Tall-trees are falling: will your credit risk profile weather the storm?

The forest is getting thinner. No surprise, then, that measuring Single-Name Credit Risk is back on everyone’s agenda. The surprise is how many firms lack robust methodologies to capture and assess their exposure.

Large Counterparty Default Events. Tall Trees. Single-Name Concentration Risk. At times like these we are all reminded that even the seemingly safest of counterparties can run into financial difficulties and when they do the costs to our businesses can be substantial. What’s more, it’s clear that comprehending portfolio credit risk exposures requires more than human intuition and needs help from analytical approaches.

The basic building blocks are straightforward. We need to know: (i) how much we’re exposed to (Exposure-at-Default/EAD: often straightforward, but sometimes not if linked to other volatile market factors); (ii) what we expect to recover in the event of default (Loss-Given-Default/LGD); and how likely default is (Probability-of-Default/PD). Credit risk modellers can typically come up with a sensible answer for each of EAD, LGD and PD for each of their counterparties, and by multiplying the three together we can get our Expected Losses. All well, good and hopefully adequately provisioned. But a quick glance at the numbers for our highly-rated counterparties will show that figure to be surprisingly low and something just doesn’t feel quite right: we’ve missed the risk!

If we’re an investment firm, bank or any other risk-conscious manager with a portfolio of credit exposures we don’t just want to know the expected outcome each year, we also need to know how bad things could get if the unexpected occurs, as is happening right now. After all it could be the difference between surviving a financial crisis with mild wounds to lick, and making headlines for all the wrong reasons.

Let’s consider a simple portfolio of 5 counterparties like the one below. Each one owes us £10m, we expect to recover 50% if they default and each of them has a 1-in-100 year probability of defaulting.

It’s a pretty straightforward example so we should be able to work out the risk profile, right? For instance, over a 10 year period what is the highest annual amount we would expect to lose? If you guessed £5m then you’re half right. If each of these counterparty’s defaults were completely unconnected then the answer would be ~£5m due to this event being the default of one counterparty (there are 5 counterparties over ten years each with a 1/100 chance of default = 5 * 10 * 1/100 = 0.5 defaults, so we should expect a default) in which case £5m is lost.

However, if the default of these counterparties is even slightly correlated, that is, if they are more likely to fail at the same time (which they undoubtedly will be) we should “expect” our highest annual loss in a 10 year period to actually be zero. How can this be: surely positive correlations should increase the risk, not make it less? Well the answer is that if default events are correlated then so are non-default events and we can actually go a few more years before seeing any losses… though of course the losses can be much greater in magnitude when they do finally happen. From this we can see that even in the “simple” case we need a view of the correlations of our counterparties (how likely they are to default together) and a model to help us blend all that information into a full risk profile, that allows us to understand likely losses at different levels of confidence (or stress).

And that was just a very simple case: what happens if our portfolio looks a bit more like this:

With this risk profile, what is the highest annual amount we would “expect” to lose over a 5 year period? Interestingly it’s actually the default of counterparty D leading to a ~£29m loss based on £40m EAD with an LGD of 72%. But we can only know this by using a model that considers the full loss profile. The loss profile for this portfolio, with a few key points marked, is shown below.

At first glance, this looks a lot to take in, but once you are familiar with the technique, it becomes an indispensable tool for understanding the overall risk profile of a set of credit exposures, particularly as probabilities of defaults are constantly on the move!

So how did we do this? We use a methodology that’s been around since the 1980s and the beginning of modern quantitative portfolio credit risk management: the Vasicek-Merton framework. Under this framework defaults are assumed to happen when the value of a customer’s assets fall below the value of their liabilities (i.e. a company cannot generate sufficient value from the sale of all of its assets to cover the company’s debts as they fall due). The movements in asset value are assumed to be correlated across different customers due to shared sensitivity to underlying market related factors. This ensures that customers in the portfolio who are facing shared sensitivities to systemic factors are more likely to default in the same period, while still allowing for idiosyncratic risks affecting each customer to impact their individual level of credit risk.

We can use the framework to apply the same level of systemic stress to all the exposures in a portfolio and understand the losses that result from the correlated market risk factors (such as central bank interest rates, and government bonds).

This methodology is widely established (even underpinning the IRB Formula of Basel/CRR Credit Risk Capital measurement), and, once you have it up and running for the credit exposures in your portfolio, all you really need to focus on is the distributional profile result, as we demonstrated above. You can then use that profile to make better business decisions, and manage or mitigate any risks that are unacceptable, keeping your overall risk profile within appetite.

Interested in knowing more? Reach out to the Deloitte Capital Clarity team via our website: Capital Clarity | Deloitte UK.