Coincidence or not, ever since "Flash Boys" came out four months ago, alleging that U.S. equity markets are rigged by high-frequency traders and crooked dark-pool operators, there's been a whole lot of regulatory attention paid to high-frequency traders and crooked dark-pool operators in U.S. equity markets. New York Attorney General Eric Schneiderman has been on quite a tear, going especially hard after Barclays's dark pool. The Securities and Exchange Commission has upped its game, fining Liquidnet $2 million for using dark-pool customer data to pitch issuer business. Even Finra has gotten involved, fining Goldman Sachs $800,000 for messing up the clock on its dark pool.
Here's another one! It's a $5 million SEC settlement with LavaFlow Inc., a big stock trading venue owned by Citigroup. LavaFlow is not a dark pool; it's an electronic communications network (ECN) that displays orders. (Basically: It's a stock exchange within Citigroup.) But it allows subscribers to submit various sorts of non-displayed orders, which are, you know, not supposed to be displayed. And the problem was with those non-displayed orders.
LavaFlow had an affiliate called Lava Trading, whose main business was running a smart order router called ColorBook. From the SEC's order:
Lava Trading’s flagship product was ColorBook, which was software that provided smart order routing services to over 100 registered broker-dealers that principally used ColorBook to route their customers’ orders to execution venues. ... An “order router” is an application that generally allows an end user to submit an order to an execution venue. A “smart order router” applies preprogrammed analytics that carry out an execution strategy for order flow provided to the smart order router.
The point here is that, if you just want to buy some stock, that's a weirdly hard thing to do. You can't just go to the New York Stock Exchange and say "buy 100 shares of IBM," because IBM shares are traded on a bunch of different exchanges and ECNs and dark pools, and you want to buy them at the lowest price offered on any of those venues. Or maybe you don't -- maybe you prefer convenience to getting the lowest price. But that doesn't matter, because Regulation NMS says that your broker can't buy you 100 shares of IBM at $195 on the New York Stock Exchange if someone is offering 100 shares for $194.95 on, say, the LavaFlow ECN. So you're forced to deal with all the venues anyway.
So what you might do is, look at all the displayed prices on all the exchanges, pick the one with the best price, and then buy there. But this again is weirdly hard. First of all, if you are a human, by the time you finish looking at all the displayed prices with your human eyes, they will have changed, because the human eye is rather famously slower than high-frequency trading. Also, there are lots of non-displayed orders, in dark pools and such, so you might want to see if you can get a better price in a dark pool before you go to the displayed exchanges. And if you want to buy 10,000 shares of IBM, just sending that big order to the exchange offering 100 shares at $195 is risky, because everyone will see that trade and move up their prices on the other exchanges, so that by the time you go there your other 9,900 shares will cost $195.25.
That's basically the plot of "Flash Boys": A human trader at a big bank, Brad Katsuyama, had trouble buying shares because he was a human, and the price kept moving. So he built a smart order router (and named it "Thor"). Smart order routers are (1) computers, so they're fast, and (2) smart, so they break up your order and send it to the places where it's most likely to execute quickly at the best possible price. Some brokers have their own order routers; others will rent someone else's, because building one is hard. ColorBook is an order router that Citi uses, and that other brokers can rent.
One thing that made ColorBook smart was that it remembered where it sent orders. So not only could it see all the displayed orders on all the displayed exchanges and ECNs, but it also could see all the undisplayed orders -- dark pool orders, or hidden orders sent to exchanges or ECNs -- that it had sent there. So if you came to ColorBook asking to buy 100 shares of IBM, and the best offer for IBM displayed anywhere was $195, but ColorBook remembered sending an undisplayed order to LavaFlow to sell at $194.95, then ColorBook would route your buy order to LavaFlow to get the better price. As the SEC puts it, "ColorBook considered its historical knowledge of non-displayed resting orders to make future routing decisions."
If you used ColorBook, that was part of the deal: It would remember your secret orders and use them to route its future orders, though it wouldn't tell anyone about them.
But as it turns out, if you just used the LavaFlow ECN without using ColorBook -- as a "direct subscriber" -- then ColorBook remembered and made use of your orders anyway:
From March 2008 until March 2011, LavaFlow allowed ColorBook to access and use its knowledge of the LavaFlow ECN direct subscriber non-displayed order flow when making smart order routing decisions for those smart order routing customers who also were subscribers of the LavaFlow ECN.
And LavaFlow wouldn't tell you that: That wasn't disclosed in its filings, or consistently disclosed in its advertising. (Though it was sometimes: This is not so much "a fact LavaFlow kept secret" as it is "a fact LavaFlow forgot to tell people." ) And you're supposed to tell people how you're using their secret orders. And if you don't you get in trouble. And so LavaFlow eventually got in $5 million worth of trouble.
So that's the story. Here are three things worth thinking about it.
First: Strangely, Lava got in trouble in 2012 for doing pretty much exactly the same thing. The story there was that ColorBook also knew about the undisplayed orders on the LeveL dark pool, which Lava also ran. That story is in some ways worse than this one -- dark-pool orders just feel more secret than undisplayed ECN orders, don't they? -- but the fine that Lava paid there was just $800,000. Now it's $5 million. So the price of market-structure abuse is going up.
Second: Did Lava's conduct hurt anyone? Well, why would you submit a hidden order? I think it's because
- you don't want people to know about your order in making their decisions -- about whether to trade, and at what price, but
- you do want people to trade with you.
So, again, think of "Flash Boys." Brad Katsuyama didn't want people to know that he wanted to buy a lot of shares, because then they'd move their prices up and liquidity would vanish. But he did want to buy a lot of shares. He wanted people to trade with him; he just didn't want them to know about his interest in deciding what prices they wanted to trade at. If he could somehow ensure that (1) nobody knew about his interest until they made their trading decision, but (2) everyone knew about his interest as soon as they irrevocably decided to sell, then that would be perfect.
And that's sort of what ColorBook did? What ColorBook did with the hidden orders is route its customers to those hidden orders. What it didn't do was tell them about the hidden orders. So the ColorBook customers couldn't use the hidden orders to make pricing decisions, or to decide whether or not to trade, or to "front-run" the hidden orders. All they could do was trade with the hidden orders: Once they submitted an order to buy X shares at Y price, ColorBook would send it toward the hidden orders. That's exactly what you want when you submit a hidden order! That's not bad! If you're a LavaFlow direct subscriber, you might be annoyed that ColorBook used your data without telling you, but I don't think you were harmed by it. You might even have been helped: Your orders were more likely to execute, and more quickly.
Third: Whom did Lava's conduct hurt? Probably no one, as I just said, but who were the "victims"? Remember, the beneficiaries were the ColorBook users, who got routed smartly to the hidden orders. Those people are basically the brokers who rented ColorBook as their order router, and the customers of those brokers. Think, basically, buy-side institutions -- hedge funds and mutual funds and asset managers and whoever -- who used Citi or a broker who rented technology from Citi to trade stocks for them. In the scheme of things, these are the "regular" investors, the victims in "Flash Boys."
The victims were the LavaFlow direct subscribers, who did not use ColorBook, and who left hidden orders on LavaFlow. Who subscribes directly to Citi's ECN, without using Citi's order router? I think the answer is basically:
- other big broker-dealers (and their routers), who feel obligated to connect everywhere, and who do their own routing, and
- high-frequency traders, who want to trade everywhere, and who also do their own routing.
Other banks and brokers probably subscribe to LavaFlow, but it seems unlikely that they'd leave a lot of hidden orders there. (Why leave orders with your competitor? ) The LavaFlow direct subscribers who were actually leaving hidden orders there were ... probably mostly high-frequency traders?
So this case looks a lot like: Citi took information from high-frequency traders and used it to improve the outcomes of regular customers. (Probably without worsening the outcomes of those high-frequency traders.) Obviously Citi wasn't supposed to do that. But if you're looking for the post-"Flash Boys" enforcement action that shows up the evils of high frequency traders and the brokers who cater to them, this isn't it.
From the SEC order:
The direct subscriber non-displayed orders submitted to the LavaFlow ECN mostly consisted of “hidden orders,” “reserve orders,” and “pegged orders.” Hidden orders are an order type in which all or a portion of the order is not displayed to other market participants. Reserve orders are those orders in which portions of the order are only gradually displayed in accordance with customer instructions as the portions are executed. A pegged order is a limit order, the price of which is automatically adjusted to follow the price movement of a reference price (e.g., best bid, best offer). Some or all of a pegged order might be non-displayed.
Also your broker is separately obligated to get you the "best execution" on your trade, which probably means that if he knows he can get those shares for $194.90 on a dark pool he should do that. Though nobody really knows what "best execution" means. And there are some exceptions to what I said about Reg NMS in the text though it's reasonable enough.
Or that is the idea! Another part of the plot of "Flash Boys" is that the big brokers who offer smart routing services are not using the word "smart" in the sense of "smart for the customers." Instead, it means "smart for the brokers." A customer says that she wants to buy 10,000 shares of IBM, and the broker says, nonono that is far too difficult for a human to accomplish, let me put it into my smart order router, and it will get you those 10,000 shares. And it will. But it will do its smart little thinking with the goal, not of optimizing the price that the customer gets, but rather of optimizing the broker's profitability. So some venues will pay brokers fees for certain types of orders, and some venues will charge brokers a fee for certain other types of orders, and the broker's own dark pool, for instance, will tend to be free. So brokers will route orders in a way that maximizes the fees that they get and minimizes the fees they have to pay.
This is part of Eric Schneiderman's fulminating at Barclays, that it routed a ton of orders to its own dark pool not because that was the best place for them but because it was the most advantageous place for Barclays (since it cost Barclays nothing to execute there, and increased Barclays's market share).
Another source of non-displayed orders came from smart order routing customers. For example, many such customers would submit through ColorBook non-displayed orders that the customers destined for one or more particular venues; in such instances, ColorBook acted as an order router to direct the order to the venue specified by the customer. In other instances, a customer would submit an order that ColorBook would determine how and where to route. In both instances, ColorBook would know that a particular order rested at a particular venue, and until ColorBook received a message that the order was canceled or executed, ColorBook could use that information to determine how to route a subsequent order so long as the customer submitting the subsequent order was entitled to trade on that venue. Because the order had a non-displayed component, ColorBook had unique knowledge of that order compared to others in the market, and ColorBook could provide a benefit to smart order routing customers by taking into account such information in determining where and how to route orders.
Marketing materials, which were shown to at least some ECN subscribers, contained language that some ECN non-displayed resting orders would be “exposed” to ColorBook. Based on these materials, some ECN subscribers might have understood that LavaFlow was granting access to, and ColorBook was retaining information about, ECN subscribers’ non-displayed order flow. LavaFlow did not have procedures in place to ensure that all ECN subscribers reviewed and understood these materials, which were used for marketing purposes, not compliance. As a result, these materials were not a fulsome or adequate means of ensuring that the LavaFlow ECN subscribers consented to, or were notified of, ColorBook’s use of the confidential trading information, and were not a sufficient safeguard or procedure to protect subscribers’ confidential trading information.
LavaFlow had in place certain safeguards, such as physical barriers and restrictions on employee access to prevent the unauthorized access to the LavaFlow ECN subscribers’ confidential information. For instance, there is no evidence that information about the ECN’s non-displayed order flow was displayed, or otherwise communicated to, customers of the smart order routing business or other third parties.
I guess this is debatable, depends on adverse selection, etc. etc. But one reason to think it's true is that ColorBook users knew that their hidden orders would be used this way; only LavaFlow direct subscribers didn't. If you're a ColorBook user, that just seems good: ColorBook will route your orders to interact with hidden liquidity, and will route other orders to interact with your hidden liquidity, so you have the best possible chance of actually trading.
If you're a big broker, you mostly go to other brokers' internal dark pools and ECNs to satisfy your Regulation NMS requirements of not trading at worse than the national best price. So you don't rest orders, you just ping them to see if you need to execute immediately. And if a client gives you a non-marketable order, you probably want to rest it as a hidden order on your own dark pool, which provides you the best economics. (See footnote 5, etc.) Or maybe for best execution reasons you might feel compelled to send it elsewhere -- say at an exchange -- if that seems more likely to get an execution than your own dark pool. But resting a large hidden order at a competitor's dark pool or ECN seems less attractive.
To contact the author on this story:
Matt Levine at email@example.com
To contact the editor on this story:
Zara Kessler at firstname.lastname@example.org