Use Case: Corporate Guarantee Fraud
Globally, the majority of commercial banks’ profit come from corporate loans. The banks use collateral to counter the risk of a potentially bad loan, and it’s called: corporate guarantee. When an entity (a corporate or a person, say, A) applies for a loan from the bank, the bank requires another entity (usually another corporate, say, B) to financially guarantee that if the original entity A fails to pay the loan back, party B will pay the loan off on A’s behalf.
A commercial bank usually deals with hundreds of thousands of, if not millions of, corporates and corporate loans. The front-office staff (sales reps) issue many loans on daily basis, and the back-office staff work on identifying potential risks associated with each corporate guarantee. This process has been traditionally laborsome and time consuming.
There are a few typical types of frauds, intentionally or unintentionally, associated with a guarantee.
- Simplest form of fraud: A ßàB, A guarantees B and B guarantees A, this is in direct violation of any bank’s loan issuance prerequisite.
- Forming of chain or ring guarantee: AàBàCàDàEàA, this is slightly harder to detect, as you have to dig all the way to E, then to find that E’s guarantee for A violates the bank’s rule.
- Forming of more complicated topologies that violate bank’s rules. For instance, an entity may be involved in many loans thus forming a forest of guarantee chains.
The immediate below diagram shows that the red-dot corporate entity guarantees 4 other corporates, forming 3 loan-guarantee triangles (Note: triangle is the simplest form a “guarantee ring”) and 1 loan-guarantee quadrilateral (4 parties involved).
Guarantee Rings – 4 Triangles of Corp Loan Guarantee
List Mode of Corporate Loan Guarantee Chain Detected -3 triangles & 1 quadrilateral
The above example is a zoomed-in investigation of a particular corporate, often times, you would run through all or a large chunk of loan data in a batch processing fashion to understand how many violations are out there. This process can be very time consuming given a large number of entities involved (often times in the range of or exceeding millions). This is where real-time fraud detection boosted by real-time graph computing does the magic.