In the previous post, I briefly mentioned some of the potential issues of the leverage ratio (LR). To be fair, these are only issues to the extent that the regulators did not foresee and did not include these consequences in the cost-benefit analysis. For instance, in the case of repo activities, it could well be that regulators are fine with banks pulling back from these activities. But in order to do any of this, we first need to understand the effects of the different policies that are being put in place.
Allahrakha, Cetina, and Munyan (2016) look at how banks reacted after the announcement of the LR regulation. They focus on the tri-party repo market and use the empirical strategy known as difference-in-differences: they look at how banks more affected by the leverage ratio reacted as compared to less affected ones. They have daily dealer-level activity information, so they are able to observe this market at a fairly granular level.
A key element in this type of papers is to indentify the ‘control’ and the ‘treatment’ groups. In this case, they define six different groups, although I will only mention four. The ‘control group’, which in this case only means the group that does not have a dummy variable in the regression, is formed by dealers that are part of a bank holding company (BHC) that is not subject to the LR. Another ‘control group’ that is looked at is formed by dealers that are not part of a BHC (and hence not subject to the LR either). The two treatment groups are: dealers that are part of a BHC with a LR below 3.5% in 2012, and dealers that are part of a BHC with a LR above 3.5% in 2012. The idea is that the former should react more as it is more constrained by the LR.
The second important element is to decide when is the treatment applied. This is fuzzier that it seems. In December 2010, the BCBS issued the Basel III framework where there was a reference to the LR. In June 2012, the US regulators issued a proposal. In June 2013, the BCBS issued a consultative document. In July 2013, the US regulators finalised the rule. And so on… So when should the treatment had taken place? They settle for June 2012, and I would have chosen the same date. But looking at the data they conclude that the series breaks on November 30th 2012. Why?
Nothing really happened that day, at least that we publicly know, about the LR. The authors claim that “This corresponds to the public comment period following the first announcement… and suggest bank holding companies may have assessed the impact of the proposed rule on their operations and responded proactively“. But the public comment period actually ended more than a month before, after it was extended. And more importantly, the fact that they are already assuming that the LR had an aggregate effect should alert anyone interested in approaching empirical research as agnostically as possible.
The authors show, using the diff-in-diff approach, that more affected banks reduced repo activities more as compared to the rest. There is no evidence that less-affected banks reduce repo activities more than not affected ones. This is important because it means that banks do behave differently if they are more LR-constrained. Moreover, the most affected banks seem to increase repo activities using riskier collateral (equities and corporate bonds). On the other hand, non-bank dealers seem to increase, relative to not-affected bank dealers, repo activities, in particular using safer collateral (US treasuries and Agency MBSs).
The authors claim that this is evidence on the unintended consequences of the LR: decrease in repo activity, risk-shifting by more affected banks towards riskier collateral, and part of the market moving towards the non-bank financial sector. All of this should be worrisome. But there are concerns about the approach. They do not use time (for instance, week or month) fixed effects even though they are not looking at the pure time dimension of the treatment. Moreover, why is it riskier that the safest part of the repo market moves to the non-bank sector as opposed to having banks with a leverage ratio below 3.5% playing a key role in this market?
I think it is reasonable to conclude that more LR-constrained banks reduced repo activities. But I have not seen even relatively clear evidence that the LR caused the reduction in repo activities, and I do not buy the “concerns” that the authors express. I understand that institutions have agendas, but it would be good for everyone to start looking at the evidence without their conclusions already made up.