Should IEX Become a Full Exchange? Incentives for Exchanges to Compete by Increasing Delays

Startup IEX Group has announced its imminent transition to a lit venue. [1] I’ve been reluctant to wade into the highly politicized waters surrounding IEX, but nobody else has publicly questioned them on a potential flaw in their market model should they become a lit exchange. [2]

John McCrank touched on this concern last summer. For exchanges with fully protected quotes, the regulation (see parts with “immediately and automatically”) very explicitly prohibits a deliberate delay like the ‘magic shoebox’ IEX employs. [3] Something akin to this delay was proposed and rejected a few years ago on a minor exchange operated by Nasdaq, PSX. Evidently the SEC is reconsidering its judgement on that:

The source familiar with the SEC’s thinking on the matter said PSX may have been ahead of its time. The regulator now will likely look at the IEX case with the perspective that Reg NMS was written some 10 years ago, the source said.

Now, there are some important differences between IEX’s model and what PSX tried to do. But I get the sense from the above that those differences are not what’s changing the SEC’s thinking. Of course, questioning regulation that may be outdated is a good thing, but it’s important to question it in a sensible way. Allowing quotes to have a delay seems perfectly harmless, until you consider making them displayed, protected quotes. To see why, it’s useful to consider the analogy with liquidity-provider rebates.

Liquidity-provider rebates are controversial for a variety of reasons. One of those reasons is that, in combination with the order protection rule, they can create a ‘race to the bottom’ where exchanges are compelled by competitive forces to have higher and higher rebates. Imagine a world where Exchange A has a $1.00 per share rebate and commensurate access fee. Exchange B has no rebate and no access fee. Say the market values QCOM stock at $50 and an investor is trying to sell 100 shares of it. Maybe there’s a bid on Exchange B for $50.00. On Exchange A, however, the bids are about $50.99 – because the bidder will be paid an extra dollar upon execution. Under the order protection rule, Exchange B is nominally offering the best bid so the investor’s sell order *must* be routed to Exchange B – even if the investor’s effective sell price after fees will be $49.99, a penny worse than the price available on Exchange A. Thus, in order to maintain market share, exchanges would continually raise their rebates to be greater than those offered by competitors. It’s not hard to see how this would spiral out of control. The SEC placed a 0.3 cent cap on access fees partly for this reason and there has been plenty of talk that even this level is too high.

I’d argue that adding an intentional delay into an exchange is the temporal version of the price distortion caused by high rebates. Consider what would happen if an Exchange implemented a 10 second delay. Any time the market moved, all incoming orders (on any exchange, marketable or otherwise) would have to be routed to the stale, inaccessible quotes displayed by the delayed exchange. [4] If the traders on the delayed exchange who deleted those now-stale quotes changed their minds and quickly re-submitted orders, they might get executed against some of those incoming orders that were required to route to the delayed exchange. Much of the time, the routed-in orders would not get executed and would have to be repeatedly tried again. The result being mostly that investors would need to wait longer for orders to get executed at worse prices. But, as a side-effect of the routing, the delayed exchange would gain market share. [5] Other exchanges, in response, would be spurred to institute their own delays, making the market completely non-functional.

Thinking of our analogy with access fees, disaster could be averted if a cap on the additional delay were established. What should that cap be? My intuition is that the cap should be, at most, 10% of the typical time it takes for orders to be routed. It’s no coincidence that this is right around the level of IEX’s delay. IEX states that the delay allows their system to update its view of the NBBO. [6] The NBBO (at its fastest) generally changes at about the routing timescale (which is complicated, but around 2ms [7]). And a large chunk of the routing timescale is due to exchanges deciding what to do with an order based on the prices they see on other exchanges. This process is similar to what IEX needs to do when updating their view of the NBBO, except IEX does not need to spend time assembling an order to send to another exchange.

I don’t want to get into the actual value of the delay. But I will say that an exchange with good technology should not have trouble updating their view of the market much, much faster than 350us. I have, on rare occasions, seen orders executing at the old NBBO after a single tick up or down. [8] This has always been on dark pools though; I’ve never seen it on a major exchange. Goldman’s Sigma-X was fined for this very reason, so it’s not just an imaginary issue. IEX rightfully wants to ensure that this won’t happen to their customers. That said, Goldman updated their system and fixed the issue. I imagine IEX could do the same. [9]

[1] IEX previously had a soft target of launching in Q2 of 2014. Their public comments and website suggest that they will succeed in their goal for a Q2 2015 launch.

[2] I want to re-iterate a disclaimer that applies to this entire blog. Nothing on this blog should be construed as trading or investment advice. This blog is the product of analysis that may contain errors. And this blog does not reflect any views but my own personal views.

[3] Adding an intentional delay is, to the best of my knowledge, perfectly compliant for a dark pool, which IEX currently is.

[4] Technically incoming orders could instead be rejected, have their price slid, turned into hidden orders, delayed in some fashion, or some combination of the four.

[5] The reasons why, as with a lot of microstructure, are complicated. Here is an illustrative example that shows how an intentional delay can both harm customers and increase market share for IEX, should they become a lit exchange with fully protected quotes.

1.  Nasdaq and IEX both show quotes for a stock at a bid/offer of 10.00/10.01.
2.  The IEX offer of 10.01 is deleted, but still shows as visible in the public data feed until its additional delay has been completed.
3.  A large IOC buy order comes in to Nasdaq at 10.01, clearing out the resting Nasdaq offer of 10.01. The remaining quantity from this buy order starts going through Nasdaq’s routing software.
4.  A small IEX offer of 10.01 is re-submitted, but is pending their additional delay.
5.  The original IEX offer disappears from the public data feed.
6.  A new Nasdaq offer of 10.01 is re-submitted.
7.  The new IEX offer of 10.01 has finished going through its additional delay.
8.  The original routed bid of 10.01 from Nasdaq finally has finished making its way to IEX and finished its additional delay, partially executing against the newly added small order resting on IEX.
9.  The original routed bid was only partially filled and, being an IOC order, leaves the customer needing to send a new order to finish their execution. The customer does so, but by the time they do the previously available Nasdaq offer has been deleted.

This is just one problematic example. Even though it is complex and 9 steps long, it could occur quite often. The pattern of cancellations, re-submissions, and large incoming marketable orders is characteristic of equity markets.

[6] From IEX:

Most importantly, the (patent pending) IEX POP architecture introduces 350 microseconds of latency for any order action taken by a participant (enter, cancel, revise an order), which ensures the IEX Matching Engine has sufficient time to receive and process changes to the NBBO before a fast participant can receive and act upon the change on IEX.

 

[7] It’s hard to find a public source for this number. One way to see the minimum timescale for routing and inter-exchange synchronization is to look at how long it takes for orders to start executing 1 cent through the previous NBBO after an incoming aggressive trade. You can see from this plot (from an earlier post) that it is several milliseconds after a trade before any trades at a more aggressive price occur.

[8] These executions are technically compliant, if the market is 10.00/10.01 and a new bid locks the NBBO, making it 10.01/10.01, then an exchange is perfectly entitled to execute a resting sell order at a price of 10.01. Many exchanges have a policy (sometimes enabled at the customer’s discretion) of not executing orders at times like these. On very rare occasions, I recall seeing orders execute even outside of the *old* NBBO. I couldn’t begin to tell you why that should happen, but I’ve only seen it on dark pools with pretty pathetic volume. It’s possible that a non-compliant execution of this sort is what caused the original customer complaints about Sigma-X.

[9] Or at least upgrade their system so it can accurately determine the NBBO in less than 100us. They could then cut their ‘magic shoebox’ delay from 350us to 100us, which feels like a less disruptive level to me.