• Related Articles

• 

  Gearing up to meet the challenges of High Frequency FX trading

FX sell-side perspectives on high frequency trading

By Irene Aldridge

Attention has surrounded High-frequency trading ever since Goldman Sachs pressed charges against their former employee for allegedly walking away with their Golden Goose - a trading system capable of producing consistent profits by trading at high speeds. A public outcry ensued accusing Goldman and other high-frequency traders of unethical behaviour that disadvantages smaller brokerages unable to afford the sophisticated know-how and high-speed computers. Various proposals to slow down or even ban high-frequency trading played out in the court of public opinion. Misconceptions, fears and public outrage poured onto the institutions engaging in high-frequency trading. Misunderstandings about the nature of high-frequency trading strategies and sell-side strategies in particular are driving much of the public response, however emotional. Most sell-side high-frequency strategies go far beyond the naïve first-to-market approach. This article attempts to fill the information void by discussing the mechanics, pros, cons and the implementation challenges of several key sell-side high-frequency strategies which can be applied to many securities, including FX and shows how all foreign exchange broker-dealers and investors, institutional and retail alike can profitably harness these strategies.

Like much of the financial services sector, high-frequency strategies, including those for FX, fall into two broad classes: the buy-side and the sell-side strategies. Buy-side strategies are driving portfolio allocation decisions and generating actual trading orders. Sell-side strategies optimize execution, securing the best possible price for each order. Sell-side high-frequency strategies have been fingered as particularly "evil," as they are thought to engage in a speed-to-market "arms race" with small institutional and retail businesses.

Successful sell-side strategies, a.k.a. algorithms or "algos," balance five key order parameters: price, time, size, costs and fill ratio to determine optimal order execution schedules. The resulting strategies continuously examine the patterns in the market data to find the optimal times to place orders; determine whether the orders should be placed as market orders, limit orders or other order types; decide whether an order should be broken down into several smaller order parcels and at what price each of the parcels should be executed; and manage the risk of non-execution, i.e. the risk of being unable to execute the order with the desired characteristics altogether.

Order characteristics as Inputs into the sell-side algorithms

First, let's look at the inputs into a sell-side algorithm in detail: price, time, size, transaction costs and fill ratio.

The price parameter specifies the benchmark execution price, the relative measure that the algorithm is designed to outperform on average. The price parameter can be a static number or a dynamic constraint.

The time parameter reflects the frequency of the buy-side strategies driving the order execution. Strategies that are rebalanced daily can be often be filled throughout the trading day; those rebalanced hourly have the correspondingly shorter execution windows of at most one hour.

The position size is an important characteristic since large trading positions can be difficult to fill within the prescribed period of time relative to a specific price benchmark. Market liquidity plays the hand that determines the order size that can be executed at the given market price: the higher the liquidity of a particular security, the bigger the trade size that the security market can absorb. Market liquidity, in turn, is provided by traders, low-frequency and high-frequency alike - the more traders present in the market, the better the chances of successful execution. To ensure that traders are able to execute their orders at market prices, in other words, to ensure high levels of liquidity, several exchanges now offer rebates to attract limit order traders, known as passive traders or liquidity providers, that guarantee existence of demand and supply.

Market liquidity can be measured using several methods. The most direct way to measure liquidity is to measure the market depth of a particular instrument: the outstanding number of passive contracts to buy and sell the security of interest at the quoted market price. The higher the number of spot contracts supplied by buyers and sellers, the higher the liquidity. Another simple indicator of liquidity is the tightness of the quoted bid-ask spread. The bid-ask spread measures the cost of the instantaneous reversal of a standard trading amount. The tighter the bid-ask spread, the easier it is to move in and out of positions at the market prices, the higher is the liquidity. Other, more complex, measures of liquidity exist that pinpoint the trader's ability to execute his trades of a given size at the going market prices.

"High-frequency trading systems are often called "mission-critical" applications, and for a good reason: they have little tolerance for error"

Transaction costs incorporate both transparent and latent charges that affect the profitability of each trade. Transparent costs are known with precision ahead of each trade and encompass broker commissions, exchange fees and taxes. Latent execution costs are not known in advance of the trade and can be only approximated through their historical realizations. Latent costs include bid-ask spreads, investment delay costs, price appreciation during execution of a large position, market impact, timing risk and opportunity cost of the trade. Bid-ask spreads compensate liquidity providers for the risk of serving as passive counterparties. Investment delay costs reflect adverse price movements during potential delays and disruptions between the portfolio allocation decisions and trade execution. The cost of price appreciation refers to the loss of value incurred while waiting to process a large position. Market impact costs measure the adverse price move due to the order execution itself. The costs associated with timing risk measure adverse price movements that are random from the trader's perspective. Last, but not least, the opportunity cost measures the cost associated with the inability to complete the order.

Fill ratio specifies the percentage of time the order is filled given other characteristics of the order, namely, price, time, size and transaction cost benchmarks. Certain algorithms guarantee a specific fill ratio. A guaranteed fill ratio of 75%, for example, implies that 75% of the requested orders are executed according to the order specifications.

Algo Implementation

Once a portfolio manager establishes the desired price, time, size, transaction costs and fill ratio characteristics of the order , the algorithms select optimal execution criteria for the order: market aggressiveness and optimal order slicing.

Market aggressiveness is measured as the percentage of volume (POV) or available liquidity that the order absorbs. It can be increased by switching to market orders, and decreased by placing limit orders at prices away from the going market price.

Depending on the objective of the trader or investment manager placing the order, market aggressiveness can be selected to minimize the costs of execution relative to a benchmark, following an algorithm known as the Strike; to maximize the probability of outperforming a specific benchmark through the Plus algorithm; or to preserve investor wealth using the Wealth algorithm.

Optimal order slicing allows investors to execute their orders without being singled out by other market participants. Such "stealth" trading has been shown to benefit investors trying to profit from their own research: market participants fail to "pick off" the large order as they cannot distinguish the resulting small order parcels from noise. The order parcels further consume little available liquidity, minimizing market impact costs.

Implementation challenges

High-frequency trading systems are often called "mission-critical" applications, and for a good reason: they have little tolerance for error: one erroneous high-paced trading functionality can wipe out significant capital just on transaction costs alone.

Building high-frequency trading strategies requires careful planning and scrupulous implementation. Like with any solid technology implementation projects, development of high-frequency trading systems follows a recurrent five-stage process:

• Planning;
• Analysis;
• Design;
• Implementation; and
• Maintenance.

The five phases are shown in Figure 1. When a version of the system is complete, new issues demand advanced modifications and enhancements that lead to a new development cycle.

Typical implementation process of run-time high-frequency trading systems.

In the planning phase of the process, the management determines the high-level goals, functionality and operational requirements of the system, including budgets and timelines. A limited-scale proof of concept study is often performed at this stage to ascertain viability of the project. The outputs of the planning phase include concrete goals and targets for the project, established schedules, and estimated budgets for the entire system.

Requirements for system functionality, the scope of the project and initial feedback from users and management are compiled and signed off upon in the analysis phase.

Next, in the design phase, project teams map out detailed specifications of functionality, including process diagrams, business rules, screenshots, other output formats such as those of daily reports, and other documents. An objective of the design stage is to separate the whole project into discrete components subsequently assigned to teams of software developers; the discrete components will have well specified interfaces that would lock in seamlessly with other components designed by different teams of software developers.

Once the design of the system is completed, the project finally moves into implementation. The individual modules are then tested and integrated into a whole.

The performance of high-frequency systems are particularly sensitive to delays in transaction speed and limitations in computer processing power. Both issues have simple and inexpensive solutions discussed below.

High-frequency trading systems make instantaneous decisions to enter and exit positions based on contemporary market conditions and rely on fast execution of transactions. According to Jim Wang of Stevens Institute of Technology, current speed of delivering an order to the exchange averages just 115 milliseconds. Buy-side customers can further speed up execution of their orders by engaging co-location services: the services of placing the computer servers in close proximity to exchanges in order to shave off the physical distance the order travels to the exchange. Contrary to the belief of some, co-location services are inexpensive and simple to create for even retail customers.

Another issue that impacts performance of high-frequency trading systems is the processing capacity of the servers running the trading programs. High-frequency systems typically analyze every tick across multiple securities and, as a result, consume significant processing power. The consumption of processing power often comes at the expense of other applications running on the same machine. One of these other applications often happens to be the internal clock of the computer - high-frequency trading systems cause delays in the clock speed, distorting time stamps on incoming data and negatively affecting their own performance. A most straightforward solution is to acquire advanced processing power. Luckily for all, computer equipment costs now are very reasonable: an advanced PC with several gigabytes of memory and fast processor can be acquired for less than $1,000.

Many software development companies offer off-the-shelf components of high-frequency trading. The components range from implementation of quote processing to canned trading decisions through order transmission and acknowledgement. While such systems are undoubtedly helpful to firms and investors with little experience in systems implementation, they may also present an unwitting Trojan horse: the unknown and unforeseen issues in the code implementation in the third party software can create considerable delays and distortions in high-frequency trading, resulting in serious losses. Care should be exercised in testing third-party software prior to its deployment on live capital.

Overall, high-frequency trading is a sophisticated automation of functions traditionally performed by human employees. Other industries, such as car manufacturing, have seen similar automation of processes take place in the 1970s and 1980s. When given competent instructions, computer technology is capable of analyzing information infinitely faster than the most advanced human brain.

Irene Aldridge is Managing Partner, Able Alpha Trading, LTD. Her book, "High-Frequency Trading: A Practical Guide to Algorithmic Strategies and Trading Systems" is available for pre-order on Amazon.com and wherever books are sold.