“Counterparty credit risk is particularly difficult” to model due to its “bilateral nature” and the fact it often covers more than one year, said Rajan Singenellore, Global Head of the Default Risk and Valuation Group at Bloomberg. He was the third of three presenters at a GARP webinar on counterparty risk held on May 20, 2014.
Singenellore divided the challenges to modelling counterparty risk into three categories. The first, the counterparty’s probability of default (PD), depends on multiple factors and requires estimates of recovery. The second category is how to estimate the future value of securities, which depends on the future states of netting and collateral. The third category of challenges is concerned with scenarios. “Stress scenarios are especially important in the portfolio context,” he said.
He distinguished between two major types of modelling default risk of counterparties. Structural models, such as the Merton model, tie together concepts including leverage and volatility, and how these could lead to increased default risk.
On the other hand, reduced form models “do not assume such a structure” linking those aspects. Variables have a random connection. Although reduced form models are often “better able to fit term spreads,” Singenellore said he prefers structural models because “there is economic intuition about what drivers define default risk.”
There are several things one must watch out for in modelling defaults. The modeller must “transform accounting to economic reality,” he said. For example, does the counterparty firm lease its stores or own them? This will affect the financial statements, since leases are off-balance-sheet, and are “a form of economic leverage.”
He advised modellers that “it is critical to calibrate models to real data.” As well, “stability of coefficients must always be monitored.” The future will not be the same as the past. “Correlations converge under stress conditions, as we found to our chagrin.”
Singenellore referred briefly to the Bloomberg structural model, DRSK, which estimates model-implied credit default swap (CDS) rates. DRSK provides “instantaneous evaluation of risk in certain scenarios.”
A second part of the counterparty risk puzzle is estimating recovery. “Bloomberg has done a lot of research into what drives recovery,” he said, citing degree of leverage, health of sector, and “position of instrument in question within the capital structure.”
Singenellore drew parallels between recovery and default. “Both are bounded by 0 and 1.” Also, models for either can be made forward-looking. “Models must be couched in an understanding of economic phenomena.”
There are challenges to the accurate estimation of counterparty risk. For one thing, it’s “inherently difficult to estimate future exposure,” he noted, and correlations can reduce netting benefits. He advised modellers to consider first the risk-neutral world and then add on real-life risk. “Keep in mind wrong-way risk,” he said.
When choosing scenarios, the question is always: simple or complex? “No matter what model we choose we must be mindful of the stability of parameters over time,” he said.
Singenellore closed his presentation on a sobering quotation: “The past teaches us about the future, but the future could be quite different from the past.”ª
Click here to read about the first presentation on counterparty risk. ª
Click here to view the counterparty credit risk webinar. Rajan Singenellore’s presentation begins on slide 18.
Click here to read the March 2013 presentation by Rajan Singenellore on another aspect of DRSK.