Resolving “Too Big to Fail”

Recent examples include Gandhi and Lustig (2015) and Acharya et al. (2016). See Kroszner (2016) for a broad review of the methodological approaches used in the extant literature.

Penas and Unal (2004), Dam and Koetter (2012), and Duchin and Sosyura (2014) study these moral hazard issues in detail.

For empirical support of this political economy rationale, Brown and Dinċ (2011) show that after elections, policymakers are less likely to support failing banks.

Penas and Unal (2004) study bond financing cost effects of bank mergers. They find differentially large benefits when medium-sized banks merge to form a large bank.

Dam and Koetter (2012) use political instruments to show that German banks are more likely to enter distress if they expect a bailout, and Duchin and Sosyura (2014) use Troubled Asset Relief Program data to show that US banks originate riskier loans and shift assets toward riskier securities after getting bailed out.

If there was a strong effect of the policy decreasing the likelihood of default, it should be observable directly in a reduced price of credit default swap contracts. However, existing evidence does not corroborate a strong countervailing effect: Berndt et al. (2018) show a net increase in CDS prices for large banks in the years after Lehman Brothers’s failure, which the authors associate with a decrease in the perception of TBTF guarantees.

Source: https://​www.​minneapolisfed.​org/​publications/​the-region/​interview-with-eugene-fama.

Note that since the outcome variables change between different tested outcomes, so too do the covariate weights and hence the synthetic control group weights.

The difference in size between banks subject or not to the new regulation remains an important challenge for this type of studies, due to the extreme skewness in the size distribution of US banks. For example, the average size of the ten smallest treated banks is about 90 billion, The average size of the ten largest of the non-treated is about 30 billion. This data characteristic advises against using traditional methods, and it is indeed one of the reasons we chose to adopt a synthetic control methodology.

While the two metrics are commonly used to gauge systemic risk, they nevertheless capture different factors contributing to overall risk. As highlighted in the literature, they may not yield similar rankings in a cross section of firms (Benoit et al. 2013). We present results using both to show that the result holds regardless of ones’ particular favored systemic risk measure, and is not sensitive to this choice.

Typical estimates range between 20 and 100 basis points. See Gropp et al. (2010), Ueda and Di Mauro (2013), Gandhi and Lustig (2015), and Acharya et al. (2016).

Nevertheless, we ran a robustness test to gauge the potential impact of the regulation on the cost of new debt issuances. To do so, we gained access to the Mergent Corporate Bond Securities database and extracted information for all debt issues with an offering date between January 1, 2009 and December 30, 2016. We then hand-matched these issues to our main sample, yielding 22,199 issues mapped to our treated entities, and 115,636 mapped to other entities. We estimated a difference-in-differences model where the dependent variable was the spread over treasury (maturity matched) of each yield at issuance, against a dummy that turned equal to one for the announcement of the living wills policy, a dummy for the treated subset of issuers, and the interaction of the two dummies, while including time and parent company fixed effects. In the attempt to match treatment to a comparable control, we restricted the issuers in the control group to U.S. financial sector firms. We find that after the announcement of the living wills regulation, the banks subject to the treatment display relatively higher spreads. Interestingly, the point estimate of that effect, about 20 basis points, is of a magnitude comparable to our average synthetic control estimates for the non-deposit debt component (as reported on Fig. 6, right panel), which shows roughly a relative higher cost of debt for treated banks between 10 and 20 basis points. Consistent with our main analysis, both treated and untreated display a common negative trend, so this positive result indicates that treated banks experienced a relatively lesser decline.