Speed Bumps Are the Hot New Thing for Exchanges
Everyone knows the story of IEX by now, but let's tell it again anyway. Some investors were sad. They were sad because they wanted to buy some stock, and their computers told them that there were 5,000 shares of the stock available on different exchanges at a price of $20 each, but when they sent an order to buy those 5,000 shares, they only got like 100. The rest instantly, magically, flickeringly disappeared, only to reappear priced at, like, $20.05, or $20.20, or $21. The investors were being front-run! High-frequency trading firms saw their order execute on one exchange, and immediately raced ahead of them to raise the prices on all the other exchanges. Those 5,000 shares at $20 were a mirage, just "phantom liquidity."
So IEX built a new kind of trading venue that fixed this. The way it fixed it is with a "speed bump." Communications to and from IEX are slowed down by a tiny little bit: When you send an order to IEX, or when you get a confirmation back that your order has executed, IEX delays the message by 350 microseconds. If you give IEX an order to buy 5,000 shares, and there are 500 shares available on IEX at $20.00 and 4,500 available elsewhere at $20.00, IEX will sell you the 500 shares at $20.00 and route the rest of your order to the other exchanges. But! The people who sold you the 500 shares on IEX won't find out for 350 microseconds, so they won't be able to race ahead of you to the other exchanges to jack up the price. You will get real liquidity, not the bogus phantom liquidity offered elsewhere.
That's more or less the most popular version of the IEX story, though it isn't necessarily accurate in every particular. I think it captures the right feeling, though. You can find some particulars in the usual place for particulars, 1 but feel free to skip them, because this post is not about IEX.
Instead it's about the Chicago Stock Exchange, which announced its own proposed speed bump this week:
The Chicago Stock Exchange just revealed plans for a 350-microsecond speed bump, similar to the famous delay on IEX Group Inc.’s Investors Exchange, which started trading last week as the 13th official U.S. stock exchange. The idea is to make trading more fair by blunting the advantage of some of the fastest traders.
But there’s a key difference: Whereas IEX slows down everyone using its exchange, the Chicago market will only slow down traders who take liquidity -- that is, those who trade against standing orders posted by market makers.
The Chicago proposal to the Securities and Exchange Commission involves a 350-microsecond delay in processing orders "that could immediately execute against one or more resting orders on the CHX book." Non-marketable limit orders, "and cancel messages for resting orders, would be immediately processed without delay." So if you are a market maker 2 on the Chicago exchange, posting bids and offers for stocks and hoping someone will come trade with you, you will still be able to quote prices and update your quotes without delay. But if you are one of the people trying to trade with those market makers -- if you are trying to take liquidity -- then you'll have to wait 350 microseconds.
This works kind of like a "last look" for the people posting the orders: If you are a market maker, and you have put out a bid to buy stock, and someone comes to the exchange to sell to you at that price, you get 350 microseconds to change your mind before you have to buy. If the stock goes down in those intervening 350 microseconds, you can cancel your bid. Unlike the true last look in some foreign exchange markets, here you don't know there's a matching sell order during the 350 microseconds, so you'll only change your mind if information from elsewhere suggests that the price is going down and your bid is stale. It's sort of a blind last look. Still pretty helpful, if you're a market maker! 3
The exchange says that this "Liquidity Taking Access Delay" will discourage "latency arbitrage" and "encourage liquidity providing market participants to make tighter and deeper markets, which is good for the investing public." Also it will help Chicago compete for trading in the SPDR S&P 500 exchange-traded fund:
In a filing made with the SEC, the exchange said its speed-bump proposal is a “direct response” to problems that have dented its market share in the SPDR S&P 500 ETF, which is often the most frequently traded security in the U.S. equity market. The exchange’s share of trading the ETF, known as the SPY, fell to 0.57% in July from 5.73% in January, according to its proposal sent to the SEC.
The exchange said market makers have become less willing to trade in Chicago because faster traders have been able to pick off their orders at stale prices.
"Pick off their orders at stale prices" means: Market makers on the Chicago exchange would post an offer to sell the SPY ETF for $218, and someone would buy it from them at $218, and the market makers would regret it. One reason to regret it is that, after the market maker sold SPY at $218, the price went up to $218.01. Another reason to regret it is that, before the market maker sold SPY at $218, the price went up to $218.01. In the sub-millisecond world we're talking about, those two reasons are almost the same reason. 4 In any case, with the new proposed delay, the market makers will have a whole luxurious 350 microseconds to see whether the price goes up to $218.01 before they have to sell at $218.
One thing to notice here is that this is the opposite of the IEX story. 5 In the IEX story, investors were sad because they wanted to buy all the stock at the price that was displayed, and sent out orders to buy all that stock at the displayed price, and couldn't -- because high-speed traders reacted to trades on one exchange by pulling their sell orders from other exchanges before the investors' buy orders could be fully executed. In the Chicago story, market makers are sad because high-speed traders tried to buy stock from them at the price they had displayed, and could -- because the market makers couldn't react to trades on other exchanges by pulling their sell orders from the Chicago exchange fast enough. The IEX story is about the sadness of being unable to get all the stock that you want before the price reacts to your demand. The Chicago story is about the sadness of being unable to react to demand before selling all the stock that was demanded.
I suppose a possible takeaway here is that someone is wrong, or disingenuous: Perhaps the Chicago Stock Exchange (or IEX) is really trying to stop latency arbitrage, while IEX (or Chicago) is secretly trying to promote it. But that seems like an overly literal reading, premised as it is on the idea that "latency arbitrage" is an identifiable thing that is bad for an identifiable reason. But "latency arbitrage" seems mostly to mean fast trading that goes poorly for you, and that can happen to anyone. 6 One can have sympathy for everyone here. Market makers want to be able to react to information; they don't want to be "picked off." Big investors want to be able to pick off market makers, to buy a lot of stock before market makers can react to their orders.
Chicago's innovation is good for market makers -- it gives them a quasi-last look -- and so it should encourage them to post tighter and deeper markets, reducing trading costs for investors. On the other hand, those tighter and deeper markets will be somewhat illusory, since the market makers get 350 microseconds to change their mind if the market moves against them. Investors should get better pricing with less certainty, a trade-off that may or may not be worth it to them. 7
None of this has any moral valence whatsoever. Market makers perform a valuable function and should be able to make a living, 8 and investors perform a valuable function and should be able to profit from their own information, and they are locked in a perpetual battle over how to divide up the proceeds, and that is how it should be. It is a fun little puzzle, an enjoyable game to watch, but it is a mistake to attribute too much villainy to anyone involved.
It was widely predicted that the approval of IEX as an exchange would create more complexity in market structure, and those predictions are coming delightfully true. Speed bumps will now be a competitive tool for exchanges, but each exchange can build its speed bump to target a different audience. Chicago's speed bump will advantage market makers who provide displayed liquidity. IEX's speed bump advantages investors who place hidden discretionary peg orders, as well as investors who want to buy all the shares on all the exchanges before the displayed liquidity can change 9 -- exactly what the Chicago plan is meant to prevent. Nasdaq's proposed Extended Life Order targets yet another group, of traders willing to leave orders in force for a while. A thousand -- or at least a dozen -- market structures can bloom, each subtly optimized for a different type of trader. It's an innovative and competitive market, in which each exchange can figure out what sorts of traders it wants to favor, and then optimize its speed bumps to cater to those traders.
And the SEC's national market system knits all of those exchanges and their disparate speed bumps awkwardly into one market, where everyone is required to route to the exchange showing the best price, no matter how delayed it might be, or for what reason. It's kind of a mess! 10
By the way, a fun fact about Chicago is that it is pretty far from New Jersey. 11 And most of the rest of U.S. equity market infrastructure -- the other exchanges' matching engines, etc. -- are in New Jersey. When IEX was seeking SEC approval of its speed bump, people complained that it would slow down the markets, and IEX quite reasonably responded: Look, even with our speed bump, it's faster for brokers to get orders to and from our exchange than from some other exchanges. By far the slowest exchange, according to IEX's data, was Chicago, just because it is so far away. But now it'll be even slower! Adding an intentional delay to the already-furthest-away U.S. exchange -- and requiring brokers to send orders to Chicago if it has the best price, even if that price might be stale -- will make market structure just that much more challenging.
This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
The most important detail that is fudged in the text is that the description of how IEX executes a routable order (execute on IEX, delay confirmations, and immediately route to other exchanges) really describes IEX's historical practice as a dark pool, and how it wanted to operate as an exchange. But the Securities and Exchange Commission said no: IEX's matching engine can't send out routable orders that skip the speed bump. So IEX re-shuffled its mechanics so that now its router sits outside the speed bump. If you send a 5,000-share order to IEX, IEX's router will send part of it to other exchanges before it hears back from IEX's matching engine, because now IEX's router too has to wait 350 microseconds for confirmation.
But the 350-microsecond delay probably still helps the investor with the routable order, because it still gives IEX's router a bit of a head start: Even if the orders don't get to every exchange simultaneously, the delay means that no one can use an execution on IEX as data to reprice orders elsewhere, at least for a little while.
- There is another type of order -- the discretionary peg hidden order -- that also benefits from the speed bump, in a different way. (More here.) That one, unlike the router, survived the SEC process. But it's less important in the popular image of IEX -- and in my Bloomberg View colleague Michael Lewis's telling of the story in "Flash Boys" -- than the routing one.
- There is perhaps -- perhaps? -- a difference between "high-frequency traders see an execution on one venue and race to cancel their orders on other venues and resubmit them at higher prices" and "high-frequency traders see an execution on one venue and race to buy shares on other venues to resell at higher prices." I guess the latter sounds worse. (Also harder though.)
- Obviously it is not actually "front-running" to see a trade in one venue and do a trade on the same side in another venue. It's even more not front-running to see a trade in one venue, cancel your orders in another venue, and resubmit them at a different price. I am putting the phrase "front-running" very much in quotation marks, because people do use the term that way.
Or, I mean, anyone. "Market maker" isn't doing anything particularly rigorous here. But the idea is that people who mostly post orders -- who mostly create liquidity, who mostly function as market makers -- will mostly benefit, while the people who mostly take liquidity will mostly be delayed.
I once described the essential benefit of last look to market makers this way:
Obviously one way to avoid trading with traders who know more about future values than you do is to wait until the future, see where the values are, and then decide whether or not you traded in the past.
Equivalently, if you shift everyone else 350 microseconds into the past, then it's like you perpetually live in the future.
Obviously if the price went up to $218.01, like, 10 minutes before the market maker sold at $218, the market maker would be pretty dumb to sell at $218. But if it went up like 10 microseconds before, the market maker just didn't react fast enough.
Remco Lenterman said on Twitter:
- Not sure if people noticed that what the CHX describes as 'latency arbitrage' is the exact opposite of how IEX describes it
- CHX describes 'latency arbitrage' as an aggressor trading on a passive quote. So the arbitrageur is the aggressor
- IEX, however, describes latency arbitrage as a passive quote fading. So the arbitrageur is the passive party
- IEX 'protects' orders that CHX calls latency arbitrage and CHX 'protects' the orders of the ones who IEX calls latency arbitrage
Item 4 there is perhaps not accurate in every particular, for reasons discussed in footnote 1, but it captures the feeling correctly. Certainly IEX wants to protect the orders that CHX calls latency arbitrage.
“It’s unfortunate that CHX has chosen to ignore the best evidence that widespread ‘latency arbitrage’ is a myth,” said Bill Harts, chief executive of Modern Markets Initiative, a trade group founded by several leading electronic trading firms.
Harts is presumably talking about studies like this one, finding "little evidence that fast traders initiate these liquidity-taking orders to pick-off stale quotes." But the bigger problem is that "latency arbitrage" can mean both liquidity takers picking off stale quotes and liquidity providers moving their quotes before they can be picked off. Which means that "latency arbitrage" is meaningless. It just means that you traded when you didn't want to, or didn't trade when you did want to. As Adam Nunes tweeted:
When price is changing, it seems all trades are latency arb. If the order gets filled, it's a pick off. If it doesn't, it's quote fade.
There is a dumb simple logic that says:
- Trading is a zero-sum game.
- The market makers on CHX wanted this.
- Therefore you should expect it to benefit market makers (liquidity providers) and harm investors (liquidity takers).
But that probably oversimplifies the categories of traders. In any case, at least one market maker seems happy:
One of the biggest market makers endorsed Chicago’s change.
“We applaud CHX for this important innovation that will slow down traders intent only on picking off stale passive quotes and enable market making firms like Virtu to provide more liquidity,” Virtu Financial Inc. Chief Executive Officer Doug Cifu said. “This change will benefit institutional investors.”
That seems to be CHX's view:
“High-frequency firms are really modern-day market makers and specialists,” John Kerin, chief executive officer of the Chicago Stock Exchange, said in a phone interview. “Just like anything else, you have some that really add to the market and provide good and public benefit and others that do not.”
Sort of; see the caveat in footnote 1.
As Larry Tabb wrote when IEX's exchange application was approved:
One or more speedbumps and/or speedbump order types will decrease the determinism of the market, as there will be aggressive quotes that won’t be actionable (as they are traversing a bump) and displayed quotes that don’t actually exist. When displayed prices are not actionable or predictable and the data that is being broadcast – especially to market makers (the firms posting their trading intentions) – is not predictable, market makers must widen their quotes, because they cannot guarantee that they will be able to effectively trade out of (or hedge) a newly acquired position.
One particular source of mess here is that Chicago's proposal really does seem like it would reduce spreads, by giving market makers a quasi-last look on their trades. But that is true if you look at it as a single venue. If you consider not just one exchange, but the entire national market system, then the last-look benefit of quoting on Chicago may be canceled out by the reduced determinism of the market as a whole. The more speed bumps there are, the harder it is to know what the right price is, so the wider your quote has to be.
It is, however, near Chicago. (I mean, it is Chicago.) This is helpful because the Chicago Mercantile Exchange, where S&P 500 Index e-mini futures trade, is also in Chicago. If you are trying to trade stocks by arbitraging between Chicago and New Jersey, it is challenging, because they are quite far away, so there is a delay in transmitting information between them. But if you are trying to arbitrage the S&P 500 e-mini future against the S&P 500 ETF, it is helpful to have them in the same place.
To contact the author of this story:
Matt Levine at email@example.com
To contact the editor responsible for this story:
James Greiff at firstname.lastname@example.org