When it comes to transaction cost analysis (TCA), the tail shouldn’t wag the dog. TCA shouldn’t supersede the careful decisions a trader makes to achieve a portfolio manager’s objective.
Indeed, getting the best fills on a trading desk could mean discounting what TCA might say is a good trade. Greg Komansky, senior trader at ClearBridge Advisors, points out that TCA isn’t always effective at valuing the quality of a trade-meaning it could lead traders astray.
A trader may beat various benchmarks such as volume-weighted average price and implementation shortfall, but still execute a trade poorly, Komansky said at industry confab TradeTech USA last month. For TCA to be most effective, he said, traders should know a portfolio manager’s reason for a particular trade, including the trade’s urgency level, and then apply that information to their trading decisions.
Whether TCA measures the man or the machine may also be unclear at times, said Michael Catalano-Johnson, head of algorithmic and quantitative trading at Susquehanna International Group, at the same conference. For example, a trader could select an algorithm that isn’t appropriate to use under certain market conditions. In that case, the algo may not be at fault for the resulting executions.
William R. Yost, senior vice president for U.S. quantitative equity products at Tokyo-based DLIBJ Asset Management, noted that traders must monitor algos in “high-impact” and illiquid names in particular. For those stocks, he said, TCA is less useful than it may be for more active stocks. Yost said his desk frequently withdraws trades from algorithms when the market is volatile.
Results from a recent TABB Group survey of money managers support these views. Among traders who use algorithms, 36 percent said they need more “decision support” from their algo providers. Many of them want more help in selecting the strategy and settings that are most appropriate for a particular order. Only 16 percent said they wanted more customization in their algorithms.
Using TCA to compare and contrast brokers’ algos may also not be reliable. TCA may not differentiate between similar variables across brokers. What’s passive, normal and aggressive for each broker’s algorithms? Catalano-Johnson asked. How these user-selected parameters are defined affects the quality of executions.
And another thing, Catalano-Johnson continued: Does the TCA throw out the high and low trades when it calculates the average price paid? Not doing so could skew results. “Know your TCA,” he advised.
ClearBridge’s Komansky added that no TCA takes into account the fact that a trader may have stopped trading for a period of time because of unusual market conditions or another reason. Making that decision, he said, is part of a trader’s job that isn’t captured in a TCA report.
But while TCA may not be ideal for assessing algos across brokers, it’s important-and relatively easy-to know whether an algo did what it was supposed to do, Catalano-Johnson said. If a trader told an algorithm to be more aggressive when the price went up, the trader should be able to see if the algo did that.