Hazard Rate

The key condition is survival to the time being measured: if a model concerns first default, entities that have already defaulted are no longer in the surviving population, and the rate describes the exposure of those that remain. This makes hazard different from simply dividing total failures by the original population, because the denominator changes as the population survives or experiences events.

Losing sight of that condition can distort comparisons between groups at different stages. The National Institute of Standards and Technology defines the failure or hazard rate for non-repairable populations as an instantaneous conditional rate, expressed as a probability density divided by the survival function.

Its reliability explanation provides the mathematical structure, not a credit forecast for a particular borrower. In notation, f(t) is the event-time density, F(t) is the cumulative distribution and S(t) equals 1 - F(t), so the hazard is h(t) = f(t) / S(t), wherever the surviving fraction is positive, and a density and an ordinary probability are different quantities.

Time units must be clear, since a rate quoted per year cannot be compared directly with a monthly rate without conversion and a consistent model, and an instantaneous rate should not be presented as the exact probability over an entire year. The survival curve connects rates through time: the cumulative hazard integrates the hazard, and survival equals the exponential of its negative.

Consequently, a rate that varies through time cannot generally be handled by multiplying one observed value by the whole horizon. A constant-hazard model is a useful simplified case that assumes the same conditional event rate at every elapsed time.

That assumption can be unsuitable when equipment ages, borrowers' conditions change or economic stress alters risk. In credit analysis, default timing is the event of interest, and a hazard framework can describe the chance of default among borrowers that have not yet defaulted, while recovery assumptions and discounting remain separate inputs when turning default timing into an expected loss or valuation.

Observed history does not automatically establish a future hazard, because the sample, observation period and event definition matter, and borrowers exiting the data for another reason also require appropriate treatment rather than being silently counted as permanent survivors. Censoring occurs when the event time is not fully observed, such as a loan remaining current when the study ends.

That record still contains information about survival up to that point, and treating it as a completed lifetime without default can bias the analysis. A model output should include its horizon and assumptions.

A hazard estimate can support scenarios, comparisons or monitoring, but it is not a statement that a named borrower will default at a particular moment, and decisions should also consider exposure and consequences.