There is a widely quoted paper re-use and re-hypothecation by Egemen Eren — Intermediary Funding Liquidity and Rehypothecation as Determinants of Repo Haircuts and Interest Rates (Institute of Global Finance, 2015) — which provides another example of the danger for researchers of disregarding the difference between the re-use of collateral sold in repo (re-selling) and the re-hypothecation of collateral pledged in margin lending (re-pledging). I have gone on about this confusion in several previous blogs (including a Bank of England paper that excited the financial press and one by Issa and Jarnecic).

The consequence of the confusion on this occasion is that the author misinterprets the post-crisis reduction in the recycling of collateral by dealers as an increase in repo haircuts, whereas much of this change would really seem to represent the withdrawal by hedge funds of rights of re-hypothecation on collateral pledged to prime brokers to cover margin lending and derivatives exposures. The confusion is not academic, inasmuch as the paper appears to add weight to a body of literature that tries to identify the pro-cyclicality of haircuts as the source of the Great Financial Crisis. This academic corpus originated with Gorton and Metrick (2010), who extrapolated from a time series of haircuts on exotic collateral from a single unidentified source in the US to make an ambitious inference about the origin of the entire crisis. It is fuelling regulatory interest in extending the imposition of minimum haircuts across the repo market. The consequences of such an initiative for market liquidity could be serious and should not be based on research that is not firmly grounded.

The model

Eren follows Infante (2014) in modelling the intermediation by a dealer between hedge funds supplying collateral and cash lenders supplying collateral but seeks to extend the analysis to show that the demand for funding by the dealer simultaneously determines both repo haircuts and repo rates. Implicitly, given his data, he is also trying to estimate the rates of re-hypothecation and the cost of collateral pledged in margin lending from prime brokers as well as the haircuts on collateral posted against derivatives exposures.

The proposition is that the dealer in the model obtains funding from the spread between the haircuts paid on repos to cash investors and the haircuts received on reverse repos from hedge funds. This ‘haircut spread’ is seen to provide an alternative source of funding to the dealer’s his own cash holdings or fire sales of illiquid assets (all sales of illiquid assets are assumed to be fire sales). As haircuts expose hedge funds to the risk of dealer default, it is argued that an incentive is required to persuade them to agree to repo (and to give rights of re-hypothecation). The incentive proposed is lower repo interest rates.

The model assumes that the haircut spread is determined by the volume of lending available from cash investors and the dealer’s need for funding, while repo rates are determined by the adequacy of the terms offered by the dealer to hedge funds and their outside options (ie alternative sources of funding, including from other dealers). The conclusions are that haircut spreads are narrow when funding is abundant and wide when funding is scarce.

The adequacy of the terms offered by the dealer to hedge funds is modelled by a comparison between:

• the returns from the investments being funded by the hedge funds plus the value of bonds pledged by the hedge funds but not re-hypothecated by the dealer and instead lodged for safekeeping with a custodian less the costs of repo funding; and
• the cost of the outside options available to hedge funds.

On this basis, if hedge funds have a net liability to the dealer at maturity, for a given haircut, repo rates will depend only on hedge fund returns and outside options (what they can afford to pay given their investment return). On the other hand, if the dealer has a net liability to the hedge funds at maturity, repo rates will also depend on the probability of the dealer’s default.

The author makes bold claims for his model, stating that it can provide an explanation for all of the empirical evidence about haircuts and re-hypothecation before, during and after 2008.

Empirical evidence

Evidence for the model is adduced in the form of the survey of haircuts in June 2007 and June 2009 conducted in 2010 by the BIS Committee on the Global Financial System (CGFS). Dealers are assumed to the ‘prime’ parties in the survey and hedge funds are assumed to be the ‘unrated’ parties. The implied haircut spreads across all securities increased between June 2007 and in June 2009, which seen as being in line with the predictions of the model.

In addition, it is noted that, in the second quarter of 2008, the fair value of collateral that Lehman Brothers was permitted to ‘repledge’ and the fair value of collateral that it did ‘repledge’ were around a half their values in the first quarter and the second quarter of 2008. Moreover, it is implied from financial reporting of collateral received and ‘repledged’ that, in the third quarter of 2008, that Goldman Sachs needed a haircut spread of 7.2% and Morgan Stanley needed a haircut spread of 7%, while in the fourth quarter, Goldman Sachs apparently needed a haircut spread of 11.2% for Goldman Sachs and Morgan 22%, which is taken to support the model prediction of a sharper reduction in lending to hedge funds by riskier Morgan Stanley than by safer Goldman Sachs.

Is the model and evidence credible?

The problem with the conclusions of this paper is that the haircut spread on repos specified by the author and implied from the investment bank data on re-pledging actually includes the withdrawal of rights of re-hypothecation on margin lending and haircuts on derivatives collateral. In practice, margin lending and derivatives are far more important for hedge funds that repo, which tends to be used for transactions with dealers other than prime brokers. But the withdrawal of re-hypothecation rights by hedge funds is ignored in the analysis.

There are other weaknesses in the paper:

• The idea of a reduction in repo rate compensating for the exposure of hedge funds to the increasing haircut spread imposed by increasingly risky dealers is based on the assumption of risk neutrality. In reality, as noted by the CGFS and others, parties responded to the crisis, not so much by raising prices and haircuts as by rationing credit. Academics tend not to realise that counterparty credit risk is the primary risk faced by dealers given that collateral does not provide a perfect hedge.
• On haircuts, the paper offers only ordinal proof (direction of change) rather than cardinal proof (degree of change).
• The model does not ‘pin down repo haircuts and interest rates jointly’. There is no empirical proof of interest rate behaviour. And the repo rate estimated by Eren is not the market rate but an internal hurdle rate for hedge funds.
• Looking at the “volume of lending by cash investors and dealers’ balance sheets as determinants of haircut spreads” is too narrow a perspective. Collateral price volatility is a key determinant. Collateral price volatility and cash supply/demand are arguably both driven by risk perception and aversion. So a linear chain of causation is unrealistically simple.

However, the hypothesis underlying the paper is not without any foundation. US dealers do appear to impose haircut spreads. So the real problem may be that the paper is too US-centric. This is certainly one conclusion suggested by the work of Issa et al (2016). On the basis of far more compelling data, these authors contradict both Eren and Infante by showing that haircut spreads in the Australian repo market actually narrowed during the crisis. In repo markets such as Europe, there is anecdotal evidence of occasional trade-offs between repo rates and haircuts but only in particular circumstances and under normal conditions. There is no evidence of a widespread or strong relationship between haircut spreads and repo rates. The two are fundamentally different. Haircuts primarily reflect collateral quality and thus price volatility, whereas repo rates primarily reflect counterparty credit risk. Both will be affected by supply/demand for funding and competition but in different degrees, given that haircuts are less transparent and more under dealer control, while the repo rate is largely exogenous.

The paper’s conclusion is that, “The financial system is still vulnerable to liquidity dry-ups and wild increases in haircuts. This paper highlights the need for further theoretical and empirical research that will be of interest to both academics and policy makers to develop tools to mitigate risks in the financial system.”

Unfortunately, the assertion is not proved, which ironically makes the recommendation for more research all the more apposite but only if that future research is based on a better understanding of the legal differences between re-sale of repo collateral and re-hypothecation of pledged collateral.