AI in OTC Derivatives: Why Connected Risks Matter More Than Single Metrics

By Kirat Singh, President of Risk and Alternative Assets, Clearwater Analytics

For artificial intelligence to help manage over-the-counter derivatives, it needs to explain how exposures connect, not just calculate them quickly. That distinction matters because a portfolio’s vulnerability can sit between individual risk measures: a shift in market prices may change collateral needs, liquidity demands and funding costs at the same time.

Derivatives work is well suited to computation. Pricing and risk analysis draw on data, models and calculations, and AI agents can coordinate those tasks at speed. But speed alone doesn’t make an answer useful to a risk manager. The relevant question is whether the agent can show the inputs, models and assumptions behind its result, and whether its analysis captures the ways exposures may reinforce one another.

One position can carry several exposures

An interest rate swap, for example, can respond to movements at multiple points along the yield curve. A bilateral foreign exchange portfolio may carry exposure to market direction and volatility, with potential correlation risk as well. Commodity derivatives can connect financial positions to physical exposures. Across these products, firms also need to account for counterparty, liquidity, collateral, margin and funding risks.

These categories don’t operate independently. A market move can change a position’s valuation, which may affect collateral and margin requirements. Meeting those demands can put pressure on available liquidity and funding. The relationships can shift as conditions change, so a snapshot of one exposure may not describe the strain on the wider portfolio.

That is why asking an AI agent, “What is our risk?” is only a starting point. A useful answer should identify the market data it used, the valuation models it called and the assumptions applied across products and asset classes. It should also make clear whether basis and liquidity risks were included, and how collateral and margin requirements might behave under stress. Without that trail, a result may be difficult to assess even if it arrives almost instantly.

Keep the calculations controlled

Models and assumptions have always shaped the pricing and management of swaps and other bespoke instruments. Agentic AI raises the stakes for disciplined model use: it can produce analysis quickly, but rapid output doesn’t establish that the underlying method is appropriate or consistently applied.

A practical design is to let an agent coordinate the trusted data, models and calculations a firm already uses. For a complex instrument valuation, the agent would work from an approved model, defined market data and explicit calibration parameters. Those inputs set the basis for the price. The agent can then examine the result alongside other portfolio information, while the financial mathematics remains under controlled rules.

This division gives firms room to use AI for inquiry and synthesis without asking it to invent the method used to value a position. Risk managers can question the combined findings, inspect the assumptions and focus attention on exposures that may require further analysis.

Infrastructure can make connections easier to see

As financial systems become more API-driven, positions, reference data, market data, valuation engines and risk models can be offered as defined services. An agent can connect these sources, helping a risk manager examine exposures across systems without manually switching among reports and tools.

The benefit depends on the quality and consistency of those services. If data definitions differ across products, or models apply incompatible assumptions, putting them behind a single interface won’t resolve the underlying gaps. The agent needs to make clear which sources it combined and where its analysis depends on particular model choices. That gives the user a basis for challenging the output rather than accepting a polished summary at face value.

The history of derivatives markets, including the losses associated with 2008, illustrates why headline risk measures can leave important connections out of view. Leverage, liquidity, basis, correlation and funding assumptions elsewhere in a portfolio can shape how a position behaves under pressure. A single figure may not reveal how those factors build on one another.

AI agents could help surface those links by bringing portfolio information together and making it easier to question. Their usefulness will depend on whether they can trace each conclusion to reliable inputs and controlled calculations. In derivatives, identifying separate exposures is necessary; understanding how they interact is what makes the analysis meaningful.