Data representativeness (pool ↔ bank)
About this app
Centrally developed method at pool level, representativeness evidence per pool bank across several dimensions.
The idea
When an IRB rating system is developed centrally on pooled data from several institutions, each participating bank must show that the pool data are representative of its own portfolio. According to the app, the EBA guidelines (EBA/GL/2017/16, paragraphs 31–40 and Annex II) require this at the level of obligors, risk and default data; pool-level representativeness alone is not enough. A common measure of distribution shifts is the Population Stability Index, PSI = Σ (hᵢ − pᵢ) · ln(hᵢ / pᵢ), with shares hᵢ in the bank and pᵢ in the pool per class.
What the app shows
A scorecard with five fictional pool banks, such as a private bank or a specialist lender, and four dimensions: size classes, sector mix, regions and risk profile, each with seven classes. Every cell shows the bank’s PSI against the weighted pool, coloured by the common rule of thumb: below 0.10 representative, 0.10 to 0.25 monitor, 0.25 and above action required. A click opens the pool-versus-bank distribution with KS distance, the largest deviation and a recommendation.
Try it yourself
Click through the banks and find each one’s weakest dimension. The bar chart shows which class drives the PSI, for example a region or sector that is much more prominent in the bank than in the pool. Recommendations range from periodic monitoring through stratification or an additional MoC to portfolio-specific recalibration. The thresholds are rules of thumb, not supervisory limits, and the data are made up.
Frequently asked questions
- What does data representativeness mean in IRB?
- Representativeness means that the data on which a risk model was developed match the portfolio it is applied to. If sectors, size classes or regions differ strongly, the estimated parameters may be biased for that portfolio. With pooled models, each institution has to demonstrate this for its own portfolio.
- What is the Population Stability Index (PSI)?
- The PSI measures how much two distributions over the same classes differ: PSI = Σ (hᵢ − pᵢ) · ln(hᵢ / pᵢ). It is zero for identical distributions and grows with the deviation. As a common rule of thumb, values below 0.10 are unremarkable, 0.10 to 0.25 warrant monitoring, and 0.25 or more indicate a clear shift.
- What is the KS distance?
- The Kolmogorov–Smirnov distance is the largest absolute difference between two cumulative distribution functions. It shows where two distributions are furthest apart and complements the PSI as a second measure.
- What can I do in the representativeness app?
- You select a pool bank and a dimension in the PSI scorecard and see the bank and pool distributions side by side, with PSI, KS distance, largest deviation and a recommendation. All banks and figures are fictional.
Subject: IRBA / banking supervision
More from IRBA / banking supervision
- CreditMetrics — credit portfolio model
- Calibration (EBA/GL/2017/16)
- Margin of Conservatism (EBA/GL/2017/16)
A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps