High–Frequency Trading in a Risk Conscious Environment
publication date: Dec 19, 2007
|
author/source: Dr John Bates and Chris Martins (December 2007/January 2008)
A common concern about algorithmic trading is the increased exposure to risk, derived from delegating trading to an automated system. While this concern has substance, it assumes that such risks cannot be managed. Actually they can - often with the same technology that is used for the algorithmic trading. The technology that monitors market conditions and identifies patterns that warrant trading actions can also be used to monitor portfolios and continually appraise value-at risk to ensure breaches of risk thresholds are identified immediately. Corrective actions can be instantly taken, such as trading to take a position back to a more risk-neutral status.
With the credit crisis looming as an ever-gathering storm, there’s been no shortage of commentary on likely cause and effect. And there’s been no small amount of attention to the issue of risk management in a market environment in which market volatility has called into question existing risk processes. Given the size of the challenges ahead and the inherent complexity of the financial instruments involved, there is certainly no easy answer. In coming months we will no doubt see considerable analysis regarding what has happened.
What has already begun to emerge, however, has been much closer attention to the role of quantitative trading strategies. Strategies that exploit quantitatively-driven models have seemed – perhaps with a clarity that exists only in 20/20 hindsight – uniquely exposed to new risk, with several high profile implosions serving as unfortunate cases in point. The results have prompted some to question whether quantitative trading is fundamentally unsound in the current market environment. If market behavior defies expectations it will likewise confound trading models whose design has incorporated those expectations. More importantly (given that one cannot predict when normalcy will return, or perhaps even what “normalcy” is), the more fundamental question has arisen as to whether quantitative strategies are appropriate at all. If you don’t know when the proverbial “black swan” might make an appearance, can you safely entrust trading and risk management systems to automated strategies that effectively assume that all swans are white?
Does Quantitative Mean Mechanistic?
For some context, one might look to Rick Bookstaber, an author and online commentator on risk. In his online analysis of the current market situation, “The Myth of Noncorrelation” (http://rick.bookstaber.com/2007/09/myth-of-noncorrelation.html), Bookstaber references the “…tight coupling [that] comes from the feedback between mechanistic trading, price changes and subsequent trading based on the price changes.” The reference to “mechanistic” trading is noteworthy, since it might easily be considered as synonymous with quantitative trading. A key question to evaluating the ongoing role of quantitative trading is to what degree are such strategies inherently mechanistic? How much have we handed over to the machines? To what extent can we delegate to the machines, but yet still manage their operation such that risks can be mitigated?
Perhaps quantitative strategies need not be inherently “mechanistic” in behavior. The degree to which a trading system’s behavior is predetermined – and thus at greater risk in volatile conditions – may be more a matter of the algorithmic tools with which the strategy is built, as well as the tools available to monitor the strategies once deployed within production environments. With appropriate flexibility of tools, one has the power to make adjustments to strategies to accommodate changes in market conditions that were not foreseen as part of the initial development of the strategy.
The challenge is to be able to sense the changing market environment and do so quickly enough to make adjustments in a timely manner. That requires technological agility – a virtue that often is lacking in trading systems that have been developed with traditional languages and supporting tools. While those systems may deliver tightly crafted trade execution they can be ill-suited to the need to make rapid changes that effectively address unexpected market conditions.
With the right set of tools one can incorporate “checks and balances” within the trading strategies that can incorporate key risk calculations that are integral to trade execution. Such in-line risk evaluations can be conducted “pre-trade” or, for those wary of latency impact, “post-trade”, taking steps to better ensure that trade execution results are properly balanced with sound risk management practices. In either instance, pre- or post-trade, algorithmic risk management can implement prospective “circuit breakers” that can either pre-empt or unwind trades when market conditions or the particular trade specifics exceed established thresholds.
The premise of incorporating circuit breakers is not new to financial markets, given that numerous exchanges throughout the world have employed them since the late 1980’s to moderate the potential price impact from spikes in volatility. But exchange circuit breakers are fairly straightforward, triggered by declines in price as monitored by the exchange. Risk management circuit breakers within particular trading strategies must incorporate complex calculations that are specific to the portfolio and the particulars of the trading situation.
Algorithmic Risk Management with Complex Event Processing
For firms that want to continue to pursue quantitative strategies, while incorporating the right levels of risk management, adoption of complex event processing (CEP) may well prove to be the answer. CEP provides a technology foundation that enables high frequency trading solutions to monitor and react to market conditions in real time. With CEP, instead of waiting for market data to be neatly marshaled within analytic systems that permit sophisticated risk-based calculations, that market data can be streamed into the system for real-time calculations, allowing positions to be evaluated and changed as part of intraday adjustments. Such real-time risk can be delivered within standalone risk management systems or incorporated within the trading systems directly.
The core principles of CEP can be summarized as “monitor, analyze and act”. Also characterized as “sense and respond” or “event stream processing” systems, CEP technology has already begun to made significant inroads within algorithmic trading systems. As expressed within the algorithmic context, CEP allows organizations to monitor market events (delivered via fast-moving market data streams), detect sophisticated patterns (as represented by event attributes and their temporal relationships), and to take action.
Much of the industry buzz around the use of CEP within algorithmic trading has concentrated on execution speeds, recognizing that extremely low latency is needed to capitalize on momentary market opportunities. But as CEP technology has matured, the need for flexibility in the tools that create CEP-driven solutions has begun to draw greater emphasis. That flexibility will prove key to the adoption of CEP for real-time risk management. Traditional “black box” algorithms, with their opaque execution, do not offer the transparency needed to meet risk management guidelines. Flexibility is required to deliver the continuous calibration that allows an organization to monitor market positions and make the necessary adjustments, allowing a firm to pursue high frequency trading opportunities while retaining sound risk management practices.
Likewise, the need to understand the execution behavior of a risk system, once in place, puts an emphasis upon transparency. Organizations will need to audit the performance of these systems and probe into execution specifics to understand what positions - and what parameters within those positions – did or did not trigger actions. Systems that fail to support such introspection will be ill-suited to risk-focused applications.
Thus, the flexibility to make rapid changes and the transparency to inspect and analyze results will become defining components of the next generation of CEP platforms, with transaction speeds and low latency execution become commoditized features that all platforms offer, with little differentiation amongst them. Real differentiation will be found in the features of the platforms and the ease/speed with which those features can be implemented. With the right complement of CEP technology features and the right logic trading/risk business logic, organizations will be able to implement quantitative trading and quantitative risk within well integrated systems.
Summary
The market interest in quantitative trading is not likely to abate, but volatile market conditions and related factors suggest that mechanistic systems that do not support real-time calibration will lack the necessary flexibility to operate effectively in such conditions. CEP offers the technology foundation with which an organization can continuously track positions and implement real-time responses to changing market evaluations. That will enable quantitative trading to remain a viable option by introducing the important risk management component. But CEP platforms must offer the flexibility of development to rapidly adjust to changing market conditions and the execution transparency that allows organizations to monitor and analyze what is happening. It will be those capabilities – rather than the “feeds and speeds” of performance and low latency – that will determine CEP’s effectiveness in supporting real-time risk management.
