OSTTRA Targets FX Timing Mismatches Behind Margin Disputes

OSTTRA, the global post-trade network, has expanded its triResolve portfolio reconciliation service with a new prediction model designed to identify one of the most persistent causes of margin disputes in the FX market: counterparties valuing the same trade using market rates captured at different times.

The scale of the underlying issue was highlighted in a May 2025 Financial Markets Standards Board (FMSB) review of uncleared margin for OTC derivatives. Its survey found that an average 45% of margin calls in June 2024 were disputed, either fully or partially. FX timing was particularly problematic: 85% of respondents rated it as an area of high or medium friction, while insufficient agreement over FX snap times was the largest individual category among respondents’ top three dispute drivers.

The problem stems from the sensitivity of instruments such as FX forwards and non-deliverable forwards (NDFs) to movements in spot rates. Even when counterparties agree on the trade itself, capturing the underlying FX rate 15 minutes apart can produce materially different valuations, particularly during volatile markets.

For large institutions, investigating the resulting discrepancies can consume significant operational resources, according to Carl Thornberg, Head of Optimisation and Analytics Technology at OSTTRA.

“For some of the larger banks, they can employ between 50 and 100 people whose job every day is to explain why there are valuation differences for trades,” he tells TradingTech Insight. “It’s manual work, where they typically talk with counterparties and try to understand why there is a difference. Is it a true booking error or is it a phantom difference? The vast majority of the disputes these teams are dealing with are FX timing differences.”

Inferring rather than standardising snap times

One apparent solution would be for counterparties to agree when they take their FX snapshots. In practice, however, standardisation has proved difficult. Firms operate across different locations and end-of-day cycles and may want to retain control over the inputs used in their own valuation processes.

OSTTRA’s approach doesn’t require counterparties to adopt a common snap time. Instead, it attempts to determine the time each is already using.

For trades that have been on the books for a reasonable period, the model compares a history of MTM valuations with the FX rate at different points during each trading day. Because firms’ valuation systems tend to take their snapshots at a consistent time, the model can look for the point at which the historical pattern of FX rates most closely matches the valuations submitted by each counterparty.

To explain how it works, Thornberg draws on a useful analogy. “It’s a bit like tuning an old-school radio where you’re trying to find the frequency of your favourite channel. You hear noise and can’t make any sense of it, and then suddenly you hit the right frequency and the music comes through clearly. We’re not listening for music, but we see the pattern fall into place.”

Once the likely snapshot times for both sides have been established, OSTTRA can adjust for the timing effect and determine whether the apparent valuation break remains.

“When we have those two in place, we suddenly understand what’s going on,” says Thornberg. “We can move these trades to lie on top of each other and neutralise the effect of the timing difference. What you then see are the real valuation differences. It’s about getting rid of this noise so that what’s left are those potentially dangerous errors that the banks actually want to find.”

Using the wider portfolio

Younger trades are more difficult because the model has less historical evidence on which to establish a consistent valuation pattern. Rather than waiting for that history to accumulate, OSTTRA has developed a way of extending the analysis to these newer positions.

“With younger trades that might only be one or two days old, we need to use the whole portfolio,” says Thornberg. “We mix in the older trades where we have information with the whole portfolio, and there is a mix of statistics, machine learning – or AI, if you like – to figure out what to do with those younger trades. That is something only we can do, rather than the banks themselves, because they only see their trades with the counterparty. They don’t have the full portfolio.”

An important aspect of the proposition is the network effect. Both counterparties need to be OSTTRA clients using its reconciliation service for the capability to operate, although the prediction model itself isn’t a separate opt-in product or additional service. Instead, its output is incorporated into existing triResolve workflows, augmenting the platform’s existing mechanisms for flagging significant valuation differences.

OSTTRA says triResolve has more than 2,500 subscribers. The new model analyses data continuously across the global FX trading week, operating 24 hours a day, five days a week.

From exception detection to root-cause analysis

The model was tested over four weeks with nine major firms before being made available to all clients. OSTTRA says it has observed successful snap-time predictions for more than 95% of applicable trades at a number of firms and is working to improve prediction rates further.

The development also reflects a wider shift in post-trade automation. Much of the uncleared-margin process is already highly automated, but the FMSB survey found that exception handling still accounts for half of firms’ resource allocation in this area. Its recommendations specifically called on industry service providers to improve analytical tools to enable faster and deeper analysis of the root causes of discrepancies.

Thornberg notes that OSTTRA has been examining the FX timing problem for some time. Increasingly sophisticated statistical, machine-learning and AI capabilities have now enabled it to turn that analysis into a feature embedded within the existing reconciliation process. With the launch of the FX Snap Time Prediction Model, removing timing-related discrepancies from the investigation queue should allow firms to direct more of their resources towards the smaller population of breaks that may point to genuine booking, valuation or risk issues.