Inside Last Monday’s ETF Crash—–SunGard’s Alleged ‘Equipment Failure’

Earlier today, we noted that BNY Mellon and SunGard were, as of Sunday evening, still attempting to sort out exactly what happened last Monday when suddenly, a system “glitch” caused widespread errors in calculating NAV for some 1,200 ETFs and mutual funds.

For its part, SunGard claims the problem has nothing to do with the flash-crashing mayhem that unfolded across US equity markets last Monday morning and everything to do with a “corruption” that apparently occurred last Saturday when someone tried to do a system upgrade. As we said earlier, “Bank of New York Mellon can’t simply come out and say that broken markets broke its accounting software because that would be to place the blame squarely where it belongs and everyone knows that is a very dangerous thing to do.”

Of course our guess is that in one way or another, broken markets were in fact the culprit here and SunGard’s system “corruption” was just a casualty of the corruption of capital markets in general.

As we said last week, no one will ever know what exactly happened to cause the ETF pricing model to break down but if one had to venture a guess, it might involve “market makers simply walking away or else putting in absurdly low bids in order to avoid getting steamrolled when the constituent stocks came off halt [and/or] ‘liquidity providers’ acting in ways that neither provide liquidity or appear to emanate from human traders, all of which ultimately conspired to cause market orders to hit absurdly low bids, which in turn served to take out stops, and somewhere amid the rampant confusion, Bank of New York Mellon’s/ SunGard’s platform simply malfunctioned.”

Or something. But who knows. As Themis Trading’s Joe Salluzi put it, all we know for sure is that “something went wrong here. Somewhere along the way, the ETF pricing model was broken today.”

One explanation that in many ways mirrors what we said above comes from FactSet’s Director of ETFs Dave Nadig. Here’s what Dave thinks might have happened, although he, like us, realizes that at a certain level this is all just “voodoo tea-leaf reading.”

*  *  *

From FactSet

So what happened? Let’s go in time series order.


On a normal day, the designated market makers at the NYSE fire pre-opening indications of where they intend to start making a market – what their best bid and offer is going to be. Because of the chaos of Friday’s trading, the NYSE opted to invoke the rarely used “Rule 48.” Rule 48 effectively lets the DMMs not tell anyone where things are going to open until they start trading, removing a level of information from the market.

But remember, not every trade has to go through the market maker, and not every trade has to go through the exchange. So before the open, we had bids and asks being put up on Nasdaq in the range of $70.50 to buy and $74.19 to sell – a very wide opening spread – which implies to me that the market makers were either waiting for the dust to settle or very nervous about the opening trade.

The Open

The market opened with an opening cross of some 70k shares at $72.52 on NYSE, right in the middle of that advertised spread and probably close to the fair value as best anyone could guess (chaos was reigning in S&P 500 stocks at the same time, so it’s a bit of guesswork). That opening auction seems to have gone off without a hitch and pretty rationally.

But that’s when things get squirrelly. While technically trades were occurring right from the start, the best bid/ask was from Nasdaq for the first 20 seconds or so of trading and very wide: $59.22 to buy and somewhere between $65 and $75 to sell.

In those opening seconds, roughly 100,000 shares changed hands in very small lots (many under 100 shares, many of a single share) between $74 and $65. That lower boundary trading triggered the first of 10 halts.

The Halts

Between 9:30:28 and 10:30:06, nearly a full hour, trading in RSP only occurred in 15 to 30 second bursts between halts. The current single stock circuit breaker rule is that any move of more than 10% within a five minute window triggers a halt. So trading halted four times on down moves and six times on up moves as it recovered.

So What?

Here’s my interpretation of what happened, recognizing that this is voodoo tea-leaf reading.

  1. Market makers stepped away. Market makers are in the business of avoiding risk. When irrational actors are at work (and let’s be clear, anyone selling RSP for half off isn’t being rational), the smartest thing to do, if you’re the actual human being watching RSP trade, is to simply step out for a cup of coffee. That means stubbing out your bid and ask to a point you think nobody will hit.
  2. HFT algorithmic trading was a big part of what we saw. I say this for a few reasons. Firstly, no human being trades one share of anything. Second, about half the trades were flagged as intermarket sweep orders (ISOs), the common form of HFT trade which values speed over price. Mom-and-pop investors aren’t flagging their 80 share trades as ISOs, in my experience. Third, someone selling RSP for $50 when it was demonstrably worth $70 is irrational and working against their profit motive. Since I believe human traders generally express their irrationality more subtly, I think this points to broken HFT models that were unloading positions without regard to price or fair value.
  3. Sleeping-stops kill. Of course, not every trade was an HFT algorithm. To the extent any “regular” investor was hurt in RSP during the hour of doom, it was likely from a stop-loss order they had sitting on the books. If you had a stop to sell if RSP crossed under $70, say, right after the open, your order became a market order, and you could have been executed anywhere in the line of terrible trades.
  4. The market structure worked. I doubt any of these trades will be canceled. They happened in an orderly, logical fashion (just at stupid prices). The invocation of Rule 48 worked, in that the market makers at the NYSE were seemingly able to step out of the way and let the market come in and fill up the book before they had to act and advertise their own prices. The halts did exactly what they were supposed to do. You can complain all you want about what happened, but you can’t claim the system broke.
  5. There are still real issues. I’m unconvinced that the current market structure is actually well-suited to this kind of activity. This is a market where information is processed in milliseconds, yet we have five-minute trading halts. This is a market that increasingly thrives on complete transparency, yet the exchange rule for dealing with volatility is to actually remove transparency. This is a market where speed is often valued over price—and we have an entire set of regulations to ensure that’s OK—yet we cry “foul” when speed is actually more important than price. Lastly, if we’re concerned about protecting individual investors, we need better education and controls around the most dangerous tool in the ETF investor’s toolbox: the market order.

*  *  *

Bonus: consider the above and then consider the following chart from Nanex