The Future Of Actuarial Risk Modeling Is Smarter, Not Just Faster, Computing
Reported by Forbes1 source ↓
New York Life CIO Sanchet Pachpute describes using distributed computing to cut one portfolio valuation from about 50 days to two hours, and says AI could further target computing resources while requiring validation and regulatory documentation.
Key points
- Pachpute says a multibillion-dollar portfolio valuation fell from about 50 days to two hours after being distributed across machines
- He says the faster process could support daily valuations instead of quarterly runs
- Pachpute proposes using machine learning to prioritize representative scenarios and approximate costly calculations
- He says AI-based approximations need testing against full calculations and documentation for regulatory review
Details
In a Forbes Technology Council article, New York Life CIO Sanchet Pachpute describes reworking a stochastic valuation workload to run across a distributed cluster. He says the same calculation and accuracy requirements were maintained, and that a full valuation of a multibillion-dollar portfolio went from roughly 50 days of sequential processing to about two hours.
Pachpute says the shorter runtime can make daily valuations possible where computational limits previously meant quarterly runs, and allow more scenarios for tail-risk analysis. He proposes using machine learning to identify redundant scenarios and trained models to approximate expensive calculations, while reserving full computation for validation and cases requiring precision. The article does not report a deployed AI system or measured results from these proposed techniques.
Written from coverage as of
Images
- Forbes