Measuring risk after a trade is no longer a viable way forward-risk must be measured in real time as the trade occurs. By Bob Giffords
Given the recent market turbulence and volatility it is no surprise that the focus inside investment firms has shifted to intra-day and real-time risk management. "It's a matter of survival," says Miranda Mizen, a senior consultant at Tabb Group.
According to Mizen, there are four key drivers: increased trading automation within asset classes; cross-asset investment strategies; changes in market structure, such as fragmentation and dark pools; and increasing product complexity.
As automation increases, traders spend their days monitoring transaction flows and their risks. "No longer are they watching every trade," explains Tim Dodd, head of product management for SunGard Front Arena. "Rather, they allow machines to manage their trading and hedging while watching for exceptional activity or opportunity alerts that may foretell exceptional risk or exceptional profit."
Peter van Kleef, senior capital partner at Lakeview Capital Market Services agrees that when it comes to cross-market and cross-asset trading, real time analytics are key. "They give you a good sense of market dynamics and are great indicators when things are 'misbehaving' from what used to be normal in the markets," he says.
"The sell side sees real-time risk as an opportunity to attract order flow, to understand their risk borders and therefore be able to push the envelope that much further," says Mizen. As an example, she cites exchange-traded funds (ETFs) trading on different markets but with overlapping components. "There may be real arbitrage opportunities, which only those with real-time risk controls can properly assess."
Achieving Control
In the last few months, Adam Mazur, global head of connectivity for Goldman Sachs electronic trading, has seen a sharp increase in its focus on electronic trading platforms and risk management at the point of execution. "We're doing many more controls for fat-finger types of errors, notional volume checks, percent of average daily volume, and so on, which are all driven by the client's risk preference profile," he says. The ultimate aim is to understand what the trader is trying to achieve. "If it's outside the norm then we'll fire off an alert," he says.
"Real-time risk management is not just technology, but requires a very different approach to front- and back-office operations," says Giuseppe Ballocchi, head of financial engineering and risk analytics at the trading division of Swiss bank Pictet & Cie, and president of the Swiss CFA Society. "Even if some of the settlement details are still to be confirmed, all the economic and risk data must be captured as soon as possible, even pre-trade."
Indeed, pre-trade analytics are essential. "Traders need to be able to see how any prospective trade will impact their risk positions," explains Ballocchi. "So we give traders direct access to our risk reports, just as we will see them, but including the simulated trades. That not only saves calls on our risk managers but leads to a better quality of trade."
"Compliance checks keep moving closer to the front office, especially for algorithmic trading," says John Bates, founder and general manager of the Apama division of Progress Software. Bates believes that individual transaction checks need to be applied in-line before the order is placed, while portfolio level checks-such as real-time mark-to-market, value at risk (VaR) or auto hedging-can be done in parallel.
Nevertheless, there may be cases, such as high-frequency trading, where pre-trade risk checking may not work. "Even in normal trading you will miss a lot of opportunities," says van Kleef, who offers a different strategy. "Algos and mechanisms that reverse trades that exceed limits and that keep position risk in check should it be exceeded are more suitable. They act like an electric fence. Like cows that get a hit by an electric fence, traders usually only stray once or twice from course if they experience immediate reversal of their trades, usually at a limited but still painful enough loss."
Similarly, dealers need to protect themselves against so-called latency arbitrage or algorithmic arbitrage. "Hedge fund algorithms might probe a market maker's defenses looking for inconsistent bid-offer spreads or testing how their auto-pricing algorithm reacts to specific market data patterns," explains SunGard's Dodd. "Then they will nudge the price in one direction only to swoop on the other side and offload the position with another dealer."
Dodd sees risk appearing everywhere, especially in the booming foreign exchange (FX) market. "Traders are often faced with huge shoals of small transactions from online or algo-executed trades mixed with an occasional big shark of a trade. We need to recognize the big or exceptional transactions in our software and fast-track them directly into the trader's position. This allows them to take appropriate action on these higher risk trades and prevents these trades from taking big bites out of a trader's profit."
Latency and Contingent Credits
Managing risk implies much-needed caution but at the same time, traders are fighting the latency that occurs when huge flows of market data delays trades. "Execution has become much more challenging depending on millisecond differences across a large number of trading venues and resulting in huge volumes of market data," says Mizen, who insists that latency and speed need to be constantly managed. "If I see latency growing, then I might conclude that someone has a problem and avoid certain venues. Am I getting fills back as quickly as my competitors? That's the real issue."
Regulation has played a direct role here with Regulation NMS leading to an explosion of market data in the US equity markets. "The need to capture the data, including orders, time stamps and protected quotes is essential when reviewing for compliance to rules or when demonstrating a regular and rigorous review," says Greg Pratnicki, product manager for SunGard's Protegent Trading Compliance platform. "It's important to understand there is a responsibility on the trading centers to implement policies and procedures to assess latency between their data and network data and address any issue if one should arise."
In this Reg NMS environment, a single quote may update a considerable number of times within a one second window. "As such, there is an exemption for 'flickering quotations,' wherein the quote in the subject security was displayed for less than one second," says Pratnicki.
A very different challenge is posed by contingent credits. "Recent events have increased the emphasis on cost of capital," says Dodd at SunGard. "Now many traders not only have to worry about funding, but also the risk capital budget." Traders will contact the firm's credit desk to get an internal charge for a potential deal prior to execution depending on their counterparty. If they go on to trade, the charge will be a one-off fixed commitment, but the credit desk takes on and manages the ongoing exposure to all counterparties in the credit market. "This management is a very complex process that needs lots of computing power plus an efficient calculation methodology," says Dodd.
Sidhartha Dash, principal, risk and analytics at HCL Technologies, explains the evolution. Over the past 18 months, he and his firm have seen the rise of rigorously modeled, intraday contingent credit processes to calculate a trader's over-the-counter (OTC) limits based on dynamic market conditions for the most liquid derivatives and structured products. "These are computationally intensive, non-linear algorithms. Some collateralized products have huge embedded liquidity risk where we really need to take more of a portfolio-based approach, but the theory here is still evolving," he says.
Dash says that some credit derivatives with multiple event types and dependencies are pretty complex to simulate. People need to ask questions like, what happens if a credit rating is dropped or a credit or rates event occurs? They will want know how that will play across their portfolio. "People only used to do their contingent credit calculations at end of day, taking hours," says Dash. "But they are now using data and compute grids to run them in minutes on a couple of hundred nodes." He sees the current challenge as extending the range of collateralized trades covered, and increasing the size of portfolios, and changing the kinds of collateral accepted and the types of event.
The tension between precision and IT practicality is ever-present. "Since risk analysis is infinite, firms have to decide what is 'good enough,'" says Tabb Group's Mizen. She concludes that there is no simple answer. You have to at least be better than your key competitors, she says.
This is particularly important for OTC derivatives that take longer to price and then only with pricing data that is available periodically, perhaps every 15 or 30 minutes from the brokers. "When it arrives, we'll fully recalculate prices and exposures," says Dodd at SunGard. "We then switch to profit-and-loss approximations that are based on first order sensitivities of the position that are 'good enough' to allow the trader to spot opportunities in real time that they would otherwise miss."
"To keep latency down, people are reusing intermediate results in subsequent calculations, designing quite complex data flows," says Bates of Progress Software. "The overriding principle is to avoid starting from scratch for each decision."
According to Pictet's Ballocchi, the name of the game is to have an accurate consolidation of all positions in real time. "That's easy for exchange-traded instruments, but for OTC derivatives, where trade capture is delayed, it could take hours or, for some firms, even days." Ballocchi notes that if a trader executes OTC but then hedges on exchange, the hedge will soon be visible in the risk system but perhaps not the underlying trade, which will give a false picture of the position.
Many such inconsistencies complicate risk management. "Throughout the day, we run extensive cross checks on data coherence and consistency to identify and address such anomalies as they arise," says Ballocchi. He emphasizes that these data filters are constantly evolving and apply not just to their order flow data but also to exchange and vendor feeds as well. "Sophisticated risk models are only as good as the data. Therefore, data quality is key for risk management. Real-time risk controls are meaningless if any data is missing," he says.
The Dawn of Re-Engineering
Intraday risk is also driving a broader re-engineering of systems inside firms. "There are a lot of problems with risk right now," says Vivake Gupta, managing director and co-founder of technology consulting firm Lab49. "In many institutions, risk and pricing are still using completely different models and systems. It's mind-boggling given the lessons we've learned through these last few business cycles."
Lab49 has been working with firms to move risk to be an application to sit on top of pricing, where it would share the same libraries, market data and securities' descriptions. He argues that this would not only make it consistent, but also fast and, in some cases, real-time. "You may want to simulate some market data to do some what-if scenario valuations with alert thresholds for the risk managers, but many of the algorithms can and should be shared," says Gupta.
"Trading and risk converge as market-makers automatically re-hedge their derivative position gammas as underlying prices fluctuate," adds Dodd.
Yet nothing is ever new under the sun. "I think for the sophisticated trading community, real-time risk management is long established," says van Kleef. "As in any liquidity crisis, it is always the less sophisticated players that get taken out of the market. That, I think, is often surprising to investors but in the end good for the market, as it ensures survival of the fittest and helps to eliminate unviable and badly managed businesses."
Real-time risk may also be revising how people see electronic trading. "Some people think algo trading is more risky than regular trading," says Bates. "But if you put risk management in the algo it can in fact be even less risky."
Clearly, risk may always be with us, but current trends are to keep it under constant surveillance.
Bob Giffords is an independent banking and technology analyst. He can be reached at .
-->