Democratizing access to high-quality displayed orders
Today, we’re announcing that IEX has filed for a new order type: Discretionary Limit, or D-Limit.
Those of you who know IEX know that we don’t take introducing new order types lightly. All new products here must set a new standard for performance, fairness, and transparency.
This one does just that. We designed D-Limit to level the playing field for all market participants who want to contribute to public price discovery and get high-quality trades.
Displayed trading today is fragmented, complex, expensive, and tiered, creating an environment where only the most well-heeled and technically sophisticated participants can thrive.
The goal of D-Limit is simple: we want to provide a tool designed to help all market participants achieve better performance in displayed trading.
D-Limit is an IEX order type that behaves like a regular displayed limit order for nearly all of the trading day. It differs from a regular limit order by leveraging the IEX Signal (i.e., the Crumbling Quote Indicator or CQI), which identifies brief moments in time when the quote is “crumbling” — a strong indication that the price is about to change. The Signal is “on” for only 0.02% of the trading day on a volume-weighted basis — but 24% of all IEX displayed trading happens when the Signal is “on.” As we’ll show in this post, performance for displayed resting orders is, on average, terrible when the Signal is “on,” and the goal of D-Limit is to avoid trading during these moments.
When the IEX Signal is “on” and predicts that the price is about to change, D-Limit reprices to 1 MPV outside of that price, and it stays there unless the Signal fires again or the Member updates the limit price. D-Limit doesn’t operate like a pegged order, so the Signal only triggers moves in a single direction when market prices are moving adversely to the resting order.
The most important feature of D-Limit is that it is usable by all participants. D-Limit is designed as a non-discriminatory solution for any participant that wants to display an order and receive a high-quality fill. It is designed to equally benefit fundamental, long-term investors, their brokers, and market makers that are subject to outsized risk when quoting in hundreds of stocks. In providing this benefit to individual participants, it will contribute to healthy pre-trade competition, transparency, and fair and efficient markets.
Below, we examine how the markets got to this point and how D-Limit provides a targeted and equitable solution.
Price-time priority is a good way to incentivize market participants to compete with one another on best price, as it rewards those willing to be first at the best price. However, we believe the race to compete in price-time has generated negative side effects. Strict price-time priority can disincentivize some market participants — potentially all but the fastest ones — to participate in the race to display and manage liquidity at the top of order book queues. The importance of speed in this race, and in general, enables exchanges to extract monopolistic rents for latency reducing products.
The problem with latency arbitrage is that microseconds determine winners and losers. Finding edge requires traders to have the very fastest (i.e., most expensive) connections and data feeds, have world-class technology acumen, and use esoteric order types such as a “post-only day ISO order.” As we demonstrate with data, very few firms have come to dominate this space, and most brokers acting on behalf of a customer/investor are at a disadvantage. They are simply not built to win this contest, much like me stepping into the Octagon to face off against Conor McGregor — the agency broker and I will lose 1,000 out of 1,000 times.
Realizing how difficult and expensive it is to compete on the microsecond and even nanosecond timescales, many investors and brokers have looked for alternative ways to source liquidity. They are forced to either trade dark (there’s been far more performance-focused innovation in non-displayed trading than displayed), pay up to jump the intermarket queue by posting on inverted exchanges, or just accept the very high risk of poor results from posting displayed on exchange.
All of these outcomes have deteriorated the state of displayed markets. The question becomes, how can you level the playing field for displayed trading in a fair and transparent way?
It turns out that that poor performance of displayed resting orders doesn’t apply to every lit trade.
As I’ve written about previously, there are tiny increments of time — we call them periods when the quote is “crumbling” — when it is possible to predict, with a fairly high degree of certainty, that a price is about to change. Some of the fastest traders have the ability to make these predictions and trade against slower participants who don’t know that they are trading at a soon-to-be stale (aka bad) price.
We developed the Signal to identify these “crumbling quotes,” and level out that informational advantage for all IEX participants. The periods when the Signal predicts that the price is about to change (i.e., when the Signal is “on”) amount to only an average of 4 seconds per symbol per day on a volume-weighted basis, or 0.02% of the trading day, but when looking at performance statistics, these are critical moments.
So far, we’ve integrated the Signal into non-displayed order types on IEX, particularly our signature order type, D-Peg, which provides excellent performance for pegged non-displayed orders. But what we’ve found is that displayed orders are also extremely vulnerable during these tiny periods when only some participants know that the price is likely to change.
Markouts of displayed orders adding liquidity while the Signal is “on” are truly terrible on average.
For resting orders, trades that execute while the Signal is “on” underperform those when the Signal is “off” by $.0077 per share just 1 millisecond after the trade. And that underperformance persists — even 5 minutes after the trade, they underperform by $.0060 per share.
Additionally, while the Signal is only on for a tiny portion of the day, the scale of the problem is huge.Fully 24% of displayed order volume on IEX trades while the Signal is “on.” Those orders are subject to significant adverse selection as shown by the chart above. Compare that to D-Peg and P-Peg orders on IEX, which have functionality that is designed to protect them while the Signal is “on.” Only 0.17% of their volume (17 basis points) occurs while the Signal is “on”!
As I wrote in the beginning of this post, D-Limit is a simple solution designed to enhance performance for better displayed trading results. In particular, D-Limit benefits from being:
We believe that D-Limit will promote competition and improve public price discovery by broadening the set of firms who can compete in displayed trading. It’s a big ambition — and one that we’re excited to take on.
The D-Limit rule filing is publicly available on our website. Please reach out anytime with any questions on D-Limit or any other aspect of IEX’s design.
 Repricing only applies to D-Limit orders whose booked price is equal to or more aggressive than the price which is predicted to crumble. For example, an order to buy $.01 below the NBB will not be repriced if the NBB is predicted to crumble.
 D-Limit orders, modifies, and cancellations all must traverse the 350 microsecond speedbump like every other type on IEX.
 “A Deliberate Strategy”: Three IEX member firms are responsible for 55% of all removing volume versus displayed orders that occurs while the Signal is “on.”
 Accepting rebates to compensate for adverse selection or paying to jump the intermarket queue for a higher quality fill both create separate information leakage concerns not covered in this blog.
 We expect that subsequent versions of the Signal will be on for longer periods of time (subject to SEC filing and effectiveness), but even if the Signal were to be on for 10 seconds per symbol per day that would be a small fraction of the 23,400-seconds-long trading day.
 Discussed in various places, including our recent blog post on D-Peg markouts.
 All data referenced is from November 2019 unless otherwise noted.