Quants dive into FX fixing windows debate

The methodologies used to calculate financial benchmarks have long been a subject of debate, with much of the back-and-forth centring on their representativeness and vulnerability to manipulation.

While Libor became the posterchild for benchmark dysfunction, concerns about daily foreign exchange fixings came to the fore at the start of the Covid-19 crisis, when unusual price swings were observed during the trading windows used to calculate the benchmarks.

The episode prompted Refinitiv to launch a public consultation on lengthening the calculation window for its widely used WM/R benchmark, which is determined over a five-minute period either side of 4pm in London.

Proponents of longer calculation windows argue it would make the benchmark harder to manipulate and significantly lower transaction costs for end-users. But that thesis has never been rigorously tested and proven.

Window addressing

A new study by Deutsche Bank’s head of the quantitative R&D Lab for sales and trading Roel Oomen and Imperial College London finance professor Johannes Muhle-Karbe tackles this question head on. They found the fixing window is indeed the most significant factor in determining the outcomes for clients and dealers, and that its lengthening may be beneficial to clients, but needs to be balanced against the dealers’ response. 

“The intuition behind why a client would benefit from a longer time window is that spreading out your execution over a longer period of time reduces transaction costs due to impact decay,” explains Muhle-Karbe.

Lengthening the window also reduces price predictability during the fixing period, making the benchmark harder to manipulate.

Widening the window makes it increasingly less viable for dealers to offer the execution service on the same terms

Oomen and Muhle-Karbe compared Refinitiv’s WM/R with two other less-established benchmarks: Bloomberg’s BFIX, which is also calculated over a five-minute fixing window; and Siren, with a 20-minute window.

The modelling framework used by Oomen and Muhle-Karbe assumes the prices being benchmarked follow a random walk, but that dealer hedging has both a permanent and transitory impact. Dealers are assumed to optimise their hedging strategy using a risk-adjusted measure of profit and loss, while their clients measure execution performance by the standard arrival price metric.

The pair replicate all three benchmarks and test each of them over a fixing window of one, five and 20 minutes, to assess the impact of each combination on the dealer’s P&L and on fixing price. The charts show how the outcomes tend to cluster by window width rather than methodology type. Benchmarks calculated over a one-minute window display a higher market impact and a higher P&L Sharpe ratio for dealers, whereas benchmarks calculated over 20 minutes lead to lower market impact and lower P&L Sharpe ratio.

The study therefore shows that the weighting scheme used for the benchmark calculation – that is, how individual price observations within the fixing window are weighted – is far less important than the length of the fixing window.

No short cuts

The authors also caution against drawing simple conclusions.

Proponents of benchmarks with longer fixing windows, such as Siren, argue that they lower transaction costs for end-users because the fixing tends to deviate less from what would be considered the representative rate over that period, thereby reducing market impact.

But Oomen and Muhle-Karbe warn that this argument fails to consider how dealers will respond to a longer window. “A key aspect in this experiment is that a change in the methodology will lead to a change in the observable data,” says Muhle-Karbe. “Indeed, with different incentives, the dealers will adjust their strategies, which will in turn change the benchmark prices that feed into the fixing calculation.”

Put simply, trading activity during the current WM/R fixing window cannot be reliably used to test what would happen if the window was extended to 20 minutes, as this would result in completely different market dynamics.

The study also highlights a potential problem with longer fixing windows. “There is a downside to widening the window in that it makes it increasingly less viable for dealers to offer the execution service on the same terms, as the hedging of fixing exposure is more risky over longer horizons, especially for smaller transactions,” says Oomen.

Research stream

Oomen and Muhle-Karbe’s study is a continuation of a stream of research that began in 2023, when they investigated pre-hedging – where a dealer covers its risk before a client order arrives.

“When we started to talk about this paper, we realised that people tend to conflate pre-hedging with hedging ahead of the fix,” says Oomen. The crucial difference is that in pre-hedging, the deal is anticipated but uncertain to take place, whereas a fixing transaction is committed, and the deal is certain to take place.

“We thought it was a worthwhile and interesting extension of the pre-hedging paper to explicitly analyse the case of hedging committed fixing exposures,” Oomen continues. “Compared to pre-hedging, the similarity is that a potential conflict of interest can arise in both these cases: when the dealer hedges ahead of the benchmark point or the point of execution, the associated market impact may (or may not) disadvantage the client.”

The pair’s collaboration in researching the effects of fixing methodologies is ongoing. Their next paper will examine cases where a third independent trader intervenes in the market around the fixing period.

“We look at the case where a third independent trader monitors the market around the fixing period, and we investigate what incentives and what trading strategy they may follow. Earlier studies find that such traders may either engage in predatory trading or liquidity provision, depending on market conditions. In the present context, novel questions arise about how the dealer (and client) should adjust their actions as a result”, says Muhle-Karbe.