Svante Hedin, global head of FX automated trading strategies, JP Morgan
"Having to keep up with the agility of ultra-high frequency traders is a real barrier to entry and a very costly exercise,"
Efforts to slow interbank currency trading down could expose players that rely purely on speed advantages gained by superior technology and simple arbitrage strategies. After years of single-minded focus on speed, the new question institutional currency traders will have to ask is this: will my strategy still make money if my speed advantage is taken away?
The theory is being tested on a recently launched FX trading platform, ParFX which employs speed bumps for trading to level the playing field among market participants. The 20-80 millisecond speed delays are randomly applied to orders arriving to the platform's matching engine instead of the usual first-in-first-matched rule applied on other institutional platforms.
The move comes after years of growing discontent among large banks that control the majority of flows in currency markets. In the past few years, the face of the interdealer market has changed considerably due to some non-bank players engaging in aggressive latency arbitrage strategies, pushing up the cost of investment and resources required to keep up with the increasingly faster market.
Nimble high frequency trading firms often rely on their superior technology and connections to trading platforms on both the execution and the data side to profit from other traders' bets - a trading profile called latency arbitrage.
ParFX portrays its approach as one that helps the little guy as well as the big guy.
"The victims of the status quo are not a narrow cluster of banks and wealthy institutions; it impacts regular people - those invested in pension funds," said Campbell Adams, a London-based consultant who initiated the ParFX platform under the working title 'Pure FX' some three years ago. Subsequently, Adams and Tradition secured the support of 11 founder banks.
"There is a need to fundamentally reassess the way the market is going and develop a way of trading that is more socially responsible and socially valid," Adams added.
Adams said internal studies showed that ParFX's randomised speed bumps and free data updates has already proven to be beneficial for participating banks, while other platforms are still operating on the first in first matched basis.
One such system is ICAP-owned EBS. The platform has been fighting discontent from core bank customers in the past two years. Gil Mendelzis, the chief executive of EBS, has revamped the way currencies are quoted on the system, only allowing quotes in the fifth decimal to be made in increments of five. That fifth decimal had made it easier for high frequency traders to make latency arbitrage trades.
Now EBS says it is in consultation with its clients about imposing speed bumps like ParFX, albeit on a much smaller scale. EBS suggested its latency floor could be a range between 3 milliseconds to 4 milliseconds. A decision is yet to be made on the exact delay. It is also unclear if EBS plans to batch incoming orders together and release them after the imposed delay, rather than applying randomised pauses to each order. EBS declined to comment.
"What's been happening in recent years is a blurring of the lines between dealer-to-dealer platforms and dealer-to-customer platforms. After EBS allowed buy-side customers such as hedge funds onto its platform banks have been torn between the desire to see those flows and their hatred of being picked off," said Fred Ponzo, managing partner at Greyspark, the financial consultancy.
"I think the proposal is an effort to try to give enough ground to the banks to keep them happy and to alleviate their main gripe, which is that customers are trading on the same platform as them," Ponzo added.
The predatory HFT play
Currency markets experience mini flash crashes every day due to the asymmetrical distribution of market-relevant information.
Ultra-fast traders have a very simple strategy: they flush out banks' positions by sending tiny orders into the market and cancelling them before the trade gets done. By doing this signalling, they gain information about where large orders are and exploit it by moving markets away from those levels and forcing other traders to execute their orders at artificial levels.
Foreign exchange traders call these events gapping. They occur when high frequency traders layer their books with bids and offers in small amounts to give an impression of liquidity in the market.
However, as soon as other players try to trade against those prices, the liquidity is pulled in within microseconds, preventing traders from executing their orders at the price they thought they were getting. Once the order is completed, the market returns to the real price or to where it had been before.
HFT firms in effect load their books up to gather market information while watching multiple market data feeds and reacting before traders can hit their prices to jump on tiny moves then cashing in. Decimal pricing was a crucial factor in gapping, as orders could be sliced at more price points within a pip.
But to do this, predatory HFTs need to be able to access data and send orders faster than anyone else. A study by Brian F. Mannix from George Washington University's Regulatory Studies Center says that racing the tape is possible because the technology used for disseminating market news and processing market orders is the same, making the two processes strongly coupled.
"The net effect can be destabilizing as trading programs attempt to outrun each other in the direction of any perceived trend, or else defensively withdraw, causing liquidity to evaporate," Mannix writes.
The use of speed bumps is aimed at eliminating the information asymmetries that arise purely from a speed advantage. By imposing a randomised auction process onto the orders, platforms allow a larger proportion of participants to absorb market data faster than the order execution process, in theory making price discovery more efficient because information is now more symmetrical.
"Because the efficient market hypothesis emphasizes the speed with which information is incorporated into prices, many people tend to confuse speed with economic efficiency, thinking that faster must always be better. This is nonsense, of course," writes Mannix.
"Real-world markets can always be made to operate a little faster, for a cost; but they can never be instantaneous. As the speed of trading approaches instantaneity, the cost will approach infinity."
Global FX volumes grew throughout the past
The Bank for International Settlement's tri-annual report on
global FX volumes, based on a snapshot ever three years in
April, shows volumes kept climbing through 2010, although the
pace slowed. The next report is due in September of this year.
But EBS has seen volumes taper off…
ICAP-owned EBS has seen declining volumes on its platform and
lost its dominant position as the system with the most-traded
flows in October 2011, spurring a raft of rule changes by new
CEO Gil Mandelzis. Data source: EBS
And FXall is catching up...
FXall still lags Reuters and EBS but it has shown the strongest
growth and is now narrowly behind EBS. Chart:
In currency markets, the costs imposed by predatory high frequency behaviour is not only a burden on HFT firms that spend vast amounts of resources on technology and market data to make sure they are the fastest, but also on all the other market participants who the HFT firms are attempting to outgun.
This execution slippage can be costly. A one-basis-point change in a $10 million order increases a firm's transaction cost by $1,000. Banks and large institutions trade billions of dollars a day, at which point execution quality becomes a paramount concern.
Latency matters on both accessing market data faster than others, and on the order placement side where traders can send multiple orders and cancel them as desired.
When a manual order arrives in the market, it shows up on the next EBS Live update. Now everyone sees this order and all the HFT algos want to trade against it. At this point there are maybe 10 different counterparties sending 2-3 orders each to hit the manual order and that's where the race for speed comes into play again.
HFTs send these orders using several order connections, at a huge cost. But the more connections players have, the more chance they get for getting their orders there first.
"EBS is making an effort to cut down this noise on the order placement side by randomising the orders, at which point it doesn't matter how many connections a player has," said a currency trader who asked to remain anonymous.
This person expressed hope that EBS at some point will eliminate the opportunity to send orders via multiple connections. These connections cost vast amounts of money, along with a raft of other expenses required to stay abreast in the arms race.
Counting the costs
These costs are something that banks are struggling with.
"FX dealers and providers recognise that it requires a very significant investment in technology every year for a market participant just to stay in the game. Having to keep up with the agility of ultra-high frequency traders is a real barrier to entry and a very costly exercise," said Svante Hedin, global head of FX automated trading strategies at JP Morgan.
And the costs to just stay standing have been increasing every year. Setting up a trading infrastructure in 2008 would have cost couple of hundred thousand pounds of investment. Now, trading executives say it's prohibitively expensive. This means that the number of trading firms has been reduced and only a few large players have the resources to keep up with the level of investment required. These firms spend a large chunk of their gross margins on resources.
One example of current costs of trading FX for HFTs can be seen in the resources required for getting the fastest connections to the platform. Network provider Spread Networks a few years ago updated its fibre optic cable connection between Chicago and New York, reportedly spending around $300 million to do so, though some have said the cost was even higher. One person familiar with the matter said anyone signing up to the service was required to sign a five-year contract to the tune of $15 million over that period. Spread Networks declined to comment on the costs of the upgrade or the terms for its service, which are not publicly disclosed.
On top of network costs, players have to shell out large amounts on a yearly basis to buy colocation services from providers such as Equinix. Colocation is paramount for enhancing latency advantages but also to defend against them.
Further to these, there is also the cost of market data, running at about $50,000 per month on EBS via a service called EBS Live, which sends updates on incoming orders and prices every 100 milliseconds. Without the data feed, trading becomes impossible.
"No one should be able to buy an unfair trading advantage. With ParFX you can't. We don't allow it," Adams said.
Due to the extortionate costs and the associated deterioration in banks' execution quality, these players have increasingly turned to internalisation. Internalisation has been a much-talked about topic in currency markets, usually in the context of banks' desire to keep information advantages arising from handling trading flows within the bank itself.
But internalisation is also expensive and adds to the overall cost of trading for clients. While large institutions have been striving to preserve their dominance, the more flow they internalise, the more capital they have to set aside for those positions.
"Internalisation has been a direct consequence of banks being unable to swap risk amongst themselves on interbank platforms for the fear of being picked off," Adams said.
To be sure, banks still have an information advantage in currency markets, according to a recent working paper from the Bank for International Settlements.
"The trend towards more market concentration observed in FX markets over recent years clearly benefits large financial institutions acting as dealers and potentially trading on this information in the inter-dealer market," the BIS paper said. "We find that order flows are highly informative about future exchange rates and provide significant economic value for the few large dealers who have access to these flows."
While banks still undoubtedly rule most currency markets, some high frequency trading firms have morphed into market makers themselves, in a step away from the pure latency arbitrage play. These players are increasingly important for the healthy working of the market.
Will the pause work? The reception to latency floor proposals is mixed.
"It's bloody stupid and it won't work," said Adrian Patten, chief executive of rival trading platform MakoFX.
The question for trading systems is where the optimum pause lies. To some, the proposals from EBS seem too little to stop latency arbitrage and allow slower players to compete, while ParFX may have overshot with its floor.
Meanwhile, high frequency traders will have to ask themselves if their strategies will still be viable if the speed advantage is taken away. Banks are sitting happy for now, but a new danger might be lurking in the form of geographical arbitrage.
"The latency floor absolutely creates geographical arbitrage opportunities," said Ponzo. He added that the apart from that possibility, the randomisation element could also lead to the loss of execution certainty and the measures won't necessarily level the playing field.
"It's more like a handicap race, where players that are too good are given extra weights to even out the arms race," Ponzo said. "I also expect the latency floor to decrease in the coming years and get back to zero. Ultimately, FX markets are moving towards an all-to-all model, more like equity markets."
It takes a signal 33 milliseconds to reach London from New York and pauses that are longer than that could open up opportunities for sophisticated traders to anticipate price movements between Chicago, New York and London.
As a result, any pause beyond 30 milliseconds could produce unintended consequences and arbitrage opportunities between geographical locations. Longer pauses also herald the danger of traders sending their orders at the very last millisecond, with significantly more information built into their orders. This problem could arise if orders are bunched together for the duration of the latency pause and then released by randomising the individual orders within the batch.
"There is materially more information available if you're talking about 80 milliseconds. The longer a batch is paused, the more room there is for exploiting the extra time. This could be done with extremely simple arbitrage strategies," a non-bank trader said.
ParFX said its randomisation applies to the pauses individual orders will have to go through, instead of batching them together. EBS has declined to give details on its plans but its aim is understood to be the creation of a pause that would level the playing field for 90% of traders using colocation services.
"A latency floor of 0.5 millisecond would be enough to level the playing field between the top 50 biggest bank and non-bank players, but the 3-4 millisecond delay will benefit a larger segment of the market," a person familiar with the process said.
For now, the most pressing problem for the FX venues that want to slow down the market is getting the level of latency pause right, both in terms of how it's applied to orders and also the randomisation element. One potential issue with randomisation is the loss of execution certainty.
"The most important consideration is getting the balance right between slowing the market down and maintaining certainty of execution," Hedin said.
The majority of observers agree that the measures will have to be tweaked in time to iron out glitches. But while the tussle for dominance rages on other platforms, one big player has been less forthcoming.
Thomson Reuters has so far refrained from suggesting a latency pause on its platforms, despite a potential integration of its two platforms with very different user profiles. Thomson Reuters bought FXall, an electronic crossing network used by asset managers and pension funds, in 2012 and said at the time that it planned to integrate the two systems. So far, the company hasn't given clear guidance on how this process will be managed.
"I expect Thomson Reuters to follow suit soon and propose something similar. Once they do that, they would be able to achieve the integration of the TR and FXall platforms, which is why they bought FXall in the first place," Ponzo said.
But even without the marriage of the two platforms, Thomson Reuters had its own problems with high frequency traders. In late 2012 it investigated and "took appropriate action" against high frequency giant Lucid Markets for using multiple data feeds for its trading activities. So far, Thomson Reuters has no known plans to impose a latency floor on its system.
"Thomson Reuters has always respected that various market participants have unique needs and as a result provides multiple venues. We regularly monitor our venue rule books to ensure that we are providing the appropriate access to liquidity to recognise changing market requirements," a spokesperson for the company said.