High frequency FX trading: technology, techniques and data

The concept of automated trading has attracted rapidly growing interest in recent years. In certain markets, such as exchange-traded futures, it has already become an everyday fact of life.

Over the past decade, while public attention was focused on the rate of electronic trading adoption in the equities market, enormous market opportunities cropped up in other asset classes. The emergence of foreign exchange (FX) as a legitimate asset class has resulted in rapid adoption of electronic trading in the FX market, led by the tireless efforts of various inter-dealer, multi-dealer, and single-dealer platforms to create greater market transparency and support for full the life cycle of an FX trade.
In recent years, the FX market has witnessed the emergence of a new trend in electronic trading: algorithmic trading strategies designed to capture execution opportunities in an increasingly automated and fragmented marketplace. But as the market reality for the FX market continues to evolve, it is important not only to assess the potential for growth in adoption of FX algorithmic trading, but to identify possible hurdles

We have developed a systematic trading environment for several our clients that executes approximately 5,000 trades per day. The platform is computer-based, where humans oversee the trading activity without actively intervening. We human beings check that the processes are running according to plan. In an emergency we will reactivate a process.

Our trading-model environment comprises the following services:

• The core trading model taking trading decisions.
• A tradingsupport service for trade execution (the actual trades)
• Customer reporting
• And a host of monitoring tools.

As is typical of modern technology, the success of our environment does not rely on one or two ideas, but builds on a complex system that has been well tuned at all levels of operation.

Algorithms

We focus on a broad range of risk measures. One of the most important is the Calmar Ratio, which compares annualized return with maximum draw down. The Calmar Ratio focuses on the worst-case scenario–the relationship between average return and extreme loss–and is thus more relevant than, say, the Sharpe Ratio.
Another risk metric we use is the “ExposureFactor,” a term we coined to describe the relationship between annualized return and maximum exposure. In financial markets, where uncertainty abounds, one fact is sure: at low levels of exposure, risk increases in linear increments; at higher levels of exposure risk increases exponentially, and so becomes very dangerous. It is therefore important to monitor exposure and to develop and pursue strategies that have relatively small exposure at all times.

In FX spreads are extremely low–approximately 0.01 percent for EUR-USD. If a trader has perfect foresight, he can earn–without taking on any leverage–approximately 2 percent of return every day, or approximately 500 percent during one year.
This return potential assumes that the trader can take advantage of every small price spike. If a trader cannot trade at high frequency (for example, only once a day), then the annual return potential is only 125 percent.
Other things being equal, going to HFT enhances the return potential of an investment strategy because a trader can take advantage of many more price spikes. For sophisticated investment managers with the appropriate computing power and know-how this is a great enticement.

Lower cost, higher frequency

Assume for a moment that the average profit per trade for a trading system should be at least ten times the cost of trade execution. On that basis, a fall in transaction costs from $50 to $5 cuts the minimum acceptable profit per trade by $450 - from $500 to $50. This makes it viable to deploy trading systems with a smaller profit target per trade but a higher trade frequency. Typically such systems will also be using very short timeframe prices (e.g. single ticks) as a data input and will typically be handling smaller deal sizes.
Yet sooner or later this increasing trade frequency runs up against human limitations. There comes a point when it is simply no longer physically possible for the trader to hit the keypad or click the mouse fast enough, not to mention managing the resulting positions.
This conflict has been a further factor in the growth of high frequency autotrading. Even where an automated trading environment generates fewer trades per market than a human trader can handle, it can of course replicate its actions across multiple markets and timeframes. Furthermore, it is far less restricted in the number of intermarket opportunities it can observe and act upon.
An automated system is also unaffected by the psychological swings that human traders are prey to. This is particularly relevant when trading with a mechanical model, which is typically developed on the assumption that all the trade entries flagged will actually be taken in real time trading.
This is sometimes hard for a human trader to do - and not just because they may be away from their desk when a trade signal is triggered. A mechanical trading system can experience long runs of losing trades, so a human trader contemplating placing a new order after suffering six losing trades in a row may be tempted to withhold the order. Mechanical systems often depend for their overall profitability on a relatively small number of winning trades outweighing a larger number of smaller losers, so this can be critical.
In futures markets this has prompted some technology vendors to deploy customer trading models on the broker/clearer’s servers within the exchange, rather than the trader’s workstation.
Furthermore, while the execution of the trading model may be automated, its design and coding are still performed by humans. Any errors undetected in the development stages will sooner or later emerge (probably with expensive consequences) in real time trading. Therefore it is essential to have a robust risk management infrastructure capable of terminating the activities of a rogue trading model that has run amok. Some automated trading environments already offer this infrastructure, with a broad range of controls that can be applied to the trading systems. The FX broker EBS has created a laboratory facility which allows customers to test their model trading algorithms in a secure environment using historical FX market data and live market rates as part of its Spot Ai trading offering.