In the discussion around our recently published execution quality white paper [1], one recurring point of contention is that our metrics evaluate displayed and non-displayed trades together, and therefore conflate marketable and non-marketable orders.
The implication here is that we did this to inflate our own numbers. The reality is, when it comes to actual broker behavior, it doesnt make sense to look exclusively at marketable OR non-marketable orders. Given brokers and investors have moved beyond this simple dichotomy, our measurements of execution quality must reflect this as well.
Real Broker Behavior: Pinging the Mid
Analysis of marketable (defined as spread-crossing) orders only misses the fact that in practice, many traders send a nonmarketable order to check for midpoint liquidity first (pinging the mid), before subsequently sending a marketable order if they dont get a fill.
At IEX, we see that 28% of spread-crossing orders sent to our market were preceded by a midpoint order (from the same MPID, in the same symbol, and on the same side of the market) within the 10 milliseconds (ms) immediately prior. And a full 44% were preceded by an order within the prior 100 ms [2].
This matters because when a broker checks the midpoint before sending a spread-crossing order, it soaks up all the available price improvement (defined as receiving a fill at a price better than what is currently quoted) that would have otherwise been allocated to the spread-crossing order. As a result, solely analyzing spread-crossing orders will reflect?-?whether by design or ignorance?-?an incomplete view of available midpoint liquidity on that venue, and will therefore have artificially lower assessments of price improvement.
Luckily, we can draw a pretty clear line of demarcation in how users of certain order types behave. The table below shows that brokers crossing the spread using the IEX router very rarely ping the mid before doing so. Conversely, those who cross the spread on their own frequently check IEXs mid beforehand.
A Modern Price Improvement Analysis
A better way to compare price improvement of spread crossing orders in an apples-to-apples way is to look at price improvement via data from the IEX router, which does not check the mid before routing out and exclusively sends spread-crossing orders. As such, it is a good proxy for how likely a venue is to provide price improvement for a random, spread-crossing order.
We reported such an analysis in our white paper. The results of this analysis can be seen in the table below.
This data shows that when it comes to spread-crossing orders that havent pinged the mid first, 10.9% of volume routed to IEX receives positive price improvement. And the difference in the PI% between IEX and the #2 venue (10.9% vs. 7.1%) is greater than the difference between the #2 venue and the last place venue (7.1% vs. 4.1%).
In todays markets, investors trade using both marketable and non-marketable orders to source liquidity. The line between those kinds of orders may not matter to some market structure analysts, but it is critical in order to evaluate venues on metrics that are reflective of an investors trading experience. We will continue producing these metrics and sharing them with the trading community so that all market participants can get the most complete picture of execution quality across exchanges.
[1] See Elaine Wah, Stan Feldman, Francis Chung, Allison Bishop, and Daniel Aisen, A Comparison of Execution Quality across U.S. Stock Exchanges, (April 19, 2017), available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2955297.
