THE BUSINESS APPLICATION

Where and how to use it

Use it to challenge close decisions, compare alternatives and decide whether a use should be limited to scenarios. A decision that reverses under modest assumption changes deserves different treatment from one that remains stable.

EXAMPLE: A LENDING DECISION

Two growth plans appear close on expected return. The team varies recovery timing, dependence and fitted parameters to see whether the ranking remains stable before relying on it.

From evidence to a decision

HOW IT WORKSConceptual diagram
  1. 01Baseline resultRecord evidence and assumptions
  2. 02Plausible alternativesResample · refit · vary inputs
  3. 03Range of resultsLabel what the range measures
  4. 04Decision stabilityWould the action change?
An estimate becomes more useful when its limits are visible.

DATA REQUIREMENTS

What records does it need?

These are the records your team would bring together for this analysis. The exact fields and history needed depend on your lending products, the question you want to answer and the period you want to assess.

Record categoryWhat it containsWhy the detail matters
Portfolio evidenceRelevant loan histories or permitted summaries, model estimates and performance checks.Keep relationships between borrowers and periods intact.
Alternative approachesDifferent modelling methods, parameter choices and assumption ranges.Document why each alternative is plausible.
Decision comparisonResults, policy thresholds and the lending decision being considered.Assess whether uncertainty changes the action.

Past loan outcomes help assess how well an estimate reflects your borrowers. For a new decision, use only the information available at that time; later repayments help you review the result afterwards.

Understand data readiness →

WHAT YOU RECEIVE

The output

Clearly labelled ranges, sensitivity rankings and evidence about decision stability under the tested alternatives.

WHAT TO WATCH

The limitations

No single uncertainty measure captures every source of error. Resampling cannot repair missing populations, unobserved outcomes or structural shifts.

FOR RISK & ANALYTICAL SPECIALISTSHow the analysis works+

The modelling approach

Resampling, parameter sensitivity and specification comparisons provide different forms of uncertainty analysis. Their results must be labelled by what they measure; an assumption range is not automatically a statistical confidence interval.

What your risk team should review

Appropriate resampling units, reproducibility, distinction between uncertainty types and interpretation of bounds.

The right approach depends on your portfolio and available history. Review the fit to your borrowers, the reliability of the estimates and the effect of missing information before using the result in a lending decision.