The decision last month by Pipeline Trading Systems to block orders from buyside traders delivered via brokers’ algorithms highlights a growing concern in some sectors of the trading industry.
Pipeline, which operates one of the largest block crossing systems, says it banned algo-borne buyside orders from its platform out of concern that money managers relying on dark algos might not be getting the fills they expect. The worry was that some brokers might have sacrificed liquidity in favor of cheaper executions.
“The profit incentives were not correctly aligned-and that’s a problem for some in the brokerage community,” says Pipeline founder Fred Federspiel. He adds that brokers’ customers did not get the best execution possible if the algorithm they were using limited executions on Pipeline simply because they cost the broker more.
“Some destinations are offered at a discount, others at a premium,” says David Brooks, head of equity trading at Boston Company Asset Management. “How could that not factor into the sequencing of brokers’ routing decisions?”
Brooks adds that he has found it difficult to determine the “preferencing logic within algorithmic strategies”-and even which pools algos try to extract liquidity from. Preferencing refers to priority that may be given to certain pools over others in an algo’s routing schedule.
Institutions can still access Pipeline directly or through direct market access platforms such as BNY ConvergEx Group’s Sonic, Banc of America Securities’ InstaQuote and Goldman Sachs’s REDIPlus.
DMA platforms, like algorithms, provide broker-sponsored access to Pipeline and other crossing systems. The difference is that the buyside customers pay the cost of those executions. With broker-provided algos, the broker pays for the executions and charges customers separately.
Merrill Lynch, Goldman Sachs, Morgan Stanley, UBS, Instinet, ITG-all these and other brokers charge customers different per-share rates for executions through their dark algos. Those rates reflect the type and overall quantity of business the client does with the firm. However, the various rates the brokers themselves pay for algorithmic executions (typically lower than what customers pay when they go directly to the ATS) vary based on where the executions occur.
For algos, Pipeline was the most expensive venue, based on its explicit per-share execution cost, while NYFIX Millennium, for example, occupies the middle range. In comparison, the fees charged by the broker consortia and exchanges (or planned fees for those that haven’t yet launched) are minimal.
The fear is that brokers will engineer their routing tables (order-routing schedules) to take into account the prices they are paying different venues. They could do that by giving preference to destinations that charged them lower rates, including their own dark pool or internalization engine.
The vast majority of brokers say their dark algos do not consider price-or do so only minimally. The goal, they state, is finding liquidity for their clients. If they don’t do that, clients would make a beeline for other brokers.
But brokers acknowledge that execution prices vary considerably. UBS’s Will Sterling, global head of electronic equities trading, points out that using a dark order type on Nasdaq might on average cost a couple hundredths of a cent per share, compared to some dark pools that are more than one penny per share. “An execution on one system could be literally 50 or 100 times more expensive than an execution on another system,” he says.
That difference could be a consequence of rapid market evolution that has produced different business models for crossing, Sterling says. It could also reflect the fact that some pools offer more value for blocks than others. He adds that UBS’s Tap algo looks for undisplayed liquidity wherever it resides, without regard to price.
ITG’s Tony Huck, head of algorithmic trading, says his firm distributes order flow equitably to external dark venues accessed by its DarkServer algorithm. DarkServer accesses POSIT Match, POSIT Now, NYFIX Millennium, Knight Match and other dark pools. Until recently, it also sent flow to Pipeline.
However, Huck notes that the price paid to different venues for DarkServer’s executions naturally affects margins. “Depending on where the liquidity is,” Huck says, “some days we’ll have better margins given a particular commission rate than other days. Getting a lot done in a venue that charges us more will hurt our margin on that day, and getting a lot done in POSIT will help us.”
ITG, JPMorgan and several other firms charge customers slightly different rates based on which pools they access. That passes on some of the execution costs to the algo user. JPMorgan, for instance, periodically discusses trade performance and costs with clients and resets the fees clients pay based on estimates of actual fill costs, says Carl Carrie, head of product development in the firm’s electronic client services group.
Some brokers steer away from the pricing issue by stressing their ability to find dark liquidity without information leakage-and charging a higher rate for it. “We look at Hydra as a premium product,” says Lehman’s Michael Bleich, head of liquidity strategy for U.S. equities. Lehman provides interested customers with periodic fill reports for various destinations, as well as other execution-quality statistics.
But Pipeline’s Federspiel says the best-execution issues associated with dark algos can’t readily be glossed over. He says his crossing system’s fees were definitely a factor in the routing decisions some brokers made-and that for brokers, the incremental costs of executions on Pipeline remained high regardless of their pricing arrangements.
In December, when dark algorithms accounted for one-fifth of Pipeline’s volume, Federspiel had a number of conversations with major brokers. He says they asked Pipeline to lower its per-share fee for algos and to lower its high minimum-order size. Pipeline has always required giant minimums for orders to prevent members from gaming the system by identifying block liquidity resident within the platform.
Pipeline declined both requests. From December through February, Federspiel says, the algorithmic share of Pipeline’s executions nose-dived 63 percent-from 20 percent to 7 percent. In that same period, overall volume executed on Pipeline increased 10 percent.