On this page
FRM® Part II retail provisioning estimates the credit losses a loan book is expected to incur, then records an allowance on the relevant accounting basis. The analyst must group exposures sensibly, compare cohorts at the same age and use a loss horizon appropriate to each exposure. A single portfolio average can hide deterioration in a small but risky segment. The core exam skill is to connect the estimate to the risk drivers and to avoid mistaking a simple expected loss calculation for the complete accounting provision.
Provisioning is a portfolio measurement problem
Retail books contain many smaller accounts with shared patterns. Rather than build a wholly bespoke forecast for every borrower, an analyst can group loans by characteristics that affect default and recovery, such as product, collateral, credit quality and origination period. Each group needs parameters and a forecast that reflect its own evidence. A mortgage and an unsecured card balance should not inherit the same recovery assumption merely because both sit in the retail book. A useful segment is homogeneous enough to measure, yet large enough to support credible estimates. Excessive splitting can leave too little data; excessive pooling can conceal important differences.
The familiar probability of default, loss given default and exposure at default components are useful building blocks. GARP's discussion of IFRS 9 probability estimates explains that lifetime expected credit loss needs a path of future default probabilities, recoveries and outstanding balances. An exam calculation may simplify that path to one period and one set of inputs. Recognise when the prompt permits that approximation. A real provision also considers timing, discounting, changing balances, multiple economic outcomes and the accounting standard. Writing a one period formula beside a lifetime allowance without explaining the simplification is a category error.
Segments show where loss originates
Suppose a hypothetical bank has two invented retail pools, both measured over the same illustrative horizon. Pool A has £100 million exposure, a 1% default probability and a 20% loss rate after recovery. Pool B has £20 million exposure, a 5% default probability and a 50% loss rate after recovery. These round inputs are assumptions for teaching, not observed bank data and not a recommended model. They deliberately make the smaller pool riskier, so that averaging the balances alone becomes visibly misleading. Assume for this short exercise that exposure remains fixed and there is no discount adjustment.
| A | £100m | 1% | 20% | £0.20m |
| B | £20m | 5% | 50% | £0.50m |
| Total | £120m | — | — | £0.70m |
Pool A contributes £0.20 million because 100 times 0.01 times 0.20 equals 0.20. Pool B contributes £0.50 million because 20 times 0.05 times 0.50 equals 0.50. The book therefore produces £0.70 million in this simplified exercise. Pool B is one sixth of exposure but contributes five sevenths of the calculated loss. An analyst who spreads £0.70 million proportionally by exposure would assign too much to A and too little to B. The calculation demonstrates why segmentation changes the answer, even before more advanced accounting adjustments are introduced.
The weighted book average also needs care. A portfolio default probability alone is not enough when recovery rates differ across pools. Multiplying a balance weighted average probability by a balance weighted average loss rate does not generally reproduce the sum of the pool products; the two risks may be related across groups. Sum each group's expected loss first. Then divide the total by exposure if a portfolio loss rate is useful for comparison. Here that illustrative rate is about 0.583% of £120 million. It is a descriptive result of the assumptions, not an observed provision rate for actual retail lending.
Vintage analysis answers a different question
A vintage groups loans by their origination period. It helps an analyst ask whether loans originated under one set of underwriting conditions perform differently from loans originated earlier. A balance segment is a cross section of current characteristics; a vintage is a history of a cohort as it seasons. Both dimensions can matter simultaneously. For example, two mortgage cohorts with similar current collateral ratios may have different borrower selection, documentation or interest rate exposure at origination. If the newer cohort deteriorates more quickly, a single current collateral band may conceal that change.
Compare cohorts at equal months since origination, not at the same calendar date without adjustment. A cohort that has been outstanding longer has had more time to accumulate defaults. For an illustrative check, imagine one cohort has cumulative losses of 0.4% at its second anniversary and a newer cohort has 0.8% at its second anniversary. The second is twice the first at the same age, but that comparison alone does not identify why. Check loan mix, underwriting, prepayments, recoveries and the economic setting. Then revise relevant assumptions where the evidence supports doing so. The exam trap is to name weaker underwriting as a fact from the curve alone.
Vintage curves can also flag model drift. If recent cohorts perform worse than forecasts after allowing for seasoning, the analyst should look for changing risk factors or a forecast that no longer reflects conditions. A management adjustment may be justified where a model misses a known effect, but it requires an explanation and should not become a plug for an uncomfortable result. Separate what has been observed from what is forecast. The allowance is forward looking; a historical curve is evidence for a forecast, not the forecast itself.
The accounting horizon changes the amount
Under IFRS 9, the IFRS Foundation's project summary distinguishes an initial twelve month expected credit loss measure from a lifetime measure when credit risk has increased significantly since initial recognition. It also explains that expected losses are updated for new information at each reporting date. A twelve month expected credit loss is not merely the cash shortfall received during the next twelve months: it concerns defaults occurring in that window and the related lifetime shortfalls. This distinction is a common source of imprecise exam answers. Use the standard's definition, not a calendar shortcut.
A significant increase in credit risk is assessed against initial recognition, not against an arbitrary absolute default probability. A loan can start risky but change little; another can start strong and deteriorate sharply. The IFRS Foundation says the assessment is holistic and may be collective when individual evidence is not yet available. The move to a lifetime horizon changes measurement even if the loan has not defaulted. Conversely, a lower reported allowance is not proof that underwriting improved: the mix of stages, exposures, recoveries and forecasts may all have changed.
Accounting regimes must be named before their requirements are compared. IFRS 9's staging framework is the context here; do not carry its stage labels into every other accounting model. Likewise, the simplified pool calculation above does not itself prove the booked allowance. For a provision, the analyst needs a reporting date, the in-scope instruments, measurement horizon, reasonable forecasts, discount convention and any specific rules for impaired assets. GARP discusses these dimensions in its IFRS 9 treatment. In a short exam answer, acknowledge the missing pieces explicitly rather than pretending an illustrative arithmetic result is a complete financial statement number.
A practical answer sequence
First identify the accounting basis and horizon stated in the question. Then define groups using risk characteristics that plausibly change default, exposure or recovery. Next calculate each group's expected loss with internally consistent inputs and add the contributions. If asked about trend, compare vintages at equal seasoning and investigate reasons for divergence. Finally, explain why the estimate could change: portfolio mix, new information, credit deterioration, recovery assumptions or stage migration. The order keeps calculation and interpretation connected. It also makes it easier to spot an assumption that the question never supplied.
One final check is to ask whether the proposed conclusion goes beyond the data. A higher allowance can reflect prudent recognition of risk; it does not automatically mean cash has been lost, that a bank has failed, or that every loan in a segment has worsened equally. A provision covers expected credit loss, while capital planning also considers losses beyond the expected amount under its own framework. Candidates can revisit the FRM programme overview for subject context and the Omni study methodology for practice structure. In the answer itself, show the segment arithmetic, the vintage comparison and the measurement horizon separately.
Sources
- IFRS 9 Financial Instruments project summary | IFRS Foundation · ifrs.org · Retrieved
- IFRS 9 and Probability of Default | GARP · garp.org · Retrieved



