose to use derivatives, the subsample is selected on the hedging behavior the paper aims to study. Without a selection correction or at least a baseline regression on the common sample, the insignificant option coefficients and the sign-reversed equity coefficient do not establish an absence of hedging; they may reflect compositional differences between derivative users and nonusers. The paper's own text concedes (Section 2) that the hypothesis is 'possible... supported over all banks, though not for the restricted sample,' which directly undermines the strength of the conclusion as stated in the abstract and Section 3.
- [§1 and Tables 3–4] The regressions omit two channels the paper itself identifies as important drivers of deposit flows: digital-banking density (Benmelech, Yang, and Zator 2023) and deposit-pricing strategy (d'Avernas et al. 2023). If these factors correlate with both derivative use and uninsured-deposit changes, the coefficients on option notional and equity are biased. This is particularly concerning because the paper's contribution is a null result; a biased null is indistinguishable from an absent effect. At minimum, the authors should add proxies for digital banking (e.g., branch density) and deposit pricing (e.g., deposit betas or average deposit rates) as controls, or discuss why they cannot be constructed from Call Report data.
- [§1 and §2, Tables 3–4] The option measures are net notional amounts without any information on net position (long/short) or option type (call/put, swaption vs. cap/floor). The paper acknowledges this on page 7, but the acknowledgment undercuts the power of the test. Drechsler et al.'s prediction is specifically about convex instruments such as swaptions; combining all options and all directions in one notional variable makes it difficult to detect a hedging channel even if it exists. The conclusion that there is 'no evidence' should be phrased as 'no evidence given the aggregation of the available Call Report data,' which is a weaker and arguably different claim.
- [§2, Table 4] The Change in Equity variable is measured over the same interval (Q4 2021 to Q3 2023) as the dependent variable. This creates a contemporaneous endogeneity problem: deposit outflows can mechanically change the equity ratio through reductions in assets or through capital raises during the crisis, so the regression does not test whether banks proactively raised equity before the hiking cycle. A cleaner test would use the equity change from, say, Q4 2021 to Q4 2022, or the level of equity relative to a pre-trend, to avoid reverse causality. As it stands, the insignificant coefficient on Change in Equity is not informative about Hypothesis 2.
minor comments (6)
- [Table 3 note] The note reports significance thresholds as *p<0.1; **p<0.05; ***p<0.01, but the table uses **** in the coefficient rows (e.g., -0.218****). Either add the **** threshold for 0.1% to the note or change the stars to match the note.
- [Tables 3 and 4] The text says standard errors are 'heterskedasticity-consistent'; this is a typo for 'heteroskedasticity-consistent.' Please fix throughout.
- [Section 1, Hypotheses] The direction of Hypotheses 1 and 2 is unclear: 'Greater levels of swaptions owned in Q4 2021 implies a larger decline in uninsured deposits' and 'Higher levels of bank equity ... implies a larger decline in uninsured deposits' read as though hedging makes runs worse. If the intended prediction is that hedging reduces the decline, please rephrase to avoid confusion.
- [Section 2, paragraph on brokered deposits] The sentence 'This makes sense given brokered deposits (being wholesale funding) have a beta near 1 with respect to changes in the short-term interest rate' is under-specified; clarify what 'beta' refers to and provide a citation or derivation.
- [Table 2] The summary statistics show N=1,172 for IR Futures Forwards and IR Swaps, but N=1,165 for IR Options Bought. Please explain why the sample sizes differ across derivative types within the same table and whether this affects the regressions in columns (3) and (4).
- [Section 2, first paragraph] The phrase 'Not hedging appears to hinder a bank’s ability to lend' is a strong causal claim based on Krainer and Paul (2023); consider softening to 'is associated with' unless the cited paper establishes causality directly.
Circularity Check
No circularity: the regression tests externally motivated hypotheses against separate Call Report variables; no fitted parameter or self-citation is folded back into the claim.
full rationale
The paper's central claim—that interest-rate option holdings and equity levels are unrelated to changes in uninsured deposits over 2021–2023—does not reduce by construction to its inputs. The dependent variable (Q4 2021 to Q3 2023 change in percent uninsured deposits) and the explanatory variables (option notionals, equity ratios, etc.) are independently extracted FDIC Call Report items. Hypotheses 1 and 2 are taken from Drechsler et al. (2023), an external theory paper, not from the authors' own prior work; the manuscript contains no self-citations at all. No parameter is fitted to the outcome and then reported as a prediction. The paper's honest caveats—'it is possible that our hypothesis is supported over all banks, though not for the restricted sample' (Section 2) and the conclusion that results are 'affected by the lack of granularity in FDIC Call Report data on bank option use' (Section 3)—are statistical and data-availability limitations, not circular reasoning. Concerns about sample selection (4,544 vs. 1,123 banks) and omitted confounders are validity risks, not identity or self-reference between input and output. Therefore the derivation chain is self-contained and the circularity score is 0.
Assumptions & free parameters
free parameters (2)
- Observation window endpoints =
Q4 2021 to Q3 2023
- Derivative-user subsample =
1,123 banks
assumptions (4)
- domain assumption The only ways to hedge deposit run risk are convex-payoff derivatives such as options and raising equity as rates rise (Drechsler et al. 2023).
- domain assumption FDIC Call Report notional amounts of derivatives proxy for hedging intensity.
- domain assumption Change in percent uninsured deposits from Q4 2021 to Q3 2023 is an appropriate dependent variable for deposit run risk and is not driven by active bank management of deposit structure.
- domain assumption No omitted variables correlate with both derivative and equity positions and uninsured-deposit changes.
Cite this review
Pith. "Pith review of Hedging Deposit Run Risk Prior to the 2023 Regional Banking Crisis." pith.science (2026). https://pith.science/paper/D2PHSPJ5
@misc{pith2026250603344,
author = {Pith},
title = {Pith review of: Hedging Deposit Run Risk Prior to the 2023 Regional Banking Crisis},
year = {2026},
howpublished = {\url{https://pith.science/paper/D2PHSPJ5}},
note = {Machine review of arXiv:2506.03344}
}
read the original abstract
In this analysis we determine factors driving the cross-sectional variation in uninsured deposits during the interest rate raising cycle of 2022 to 2023. The goal of our analysis is to determine whether banks proactively managed deposit run risk prior to the hiking cycle which produced the 2023 Regional Banking Crisis. We find evidence that interest rate forward, futures, and swap use affected the change in a bank uninsured deposits over the period. Interest rate option use, however, has no effect on the change in uninsured deposits. Similarly, bank equity levels were uncorrelated with uninsured deposit changes. We conclude we find no evidence of banks managing run risk via their balance sheet prior to the 2023 Regional Banking Crisis.
Figures
Reference graph
Works this paper leans on
-
[1]
Bank runs, deposit in- surance, and liquidity
Benmelech, Efraim, Jun Yang, and Michal Zator (2023).Bank branch density and bank runs. Tech. rep. National Bureau of Economic Research (cit. on p. 2). d’Avernas, Adrien et al. (2023). The Deposit Business at Large vs. Small Banks. Tech. rep. National Bureau of Economic Research (cit. on p. 3). Diamond, Douglas W and Philip H Dybvig (1983). “Bank runs, de...
work page 2023
-
[18]
Monetary transmission through shadow banks
url: http://dx.doi.org/10.24148/wp2023-18 (cit. on p. 5). Xiao, Kairong (2020). “Monetary transmission through shadow banks”. In: The Review of Financial Studies 33.6, pp. 2379–2420 (cit. on p. 3). 11
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.