Probability that an issuer fails to meet its obligations over a given horizon, estimated either from historical frequencies by rating class or from market prices.
Probability of default is the first term of the credit loss triptych, which multiplies probability of default by loss given default and by exposure at default. Two families of estimation coexist and never return the same figure. The historical, or real-world, probability is read from observed default frequencies by rating class over several decades: it describes what happened. The implied, or risk-neutral, probability is derived from market prices, bond spreads or credit default swap premiums: it additionally embeds the premium investors demand to carry that risk, and it is therefore systematically higher, often by a factor of two to five on good ratings. Confusing the two is a frequent and costly error: you provision on the historical probability and you value on the implied one. Three further precautions: the definition of default varies across sources, the probability is highly sensitive to the economic cycle, and it says nothing about the amount lost, which only loss given default supplies.
On the agencies' long-term statistics, the cumulative one-year default rate of an A-rated issuer is on the order of a few hundredths of a percent, while the observed spread on those same issuers implies a markedly higher risk-neutral probability. The gap is not a measurement error: it is the credit risk premium, and it is exactly what the fundamental spread seeks to isolate to stop an insurer from booking it as earned income.
PD, probability of default, taux de défaut, défaut à un an