Mar 3, 2021

March 3, 2021

D-Limit Performance & the Fill Rates Race


Measuring fill rates vs. price improvement

In our last blog post, we demonstrated that D-Limit outperforms other exchanges’ offerings when measured in markouts. Even when factoring in fees and rebates, D-Limit is a clear winner.

Of course, the obvious follow-up questions are around fill rates. If you are giving up potential volume in order to get higher-quality executions, is it worth it?

The below analysis aims to answer that question quantitatively.

We should first explain that D-Limit does not only provide markout performance, but also the opportunity for displayed price improvement (PI). Typically, price improvement is thought of from the perspective of aggressively-priced orders. For example, if a buyer is trying to lift an offer and runs into a midpoint, that order gets price improved. There’s also the concept of “pre-trade price improvement” where a resting order can “float with the quote.” IEX’s D-Peg is a good example. D-Peg rests 1 tick outside the quote and its resting price adjusts as the NBBO changes. This allows you to potentially execute at prices better than when the order was sent. Before D-Limit, this pre-trade price improvement was only possible for dark orders. This is because, for displayed orders, the best price you can achieve is your stated order price. With any other displayed order type, if you post a bid at $10.00, the best case is you trade at $10.00 (unless you were to take an action like a cancel order).

The way D-Limit works is that when the IEX Signal (i.e., the Crumbling Quote Indicator or CQI) predicts that the price is about to change, it triggers D-Limit orders to automatically reprice by one tick outside that level.

This functionality is what drives the high quality of D-Limit executions — put simply, it is designed to help you avoid getting run over when the price is unstable. Said another way, D-Limit affords the opportunity for your $10.00 bid to be moved to $9.99 and now trade at a penny better than your price on entry. This functionality has proven to provide a significant benefit. Approximately 18% of all D-Limit volume from institutional brokers received PI (an average of nearly 2 cents per share).[1]When averaged out across all D-Limit volume, that’s 36 mils of savings per share.[2]

The Flip Side of Good Markouts and Price Improvement

The flip side is that this same functionality — by design — prevents trading at specific moments. After a D-Limit order has been repriced, there are two potential scenarios, both of which might result in lower fill rates for a resting D-Limit order compared to a standard limit order:

  1. The price moves into you (i.e., the Signal prediction was correct): Your D-Limit order did not trade at the old price, and is now eligible to trade at a better price from the perspective of the resting order. However, the order still needs a counterparty to trade against it at the new price.
  2. The price does not move, or moves in the opposite direction (i.e., the Signal prediction was incorrect): Your D-Limit order did not trade at the old price, is now resting 1 or more ticks outside the NBBO, and is much less likely to trade.

So how do these dynamics stack up against one another?

Spending Your Savings

One other way to look at the question “Are D-Limit’s performance and price improvement worth its lower fill rate?” is this: Are the savings big enough that you can make up the lost volume in the open market and come out on top?

To test this question, we started by looking at the fill rate difference between D-Limit and traditional displayed limit orders for institutional broker-dealers (full-service and agency) on IEX. We use their data because fill rates of market makers are almost an irrelevant metric due to the number of order cancellations and replacements.

Next, we use the results from our previous blog, which demonstrated that the “all-in” D-Limit outperformance vs. traditional maker-taker exchanges is approximately 80 mils/share, measured in 1-second markouts. To that, we add 20 mils/share of price improvement, a conservative estimate given that D-Limit volume has received an average of 36 mils/share of PI (noted above).[3] Together, D-Limit markouts and PI represent significant savings for D-Limit executions.

We now look to “spend” these savings to acquire additional shares. While we cannot replicate the “parent order experience” from our seat as a venue, we can determine how many shares we can capture after restating an order correctly.

To do this, we took each order restated by D-Limit, looked forward to the later of one second or the order cancellation, and determined if the “far side” of the NBBO was within that order’s limit at this time. For example, if a buy order was resting at $10.00, we restated it to $9.99, and 1 second later the NBO was $10.00 or lower, we executed any remaining interest on that order. In order to determine how many shares can be executed, we took the minimum of the order’s remaining shares and the average quote size of the symbol.

In the table below, you can see that this results in an additional 64 million shares, bringing the fill rate of D-Limit above that of the traditional displayed limit order on IEX.

Of course, those shares don’t come “for free.” Since we are comparing D-Limit not only to IEX but to the experience on other venues, we have to reconcile the fact that, for shares that are captured this way, traders could potentially be foregoing rebates associated with trading on another venue and may also have to pay a fee of up to 30 mils as the “aggressor” of the trade in capturing this available liquidity. Therefore, we calculate the cost of crossing the spread as anywhere from 30 to 60 mils/share (we conservatively assume a 30-mil rebate is always foregone, and the cost to cross the spread varies depending on where it is done). This cost is subtracted from the benefit we start out with from the better markouts and price improvement.

Data from Q4 2020

The end result — even at the most conservative end, where we assume a fee of 30 mils/share as well as the 30 mil foregone rebate — is a substantially higher fill rate, with a substantial amount of outperformance “left over.” These remaining savings can in turn be converted into acquiring additional volume, bringing the effective D-Limit fill rate even higher. After all, D-Limit isn’t just being compared to the IEX displayed limit order experience, but also to what occurs on other venues. The remaining outperformance can therefore continue to be “spent” acquiring additional shares to further improve the effective fill rate of D-Limit.

Conclusions and Assumptions

D-Limit poses an upfront cost to fill rates, but the performance more than allows users to catch up to their needed volume, even when having to cross the spread and accounting for the forgone Tier 1 rebates at other exchanges. If incorporated appropriately, D-Limit can provide its market-leading performance with no material cost to fill rates.

We know that this example includes a number of assumptions that may or may not hold for your trading strategy. We hope you will take this framework and try it out it for your particular context. As always, thanks so much for your time. Please reach out to me or your IEX contact anytime if you would like to collaborate on an analysis that factors in your particular circumstances.

[1] “Institutional Brokers” defined as Full-Service and Agency broker-dealers, based on IEX classifications on a best efforts basis by Member firms’ trading sessions.

[2] Based on Q4 2020 IEX market data.

[3] Price Improvement is highly dependent on rest time post-restate. To reflect the “median” experience of institutional brokers, we used 20 mils.

Appendix: Extra shares captured are capped at 100 shares

To be conservative we also repeated the analysis assuming that we capture no more than 100 shares at a time when “catching up.” In this analysis, the fill rate is slightly lower, but there are more savings “left over.”

Data from Q4 2020