Using Loss Given Default in Bond and Credit Risk Measures
Summary
The document explains how credit desks use loss given default (LGD) to estimate jump-to-default exposure in bonds, loans, and credit default swaps. Because recovery value is hard to predict before default and can remain uncertain afterward, desks may report both a zero-recovery case and a recovery-based case, or calculate a range of default P&L outcomes across assumed recovery levels.
For a bond or loan, default P&L is the assumed recovery minus its dirty price. For a vanilla CDS, the protection buyer’s result depends on the mark-to-market, recovery, and notional; a CDS with fixed recovery instead uses that fixed assumption. The document also describes recovery sensitivity for model-priced products and implied default probabilities derived from bond prices. These measures have limits: recovery estimates can be unreliable, and recovery sensitivity is more useful for comparing bond and CDS hedges than for bonds alone. Complex instruments such as nth-to-default baskets require more involved calculations.
Key ideas
- Recovery is difficult to estimate before a credit event and may remain uncertain after default.
- Credit desks may track jump-to-default P&L under zero recovery and other assumed recovery levels.
- Bond and loan default P&L is calculated as assumed recovery less the dirty price.
- CDS default exposure depends on the contract structure, mark-to-market, recovery, and notional.
- Recovery sensitivity can help assess bond and CDS hedges, while implied default probability can be inferred from bond price and an LGD assumption.
Tags
Full text
# Using LGD (Loss Given Default) in the trading of bonds
# Using LGD (Loss Given Default) in the trading of bonds
Do you use the LGD as a risk management tool when trading credit cash bonds? Or it's more for trading the loan product.
## Answer by Dimitri Vulis (score 1)
https://quant.stackexchange.com/a/73790
One problem with using LGD is that unless the bond issuer is on the verge of default, you don't have a good estimate of what the defaulted bond would be worth. Sometimes it is not clear even for some time after a default. Some regulations may insist that you guess for Current Expected Credit Loss (CECL) and for IFRS 9 expected credit loss and impairment analysis, but if a default occurs, anyone's guess is unlikely to turn out to be even close. With that in mind:
It's pretty common for credit trading desks to report at least two "jump to default" risk measures - one using zero recovery (i.e. 100% LGD; the most conservative if you're just long some bonds) and also one using the some recovery assumption (more meaningful if you have some bonds, CDSs, and other credit products). More comprehensively, you can have a grid of the P&L impact if the credit defaults and the recovery is 90%, 80%.... 10%, 0%.
Estimating the P&L from a credit event with recovery $R$ is fast and straightforward for vanilla products. For a bond or a loan, the P&L is $R-$ the dirty price. For a vanilla credit defaut swap, the protection buyer's P&L is $-pv - R + \textrm{notional}$, where $pv$ is the mark to market. For a credit defaut swap with fixed recovery $F$, it is $-pv - F + \textrm{notional}$, does not depend on $R$.
The risk measure gets a little complicated for things like (options om) $n$th to default baskets, but few people trade those anymore.
For products marked to model using a recovery assumption, such as CDSs, it is also common to calculate a "recovery assumption 01" - sometimes with a term structure - showing the sensitivity to changes in the recovery assumption. For products with observable price, such as bonds, this risk measure is not very useful except to see how well the bonds' recovery01 hedges the CDS's recovery01, so some people calculate it for bonds for this reason.
Not a risk measure, but sometimes it is also useful to calculate the probability of default (and CDS spread) implied by a bond's price and some LGD assumption. You'd also need this calculation for a bond recovery01.Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.