In today's hyper-competitive, cost-conscious trading environment, fund managers and buy-side traders are obligated to watch every penny, and to do so, many have turned to computerized algorithms provided by brokers at cut-rate commissions. Algorithms have become a must-have for brokers seeking to gain new business and retain current clients - TowerGroup estimates that buy-side adoption of algorithms will triple between 2003 and 2006. But now that a plethora of algorithms is available, the buy side is looking for more quantitative support than just post-trade transaction cost analysis (TCA).

To meet that demand, the sell side and the TCA vendors are scrambling to provide pre-trade analytics, which, through the analysis of historical and current price and volume data, attempt to help clients determine where to send orders and when; whether to use algorithms or manually trade an order; and what the opportunity cost of not acting on information might be. But brokers have been struggling to produce pre-trade analytics that are meaningful for clients and have encountered trouble in distributing these analytics to client desktops, both for cultural and technological reasons.

At the moment, most pre-trade analytics products cannot accurately predict the price of a stock for a given day, which is what traders ultimately want, according to Linda Giordano, product specialist, Quantitative Services Group (QSG), a TCA vendor. Most compare the spread between bid and ask prices, reference that against the volatility of a given stock and attempt to create a range of potential outcomes.

Usually, the analytics predict the expected cost in trading fees that clients would pay if they executed a given trading strategy at a given point in time, says Gavin Little-Gill, author of a TowerGroup report on algorithmic trading.

"Pre-trade analytics is a Holy Grail that has been very elusive," QSG's Giordano says. "It is tough to predict in a way that is relevant or practical, and it is almost impossible to do in the way the market expects to do it."

To Build or Not to Build

Most buy-side clients get their analytics from brokers as part of an overall service package. The benefit to this is that, having created the algorithms, brokers know them backwards and forwards. However, there is a perception that brokers offer biased analysis that favors their own algorithms. Additionally, buy-side firms are concerned with disclosing too much information about their proprietary trading strategies. As a result, buy-side firms often choose to use third-party TCA vendors or build pre-trade analytic capabilities in-house.

Most TCA firms, however, provide only post-trade analysis, though QSG is developing a pre-trade offering, Giordano notes. These providers can be valuable independent technology sources, but their solutions are limited by the information that brokers and the buy side are willing to share.

So, some of the larger, more technologically savvy buy-side firms conduct their own analysis, which requires a tightly integrated research and trading team and substantial amounts of real-time data. For example, Barclays Global Investors (BGI) conducts all of its analysis in-house, updating stock-specific models daily with real-time data feeds and trade histories from its in-house order management system (OMS) and its FlexTrade execution system, according to Ananth Madhavan, global head of trading research.

"The concept of more sophisticated pre-trade analytics is important," says Michael Sobel, head of U.S. equity trading, BGI. "But the essence of brokers with TCA is that they do have advance knowledge of what you are doing - we try to avoid that."

To limit disclosure of their trading strategies when using pre-trade tools provided by brokers, some buy-side firms run the software in-house rather than send guarded data to the brokers. "Any [analytic system] that we use is used on-site, because if we shot a list of [intended trades] to a broker and had them analyze the list, we'd risk exposing the list. I feel we can get sniffed out in the market," says John Wheeler, director of U.S. equity trading, American Century Investments. Wheeler adds that, since it has built its own trading systems with direct data feeds, American Century can perform its analysis on raw market data, rather than rely on a broker- or vendor-supplied trading system to supply and interpret that data.

Benchmark Ambiguity

It's not only the suspicion of vested interests that has made marketing analytics a challenge for brokers; there is also no prevalent standard benchmark for evaluating algorithms against each other - and some say there shouldn't be.

The most prevalent trading benchmark in use today is volume weighted average price (VWAP), which is calculated by adding the dollars traded for every transaction in terms of price and multiplying that by shares traded, then dividing that by the total shares traded for the day. VWAP is popular because it's relatively easy to measure and provides comparative results, but it isn't as useful for evaluating strategies that are trying to do something other than follow the market midpoint; such strategies often actively trade single stocks, using algorithms that can be altered on the fly.

"VWAP is really the only agreed-upon measure out there," TowerGroup's Little-Gill says. "The problem with applying that to algorithms is that it doesn't take into consideration anything other than indexing. A thousand shares of IBM traded four times in a one-week period will get a different execution quality each time. People talk about comparing algorithms across different broker-dealers, but it is really tough unless you have significant quantities traded, and it is as much an art as it is a science."

VWAP is often incorrectly applied to measure the performance of algorithms that have different trading goals, according to Brian Fagen, managing director at Morgan Stanley, who adds that algorithms should be evaluated as part of a firm's overall strategy to achieve a goal, not as separate entities. The desire for a benchmark specifically for algorithms "calls into question those algorithms that do not have a goal," he says.

Brokers have begun to focus on other standards, including market-on-close (MOC) and arrival price. MOC measures the last price obtained by a trader at the end of the day against the last price reported by the exchange. Arrival price is the midpoint of the bid-offer spread at order-receipt time, and it also notes the speed of the execution.

Rob Flatley, managing director of Bank of America's Electronic Trading Services (ETS) unit, says the market is moving toward these new standards. "The quant-est shops on Wall Street have already moved to arrival price," he says.

There are algorithms geared to all three benchmarks, but a "pure" benchmark of all the algorithms in the market is unlikely to be created, since brokers aren't keen to part with information about their trading partners or strategies. "We do not collect the isolated data that would be needed to rate algorithms against each other, between brokers," says Marie Konstance, director of sales and product management at Plexus Group, a TCA vendor. "Brokerage firms do not provide algorithmic trading data on its own; you only can analyze how well they are doing overall."

"It would be hard to maintain [a generic algorithm benchmark] and it would have very little relevance," Morgan Stanley's Fagen says. "If my algorithm is benchmarked to the market close, does it make sense to measure that algorithm up against some other benchmark?"

Integration Obstacles

A lack of technological integration with buy-side OMSs also has restrained the use of pre-trade analytics. Although broker-sponsored trading systems have algorithms and analytics built in, very few vendor and in-house OMSs support the real-time tick data that allows for informed, on-the-spot decisions.

"We've all had a tough time [linking decision tools to] our execution capabilities, because we are dependent on OMSs [that] mostly weren't built at the time algorithms were invented," Bank of America's Flatley says. But that is beginning to change.

BofA now is integrating decision support tools with buy-side order management systems, rather than delivering the tools in a stand-alone screen off to the side, a job that is about 50 percent finished, Flatley relates. He says he expects to integrate BofA's analytics application with seven to 10 vendor-built OMSs by the end of the year. But that's only the beginning of the integration effort: Flatley estimates that 80 of the top 500 U.S. mutual funds run proprietary OMSs that would require a broker-dealer to make individual configurations for each client.

Even as the industry struggles to deliver effective pre-trade analytics, the demand for accurate predictive information isn't going away. Rather, in the current high-frequency trading environment, the demand for pre-trade analytics will only grow. Though tighter technological integration is bound to come in time, if pre-trade analysis is to fully come to fruition, firms may also need to find a way to share more information without putting their trading strategies at risk.

---

What They Have to Say About Pre-Trade Analytics and TCA

"A big area of growth for us is clients coming to us asking how to best use the algorithms we have," says Carl Carrie, vice president of new product development at JPMorgan Securities. "Once you are using an algorithm, under what context do you use it? That whole side is becoming very useful. We offer pre-trade analytical products over the Web and embedded into execution systems."

"A lot of the broker-sponsored algorithmic trading systems attempt to measure or project the trade costs as well, but they are pretty rudimentary," says John Wheeler, director of U.S. equity trading, American Century Investments.

"Pre-trade analytics attempt to predict how close you could come to some kind of strike price, like the prior day's close or today's open," says Eugene Noser, president of Abel/Noser. "None of them are very good. Most pre-trade analytics overestimate the cost of a routine or benign trade and greatly underestimate the cost of difficult trades."

"If you think of basic time-slicing, the ability to map that activity and maintain that benchmark is what separates people," says Jana Hale, global head of algorithmic trading at Goldman Sachs. "You need to do a little [analysis] in-house, a little with the third-party TCA and a little with the sell side, and within that triangle of resources you come up with best execution."

"In a stock that is not so liquid, if you trade a large position over the course of the day, if you are measured against VWAP, you become the VWAP," says Linda Giordano, product specialist at TCA firm Quantitative Services Group (QSG). "There is behavior that goes on in the microstructure of a trade that makes VWAP a very poor, yet prevalent strategy. It's become like a reflex."

"It's as critical to have a decision support system in an algorithmic suite as it is to have execution capability," says Rob Flatley, managing director of Bank of America's Electronic Trading Services (ETS) unit. "From the pre-trade perspective, we think arrival price or market-on-close (MOC) are the two standards that people have already started to focus on versus VWAP."