Skip to content

New to Omni?·Get 20% off your first purchase with code WELCOME20 at checkout.See plans

FRM® loan portfolios: why do more loans leave default risk?

Adding borrowers reduces loan-specific noise, but shared economic shocks still make defaults cluster. See a worked portfolio example and the correlation trap.

ConceptOmni Curriculum Team7 min read

Rows of blank loan folders on a navy surface, with several adjacent folders displaced together.
On this page

In an FRM® loan portfolio, adding many different borrowers reduces the chance that one unusual default dominates the result. It does not remove exposure to conditions that affect borrowers together. Default dependence means the realised fraction of loans that fail can remain volatile even in a large portfolio, so an independence calculation can understate the probability of a bad credit year.

The question behind default dependence

A lender can have a low expected default rate and still face severe clustering. The average probability that an individual borrower fails is one input; the tendency for several borrowers to fail in the same economic state is another. If each loan were an independent draw, the fraction defaulting would settle near the average as the portfolio expanded. Actual exposures can share customers, employment conditions, funding markets, collateral prices or industry demand. Those shared drivers mean that a shock may weaken several borrowers at once. The issue is the distribution around the average, particularly its adverse outcomes.

GARP identifies Valuation and Risk Models as a Part I area in its FRM study materials, while its exam FAQ distinguishes the measurement tools in Part I from their application in Part II credit-risk management. The assigned reading uses a loan portfolio to connect individual default probabilities to portfolio behaviour. For an exam answer, begin by separating the probability of one borrower failing from the dependence among default events. An identical marginal probability does not imply identical portfolio risk if the joint behaviour changes.

Idiosyncratic and common drivers

A borrower may default for a reason specific to its business, such as an operational failure that does not affect its peers. Spreading a portfolio across genuinely different names reduces the influence of that type of event. Another borrower may fail when the economy contracts, rates rise, refinancing disappears or collateral values decline. Those conditions can affect many names simultaneously. A large portfolio is still exposed to the common component because adding names also adds exposure to the same economy. The BIS illustrates the distinction with systematic and loan-specific factors in its analysis of credit-risk estimation.

Do not turn “common factor” into the claim that every borrower must default together. Firms have different sensitivity to a downturn, and some may remain healthy. Default dependence means that learning one borrower failed changes what one should expect about others when that event reveals a shared adverse state. It can also reflect direct links: one firm’s unpaid bill weakens its supplier, or funding stress spreads through connected institutions. The distinction matters because a pairwise correlation is a summary of joint behaviour, not a description of one unique causal channel.

Credit loss has three separate components. Probability of default describes the chance of failure over a stated horizon. Exposure at default describes how much is owed when that happens, which may differ from today’s balance for a credit line. Loss given default describes the fraction not recovered after default. Dependence primarily changes how many borrowers may fail together, though a downturn can also affect recoveries and exposures. A calculation that multiplies a default count by an assumed loss per failure must say which assumptions are fixed and which may worsen in stress.

A small worked portfolio

Take an invented portfolio of 100 equal-sized loans. Assume each loan has a one-year default probability of 2%, and treat exposures and recoveries as identical solely to isolate default dependence. The expected number of defaults is two whether defaults are independent or positively dependent; correlation does not alter that marginal expectation. What changes is the spread of possible default counts. Under independence, the variance of the count is 100 times 0.02 times 0.98, or 1.96. The standard deviation is 1.4 defaults.

Now assume, illustratively, that every pair of default indicators has correlation of 0.05. The variance of the default count gains a covariance term for each distinct pair. There are 100 times 99 ordered pairs, and each pair contributes 0.05 times 0.02 times 0.98 to the variance. The result is 1.96 plus 9.702, or 11.662, giving a standard deviation of about 3.42 defaults. The expected count remains two, but the uncertainty around it is much wider. This is an invented teaching example, not an estimate of any bank or year.

The same result can be expressed as a default-rate standard deviation. Divide the count standard deviations by 100: 1.4 percentage points for independent defaults and about 3.42 percentage points for the correlated illustration. That comparison is about variability, not a claim that exactly three or four defaults will occur. A normal approximation is especially unreliable at the low end, because a realised default rate cannot fall below zero. A correlation parameter alone also does not completely specify an extreme tail; model form and changing conditions matter.

Why adding loans has a limit

For equal default probabilities and equal pairwise correlation, the variance of the fraction that defaults can be separated into an individual term and a common term. The individual part shrinks as the number of loans increases. The common part approaches the default-indicator variance multiplied by their correlation, so it remains after a very large number of loans is added. This is the mathematical reason the “large portfolio means no risk” answer fails. The BIS discussion similarly notes that systematic movements affect all loans while individual shocks become diversifiable in large pools.

Suppose the illustrative portfolio expands from 100 to 1,000 loans without changing either the 2% marginal default probability or 0.05 pairwise default correlation. Independent default-rate standard deviation falls from 1.4 percentage points to about 0.44 percentage point. With dependence it falls from about 3.42 to about 3.16 percentage points. More loans do reduce the loan-specific part, but nearly all remaining variation is shared. These figures are derived from the deliberately simple assumptions; they are not a forecast for a real lender.

This model needs careful interpretation. Equally sized loans with a common default probability and the same correlation for every pair are convenient for teaching, but real portfolios are heterogeneous. Concentrations in one industry or region can create stronger dependence than a broad average suggests. A downturn may also cause correlation to change rather than stay fixed. A risk manager therefore needs to know how the assumptions were calibrated and what stresses they omit. The BIS warns that estimates based on rare defaults carry material statistical noise even before model uncertainty is considered.

Joint versus conditional probability

Two borrowers each having a 2% default probability do not by themselves tell you the chance both fail. If they are independent, that chance is 0.02 multiplied by 0.02, or 0.04%. With default-indicator correlation of 0.05 in the illustrative homogeneous model, their joint probability becomes 0.04% plus 0.05 times 0.02 times 0.98, or 0.138%. Dividing that joint probability by the 2% chance of the first default gives 6.9% for the second borrower conditional on the first default. These values are internally consistent within this pairwise model, not empirical observations.

The conditional result should be read with care. One default is evidence that a common adverse state might be present, but it is not proof that the first default caused the second. The reason the conditional probability exceeds the unconditional figure is that the joint event is more likely under positive dependence. If an exam question explicitly states independence, multiply marginal probabilities. If it supplies a correlation or common-factor structure, use that information. Do not import an unstated assumption that the loans are independent merely because there are many of them.

What to write in an exam answer

First identify what is held constant: individual probability, exposure, recovery and horizon. Then ask whether the question concerns expected loss, the variance of defaults, or the adverse tail. Expected default count can stay the same while the likelihood of clustered losses rises. State which risk diversification can reduce: borrower-specific variation. State which it cannot remove: exposure to shared shocks. If the answer involves a portfolio capital buffer, explain that a buffer designed only around the mean will not describe losses in a bad common-factor state.

The most common mistake is to treat “many loans” as a substitute for “independent loans”. Another is to conflate a default probability with loss severity or to call a pairwise correlation a complete stress model. The FRM programme page gives the broader exam context, and the study methodology is a place to organise practice by these distinct reasoning steps. Keep the portfolio question narrow: what happens to the average, what happens to variation, and which assumption permits your calculation?

Sources

  1. FRM Study Materials | GARP · garp.org · Retrieved
  2. FRM Exam FAQs | GARP · garp.org · Retrieved
  3. V. The road to a more resilient banking sector | BIS · bis.org · Retrieved

Explore FRM study resources

Educational content, not investment advice. This article is written to help candidates prepare for professional exams. It is not a recommendation to buy, sell or hold any security, and it does not take your personal circumstances into account.

Omni Finance Academy is an independent exam-preparation provider and is not affiliated with or endorsed by CFA Institute, GARP, CAIA Association or CFP Board. CFA®, FRM®, CAIA® and CFP® are trademarks of their respective owners. Full disclaimer.