Routing Validated Risk Intelligence Without Sharing Proprietary Data
This article discusses QIS (Quadratic Intelligence Synthesis), a distributed intelligence architecture that enables financial institutions to share validated risk model performance without revealing proprietary data or exposures.
Why it matters
This technology could help financial institutions better monitor and improve their risk models, leading to more resilient financial systems.
Key Points
- 1Risk models at financial institutions failed during the 2008 crisis due to siloed validation, not bad models
- 2Risk models degrade over time as market conditions change, but have no mechanism to detect their own staleness
- 3The blind spots in risk models are correlated across institutions due to reliance on similar historical data
- 4QIS outcome packets allow sharing of validated model performance without revealing underlying positions or trades
Details
The article explains that the failure during the 2008 financial crisis was not due to individual firms having bad risk models, but rather the lack of a mechanism to share validated model performance across institutions. Risk models degrade over time as market conditions change, but have no way to detect their own staleness. Additionally, the blind spots in risk models are correlated across institutions since they are calibrated on similar historical data. The QIS architecture resolves this by transmitting 'outcome packets' that carry the predicted value, actual observed outcome, a validation score, and a semantic fingerprint of the exposure profile - without revealing the underlying positions or trades. This allows validated risk intelligence to propagate across institutions without sharing proprietary data.
No comments yet
Be the first to comment