Hunt for toxic flow hits one of banking’s old problems

Wider use of mark-out tools helps banks spot ‘bad’ trades – and also stirs debate on pay

  • Foreign exchange dealers have been using mark-out tools for years, allowing them to track how much spread they kept from a trade.
  • Where clients break a large order into chunks and execute with multiple dealers simultaneously, the dealers’ spread can disappear almost instantly, leaving them underwater.
  • Banks universally refer to this kind of business as “toxic flow”, but the term is associated with bad behaviour on the part of dealers as well as clients – misuse of mark-out tools contributed to a $135 million settlement by Credit Suisse last year.
  • From spot forex, the tools have been applied elsewhere – in forex options and interest rate swaps, for example.
  • This has generated debate about pay schemes for salespeople. Instead of giving out volume-linked credits, some banks have tried to base schemes on mark-out – reflecting the spread captured.
  • Others have gone further into complex, all-encompassing measures of profitability.

This article is the third in a four-part series on front-office digital change. The first two are available here and here.

To illustrate what his business has been working on for the past year, the trader holds up a graph on a single piece of paper: the y-axis reflects the spread for a single transaction, positive numbers dropping to zero and then into negative territory, while the x-axis shows time passing. On this background is a single, jagged blue line, starting high up on the extreme left-hand side – the moment of execution – and dropping down, down, down into loss.

“We got absolutely killed on this one,” he says, then turns to a colleague: “Which client was this?”

That’s precisely the point. This kind of analysis, known variously as mark-out, alpha profiling or client profiling, has been in use for years in electronic trading of spot foreign exchange. It allows banks to identify “toxic flow” – customers with a trading style that typically results in a dealer losing money – and then adjust pricing, have a stern word with the client, or delay and sometimes reject the trade.

Growing electronic volumes in other products – from credit indexes to interest rate swaps – have seen banks extending their hunt for toxic flow into new areas. Now, some are rolling it out even further, into options and voice-traded products, where the technical and practical challenges are greater.

“It’s a nearly universally applicable construct,” says Konstantin Shakhnovich, head of systematic market-making at Goldman Sachs in New York. “Extending it to other products involves less the extension of an idea, and more an extension of the habits, trading structure and approaches that are needed to systematically make use of it. The idea itself is pretty trivial.”

As use of the tool spreads, it has done more than help traders dodge bullets, also focusing attention on old-fashioned sales commissions that reward quantity over quality. At some banks, in some businesses, these schemes live on – at others, they have been scrapped or modified, but dealers that grasped the nettle have also been stung (see box, Air miles vs economic value).

The former head of e-commerce at one European dealer recalls his “18-month campaign” to scrap spot forex commission schemes based purely on volume. Salespeople for the business fought to keep the status quo.

“Eventually, we were able to show that volume credits were actually counter-correlated with P&L. In other words, the people who were getting the most volume credits were bringing in the worst business,” he says.

The common-sense explanation for that is that some buy-side firms – the finger usually points at hedge funds – will chop up large orders into pieces and then simultaneously execute with a number of dealers, a tactic known as ‘spraying the market’. When these dealers try to hedge, they may find the market is already moving against them – other counterparties of the fund got there first. Within milliseconds or seconds, any profit the bank had hoped to lock in has been wiped out.

Eventually, we were able to show that volume credits were actually counter-correlated with P&L. In other words, the people who were getting the most volume credits were bringing in the worst business
Former head of e-commerce at a European dealer

In theory, mark-out tools can help banks identify this kind of trading and, in doing so, provide evidence that not all client flow is good flow. Traders would be better able to manage dangerous clients, while salespeople would be encouraged to bring in more benign business. An options trader at one European bank believes its new mark-out tools will boost P&L for his desk by millions of euros a year. Unfortunately, applying the tools outside spot forex can be tricky.

“We would love to be able to do this on the derivatives side, but we’re finding it more problematic,” says the head of forex derivatives with one large European bank. “We’re having another shot at it now, but I’d be surprised if anyone in derivatives is doing it very well.”

The anatomy of mark-out

Mark-out tools are simple in outline. At the point of execution, they capture both the price at which the trade was done, and the level of the market – the mid-point between bid and offer. The spread between the two is the amount of profit a bank trader could theoretically capture, if it was possible to instantly execute a perfectly offsetting trade at mid.

Life is rarely that simple, of course, so choices have to be made about when to hedge and what to hedge with. And as time goes by, the spread could widen, stay the same, or narrow.

If the spread vanishes quickly, it may be bad luck – other, unconnected trading was to blame – or it may be a result of the client spraying the market, or trading in other ways that banks dislike. But if trades with a certain client display the same kind of behaviour repeatedly, then it becomes harder to blame the market.

“If you have a hedge fund in the cash forex market, hitting loads of venues simultaneously, then you might see a really sharp decay in P&L – it will just die straight away as the market moves against you,” says Allen Li, head of forex options automated trading and e-risk with HSBC in London. “Of course, you don’t know for sure that’s what’s happening. The idea is to build up patterns over long periods of time. If the decay is really sharp on a consistent basis then you can infer things about the way they are trading.”

This simple explanation hints at why the tools work well for spot foreign exchange: because it’s electronically traded, the price is automatically captured at the point of execution, and because it’s liquid, a mid-price is widely available.

When you’re doing tens of billions of dollars in volume a day via a machine, then you need to quantify it systematically
Nigel Khakoo, Nomura

The tools have been used in spot foreign exchange “for as long as I’ve been involved in that market” says HSBC’s Li. Nigel Khakoo, wholesale chief technology officer at Nomura in London, says mark-out tools gained popularity as electronic trading volumes took off in the second half of the last decade.

“Voice traders have a good sense of client profitability – good clients, bad clients, those who execute well, those who execute badly – but obviously, when you’re doing tens of billions of dollars in volume a day via a machine, then you need to quantify it systematically. I’d expect all banks to be doing this in some form as part of their electronic market-making,” says Khakoo.

More recently, banks have started trying to apply that systematic approach in other products – sniffing out toxic flow wherever it may be hiding – but the factors that made it easier to apply mark-out in spot forex are less prevalent elsewhere.

For a start, other markets have fewer trades, reducing the amount of evidence a dealer can collect. It’s not a fatal flaw, because there are other ways to be sure that a client is a problem.

 

“If you see four client trades executed with no regard to the post-trade market impact and find you’re offside in less than a second, then OK – it gives you a high degree of confidence, even if you don’t have a large number of trades. But if you’re ending up offside after 25 minutes, then you need a lot of them to draw any conclusions,” says Chris Purves, head of UBS Investment Bank’s strategic development lab in London.

As a result, banks typically apply a cut-off, after which the results are given less weight – in spot forex, it’s typically 30 seconds, says Purves. UBS uses mark-out analysis for US Treasuries and interest rate swaps as well as foreign exchange, he adds.

Another problem is that it becomes harder to find the right point of comparison for each trade. Of the thousands of forex options trades executed daily, around 30% of institutional business is traded electronically, according to Greenwich Associates, making it possible to capture the moment of execution and set the clock running. But for many instruments, the mid is a matter of educated guesswork – meaning the starting spread is also a best guess.

The age-old challenge for traders in these products is to build a surface showing market-implied volatilities at different maturities and strikes, from the handful of products where a market price exists.

“Who knows what the mid is for a six-week tenor over-the-counter trade? Well, nobody does,” says HSBC’s Li. “You may have prices for one-month at-the-money, and two-month at-the-money, for risk reversals and butterflies, but you won’t have a price for the exact thing that has traded. The interpolation methodology you use to get from that handful of prices to a full volatility surface – every bank does that slightly differently.”

Another challenge for options is that the P&L is a function not just of the spot price – it also depends on vega, gamma and theta, or the volatility of spot, its rate of change, and time decay. All of this makes construction of the tool far more challenging. So much so, that at least one bank has had repeated attempts to solve the problem.

“We’re having another crack at it now,” says the European bank’s head of forex derivatives. “We did have a go a year or so back, but possibly the framework was too simple. Now, our quants have come up with a new framework – it seems extremely complicated, which raises questions about implementation.”

Despite the challenges, he says, the project is incredibly important: “Who are the clients that really matter to us? We need to know that. So much has changed in the spot market, but in options nothing has changed in 10 years.”

Who knows what the mid is for a six-week tenor over-the-counter trade? Well, nobody does
Allen Li, HSBC

It may now be on the cusp of its own technological and cultural shift. Of five forex dealers that Risk.net spoke to for this article, three claim to be using mark-out in forex options – with two of them starting in the last 18 months – while another two say they are currently building the analytics.

The tools are used in a variety of ways, advising traders how to price, how quickly to hedge, and giving salespeople a basis on which to challenge problem clients.

Credit Suisse has a reputation for being ahead of the electronic trading game – alumni of the bank have spread across the Street in recent years, taking their ideas with them. One example is New York-based Yufang Xi, who joined Mizuho International in November last year. As head of US dollar interest rate swap trading, he aims to help the bank build its business, in part by creating his own version of tools that used to guide his trading at the Swiss bank.

One of these is mark-out analysis for interest rate swaps. Again, only part of the market is traded electronically at present – for institutional clients, around 30%, according to new analysis by Greenwich Associates – and trade volumes are a big step down from those in forex options. So far this year, daily ticket numbers for 10-year US dollar interest rate swaps executed on Tradeweb have hit a high of 368 and a low of 46, for an average of 137. For the same maturity in euro swaps, the average is 58. This limits the amount of evidence a bank accumulates, but the tools can still be used in broadly the same way.

“You measure the immediate market movement. From there, you can deduce some pattern of whether a certain client’s flow tends to have significant market impact. That gives you an idea of how you should approach this flow,” says Xi.

Individual needs

At Credit Suisse, the market impact analysis helped Xi and his fellow swaps traders work out how to handle each client – from pricing to risk management. “We had an engine crunching the data and giving us a suggestion of how to respond to an individual client. If the client’s past enquiries exhibited no noticeable market impact, then we might price aggressively to win their business; otherwise, my engine would tell me to be cautious – if you need to be aggressive to win, you also need to be aggressive in hedging,” he says.

In that second scenario, the mark-out analysis prompts a trader to show a tighter-than-usual price – but also to try and exit the trade before the spread vanishes. In other cases, the trader might opt for a wider price instead, notes a forex veteran at one UK bank, the argument being that the bank needs to reflect a higher expected cost of hedging.

The analysis could also be used to start a conversation with the client, banks say.

How do those conversations go? “Well, ‘this isn’t working for us’ is one way to start,” says a senior foreign exchange options trader at one investment bank.

 

A trader at a European bank recalls getting involved in one of these meetings. The client – a hedge fund – was told its trading style was costing the bank money and shown the evidence. The trader warned the client that it would no longer be getting good prices from the bank. Initially, nothing changed, the trader says. Then, six months later, the client came back with its tail between its legs – other banks had come to the same conclusion, and the fund said it had decided to change the way it executed.

“We’ve done a few trades with them since and there’s been no sign of bad behaviour. If every bank widens its spreads, then it’s death for a hedge fund,” the trader says.

Even though traders have a clear idea of what constitutes bad behaviour, none that spoke to Risk.net for this article was willing to estimate what proportion of their volume was toxic.

The head of foreign exchange options at one dealer says it is “pretty standard, pretty common”. Another forex dealer says problem clients are in the minority, but can be among the more active traders: “It tends to be the aggressive, short-term-focused hedge fund guys and aggressive macro guys who trade in a way that moves the market sharply. As a percentage of clients it is probably really low, but as a percentage of volume it would be a fair bit higher.”

The ethics of client profiling

Mark-out tools took off in electronic currency trading not just because timestamps and mid-prices were easier to capture, but also because there was a way to put the information to immediate use: many banks will delay execution of an order if it comes from a client thought to be toxic. If the market moves against the bank during this delay, then the trade could be rejected.

Banks justify these ‘speed bumps’ as a form of self-defence, but like many forms of self-defence, it has the capacity to be used offensively as well.

In a report accompanying a $135 million settlement with Credit Suisse in November 2017, the New York Department of Financial Services described how the bank started using speed bumps – against the better judgement of some staff – to boost its revenues. Rather than subjecting only toxic customers to the delays, and rejecting trades as a matter of self-preservation, the bank began applying the delays widely, backing out of more trades in which the market moved against Credit Suisse.

The speed bumps had initially been introduced in early 2012, the report states: “Starting first with a few clients, Credit Suisse eventually expanded its application to a group of customers that it legitimately believed justified use of the function to defend against toxic flow.”

Around a year later, with the business under pressure to grow its profits, the bank changed its policy. Internal emails captured a debate between staff, in which one argued Credit Suisse should “use our rejects to improve P&L” while another countered that “P&L pressure should never be a cause for reversing matters of principle.”

Soon after, P&L won and principles lost. The bank began rejecting a wider array of customer trades in 2013. It did so without telling clients why the trades were being rejected. Customers whose orders were not filled instead received a generic error message, prompting regular complaints.

This continued for a further year, with Credit Suisse only changing its stance once journalists and regulators started taking an interest, according to the New York Department of Financial Services. The report found the bank had “failed to tailor this mechanism to ensure it was used to defend against toxic flow, and not unfairly profit at customer expense.”

Mark-out tools can be used fairly, the report accepts – but some observers see them as an inherently bad idea.

“Banks might not like customers spraying the market, but reserving the right to drop a trade just because it is no longer immediately profitable is a little obscene compared to the robustness of their manual pricing in the olden days,” says a former proprietary trader with one European bank.

An e-commerce specialist who has left the industry has a different problem: “If there was a written policy that said ‘When we see this behaviour, we’re going to apply these kinds of measures’ then I’d feel better about it. But that’s not what happens. What happens is that someone sits there and goes ‘Oh, that guy needs a 200-millisecond delay, and that guy is 300-milliseconds, and this guy – oh, we’ll give him a pass’. It’s not applied forensically, and it’s not used fairly,” he says.

In fact, the forex market’s scandals have prompted some banks to publicly discuss their policies. Goldman Sachs, for example, issued updated dealing terms in January that describe how the bank separates customers into tiers, with each receiving a delay ranging from zero to 200 milliseconds. The tiering is based on “reviews of the counterparty’s trading history, including a review of the amount by which Goldman Sachs’ internal prices have moved for a short period of time following receipt of a counterparty’s electronic trading requests”.

This new round of disclosures only applies to currency trading, however, and some dealers remain silent on the subject. As mark-out is applied in more products, transparency appears not to be keeping pace.

Air miles vs economic value

As the use of mark-out tools spreads, they are challenging old-fashioned sales pay schemes in which a salesperson is typically given as a credit one-quarter or one-eighth of the bid/offer spread on the trade, and then paid a bonus tied to that sum at the end of the year.

“We don’t currently set sales credits on the basis of mark-outs, but do I see a future in which that happens? Yes, I do,” says a foreign exchange veteran at one UK bank.

Most traders will argue that volume-based credits undermine profitability by rewarding quantity, rather than quality. One, recalling his time at a US bank, says “sales didn’t really care if you didn’t make any money out of the flow, because they got the credits anyway. They just want the flow – get every deal done.”

That may seem a strong argument for salespeople to be paid on the basis of actual profit – known at some dealers as economic value, or EV – but there is another side to the story.

“The problem with a pure EV system is that it assumes perfection on the part of the trader – and I don’t think you can assume perfection on the part of every trader, because they are human beings. You can have situations where sales brings in the kind of business a bank wants to see, and then a trader makes a bad decision – and if a trader screws it up, then the trader screws it up. You shouldn’t penalise the salesperson,” says one former sales head.

He sketches out a situation in which a salesperson spends months “trying to get a nice trade from BlackRock” and the asset management giant eventually does the trade. What can happen next is that “the trader messes it up, gets run over, misjudges the risk, and in an EV system the salesperson ends up with a negative against his or her name. That happens an awful lot, and it’s incredibly difficult to manage salespeople in that scenario. They get despondent and disillusioned, as you can imagine.”

As a result, some banks are wary of rocking the boat – a trait that can be found right up to the head of markets at one large dealer: “If I was king for a day, I wouldn’t change too much – it’s an emotive issue at most firms. You can get lynched for changing sales credits,” he says.

Another problem is that EV can be complex. In some cases, banks have tried to take it further than a measure of how much spread the trader was able to keep, and instead capture something resembling actual profit, which can means totting up the costs of everything from balance sheet, capital and margin consumption, to the use of human resources: the time lavished on the client by salespeople, economists and traders. It could also mean attributing a tiny slice of the costs involved in running the business – for example, IT spending, clearing fees, and the support provided by corporate functions like marketing and finance.

What complicates this is the level at which these various costs can be assigned. Consumption of financial resources can be calculated for an individual trade, of course, but when that trade is added to a portfolio, it may turn out to be capital-reducing rather than capital-increasing. The consumption of human resources, meanwhile, is something that is typically tracked at the level of an individual client. And broader support costs are something banks struggle to assign even at the level of a business – attributing them to trading desks is too granular, some argue.

Inevitably, then, some banks have found a different answer. At one dealer, the forex business is measured in terms of both volume-based credits and EV, with credits used to incentivise salespeople to bring in certain types of business – a quarter or an eighth of the spread is typical – and traders being told it was their job to retain at least that much of the spread. If the EV analysis showed the trader had exceeded the sales credit minimum, it suggested two conclusions: “Obviously, you have decent traders, but also the client is giving you nice, relationship-style business,” says a former salesperson at the bank.

  • LinkedIn  
  • Save this article
  • Print this page  

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an indvidual account here: