Intraday risk: moving from end-of-day VaR to continuous exposure monitoring

The Basel Committee's revised market risk framework (FRTB) replaced value-at-risk at 99% with expected shortfall at 97.5% for internal models, to better capture tail risk ^[1]. BCBS 239 sets principles for risk data aggregation and reporting, including that banks can produce accurate risk data quickly in times of stress ^[2]. Neither is satisfied by a report that arrives the next morning.

According to the SEC, Archegos Capital grew from about $10 billion of exposure in March 2020 to $160 billion at its peak in March 2021, mostly through total return swaps with several banks, and misled counterparties about its concentration ^[3]. When the positions fell, its default caused billions in losses. Each bank saw only its own slice; the concentration was visible only by combining them, and moved faster than daily reporting.

• Streaming positions and market data into one exposure view by counterparty, issuer and factor.
• Incremental risk calculation, so a new trade updates sensitivities without a full revaluation.
• Margin and collateral alongside exposure, using standard models such as ISDA SIMM for uncleared derivatives ^[4].
• Alerts on concentration and limit usage as they build, not after.

Machine learning helps with anomaly detection on flows and exposures, generating plausible stress scenarios and explaining in plain language why a desk's risk moved. Every such model is a model under the Federal Reserve's SR 11-7 guidance, which requires validation, documentation and ongoing monitoring ^[5].

Found this useful?

Share the key finding on X.