On this page
The choice between KMV and a reduced-form intensity model starts with the question and the available observations. KMV links a firm's estimated asset cushion to an empirically mapped default frequency, making the firm's finances central. An intensity model treats default timing as a random event and can use market credit prices to infer a risk-neutral default rate. These outputs should not be read as interchangeable probabilities.
Why a CAIA candidate meets both approaches
CAIA® Level II asks candidates to evaluate credit models rather than merely name them. The two approaches are useful precisely because they organise the same uncertainty in different ways. A structural account starts from the issuer's asset value, obligations and volatility. Default becomes more likely as the asset cushion over a chosen default point shrinks or becomes more volatile. A reduced-form account begins with a hazard, the conditional rate of default during a small interval given survival so far. It can describe the timing implied by a bond or credit default swap price without claiming that the model has identified the economic cause of failure.
The distinction matters for a portfolio analyst reviewing a private credit exposure, a public borrower's bonds or a credit derivative. When traded equity and a usable liability picture exist, a structural view can show how weakening asset value and leverage might alter estimated credit quality. When bond or CDS prices are the primary observation, intensity is an operational way to fit and compare those prices. Neither model removes judgment about data quality, recovery, liquidity or the horizon. The CAIA program page provides the program context, and the methodology page explains how to work through such comparisons.
What KMV changes in the structural story
In the basic structural intuition, shareholders have a residual claim on firm assets after promised debt payments. Equity value and equity volatility can therefore help infer the less visible value and volatility of the firm's assets, provided the model's assumptions are accepted. KMV-style work compares an estimated asset value with a default point related to liabilities. The resulting distance to default is a standardised cushion: a larger cushion, other inputs equal, indicates more room before the threshold is reached. This is a model score, not itself a default probability.
The consequential extra step is the map from distance to an expected default frequency, or EDF. A Federal Reserve discussion of Moody's KMV describes estimating the relation between observed distances and subsequent historical default or survival in a large sample. That empirical mapping is different from simply plugging the distance into a standard normal tail. It also explains why a reader cannot calculate a proprietary EDF from a distance alone unless the mapping is supplied. A case giving a distance but no empirical curve supports a directional conclusion; it does not support a precise EDF.
For an illustrative arithmetic exercise, suppose estimated assets are 180 units, the case's default point is 100 units, and estimated asset volatility is 20% over the chosen horizon. In a simplified one-horizon cushion calculation, the distance is the 80-unit gap divided by 36 units of asset-value volatility, or about 2.22. If assets fall to 160 while the point and volatility rate remain fixed, the gap is 60 and the volatility amount is 32, so the score falls to 1.875. That says the cushion has deteriorated. It does not justify announcing that default probability rose by a particular number of percentage points; the calibration has not been provided. These values are invented teaching inputs, not market observations.
The exercise also shows a subtle denominator effect. A lower asset value shrinks both the numerator and the amount obtained by multiplying asset value by a fixed volatility percentage. Here the numerator contracts more in relative terms, so the distance falls. Candidates who subtract the default point correctly but leave the original volatility amount in the denominator are no longer evaluating the stated scenario. The case should specify the horizon and conventions; a more complete model can include drift and a log-based expression, so a simple linear score should be labelled as such.
What an intensity model says instead
An intensity or hazard rate is conditional on the borrower still being alive immediately before the interval. Under a deliberately simplified constant hazard assumption, survival to time t is the exponential of negative hazard times t. Cumulative default probability is one minus that survival figure. The model can allow hazard to vary with time or market state, but the constant case makes the difference between conditional rate and cumulative probability clear.
Imagine an illustrative annual hazard of 2%. A borrower has one-year survival of about 98.02% and a one-year cumulative default probability of about 1.98%, using the exponential relation. At three years, survival is about 94.18%, so cumulative default probability is about 5.82%. The chance of default specifically during the third year, measured from the initial date, is survival to year two less survival to year three: about 96.08% less 94.18%, or 1.90%. All figures are rounded from an invented constant-hazard assumption. The conditional chance during year three for a borrower that actually reached that year remains about 1.98%; it is a different denominator.
One reason traders use intensity models is that observed spreads provide calibration information. In a very simplified setting with constant hazard and fixed recovery, spread is approximately hazard multiplied by loss given default. Suppose an illustrative spread of 1.2% and an assumed recovery of 40%. The assumed loss fraction is 60%, so dividing 1.2% by 60% suggests a hazard of about 2%. This is an approximation for a teaching example, not a general equality or a real spread-based forecast. A spread can compensate investors for liquidity and other risks, and its pricing probabilities are risk-neutral rather than automatically the observed frequency of defaults.
That final distinction is often missed. A historical EDF from a structural model is intended as an estimate tied to actual observed defaults. A hazard inferred from market prices is embedded in the price investors demand for bearing risk under pricing assumptions. Even if the numerical outputs happen to match, the question each answers differs. To translate between them, an analyst would need assumptions about risk premia, recovery and potentially market liquidity. A direct comparison without those adjustments can lead to a false claim that one model has disproved the other.
Sensitivity analysis makes the difference more useful. In the KMV-style example, hold liabilities fixed and raise estimated asset volatility: the same asset cushion becomes less reassuring because adverse moves are more plausible within the horizon. Hold the market spread fixed in the intensity example but lower assumed recovery: a smaller hazard can explain the same price compensation because each default loses more. Those are different mechanisms. They also warn a candidate against reading a deterioration in a model output as an observed default event. Each result is conditional on its input estimates and modelling choices.
The horizon must be kept visible as well. A one-year EDF cannot be compared directly with a five-year cumulative default probability. Even within the constant-intensity exercise, the hazard is a rate conditional on survival, while the five-year cumulative figure collects default events across five years. State the horizon alongside every probability and, if a comparison is required, put the outputs on a consistent period and probability basis first. A credible comparison may still remain impossible when the empirical mapping or risk premium is unavailable.
Match the evidence to the conclusion
The Federal Reserve's discussion of KMV notes its use of equity information and historical default mapping. CFA Institute's credit analysis overview separately describes structural models as using the position of the firm's claimholders and reduced-form models as modelling when default may occur without explaining why. Those descriptions guide the exam decision. If an issuer has traded equity, usable liabilities and an available historical calibration, KMV can connect deteriorating fundamentals to a change in a default signal. If a liquid credit instrument is priced and the task is to fit or stress that price, an intensity framework may be the relevant language. If neither market is informative, neither label magically creates reliable inputs.
The common exam errors are to convert distance to EDF without a supplied map; to treat a market-implied hazard as a real-world default frequency; to call a period-specific probability the cumulative probability; and to forget the assumed recovery rate when interpreting spread. A strong answer names the model, the observed input, the horizon, the probability measure and the missing assumptions. For further study, the how it works page shows the learning flow, but the decision in a case should be made from the evidence the case actually gives.
Sources
- Curriculum and Study Tools | CAIA Association · caia.org · Retrieved
- Finance and Economics Discussion Series 2007-29 | Federal Reserve · federalreserve.gov · Retrieved
- Credit Analysis Models | CFA Institute · cfainstitute.org · Retrieved
