Pre-Crisis, there was very little academic writing on clearing. Post-Crisis, with questions about the role of derivatives in creating systemic risk, and the mandating of clearing of derivatives as a means of mitigating this problem, this is changing. This is a good thing, but unfortunately, this burgeoning academic literature is at risk of irrelevance, and worse, of being misleading, because the theoretical models of clearing are nothing like clearing as it actually is. These models tend to focus on the mutualization of risk within CCPs. That’s important, but as I’ll discuss in more detail below, mutualization is not the most important feature of CCPs. Collateralization is.
I’ll just talk about a couple of papers in detail, Adam Zawadowski’s Entangled Financial Systems in the most recent RFS, and Clearing, Counterparty Risk, and Aggregate Risk by Biais, Heider, and Hoerova.
These papers have some important insights, and I don’t want to seem overly critical. I just want to persuade scholars that the focus of these papers, and many others, is misdirected, and to suggest where they should direct their attention.
“Entangled Financial Systems” presents a model of the periodic collapse of the financial system through the channel of inter-bank derivatives exposures. The paper is rough sledding, in part because it tackles a complicated issue, and in part because the exposition and especially the proofs are hardly paragons of clarity: I have my doubts about some of the proofs. That said, the story is a plausible one. Banks use fragile capital structures (short maturity debt to fund long-lived assets) to solve an agency problem. They use derivatives to manage risk in order to protect non-pledgeable income. Banks are at risk of blowing up due to an idiosyncratic, exogenous shock: perhaps an operational risk, like a rogue trader. If a bank blows up, its counterparties don’t get paid in full on their derivatives. If these counterparties don’t insure against this risk, they may fail, and so on, with the result being a daisy chain of failures. Thus, one idiosyncratic failure can lead to the collapse of the entire system.
In the model, the original risk of a blowup is idiosyncratic and insurable. But in equilibrium, banks don’t buy insurance against a counterparty failing because they don’t internalize the impact of their failure on their other counterparties. Thus, there is a “market failure”: the system blows up periodically because it is privately efficient but socially inefficient not to buy counterparty insurance.
One crucial issue with the paper is that most things are, laudably, endogenized, but one crucial thing is not: each bank can trade with only two counterparties located adjacently on a circle. The choice of counterparties is exogenously specified. This concentration of counterparty exposure is crucial in making the system vulnerable to collapse.
Zawadowski recognizes that greater diversification of exposures across counterparties reduces the fragility of the system. Although the externality may induce insufficient diversification across counterparties (because the systemic benefits of this aren’t internalized), market participants in reality have a variety of reasons to spread their trades across many counterparties. Meaning that real financial systems may be less fragile than in the model, with its exogenously imposed concentration of exposures.
I’m also skeptical that an idiosyncratic risk at a single institution can bring down the entire financial system. Look at some of the rogue trader losses-Kerviel at SocGen, Adoboli at UBS. These guys cost their banks billions-but even such huge losses didn’t lead to a financial system meltdown via any channel, including a derivatives channel. Instead, the 2007-2008 Crisis was related to a systemic shock-a decline in US real estate prices-that hit multiple financial institutions and investors. It’s hard to identify an actual episode where the channel analyzed in the model-an idiosyncratic shock at a single financial institution-led to a financial meltdown. The idiosyncratic nature of the risk is important: that’s what makes the risk insurable. The paper therefore has little to say about non-insurable risks.
Where does clearing come in? In the paper, clearing is a form of counterparty insurance. Mandating clearing internalizes the externality.
There are several problems here. The first is that in the model, counterparty insurance is supplied by a continuum of investors who can diversify away the risk. A CCP in theory can also diversify the idiosyncratic risk by mutualizing it. But note that CCPs are voluntary cooperative arrangements among financial institutions. If there is a gain from collective action-internalizing the externality through cooperation-why don’t financial institutions voluntarily cooperate to form CCPs? (Though in the context of the model, sharing the risk among financial institutions is more costly than passing the risk to investors. Still, there is a collective benefit from cooperation among the financial institutions.) Such cooperation increases joint wealth: why don’t market participants cooperate to internalize the externality? What stands in the way of consummating such mutually beneficial bargains? Moreover, why does it happen in some markets, not in others?
But the mutualization issue generally is the bigger problem, and the root of my qualms about the developing literature on CCPs. Most of the models, including the Biais et al paper, formalize CCPs as a mutual risk sharing/insurance mechanism. (Hell, I’ve done that myself.)
But mutualization is only one of the functions of CCPs As We Know Them. Indeed, I am increasingly leaning to the view that it is the most problematic of their functions.
CCPs operate on the “defaulter pays” principle. That is, real world CCPs attempt to choose margins (collateral) and default fund contributions so that they almost always cover a defaulter’s losses, and that non-defaulters’ default fund contributions are seldom used to make good a defaulter’s losses. That is, default funds are tapped-and risk mutualized-only in rare, and arguably extreme, situations.
Put differently, only tail risk is mutualized in real world CCPs, and the primary function of CCPs is to set margins so that default losses are NOT mutualized, except in extreme circumstances. CCPs are like monoline insurance of supersenior positions well down on the default waterfall.
In the context of the _ paper, this is particularly problematic, as he shows that collateralization is an inefficient way to address the externality problem. It ties up valuable resources that could be used to fund positive NPV projects. This is just one problem with collateral: the “initial margin problem”, if you will. There’s also the variation margin problem that I’ve written about over the years.
I consider the tail-risk mutualization aspect of CCPs highly problematic because of the wrong way risk problem. Like super-senior tranches of a CDO, losses hit the default fund during systemic episodes when those exposed to the default fund (the members of the CCP) are under stress.
This all means that the academic literature, which is modeling CCPs as mutual insurers, has two big problems.
The first problem is one of positive economics. Existing models are not able to predict (a) why clearinghouses form, and (b) why they are primarily mechanisms to net and collateralize exposures, and only mutualize extreme risks. They do not predict why market participants sometimes cooperate to implement a “defaulter pays” model, and sometimes don’t-and why they have never implemented a fully mutualized insurance scheme.
We need models that help us understand why market participants sometimes cooperate to implement a defaulter pays mechanism, supplemented by mutualization of the extreme risks; why they sometimes don’t; and why they never fully mutualize. That is, we need to understand why so few risks are mutualized, even when market participants choose to form CCPs.
With all due modesty, I think the answers will will be found in my original analysis from the 90s: that the usual bugbears of insurance-adverse selection and moral hazard-make it uneconomically costly to mutualize risk, and that these problems also make centralized/delegated setting of collateral levels more costly than bilateral arrangements for doing so, depending on the characteristics of the traded instruments and those trading them.
The second problem is one of normative economics and policy prescriptions derived from models. Policy recommendations based on models of CCPs that are flatly contrary to the way CCPs really operate are highly misleading, and dangerous. Eliding from a model that says “mutualization of risk is socially efficient but privately unprofitable” to prescribing a policy of mandating CCPs is deeply flawed, when in practice CCPs won’t mutualize risk as in the models, but will instead implement a defaulter pays model. (NB: of late regulators are telling CCPs that their main source of concern is that CCPs will set margins too low. That is, regulators want to make sure that CCPs really make defaulters pay. So in practice, mandated CCPs will mutualize only extreme risks, and collateralize the rest. Nothing like what’s in the models.) It’s a sort of bait-and-switch.
Prudent normative policy recommendations need to be based on a model with good positive content. We need to understand much better why market participants eschew implementing the defaulter pay model before mandating it. For that’s what clearing mandates do: they impose the implementation of defaulter pays, NOT the implementation of the kinds of mutualization in many formal models of clearing. Given the costliness of collateral (IM-not to mention the destabilizing effects of VM), it is particularly misleading to advocate policy measures that will lead to increased collateralization based on models in which collateral plays no role whatsoever. Again-a bait-and-switch.
In sum, I’m pleased that many talented scholars are turning their attention to clearing. Especially when they cite me 😛 But it will be a shame if these scholars go on a wild goose chase, and construct models of CCPs that are completely disjoint from real world CCPs. This goes double when they make policy prescriptions based on their models.
We need to understand the costs and benefits of defaulter pays, with the mutualization of only extreme risks. For that’s what CCPs are really about. We need models that are based on an understanding of collective action issues-because non-mandated CCPs are institutions that arise from collective action. Only when we understand these issues should we have much confidence about making policy recommendations.