Risk Solution Network
How RSN, a subsidiary of Swiss cantonal banks and an outsourcing structure for credit-risk management, uses DetectX® Credit Risk Rating to create, validate and calibrate the rating models used by its member banks.
A rating model, and then the tooling to keep rebuilding it
The work runs in two steps. Prospero optimised the rating model first, as a service project; RSN then took the same tooling in-house so it could create, validate and calibrate models itself.
At a glance
- Sector
- Banking, credit risk
- Location
- Switzerland
- Solution
- DetectX® Credit Risk Rating
- Website
- rsnag.ch
The story, as Prospero publishes it
About Risk Solution Network
A shared data pool for more than 30 banks
- Risk Solution Network AG is a subsidiary of the Swiss cantonal banks of Basel, Lucerne and St. Gallen. Aiming to be the leading outsourcing structure in credit-risk management for small and medium-sized Swiss financial institutions, it provides more than 30 banks with a set of tools designed to measure and validate hedge credit risks: financial analysis, rating, loss given default and pricing concepts. Those tools constitute a common pool of data, which is the basis for enhancing and validating the models and for tailored services such as reporting and benchmarking. RSN also offers its customers a platform on which to exchange knowledge and experience and to represent their interests together towards external parties.
Starting position
Find the factors that predict the long term, not just the next quarter
- From financial analyses, default data and other qualitative criteria, RSN identified a set of distinguishing key indicators and qualitative factors. It was not only looking for indicators of short-term risk but for factors that lead to positive constant company development. Weighting all of those factors correctly is what an efficient rating model needs, both as a basis for credit approvals and for fair risk pricing, and that requires highly developed and stable optimisation algorithms.
Step one, the service project
Optimising the rating model under RSN constraints
- Prospero improved RSN's rating processes with the DetectX® Credit Risk Rating solution. The challenge was to find scoring processes that would improve the Gini coefficients or the ROC curve while abiding by several restrictions predetermined by RSN. The optimised scoring calculations constitute the basis of the rating models used at RSN banks to rate customers in the small and medium enterprise segment.
Step two, the tooling in-house
Three days of training, then RSN builds its own
- In the second phase there was a three-day training for RSN employees. They use DetectX® Credit Risk Rating for the creation, validation and calibration of their rating models.
The solution
Fewest attributes, most separation, everything recorded
- With the DetectX® Credit Risk Rating solution, Basel III requirements can be implemented in a secure, consistent and law-abiding way. Users benefit from the system's learning process, which comes from creating forward-looking rating models and from validating or calibrating existing ones: the system itself finds error-minimised rating models with maximised separation power. A user can simulate different risk strategies and optimise the credit business. The models created are based on the fewest possible attributes, so the reasoning behind a rating stays easy to explain, and all activities are tracked and may be recorded at any time.
What RSN runs today
Two counts and the segment they are used on. None of the three is a measurement of how well the models perform, and none carries a date.
Published result
30+ member banks
Banks relying on RSN credit-risk tooling and on the shared data pool that the models are enhanced and validated against.
Three days
The training RSN staff completed before building, validating and calibrating rating models in DetectX® Credit Risk Rating themselves.
SME ratings
The optimised scoring calculations are the basis of the rating models used at RSN banks to rate customers in the small and medium enterprise segment.
Prospero's solution has enabled us to develop an optimally separating rating function, which has proved to be very reliable not only by its application in the banks but also in the validation of the model.
No Gini coefficient, ROC value or before-and-after figure is published for this work. The source states that the coefficients improved; it does not say by how much, and nothing here fills that in.
Where to look next
The mechanism underneath the rating models, the sector page behind it, and the other banking stories.
- Profiling and scoring
- How a score is built and what moves it, which is the capability a rating model rests on.
- Banking
- Why banks choose DetectX®, and where credit risk rating sits among the modules.
- Finnova Analytical Framework
- The same analytical engine, embedded in banking software used by more than 100 banks.
Prospero also publishes this case study as a PDF. That file is not held in this site's assets, so there is no download here yet, and nothing above is taken from it.