Before casting a critical lens on Gregory Crawford's March 21 article "Algorithmic trading saves money, report says," I would like to acknowledge two things. First, as a firm, we are neither hostile nor strangers to algorithms, already employing them on our desk. Second, I applaud ITG's efforts to get the ball rolling, measuring if algorithmic performance actually lives up to its recent hype. While the study is a good first step, it raises more questions than it answers due in part to limitations of the empirical data, but also to unjustifiable logical leaps. Let me explain the three most serious shortcomings.
The study uses volume-weighted average price, or VWAP, and implementation shortfall to compare trading costs, but the data from the control group cannot be fairly measured against either of these benchmarks because the strategy used to execute the trades is generally unknown. As a broker, the client-specified benchmark is crucial to how I choose to execute. In fact, if I am being measured by implementation shortfall, trouncing VWAP may still mean a terrible performance. The alpha the client was trying to capture may have evaporated as I "successfully" pieced out the order all day. Similarly, the algorithmic trades should not be measured against anything other than VWAP since the "vast majority" of the algorithmic trades in the study followed a VWAP strategy. Thus, the measurements used cannot be considered at all appropriate (and make the results meaningless) until more information is known about the strategies behind the executions and an apples-to-apples comparison can be made.
Moreover, while missing VWAP on average by only two basis points may sound like great performance by the algorithms, it's actually a miss by over 0.75 cents per share. Ironically, if the passive algorithmic strategies are on average missing VWAP, active strategies must on average be beating the VWAP because it's a zero sum game — all trades together create the daily VWAP.
Lastly, it's not disclosed what types of brokers comprise the control group. It's possible that any type of direct market access, whether algorithmic or not, might score better than trading with certain categories of brokers. Avoiding order exposure to brokers who may cause market impact from principal trading or information leakage/delays in execution from order shopping could possibly result in superior performance apart from any inherent benefits of the algorithms themselves.
Joseph C. Gawronski
chief operating officer
Rosenblatt Securities Inc.