Overview

This chapter explains how the New York Fed study measures the causal effect of stricter U.S. banking supervision during the 2008 Global Financial Crisis. Ordinary comparisons are confounded because troubled banks attract more examinations and are also more likely to fail. The study instead uses a pre-scheduled federal-state examination rotation after Lehman’s failure, comparing eligible banks more likely to receive a federal-led exam with those more likely to receive a state-led exam. Its main causal sample contains 3,658 mostly smaller state-chartered banks, with 116 failures. The design identifies a local, conditional supervisory path—including feedback from ratings to later scrutiny—not a universal effect for all banks, shocks, or individual reporting decisions.

Learning outcomes

Comparing heavily and lightly examined banks confounds supervisory selection with supervisory effect.
The pre-scheduled rotation changes early federal-led examination exposure for otherwise similar eligible banks after Lehman’s failure.
The estimates are local and conditional: they include feedback from early scrutiny and do not generalize automatically to large banks or other shocks.

Keywords

CausalityRotationInstrumentSelectionStrictnessCrisis

The Crisis Experiment: Measuring Supervisory Strictness

The central research problem in banking supervision is causal: do stricter examinations change a bank’s path, or do they simply appear more often around banks that were already deteriorating? This chapter focuses on how the New York Fed study addresses that problem during the 2008 Global Financial Crisis. Understanding the design lets readers interpret the later findings as evidence about a defined supervisory intervention rather than as a general correlation between regulators and bank failure.

The distinction is essential because the information process runs in both directions. A weak bank may receive a worse rating, more examinations, and demands for revised loss estimates. The same bank may also be more likely to fail. If the stricter supervision and the failure are observed together, the association could reflect the bank’s weakness rather than an effect of the supervisor. The study’s contribution is to find variation in supervisory exposure that was arranged before the crisis outcomes were known.

Why ordinary comparisons mislead

Imagine comparing banks that received many examinations with banks that received few. The heavily examined group might look worse on every outcome: lower capital, poorer ratings, more revisions, and more failures. That would not establish that examinations caused the deterioration. Supervisors deliberately spend more time on institutions that look risky, and the paper shows that failing banks receive more frequent examinations, longer examinations, and a larger share of unscheduled or non-routine examinations as failure approaches.

This is a form of selection, not a minor statistical nuisance. The supervisor is responding to an early signal of trouble, while the researcher is tempted to treat the later supervisory action as the source of the trouble. If the bank’s initial weakness is only partly observed, adding more measured controls may not remove the bias. A bank that receives a written agreement or a long examination may have an unrecorded problem that also predicts its later failure. The comparison confounds the decision to supervise with the effect of how supervision was conducted.

The same problem affects financial reporting. A distressed bank is more likely to revise its Call Report after an examination, but that revision may be both a response to the examination and a symptom of the bank’s worsening condition. Ratings, capitalization, examination intensity, and closure decisions are jointly determined. A statistical relationship among them is informative about the system, but it does not by itself identify the direction of causation.

The paper therefore distinguishes several questions that are often blurred together:

Selection: Which banks are assigned or attract more supervisory attention?
Strictness: Given comparable conditions, how demanding is the supervisor’s assessment?
Mediation: Does strictness change loss recognition, reported capital, or other bank behavior?
Outcome: Does the altered information or behavior change ratings, closure, resolution cost, or lending?

A credible design must separate the first item from the other three. It also needs to avoid treating the supervisor’s rating as a pure measure of bank health. A rating is partly an outcome of the examination itself, especially when the examination causes the bank to report a more conservative condition.

The study uses differences between federal and state supervisors. Earlier research found that federal supervisors were, on average, stricter in their assessment of bank riskiness. But simply comparing banks examined by federal agencies with banks examined by state agencies would still be inadequate: federal supervisors may be more likely to lead exams at banks that are already in trouble. The useful variation comes from institutions for which the lead agency was determined by a pre-scheduled rotation.

Published by DeepMox

FAQ

Why can’t a simple comparison of examined and unexamined banks identify supervisory effects?

It cannot because supervisors direct more attention to banks that already look risky, and those banks are also more likely to revise reports or fail. The comparison mixes the effect of supervision with the selection of institutions that receive it.

How did the examination rotation create a causal comparison?

It used the prior federal-state sequence to predict whether an eligible bank’s next early-crisis examination would be federal-led or state-led after Lehman’s failure. Because that schedule was set before later crisis outcomes, it provides plausibly external variation in supervisory exposure, subject to the study’s assumptions.

Which banks does the study’s causal estimate apply to?

It applies mainly to eligible smaller state-chartered commercial banks on the rotating schedule during the Global Financial Crisis: the main sample had 3,658 banks and 116 failures. It does not directly estimate effects for national banks, large complex institutions, or other financial shocks.

Previous11 minU.S. Banking Supervision: Capital to Legal Action

Next9 minHow Stricter U.S. Banking Supervision Moves the Failure Clock

U.S. Banking Supervision: Measuring Strictness in Crisis

The Crisis Experiment: Measuring Supervisory Strictness

Why ordinary comparisons mislead

FAQ