• Related Articles

• 

  FX sell-side perspectives on high frequency trading

Gearing up to meet the challenges of High Frequency FX trading

By Erik Lehtis

Having been in the FX market in one role or another since around 1980, I've seen the market expand (California-based bank trading desks and investment bank participation in the interbank market in the 80's, non-bank participation via prime brokerage in the 90's) and contract (closing those Calif.-based branches in the late 80's, bank merger mania, Y2K and Euro fear in the late 90's), volatility highs (during the 80's) and lows (the mid-late 90's), cycles of coordinated central bank intervention to knock the dollar down (the Plaza Accord in 1985) and to prop it back up (the Louvre Accord of 1987), and the impact of technology on how business is done, starting with Reuters Dealing (1981), EBS and Reuters matching in the early 90's,. However, nothing in my opinion has had the transformative impact on the FX marketplace the way that high frequency trading has, and in particular HFT participation by non-bank proprietary trading firms.

Liquidity transfer

It truly was the growth of the non-bank high frequency community that gave the FX market a much-needed shot-in-the-arm starting around 2003. Volatility in the market had become moribund beginning in the late 90's, as fear of Y2K systems failures, the advent of the Euro, and the massive bank consolidations all worked to shrink opportunity for traders. The old risk:reward equation no longer seemed to hold sway, as the market lost any ability to follow through directionally, and slippage due to poor surface liquidity became hazardous to the bank traders who had to make a living making markets on large transactions. The high frequency community arrived and brought the ability to transfer liquidity from areas of over-supply to those areas of under-supply. The market came to life, dynamically transformed.

Beginning around 2000 with the launch of the Currenex, FXall and Hotspot platforms and the ensuing exposure by EBS of their platform to API access in 2003 and Reuters Matching in 2005, automated trading in FX has changed the way liquidity is provided and by whom. Prime brokerage has been an essential element all along, due to the bilateral counterparty settlement model that underlies the very foundation of the interbank FX market. All the aforementioned ECNs operate within that context-there is no central clearing counterparty-oriented platform in spot FX. Without banks serving as prime brokers and giving clients access to their balance sheets and the ability to trade in their names, non-bank participants would not be able to have direct access to the FX marketplace, and would instead have to get their liquidity from a salesperson on a bank FX desk.

HFT and FX trading venues

A few notes about the various trading venues and the market segments they serve: EBS and Reuters are at the top of the heap from a volume standpoint. They form the backbone of the interbank liquidity pool, having supplanted the voice broking network and the direct-interbank dealing conventions that dominated liquidity until the mid-90's. Due to their long presence in the market, they offer less functionality for screen-based traders, which are almost exclusively bank traders, and they have the most restrictions on access for API traders.

They nevertheless are the most important pools of multi-participant liquidity, and they serve as the de facto primary sources of price discovery. The manner in which liquidity is divided between them is also unique: EBS has the market cornered for liquidity in the non-Commonwealth majors (EUR/USD, USD/JPY, USD/CHF, EUR/JPY, and EUR/CHF), while Reuters remains the primary liquidity for GBP/USD, EUR/GBP, AUD/USD, USD/CAD, and NZD/USD, as well as the other Asian pairs, USD/MXN and the other Latin American pairs, and the Eastern European and Scandinavian pairs. Most of these situations owe themselves to long-standing convention and the unspoken determination of the interbank market to preserve both platforms, so that neither can develop a monopoly.

FX Venues Competitive Landscape (As of end of 2009) Source: ECNs, Aite Group

The other major platforms, including Currenex, Hotspot, FXall, Integral, 360T, FXCMpro, and Bloomberg Tradebook, are to one degree or another geared toward the needs of the buy-side community, which includes corporate treasury, hedge funds, fund managers, and retail aggregators. (FXCMpro is a white-labeling of the Currenex platform). Many HFT firms will want to extend the reach of their market-making operations to these platforms, interacting with a wider spectrum of buy-side customers. But beware: HFT, both bank-side and non-bank, can often run into each other on these platforms, and therein lies a commonality between FX and
other asset classes--trades between two HFT firms usually end up with one of them being unhappy.

Lessons from other asset classes

In fact, many of the challenges faced by a firm seeking to enter the high frequency FX arena are familiar to those with a HFT background in other asset classes. Setting aside any assumptions about this familiarity, here are the specific challenges that will need to be met in order to participate effectively in this very competitive marketplace.

First and foremost, there needs to be a substantial holding of intellectual capital. This is essential-without a clear and pertinent idea about how you intend to trade profitably in FX, you simply will not. Algorithms that have demonstrated a track record of consistent success in other asset classes are not necessarily sufficient in this regard-they must be cognizant of and attuned to the unique characteristics of spot and forward FX, and currency futures.

Therefore, you must have a solid working knowledge of:

• the settlement process
• the prime brokerage model
• the Net Open Position calculation used by PBs and the ability to perform that calculation in real time
• how daily currency cash flows are calculated and managed

among other FX-specific financial details.

Your algorithms will take into account these facts, and will also take note of the FX trading landscape: each venue is unique in how it formats rates, its minimum trade sizes and trade size increments. Each exchange also has its own message policy: they vary widely and can be unexpectedly restrictive to the newcomer from equities. Failure to take heed of the exchange-imposed throttles on your behavior will result in undesirable trading consequences and in some cases steep fines. Most exchanges provide FIX connectivity for order management, but each implementation has its unique characteristics. Request For Stream, order ladders, two-sided quotes, and other complex order types are all implemented to one degree or another by the various exchanges.

Market data

Each exchange has its own unique distribution of market data as well. They vary from exchange to exchange but all have one thing in common: to the equity trader, they are paltry. They range in cost from free to exorbitantly expensive, and will test the patience of the experienced quant with their sporadic nature and limited scope (not one exchange publishes every trade, even in aggregate). Modeling the liquidity of the market by building a consolidated order book is not difficult, as the inter-exchange discrepancies in price publishing formats are easily normalized, but the fact that most prices are pulsed at regular intervals, rather than streamed, makes for some uncertainty as market snapshots age-the fact you haven't received an update does not mean the true state of the order book hasn't changed, and actual trades tend to be published (if at all) by the exchange well in arrears of the order book snapshot reflecting the aftermath of those trades. There is some data cleansing that needs to take place as well, because much of what is published can be redundant.

High Frequency Trading in FX Source: ECNs, Interviews with bank and high frequency trading firms, Aite Group

On top of all that, there are the single-bank e-platforms. You can trade directly via a GUI interface with over a dozen banks, and many will now also permit connectivity via FIX API. These banks are becoming more interested in, and technologically capable of, handling the flows from high frequency traders. One must be careful, though, in that this is a relationship business, and no one likes his or her liquidity being aggregated. A HFT firm that attempts to exploit bank liquidity will soon find that it has lost access to that liquidity.

Development environments

In order to properly and confidently develop to this unique distributed execution environment, the HFT firm's algorithms will need to demonstrate they can interoperate correctly. This means establishing a simulated environment that mimics the production environment soon to come. Among other things, a realistic Sim environment will account for expected inter-locational latencies as well as the specific trading characteristics and requirements of each exchange.

Trading engine logic and performance are not the only elements of the trading system that need to pass certain benchmark tests at this stage: the firm's risk management framework should at this stage also demonstrate its ability to keep up with trading activity across the global deployment. Real-time positions, P&L, and other risk metrics should prove their accuracy now. If your strategies are latency-dependent, you will want to have advanced network monitors tracking network performance in real-time, with particular focus on the connections between your feed handlers and gateways, and the relevant exchanges.

The software development environment is of course one of the most critical aspects of the enterprise, and in almost every case will constitute more of the firm's intellectual capital than any other single aspect. Without attempting a full-blown discussion of the various characteristics of that environment, it may be worth noting here a couple of things that are common to all successful HFT firms, particularly in the FX space. First, each firm will have a tightly integrated feedback loop between the technical and business portions of the trading team. Full and free communication between these functions in an atmosphere of mutual respect is sometimes taken for granted but cannot be over-emphasized for the success of this particular endeavor.

A dysfunctional development process will produce a nonfunctional trading system. The stress of developing a high-performing trading system in an ultra-competitive environment should not be overlooked, and when the participants are individuals who may have spent their careers developing their non-interpersonal skills, it is normal for difficulties to arise. Failure to manage these conflicts can be fatal to an incipient trading operation.

Second, the firm needs to establish a rigorous, formal process for software development. It has been said
that trading firms are really technology firms, and there are no successful technology firms that don't have a well-defined and subscribed-to production culture. Your developers must adhere to professional software production standards of some kind if work is to be performed with any degree of consistency in quality or timeliness.

Deployment issues

Providing the entrant has achieved a solid understanding of the unique post-trade settlement, back office, and exchange-specific characteristics of the FX market and has baked all this into some kind of multi-layered trading system cake, physical deployment issues can be addressed. The question of co-location is not an "if" but a "where". Most of the exchanges you will want to connect to have matching engines located in the greater NY-NJ area, and offer proximity location connectivity at Equinix in Secaucus, so that is a very logical and widely chosen option. CME's Globex platform is in Chicago, naturally, and Reuters has their matching engine in London. Additionally, EBS has matching engines in London and Tokyo, and for the ambitious global player, co-location at the Equinix sites in those centers as well will be mandatory, due to the advantages accrued executing as close to the matching engine as possible.

Therefore, high-speed circuits connecting all of these centers for the purpose of transporting market data and order traffic is essential (and costly).

Having secured rack space, powered up and cooled down their servers, leased all the appropriate circuitry, and tested all the connections, the firm can now begin the task of testing interoperability. This is where the trading engine logic is harmonized to the deployment framework, and things are demonstrated to work in their production locations as expected based on outcomes in the Sim environment. Once trading engines, feed handlers and gateways have all been installed at the various sites, a comprehensive suite of unit tests should be conducted. This suite will encompass simple functionality, stress testing and various forms of imaginable extreme market conditions, and a range of test trades that should deliver the full spectrum of expected trading opportunities.

Once the tests all pass with green lights, the system is go. From that point forward, it is all about analyzing actual trades, figuring out what went wrong, as well as what went right, and isolating issues that need to be focused on. Then it's just a matter of scaling up and out. You are now well on your way to discovering the world of high frequency trading in foreign exchange.

Erik Lehtis is the President of DynamicFX Consulting (http://dfxconsulting.com), a practice that draws
on his expertise to assist banks and exchanges operating in the high frequency space in foreign exchange.