The U.S. Securities and Exchange Commission (SEC) is openly looking at changing the order protection rule (OPR, Rule 611).
We recently looked at how often markets are actually locked and crossed. It was rare, and extremely short lived.
Today, we look at how often trades print to the Securities Information Processor (SIP) at a price that looks like a trade-through. It’s also rare, and perhaps mostly due to the time it takes to route trades to the SIP.
What are we talking about?
When you look at the SIP, you do often see trades arrive at the tape, at old (stale) prices, after the National Best Bid and Offer (NBBO) has changed.
In theory, if the NBBO is “protected,” we should never see a trade happen at a worse price than the NBBO. So, what is going on?
Chart 1: In many markets, trades hit the consolidated tape after the BBO changes
The chart above, from a study in Europe, shows what we are talking about. In that study, the trades were found to mostly be latency arbitrage opportunities, where orders hitting the dark pool were routed using faster microwave links.
However, this could also happen because of the physics of routing orders – in short, the midpoint fill was “traveling” to the consolidated tape as the BBO updated. That’s one reason why physics makes a pre-trade consolidated tape impossible.
Only a fraction of midpoints print after the NBBO has changed
The first question we ask is how often does this actually happen in the U.S.?
And the answer is almost never.
As the data in Chart 2 shows, around 19.4% of all off-exchange trades occur at midpoint. However, only 0.5% are at “stale” midpoints – where the NBBO has already changed – and almost none of those are reported outside the NBBO.
Chart 2: Midpoint trades are almost 18% of all off exchange trades, only 0.04% print through the NBBO
Did these trades occur before dark pool saw the NBBO change?
The next question is how “stale” are these trade prints?
If we track all the Trade Reporting Facility prints that occur at the “old” NBBO mid, after the NBBO has changed, we see almost all arrive within 2,500 microseconds (2.5 milliseconds). To put this in perspective, this is all pretty fast (we blink in 250 milliseconds), but it’s not faster than the speed of light.
Chart 3: Distribution of “stale” mid-point prints from off-exchange venues
The fact that there is elevated activity right after a quote changes seems to confirm that dark pool midpoints get actively probed for liquidity at the same time as the lit quote is removed.
You may remember our prior study where we showed how trades speed between venues (and the SIP) and found that trades and quote updates acted, traveled and reacted in a flurry – over about 1 millisecond. So, it is quite possible that some of the “stale prints” actually traded before the dark pool knew the NBBO had changed.
Consider this example:
Looking only at Nasdaq listings, which means the trades have to report to SIP at Carteret:
- We assume that most dark pool trades happen at data centers in Secaucus (Form ATS-Ns show where this is true).
- We also assume all orders and SIP messages travel by fiber (which is most conservative).
- Finally, for simplicity in this thought experiment, we assume the dark pools are using SIP feeds not direct feeds for their NBBO updates (although we know that’s not true for all, this does create the longest reaction window).
Once the SIP NBBO updates, it will take around 143 microseconds by fiber for the NBBO update to arrive at the dark pool in Secaucus to update the midpoint price in the dark pool.
If a trade happened right before that, it would still take another 143 microseconds by fiber for that “late” trade to arrive at the SIP in Carteret. We also need to consider some compute times — as a guide, the SIP takes around 15 microseconds to process quote changes.
That represents a round-trip time of roughly 300 microseconds, which we shade blue in the chart above. We can see that the window includes a lot of late-arriving trades, but not nearly all.
Stale trades happen, whether there is latency arb is harder to tell
In reality, all of these prints are a lot faster than the amount of time dark pools are allowed to wait to post a trade (which is 10 seconds).
But it’s clear some of the reports are a long time after the 300 microseconds that the NBBO update should take to travel to the dark pool, process and return.
What we can see is stale trade-prints happen, even with OPR in place. Whether there is latency arbitrage occurring is harder to tell.
