Statistics on trading costs in dark pools have been in short supply. Now, a new study by agency broker ITG analyzes the market impact costs associated with 10 dark venues and knocks down several assumptions about executions in non-displayed venues.
The study, undertaken by ITG managing director Ian Domowitz and two colleagues, argues that institutions can trade blocks more cheaply by executing directly in dark pools rather than through algorithms that aggregate access to multiple pools. The study also quantifies how much actual performance degrades in specific dark pools as the time to complete an order increases.
Domowitz noted that information leakage could be the culprit behind increasing trading costs over the life of an order. The likely presence of “information leakage, even though everyone believes they are trading in the dark, has some legs,” he said at a press briefing yesterday.
The study analyzed 10 dark pools, including three from ITG: POSIT Now, POSIT Match, POSIT Alert, Pipeline, LeveL ATS, the ISE Stock Exchange’s MidPoint Match, Knight Match, Morgan Stanley’s MS Pool, NYFIX Millennium and UBS’s Price Improvement Network. All are venues that ITG’s dark pool algo accessed last year. The study’s report, called “Cul de Sacs and Highways,” is based on execution data from 21 million buyside orders in 2007. (The report refers to dark pools as cul de sacs and algos as highways.)
ITG calculated trading costs relative to an arrival price benchmark. For buy orders, the benchmark price was the national best offer at the time the broker received the order. For sell orders, the benchmark was the national best bid.
The study found that executions in dark pools add value by reducing trading costs, compared with trading in the displayed markets. ITG’s own dark algo resulted in an average trading cost of 4 basis points for orders, compared with 12 basis points for the “ITG peer universe,” which refers broadly to trades resulting from direct-market access, crossing systems and other trading styles. However, the study found that executing in POSIT Match, ITG’s point-in-time crossing product, outperformed the benchmark by 4 basis points.
“Dark pool execution is beneficial,” whether the execution occurs in a single pool or through a liquidity aggregator, Domowitz said. But aggregators, or dark pool algos, do not “increase the ability to trade in size or reduce costs.” This finding contradicts arguments by brokers that their algos help institutions execute block orders by efficiently scooping up liquidity residing in multiple dark pools. About two-dozen brokers have dark pool algos, according to Traders Magazine research.
Dan Mathisson, head of Credit Suisse’s Advanced Execution Services unit, whose popular Guerrilla algorithm accesses multiple dark pools, cast a skeptical eye on some of the study’s results. “The paper concludes that traders are better off parking their trade in a dark ‘cul de sac’ versus driving around on a data aggregation ‘highway,'” he said. “But the old word for cul de sac is ‘dead end,’ which is what going to only one destination often turns out to be.” Credit Suisse’s dark pool was not among those analyzed in ITG’s paper.
The ITG study drilled down to the execution quality in specific pools. It found that trading quality in all pools worsens the longer an order remains unfilled. “You’re paying the price for the life you live,” Domowitz said. Market impact costs across all 10 pools, including ITG’s three intraday pools, increased the longer the order was in play.
However, those costs varied significantly across pools. In the first 30 minutes of trading, MS Pool provided the best executions, outperforming the arrival price benchmark by 7 basis points. Next in line, respectively, were POSIT Match, POSIT Now, Pipeline and POSIT Alert, all of which bested the benchmark. In the second half-hour period, Pipeline and POSIT Match were the only two that beat the benchmark. ISE’s MidPoint Match logged a cost of 7 basis points, on par with most brokers’ dark pools, while LeveL’s cost was 16 basis points. In the third half-hour bucket, Knight Match displayed the worst performance, with 33 basis points in transaction costs. In the fourth time bucket, LeveL’s cost was 24 basis points, while NYFIX Millennium was the second-worst, at 20 basis points. Only POSIT Alert outperformed the benchmark in that period.
The study was based on 12.6 million buyside orders ITG executed in the first three quarters of 2007. Those orders resulted in 75 million trades. ITG supplemented that data with 8.2 million orders from its transaction-cost-analysis data set. Domowitz noted that preliminary data for this year bear out the study’s findings.
The bulk of ITG’s trade data was for executions that took place within two hours of receipt of the order. Executions in the third hour were included in the study, but were “more iffy,” Domowitz conceded, because less data was available for that time period. He noted that there were several “vagaries” in the data results about specific pools, but that the study’s conclusions were sound. The study, he said, also supported findings about dark pool executions in two earlier reports by Goldman Sachs and Quantitative Services Group, although those studies were narrower in scope.
Domowitz cautioned that the results of the study should not be viewed as a ranking or “league table” of dark pools. It is unclear, he said, exactly what accounts for the execution cost differences in particular dark pools. However, he pointed out several potential factors. These include the possibility that some dark pools are accessed primarily by algos that reach multiple pools, or by algos that also execute in the displayed markets. In those cases, information leakage could adversely affect executions in dark pools. Other factors that could impact trading costs are the small average size of executions in some venues and gaming in dark pools.
There are smart ways of executing blocks in dark pools, Domowitz said. Going to pools with 600-share executions, for instance, “isn’t the right thing to do.” He added that the probability of getting lower-cost executions is higher by executing directly in POSIT Match and POSIT Now, for instance, than by going through ITG’s dark algo.
A spokesman for Liquidnet, the buyside crossing system operator, said his firm agreed with ITG’s findings that using multiple pools increases the risk of information leakage. “Most traders know not to give the same order out to multiple brokers, because they would compete with themselves in the market and move the price,” the spokesman said. “But they haven’t yet equated that with using an algorithm to put the same order into multiple dark pools.” Liquidnet was not included in ITG’s study because it does not allow brokers to access its crossing system.
