There was nearly instantaneous condemnation of broker routing practices in the wake of a recently published paper by the FINRA Office of the Chief Economist. The study has a robust sample set (330 million institutional orders across 43 brokers across 273 stocks), and uses widely accepted execution quality metrics. Therefore, the end of the first paragraph should be taken very seriously:
“Trading costs based on the implementation shortfall approach are higher when clients select a broker with high affiliated ATS routing. Broker outcomes are highly persistent suggesting that improved disclosures on order handling could help institutional clients with broker selection.”
Concerns over preferential routing have been around for years. It was an underlying theme in the series of SEC fines levied against Alternative Trading Systems: several brokers failed to disclose certain practices used to attract order flow into an affiliated ATS. The recent FINRA working paper references a study from 2017 that examined sub-optimal preferential routing practices. Search an institutional-focused trading website and you will find numerous of articles on the issue.
There has also been a lot of scrutiny on Exchange routing, not just affiliated venues. A primary component of the SEC’s rationale for the Transaction Fee Pilot is to study, “the potential conflicts of interest faced by broker-dealers when routing orders as a result of transaction fees and rebates…”[AS1] . Whether or not a broker owns an ATS, there is a potential conflict of interest with routing orders based on client outcomes versus the economics to the broker.
It is interesting to note that the SEC has excluded ATSs from the Transaction Fee Pilot. The implication is that there are more important issues to resolve than broker affiliated ATS routing. The broker that routes to an affiliated ATS to maximize its own economics would be routing to other venues using that same logic. In other words, routing to an affiliated ATS is a symptom of a larger question that the SEC is trying to address.
Nevertheless, the results of the report should be taken seriously, even after we address the caveats and limited statistics made available. First, the average “top order” size in the report is 1,370 shares; “top order” is defined as the order routed to the broker from the client, not the child orders sent to the venues. In other words, the report is capturing a lot of smaller orders that clients are routing to institutional brokers. Clean-up orders are one explanation for the small average order size.
If the intent of the report is to capture institutional orders, why is there such a small top order size?
First, it doesn’t take much to drive an average down. If the report is capturing a lot of clean-up orders off the back of a larger execution, that could easily drive the average down. Second, the report does not distinguish between held and not held orders, in other words orders with specific instructions. Some buy-side firms are now automating orders below a certain percentage of average daily volume (ADV) and sending them to brokers with specific instructions. This practice would drive down the average top order size. In addition, the report would capture this activity as the broker routing it to an affiliated ATS, even if the client gave a specific instruction to route it to that destination.
My last caveat is that because the report is based on OATS data, it is missing the increasing usage of conditional order routes. Conditional orders are not OATS reportable. However, when an ATS finds a potential cross for a conditional order, it sends a firm-up request to the order sender. The order sender then responds with a firm order, which is OATS reportable. Thus, the numbers in this report could be misleading in two senses. There are probably even more affiliated and unaffiliated ATS routes than captured in the report and lower fill rates for those child orders, even after correcting for the inclusion of not held orders. These caveats should be taken into consideration when examining the statistics. It doesn’t excuse any weakness in performance, but it may shed light on the kind of actions that could be helpful.
Try, try again?
While the report contains a lot of interesting statistics I am going to highlight a series that surprised me the most. First, one would expect that a broker who owns a dark pool would route orders to dark pools. But the degree of difference shown in the data below, from Table 3 from the report, is shocking.
The T3 (most % affiliated ATS) group has six times as much quantity routed to an ATS compare to T1, nearly all of it routed to an affiliated ATS. This happens despite only 27% of the routed quantity being filled at that ATS. That’s a lot of wasted routes.
Below are the execution quality stats for each of the groups for the Medium stocks (based on market cap). I chose the Medium names because the Arrival spread (bp) across the T1-T3 is most similar, so you can’t blame exogenous factors like the toughness of the names. First, the fill rate for T2 and T3 groups is substantially lower compared to T1. This helps explain the similar weakness in Shortfall close (bps) which is the sum of the Effective Spread and Opportunity Cost (or simply, lack of fills). We are faced with a clearly unacceptable 7bps difference in Shortfall performance between T1 and T3.
What’s interesting is that there isn’t that much of difference between the performance of T2 and T3 brokers (~3bps), even though T3 brokers sent substantially higher percentage of shares into their affiliated ATSs. If a very small percentage of routes to an affiliated ATS is responsible for the majority of slippage, then the issue may be not how much is routed to an affiliated ATS but that it happens at all. Said differently, information leakage may have a diminishing cost.
These results also point back to the earlier caveats. If there are clients that specifically use a broker because of its affiliated ATS, or an algorithmic strategy that is inherently tied to the ATS, then the % routed numbers are more understandable. For example, if you’re using a broker’s VWAP algorithm and the affiliated ATS is a trajectory cross, one would expect every order to go into the trajectory cross. It doesn’t excuse the performance, but it does explain the vast gap in routed quantity. Another example is a buy-side firm has its own proprietary router and pings each of the broker-owned dark pools before going to an Exchange.
Still, the difference in shortfall close performance is dramatic. Putting aside who is responsible for the routing and any vagaries in the routed quantity numbers, it would be hard to ignore the poor performance associated with ATS routes for smaller orders.
In some sense, these numbers support the view that the primary value proposition for the dark is executing in size. And while we wouldn’t support the draconian and byzantine order handling regulation of MiFID II, the data above certainly highlights the need for increased transparency and reporting requirements. The industry can take some comfort that by the second half of 2019, Reg ATS-N and the Order Handling Disclosure will dramatically increase the amount of information and data on ATS functionality and routing behavior.
Adam Sussman, Global Head of Market Structure