Something something Cyber Monday something something.

Apple Had a Rough Morning

Matt Levine is a Bloomberg View columnist. He was an editor of Dealbreaker, an investment banker at Goldman Sachs, a mergers and acquisitions lawyer at Wachtell, Lipton, Rosen & Katz and a clerk for the U.S. Court of Appeals for the Third Circuit.
Read More.
a | A

Apple's stock went down a lot between 9:50 and 9:51 a.m. today, and then went up again, though it still closed down more than 3 percent. If you want, you can probably find a bunch of explanations for why Apple went down, though none more poignant than this:

Lou Basenese, founder of Disruptive Tech Research, said that while he's not hearing anything specific, the most reasonable explanation is that investors are just ready to cash out.

"I think the most logical explanation is profit taking. Shares were up about 25 percent off the October lows, compared to an 10 percent move for the Nasdaq," Basenese said.

He also added that an upgrade Barclays made on Monday could also be having a reverse effect.

"Today's price target increase from $120 to $140 at Barclays might actually be negatively impacting the stock, as investors fear analysts are getting too bullish," Basenese said.

Why did Apple's stock go down today? Maybe because it went up too much last month! Or maybe because some people thought that other people thought it would go up too much this month. Or both. Or something else. But nothing specific. This tech research left me feeling pretty disrupted.

Of course I don't know why Apple went down either. On the other hand it's reasonable to say that it went down too much. Here's a chart:

Between 9:55 a.m. and the close today, Apple traded between $114.42 and $116.50, a $2.08 range, and closed at $115.07. Between the open and 9:50, it traded between $116.78 and $119.25, a $2.47 range. But from 9:50 to 9:51, it fell from a high of $117 to a low of $111.27, a range of $5.73, moving more than twice as much in that one minute as it did during the rest of the day. Or half a minute, really; the first 30 seconds of 9:50 were a relative snooze.

So what happened? Well let's posit that a Bad Thing happened to Apple -- a popular culprit is a Morgan Stanley recommendation to cut Apple exposure, though, meh -- and so some people wanted to sell. (Or, a Bad Thing happened to an Apple investor, who then needed to sell. Or, a Bad Thing happened on an Apple investor's keyboard, triggering a big accidental sell order. Etc.) When some people want to sell, market makers buy. That is the job of a market maker: They provide liquidity, selling when others buy and buying when others sell. So the initial effect is that market makers dampen volatility, stepping in to buy when others sell. "Market makers," in the modern equities market, is of course a euphemism for high-frequency trading firms.

But market makers only buy so much. If Apple is trading at $118, and you can buy 100 shares at $117.95, you probably should, because odds are it will go back to $118. And if you can buy 200 more shares at $117.90, hey, sure, why not? Another 200 shares at $117.85? Sure, sure. You keep buying for a while, but at some point you have several thousand shares and the stock is at $117.65 and you start getting worried. You've got kind of a lot of shares, for one thing, and you probably have some position limits designed to limit your risk. For another thing: You've lost money on all the trades you've done so far. So maybe you stop buying.

And so you stop dampening volatility and the price starts falling faster. And so it gets down to $117, and now things are awkward. You've lost several thousand dollars on your Apple trades. Maybe you should cut your losses and get out? Again, you are not, like, pondering this in your heart of hearts: You are an algorithm, and you are programmed with some loss limits, so you cut your losses and start selling. So instead of dampening volatility, you actually start increasing it.

One criticism of modern markets is that modern market makers -- high-frequency traders -- aren't particularly brave about providing liquidity and dampening volatility. In the story above, that looks like: HFTs are quick to stop buying, and also quick to start dumping shares again. Is that true? Meh, probably? Remember that high-frequency trading is a probabilistic business with pretty low margins. Here's a paper about Virtu Financial's business. Virtu, remember, is the HFT market maker that lost money on just one out of 1,278 trading days. But it makes an average of $440,000 a day trading U.S. equities. If you divide that among just S&P 500 stocks -- and Virtu trades a lot more than just S&P 500 stocks -- then that's around $900 a day in each stock. If your expected daily profit in Apple is only $900, then losing a couple of thousand dollars in a minute seems like a pretty bad result, certainly enough to start a panic in your little algorithm brain. So, sure, why wouldn't you cut your losses pretty fast?

On the other hand, as it happens, here is a new paper from four finance professors finding that "during extreme price movements high-frequency traders act as net liquidity suppliers, while non-high-frequency traders act as net liquidity demanders." This too makes intuitive sense: Even in the scenario I described above, in the worst case, the high-frequency trader is a net liquidity nothing. It buys and provides liquidity, then does nothing, then eventually sells and takes liquidity, but on net it's sold no more shares than it bought. In the not-worst case, you'd assume that high-frequency traders as a class do dampen volatility.

But that doesn't necessarily mean that they dampen it all that much. Something like $767 million worth of Apple stock traded in the one minute between 9:50 and 9:51 this morning, and if you are making $900 a day on Apple stock you probably aren't buying $700 million of it all at once.

One other volatility enhancer is stop orders:

Declines in Apple may have accelerated today when the shares reached prices at which investors had placed automatic instructions to sell, known as stop-loss orders, according to Michael Block, chief equity strategist at Rhino Trading Partners LLC.

“It looks like some big stops went off,” Block said in an e-mail. “Looking at the chart, $117 looks like someone’s key level here. When something like that accelerates it sets off more stops and then others take notice. It causes a lot of damage.”

So the market makers get out of the way, the stops get triggered and it's a mess. Down we go, down down down, pretty much all at once, to $111.27. We pause there for a few seconds, and then the recovery slowly begins.

What is special about $111.27? Plenty, as it happens. That was the lowest price at which Apple was allowed to trade at the time, under the stock exchanges' "limit-up/limit-down" rules, which basically prevent quick moves of more than 5 percent at a time. So $111.27 is the lowest price it traded at. If you came to the stock exchanges and bid $111.27 for Apple stock, you could buy all you wanted. If you bid $111.26, you couldn't. If you wanted to sell, you had to find a buyer at $111.27 or higher. If you wanted to sell at $111.26, and someone wanted to buy at $111.26, you were both out of luck; the exchanges would ignore your orders. If there were no trades at or above $111.27 for 15 seconds, then there would be a five-minute pause followed by an auction to restart trading.

But there were trades at $111.27: Any time someone came in with a bid to buy shares at $111.27, that bid got hit and the shares traded at $111.27. When there was no $111.27 bid, there was no trading. The stock bounced around there for a little while, and then higher bids came in and the stock righted itself.

What does this tell you? The main thing it tells you is that whoever was selling on the way down from $117 to $111.27 kept selling at $111.27, and would probably have been happy to keep selling at $111.25 or $111 or $108 or who knows how low. (It seems unlikely that there were big limit sell orders at $111.27.) So the limit-up/limit-down rules helped that seller, anyway, who otherwise would have sold a lot of shares at a big loss but instead sold fewer shares at a smaller loss.

And since the chart looks like what it looks like, that means that the limit-up/limit-down rules "worked": The "right" price for Apple during that weird minute might have been $114 or $115 or $116, but it surely wasn't $110. You couldn't necessarily have known that at the time, but by five minutes later it was obvious. The price rebounded quickly, and $111.27 was just an unpleasant memory. It would be hard to argue that there was a fundamental justification for Apple to be at $111.27 at 9:50, when it was $5 higher a minute earlier and $2 higher 30 seconds later, with no news on either side.

So anyone who had sold at, say, $108 would feel aggrieved. There's a technical term for that grievance: In previous "flash crashes," stock exchanges have broken trades that were "clearly erroneous." This has always struck me as strange: If a trade was "clearly erroneous" at the time it happened, then the exchanges shouldn't have let it happen, rather than going back later and breaking it. Here, the system worked: The exchanges didn't let bad trades in Apple happen, so there were no "clearly erroneous" trades to break later.

Of course the right price wasn't really $111.27 either, so the rules didn't work perfectly. The bands could have been tighter; if the stock's fall was halted at $112.27 or $113.27 then that probably would have been just as good for price discovery and market efficiency and all that. But that's just true ex post; you couldn't have known it at the time. Maybe Apple really was on the way down for fundamental reasons. Maybe here a circuit-breaker at 3 percent would have led to "better" prices, but in other cases 7 or 10 percent would be better. Five percent is just the rough compromise that the exchanges came to.

The story is, there's no story. Perhaps there was a fundamental catalyst that kicked off this move, or a technical catalyst, or just a fat finger. And then market dynamics took over: The equity market is set up to dampen small movements (market makers tend to buy when others are selling), to exacerbate mid-size movements (as market makers capitulate and stops get triggered), and to dampen large movements (as circuit breakers cut off trading and let algorithms catch their breath). All of those things seem to have happened here, in the space of a minute. You can see why profit taking might be a more satisfying explanation.

  1. Though consider the messenger:

    “While we plan on buying several Apple Watches, the stock has nearly doubled over the past three years and our… simulations have consistently recommended a modest reduction,” the bank said.

    I for one will not be taking investment advice from someone who's planning to buy several Apple Watches.

  2. Is that true? It's pretty much the foundation of market making.

  3. As in so many financial things, there is a trade-off between usefulness and risk. A market maker who takes a lot of risk is generally more useful in dampening volatility than one who doesn't. But a market maker who blows up and goes bankrupt is very, very bad. So from a social perspective you should want high-frequency trading firms to take enough risk to support markets but not enough to put themselves in danger, which is a non-trivial balance to strike.

  4. Obviously this example is very stylized, but it could go either way. The algorithm probably doesn't just turn around and start dumping shares, so it might end up net long (and a net liquidity provider). On the other hand it's programmed to make money, not to provide liquidity, so in the right circumstances you could imagine it net short too. It's just an empirical question which one wins out, and the answer seems to be that liquidity provision does.

  5. Eric Scott Hunsader pointed this out on Twitter. Here is Nasdaq's explanation of limit-up/limit-down, and it is not, like, super breezy and straightforward. Basically you can't do a trade more than 5 percent away from a reference price that is the five-minute rolling average price, updated at certain intervals. Between 9:50:28 and 9:50:58 today, when Apple was at its low point, the limit-down level was $111.27.

  6. What about the buyers? Why would you rest a buy order at the limit-down level? If that order gets hit, either you'll make out great in a situation like this, or you'll catch a falling knife on a company that is trading way down for fundamental reasons. Seems like a big risk. Two answers:

    • First, registered market makers (HFT firms) are required to rest orders inside the limit-up/limit-down levels. (See, e.g., Rule 11.8 of the BATS rulebook). This supports the price when it gets near the limit-up/limit-down levels, but the required quotes seem to be well inside of those levels, so it's not like there were a bunch of regulatorily-required quotes resting at $111.27 for Apple.
    • Second ... there mostly weren't resting orders. Basically the stock hit $111.27 and stopped, and then some people started coming in offering to buy at the minimum price. The theory I suppose is: The stock has hit the limit-down, you look around with your little algorithm eyes and see what's going on, and you find nothing. The broad market hasn't crashed, there's no obvious fundamental news, so it's reasonable to think that it's just a flash crash and will quickly recover. So you start the recovery.

    So this isn't a story of the stock dropping to $111.27 and then finding a bunch of resting buy orders that kept it at $111.27. Rather, it's a story of the stock dropping to $111.27 and then sitting in limbo because there were no legally executable buy orders. And then, as nothing happened, people (algorithms) were emboldened to throw in some legally executable buy orders. And it slowly recovered from there. Of course all of this is over the space of a few seconds, but the algorithms were thinking deep thoughts in those seconds.

  7. From Bloomberg News:

    None of the major stock exchange operators reported clearly erroneous trades following the move in Apple. As of 11:55 a.m. New York time, neither Nasdaq OMX Group Inc., NYSE Group or Bats Global Global Markets had received any such filing, which is made when a broker mistakenly executes trades and wants to cancel them.

    There are numerical standards for "clearly erroneous" trades, and those standards are quite sensibly wider than the limit-up/limit-down standards, so they mostly shouldn't be triggered when limit-up/limit-down is working. But the rules fit together in somewhat messy ways, so that isn't guaranteed.

This column does not necessarily reflect the opinion of Bloomberg View's editorial board or Bloomberg LP, its owners and investors.

To contact the author on this story:
Matthew S Levine at mlevine51@bloomberg.net

To contact the editor on this story:
Zara Kessler at zkessler@bloomberg.net