Estimating Default Probabilities from Market Prices
Credit Ratings and Default Probabilities
built a hazard rate \lambda out of a rating agency's historical
default table — a slow-moving, backward-looking average across many economic cycles. A trading
desk pricing a bond today wants something else: what does the market, right now, think
the default probability is? The answer is sitting in plain sight, in a number every bond
screen already shows — the credit spread — and this lesson shows how to
extract a hazard rate from it directly, without an agency in sight.
The credit spread
A corporate bond yields more than a government bond of the same maturity — investors demand
compensation for the chance the corporate issuer doesn't pay. That extra yield is the
credit spread:
s = y_{\text{corporate}} - y_{\text{risk-free}}.
A spread is quoted daily, for every liquid bond, updating in real time as the market's
assessment of the issuer changes — a merger rumor, a weak earnings call, a sector-wide scare
can move it within minutes. That immediacy is exactly what a historical rating table can't
offer, and it's why desks lean on spreads for anything that needs to be priced today.
From spread to hazard rate
Why should a spread relate to a default probability at all? Think of the bondholder's expected
loss over one year. With hazard rate \lambda, the (approximate)
probability of defaulting within a short period is \lambda per year,
and on default the holder recovers only a fraction R of face value —
the recovery rate — losing 1-R, the
loss given default (LGD). For the bond to be fairly priced, the extra yield
the investor earns for bearing that risk must roughly compensate for the expected loss it
creates per year:
s \approx \lambda (1 - R) \qquad \Longrightarrow \qquad \lambda \approx \frac{s}{1-R}.
- Credit spread over the risk-free curve: s \approx \lambda(1-R).
- Implied (risk-neutral) hazard rate: \lambda \approx s / (1-R).
- This is an approximation — it drops discounting and accrual effects that the next
lesson's exact CDS pricing restores, but it is close enough that practitioners use it as a
quick, standard back-of-envelope conversion.
Worked example
A five-year corporate bond trades at a yield of 5.20\%, while the
matching five-year government bond yields 3.20\%. Assume a market
standard recovery rate of R = 40\% for senior unsecured debt.
Step 1 — the spread.
s = 5.20\% - 3.20\% = 2.00\%.
Step 2 — solve for the implied hazard rate.
\lambda \approx \frac{s}{1-R} = \frac{0.0200}{1 - 0.40} = \frac{0.0200}{0.60} \approx 0.0333 \; (3.33\%\text{ per year}).
Compare this with a BBB-rated bond's historical hazard rate from the previous lesson —
roughly 0.36\% per year. The market-implied number here is nearly
ten times larger. Either this particular issuer is genuinely far riskier than a
typical BBB credit, or — as the vignette below explains — risk-neutral and real-world default
probabilities are simply not the same number, even for an identical borrower.
Recovery rate drives the answer — see it move
The assumed recovery rate R is not observed directly at the moment
you price the bond — it's an assumption, usually anchored to historical averages for the
issuer's seniority and industry (senior secured debt recovers more than subordinated debt, for
instance). Because \lambda is divided by 1-R,
a higher assumed recovery rate increases the implied hazard rate for the same observed
spread — the market has to be pricing in a bigger chance of default to justify the same spread
if each default is assumed to be less costly.
The hazard rate you just computed (3.33\%) is a
risk-neutral probability — it's whatever number makes today's bond price
consistent with discounting expected cash flows at the risk-free rate. The
real-world (physical) probability — the one a rating agency's historical
table estimates — is typically smaller. The gap is a risk premium: investors
don't just want to be compensated for expected losses, they demand extra compensation for
bearing a risk that tends to show up exactly when they can least afford it (defaults cluster
in recessions, when everything else is going wrong too). That premium inflates the
spread-implied hazard rate well above the "true" historical odds of default — which is also
exactly why credit spreads are a rich, tradeable source of risk premium, not just a mechanical
readout of default risk.
-
Don't use the raw spread as a default probability. A common shortcut error
is to read a 2\% spread as "a 2% chance of default." It isn't —
you must divide by (1-R) first. Skipping that step understates
the implied default probability by roughly the recovery rate's worth.
-
The recovery rate is an assumption, not a market-quoted fact. Small changes
in the assumed R move the implied \lambda
meaningfully — always state which recovery assumption a quoted hazard rate depends on.
-
Risk-neutral ≠ real-world. A market-implied hazard rate answers "what does
today's price imply," which is the right number for pricing another instrument on
the same credit — but it's the wrong number if what you actually want is an honest forecast
of how often this issuer will really default.