Streetwise Professor

John Parsons and Antonio Mello at Betting the Business argue that the dominance of OTC derivatives is not indicative of their efficiency, but instead resulted from the banks underpricing credit risk:

The logic that OTC markets are superior only works on the premise that banks are competing with exchanges on a level playing field. But prior to the financial crisis of 2007-2008 (and until the effort to reform finance is successfully completed), the playing field has been not level: banks have enjoyed some significant privileges. Key among them was the failure of some banks and regulators to properly price the credit risk embedded in derivatives.

This is a very strong statement, and one which John and Antonio do not provide persuasive evidence.  Their evidence is a post at FTAlphaville by David Murphy, which details the history of credit value adjustments (CVA), and a BIS report.  David notes that although in the early days of OTC markets, banks did not charge for the credit embedded in derivatives, they eventually did so.  Parsons-Mello argue, based on the BIS report, that regulators lagged these developments, and did not impose adequate capital charges on OTC derivatives exposures, particularly on CVA exposures which were larger than actual default losses.

There is some empirical evidence that speaks to this issue.  Papers dating back to the 1990s demonstrated that interest rate swap prices vary with counterparty credit risk, although the effect was small primarily because netting and the fact that principal is not at risk in an IRS made the credit exposures relatively small.  Counterparty credit risk in CDS is much larger.  Recent work by Gandhi, Longstaff, and Arora found small pricing effects in intedealer CDS trades, that became larger in the aftermath of the 2008 Crisis.  They conclude that the small effect reflects the fact that most interdealer traders were collateralized, though by variation margin rather than independent amounts (the OTC equivalent to initial margin).  The use of collateralization to reduce credit exposure is inconsistent with the hypothesis that credit risk embedded in derivatives is underpriced.

No doubt Parsons-Mello would interpret the evidence differently than Gandhi et al.

Collateralization is one way to control credit exposure.  Credit limits are another.  Reducing credit limits of firms that suffer adverse shocks to creditworthiness would tend to reduce the impact of such shocks on pricing.  Since credit exposure can be managed on a variety of margins-pricing, collateralization, credit limits-just looking price effects is not sufficient to determine whether credit risks embedded in derivatives are overpriced or underpriced.

The Mello-Parsons argument depends crucially on inconsistent capital treatment across equivalent categories of risk.  It is also somewhat in tension with an argument that they have made before that collateralization mandates do not raise the costs of trading OTC derivatives.  As they have pointed out before (as have I), credit that is embedded in a derivatives trade can be unbundled into a default free (collateralized) derivative and an unsecured credit line.  If capital charges are set consistently across equivalent risks, regulations that reduce credit exposure in a derivatives trade (e.g., a collateral or clearing mandate) can and will be offset by a substitution of an equivalent form of credit, leading to little or no change in overall exposure, just a change in its composition.

There is no reason to believe that regulators set capital charges consistently.  I think that John and Antonio are arguing that they should.  I would suggest that this is an impossibility.  This has been a theme of mine in several posts relating to Basel rules (you can find more related posts by typing “Basel” into the search bar).  The capital charges are inevitably make crude distinctions between different sources of risk.  Banks who understand the true risks far better optimize their activities to take advantage of disparities between their more informed estimates of actual risk and the regulatory risk charges.  Moreover, they tend to do so in a way that is highly correlated across banks because they all face the distorted capital cost structure.  Maybe making derivatives-related capital charges more sensitive to counterparty credit risk will reduce the subsidy of risk arising in derivatives books and the risk exposure there will fall commensurately; but the resulting structure of capital charges are likely to subsidize another risk, leading banks to pile into that activity.

In other words, risk based capital charges are an extremely blunt tool. They are a form of regulated pricing, and like virtually all regulated price structures, lead to acute distortions.  I am skeptical that they are a reliable means of reducing systemic risk.  Indeed, by inducing correlated risk taking across financial institutions, they can actually exacerbate that risk. (Mello-Parsons do recognize the difficulties of setting the capital charges: I think the difficulties are so acute as to make it virtually impossible to get anywhere near right, and very easy to make big mistakes.)

Assessing a CVA-based capital charge also creates the very serious risk of inducing vicious cyles/feedback loops, especially during times of crisis.  CVA charges are based on CDS prices, and CDS prices vary not just with expected default rates and losses, but also embed risk premia and liquidity adjustments that depend on market frictions and market-wide conditions.  Put differently, CDS prices reflect default rates and losses in the equivalent measure, and the market chooses the measure.  During crisis periods in particular, as liquidity dries up and risk limits (e.g., VaR limits) bind, expected default losses in the equivalent measure rise relative to expected default losses in the physical measure, meaning that CDS spreads widen absolutely and relative to what the spreads implied by expected default losses.

With CVA-based capital charges, such widening in CDS spreads will lead to higher CVAs and more variability in CVAs, which depending on how the CVA charge is implemented, could lead to higher capital charges, or to actions intended to reduce capital charges.  Increases in capital charges raise the costs associated with outstanding positions and new deals, thereby leading to reductions in positions, increases in collateral postings, or both.  These will tend to exacerbate market stresses, as firms dump positions to reduce exposures (and CVA) and sell assets to raise cash to meet collateral calls, which will lead to additional price movements that can affect CVAs, and on and on.

Put another way, it binds regulatory capital very tightly to conditions in the CDS market.  The more concerned you are about the functioning of that market, especially during times of stress, the more concerned you should be about hardwiring regulatory capital requirements to the CDS market.

That is, like any market-price based mechanism for setting collateral or capital charges (think of VaR), a CVA-based mechanism is highly likely to be procyclical, and perhaps severely so.  This makes it a major potential source of systemic risk.

In sum, I do not discount the possibility that capital requirements led to an underpricing of credit risk in OTC derivatives, and hence caused these markets to be larger than they should have been, or would have been given requirements that discriminated accurately among different risks in banks’ books.  The evidence one way or the other is rather thin.  I am highly skeptical, however, of the ability of regulators to address this problem without creating others equally severe, given the information advantages that banks have and their ability to optimize their risk taking subject to the (distorted) structure of capital requirements.

Moreover, I am downright frightened of basing derivatives capital requirements on CVA, or any market-price based measure.  It is individually rational for banks to price using CVA, but tying capital charges to this measure will almost certainly lead to procyclical capital charges that are highly correlated across major financial institutions.  That is a very bad-and very scary-idea.

I therefore do not dismiss the Mello-Parsons diagnosis, though I am not completely sold.  I am particularly concerned that any “cure” could be worse than the disease.