WM/R fix and the curse of predictability
Increased volatility and the low liquidity environment witnessed during the Covid-19 selloff has renewed attention to FX execution around the WM/R benchmark’s 4pm London fix. When adverse price moves become bigger, they are more difficult to ignore.
We previously addressed EUR/USD predictability around month-end moves and observed how levels around the 4pm fix became less adverse for month-end execution compared to morning levels. From a daily perspective, execution at 4pm was reasonably good in 2019. But everything changed this year.
In this month’s column, we look at the most recent evidence for the first half of the year, as the month-end signal became useful again. We estimate that in 2020 executing at 8am yields better results than doing so at 4pm by around 30 basis points per day on average in the last trading day of the month. We also present the results for a basket of currencies to check the consistency across instruments.
The foundations of predictability
It’s no secret that FX moves at each month-end are reasonably predictable, and that the reason for this is structural: FX order flows are derived from asset-rebalancing flows, such as equities and bonds, and these flows can be predicted with a decent degree of certainty on month-end days.
Consider a euro-based investor holding US dollar-denominated and FX risk-hedged assets. Any increase in the value of those assets on a monthly basis results in more US dollar hedging demand. With more US dollar to hedge, our investor would have to execute US dollar spot selling on top of merely rolling the existing forward hedge, thus pushing US dollar down. The situation is mirrored for a US dollar-based investor active in euro-denominated assets.
If we aggregate the FX demands of the two investors, the resulting net flow shows the direction of EUR/USD.
The question is, how do we know who holds what? Because passive capital follows well-known benchmarks, those exposures to different asset classes and hedge ratios can be estimated.
Two of the most relevant benchmarks tend to be the Morgan Stanley Capital International (MSCI) and the Bloomberg-Barclays bond index, which have the WM/R 4pm London fix as their index specification. However, the development of exchange-traded fund index trackers means other indexes are gaining in importance. For the sake of simplicity, our analysis will focus only on the S&P 500.
The rules of the game are simple: if the S&P 500 is up month-to-date at the beginning of the last business day, we forecast US dollar selling. If the opposite is true, we forecast US dollar buying. The action is to either sell or buy the US dollar against a basket of currencies.
If we consider all month-end days since 2006 and classify them into two regimes (US dollar sell and US dollar buy), the average intraday dynamics will look like figure 1.
A number of things stand out. First, the directionality is quite consistent, and if the model forecasts US dollar going up based on the equity close, then US dollar on average trends up through the entire last day of the month.
Second, there is a significant acceleration – both uptrend and downtrend – around 4pm. While the day-long trend might be driven by regular flows, but more likely by rebalancing flows executing early, the 4pm spike is clearly driven by the fix execution demand correctly predicted by the model. This means that some people leave it to the very last moment to minimise index-tracking error.
Third, the average moves are not economically insignificant. For example, the difference between selling EUR/USD at 8am and 4pm can be 20bp on average. Lastly, 4pm is on average the worst time in the day, price-wise, to execute across many currencies. The odds of this happening randomly are very low.
The next step of our analysis is illustrated in figure 2, which shows more recent intraday dynamics after the WM/R window was widened from one minute to five minutes in December 2014.
Nothing fundamental has changed. The results for AUD/USD have become more volatile but the magnitude and dynamics of the other currency pairs remained unchanged. Moreover, the same spikes of roughly the same magnitude are still observed around 4pm, in a period of supposedly improved fix benchmark. This could mean that the fix flows are simply too big to be accommodated within a five-minute window.
Economic costs of the 4pm execution
The above intraday pattern is averaged across a long period. Let’s now look at how we can track the cost of trading at 4pm over time.
Consider a simple trading strategy that could be run by a fix arbitrageur: sell US dollar at 8am against a basket of currencies at month-end if US equities are up month-to-date the day before month-end. The cumulative P&Ls of this strategy across different currency pairs are shown in figure 3.
Our analysis highlights two important aspects. First, there is a strong uptick in trading strategy P&L curves across all currencies in 2020. While we have previously speculated that this effect can and should be arbitraged away, it is not actually happening.
Second, the strategy represents a small return for a small risk, although the return to risk ratio is decent. The strategy we put in place holds a position for only eight hours a month – from 8am to 4pm on the last day. The strategy averages 2% return a year, or around 16bp per month. The roundtrip costs in those very liquid currencies are definitely a small fraction of that – 0.5bp in small size – meaning it is profitable for potential arbitrageurs.
Looking at the yearly performance shown in figure 4, it is clear that the first half of 2020 has been remarkably successful. If this trend continues, it could be the best year since 2006.
FX flows starting point
The past seven months have been very unusual on many accounts, but why the renewed predictability of FX trading? If the flow is predictable on the last trading day of the month, the next question involves establishing when the price move actually starts and when it stops. To show this, we consider a three-day period before and after each month-end and track the average EUR/USD performance in the standard ‘up’ and ‘down’ scenarios as previously defined.
As figure 5 shows, there is very little movement in EUR/USD on the day before the end of the month for both scenarios (green area). At month-end (white area), the divergence happens, and the 4pm fix becomes the worst point at which to execute. On the next day (red area) the price action is fairly muted. This pattern suggests the presence of an imbalanced flow towards 4pm and a balanced one afterwards.
The imbalanced flow at the beginning of the month-end day could be due to some passive funds executing rebalancing flows early and being willing to accept tracking error. This seems the most logical way to explain the reasonably smooth intraday trend. It also means that funds executing at 4pm are getting the worst deal and would do better by going in earlier.
The curse of predictability
Despite being quite striking, the patterns described above are actually well known. Based on publicly available research and information, it is reasonably simple to obtain a decent approximation of what the month-end FX flows would be. Passive capital following well-documented benchmarks is likely to generate predictable patterns, as anyone can have access to rebalancing rules.
In other words, they are cursed by their own predictability. Of course, this isn’t unique to FX flows. However, it is interesting to note that this predictability has not been arbitraged away – a simple arbitrage strategy is still reasonably profitable.
Redesigning the WM/R benchmark by widening the window from one to five minutes might have improved the ‘local behaviour’ – the minutes around 4pm – but the 4pm levels are still very unattractive for execution from the perspective of a daily range.
Widening the WM/R window even further is unlikely to help. Creating new benchmarks by applying elaborate weighting schemes to prices around 4pm is also unlikely to do much for the global daily pattern – as per figure 5, the trend tends to last hours.
As long as 4pm remains the worst point in the day, any benchmark would do better than the 4pm ‘point in time’. But it does not mean it is better. It is just giving more weight to prices away from 4pm and hence expresses the same pattern as documented in this article.
On the flip side, if this intraday pattern continues, the current WM/R benchmark is reasonably easy to beat. As figure 6 shows, the cost of executing at 4pm compared to alternative times is quite significant.
The ongoing 4pm predictability can be considered a problem for benchmark designers such as MSCI or Bloomberg, and by passive funds or ETFs when selecting an appropriate benchmark. According to MSCI’s website, $13 trillion of capital is benchmarked against its indexes. If all of that is rebalanced at suboptimal levels, then the total opportunity loss is quite significant.
A few things can be proposed. The creation of a set of possible benchmarks in which rebalancing does not occur at the end of the month might be the ultimate solution. However, this is costly and would break current asset management workflows, as many of the other index changes happen around the same time.
A risk budget allocation by big asset managers aimed at capturing this predictability gap between investment strategies and execution may be useful and much more practical. This would lessen the predictability and arguably move the returns from an asset manager’s ‘left pocket’ to the ‘right pocket’, similar to what we proposed previously to capture different intraday predictability patterns.