Wall Street

Fast Trader Pulled a Fast One on Some Customers

Sometimes the weirdest theories about high-speed trading turn out to be true.

Where your orders don't go.

Photographer: Michael Nagle/Bloomberg

The basic way that regular people buy stock is like this:

  1. You go to your broker's website and click a "buy" button.
  2. Your broker sends your order to Citadel Securities LLC. 1  
  3. Citadel looks at the price that the stock is selling for on the public stock exchanges.
  4. Citadel sells you the stock itself, for a price a tiny bit lower than that. 

This rubs some people the wrong way. Who asked for Citadel? Why is Citadel cutting in and insisting on selling you the stock, instead of leaving you alone and letting you trade on the stock exchange? Presumably Citadel has to get the stock from somewhere, in order to sell it to you, and since it's a profitable business, it's probably paying less for the stock than you are. So why can't you just get that deal? What value is Citadel adding? What's even more suspicious is that Citadel is probably paying your broker for the right to do this trade with you. Again: Citadel is profitable, so if it's paying your broker for the right to sell you stock, that probably means it's making a profit on selling you the stock.

But the weirdest part is that Citadel is selling you the stock for a lower price than what's available on the public exchanges. For instance, if the public exchange shows the stock offered at $40.02, Citadel will sell you that same stock for $40.016, saving you like 40 cents on a 100-share order. 2  (This is called "price improvement.") It's giving you a (slightly) better deal than you'd get elsewhere! And paying for the right to do it! And still making a profit! It is just suspicious. It looks like a trick. How does it work?

There is a straightforward boring market-microstructure answer, but that answer has not exactly ignited the public imagination. (It goes like: "Retail order flow involves less adverse selection than public markets, so market makers who can segregate that flow can profitably trade at tighter spreads." 3 So boring!) But there is also a sexy conspiracy-theory answer, which has become pretty popular. It goes like this:

  1. The public stock exchanges provide two electronic "feeds" showing the prices of shares.
  2. One is called the "SIP," and it's slow, but sort of official: It's the price that your retail broker checks against to make sure you got the best price.
  3. The other is called the "direct feed," and it's fast. It's also expensive: Exchanges make a lot of money selling their direct feeds to high-frequency traders.
  4. Citadel pays to access all the direct feeds, so it always knows the "real" prices of stocks, accurate to the microsecond.
  5. But it fills your order based on the SIP, which is slow and stale.
  6. So when the SIP says that the price of a stock is $40.02, Citadel knows that it's really $40.01, so it can buy the stock at $40.01 and sell it to you for $40.016 for a guaranteed risk-free profit.
  7. Citadel makes its money by constantly cutting in front of retail orders, selling shares at (a discount to) the stale SIP price and buying them at the accurate direct-feed price.

This is called "SIP latency arbitrage," though if you call it "front-running" instead, I guess no one will stop you. Whenever I mention it to electronic traders, they laugh, but it's a tired, rueful laugh. Why would this be a thing? The SIP latency -- the time delay between the direct feeds and the SIP -- just isn't big enough (and it's getting smaller) for this to work very much. The odds that an order will come in when the SIP has one price and the direct feeds have another are pretty small. About 97 percent of trades occur when the prices match. It would be weird to build up a whole retail market-making strategy -- and pay to interact with retail orders -- just for the 3 percent of time that you might make an extra penny from SIP arbitrage. The SIP latency arbitrage theory makes almost no sense, but it is weirdly, enduringly popular.

It's also true! True-ish. Sort of. Friday's $22.6 million settlement between the Securities and Exchange Commission and Citadel -- over SIP latency arbitrage -- has something for both sides in the debate. On the one hand, the SEC seems to have concluded that Citadel's SIP latency arbitrage was infrequent, not particularly economically significant, and ended seven years ago. On the other hand: The SEC concluded that Citadel did SIP latency arbitrage! (Sort of.) And made money on it. And even had a specific algorithm, with a specific name, to do it.

Actually two algorithms with two names. If you think about it for a minute, you'll realize you need two strategies to do SIP latency arbitrage:

  1. If a customer buy order comes in when the SIP price is $40.02 and the direct-feed price is $40.01, you fill the order at $40.02 (or $40.016, or whatever) and then go buy stock at $40.01, making a profit.
  2. If a customer buy order comes in when the SIP price is $40.01 and the direct-feed price is $40.02, though, you can't just fill the order at $40.02 and buy at $40.01. For one thing, the SIP price is in some sense the "official" price, so filling the order at a worse price, just because it's what you see on the direct feeds, doesn't work. For another thing, the direct-feed price is in some sense the "real" price -- it's where the stock is actually selling now -- so you can't just go buy the stock for $40.01. You need some other strategy to deal with the fact that the customer expects a better price than is actually available.

The first strategy is simple enough: Sell to the customer at the high SIP price, when you know you can buy in the market at the lower "real" price. Citadel called this strategy "FastFill" 4 :

One strategy, known as FastFill, was triggered when the best price from one or more of the depth of book feeds that FastFill referenced was better than the best price disseminated by the SIP feed. Assuming all other eligibility conditions were met, FastFill immediately internalized a marketable order at the SIP NBB or NBO, as applicable, or better.

For example, if CES was handling a marketable order to buy shares, and the SIP best offer was $10.01, and the best offer from one or more of the depth of book feeds was $10.00, FastFill immediately internalized the order using the SIP offer of $10.01 per share. FastFill did not internalize at or seek to obtain through routing the better $10.00 price from the depth of book feeds.

The second strategy is sort of the reverse: If the SIP is better for the customer than the direct feed, you don't want to sell at the SIP price; you want to sit around twiddling your thumbs until the SIP price catches up to the direct-feed price. Citadel's algorithm for this was called "SlowFill." No, I'm kidding. It should have been. But it was called "SmartProvide": Citadel would send the order to the market at a price that was lower than the SIP price (which, in turn, was lower than the "real," direct-feed price). Because it was pretty far away from the market, it probably wouldn't execute. 5 (If it did -- and it did 18 percent of the time -- then, hey, great for the customer.) Then it would come back to Citadel -- at which point the SIP would probably have caught up to the direct feeds, and Citadel could internalize normally.

Honestly it's hard to know what Citadel was supposed to do in that situation. The SEC says that "some of the orders CES internalized after SmartProvide displayed an order in the market on their behalf received a price that was worse than they would have received had CES immediately internalized at the SIP" price, but the point is that the SIP price was stale, and it's a bit rough to expect Citadel to take a loss trading at the stale SIP price. 6

But SmartProvide is just an intriguing anomaly; the real action is in FastFill, which is literally, exactly, the paranoid fantasy of SIP latency arbitrage: Citadel had a specific named algorithm designed to profit by selling to customers at $10.01 when the real, up-to-date price that Citadel saw was $10.00, giving it a risk-free profit. Usually this profit was sort of hypothetical; it's not like Citadel literally sold shares to the customer at $10.01 and then immediately raced out and bought those shares in the market at $10.00. 7  So it's not exactly the risk-free profit you hear about in latency-arbitrage theory, though it's in the same general area.

It's not quite as bad as all that. For one thing: It was pretty small potatoes. FastFill ran from June 2008 through January 2010. At that time, Citadel's retail internalizing business traded about 1.2 million orders for 2.3 billion shares per day. FastFill was about 0.4 percent of Citadel's retail order flow, a total of about 2.7 million orders over that time. (SmartProvide executed another 690,000, starting in late 2007.) Citadel's $22.6 million settlement includes only $5.2 million of "disgorgement," which is presumably the SEC's best guess at how much money Citadel made on these two strategies. By comparison, Citadel reports statistics on how much money it saves retail investors with price improvement. For 1,000-share-ish orders -- which seem to be around Citadel's average order size -- the savings are on the order of $5 or so per order. 8 That means that Citadel pays retail investors more in price improvement per day than it made on SIP latency arbitrage during the whole two-year period at issue here. The SIP stuff -- "the SIP stuff that the SEC caught," if you're paranoid -- is just teeny tiny stuff.

For another thing, this wasn't exactly a risk-free profit. Even FastFill trades were sometimes good for customers. The SEC grants Citadel this concession in the settlement order:

When FastFill internalized an order, it filled the entire unexecuted portion of the order at the SIP NBB or NBO, as applicable, and did so (other than in the circumstances described in Paragraph 31), regardless of the number of shares associated with the SIP NBB or NBO or the best bid or offer from one or more of the depth of book feeds. Because of this feature, which provided a price benefit as compared to the liquidity displayed in the market and price improvement that CES added, FastFill improved the overall execution price for a substantial number of (predominantly larger) orders.

That is: If the SIP displayed 200 shares available for sale at $40.02, and the direct feeds displayed 100 shares available for sale at $40.01, and you came to Citadel with an order to buy 1,000 shares, FastFill would sell you all 1,000 shares (at $40.016, or whatever). That's worse than selling you all 1,000 shares at $40.01, sure, but it might still be better than you'd have gotten in the public markets, because those markets weren't really offering all 1,000 shares at $40.01, or even $40.02. 9  Citadel was taking a risk to fill its customers' orders, though it was using the direct feeds to make sure that that risk wasn't too big.

Also: The SEC never quite comes out and says that this is illegal. Instead, it criticizes Citadel's disclosure:

During the relevant period, CES provided a written disclosure to certain retail broker-dealer clients that described a market order as an “[o]rder to buy (sell) at the best offer (bid) price currently available in the marketplace,” and made other, similar representations to its clients. As discussed above, these statements suggested that CES would either internalize the marketable order at, or seek to obtain through routing, the best bid or offer from the various market data feeds CES referenced. These statements were materially misleading in light of the way that FastFill and SmartProvide functioned.

Was it illegal? Who knows? 10 Brokers have a general obligation of "best execution," but it was long a mystery whether that should be measured against the official SIP feed or against the direct feeds. 11 The SEC never determines here that Citadel's executions weren't "best execution," and in some cases -- like selling 1,000 shares at the SIP price for 200 -- they probably were. But even if you're not sure about the substance, you can always go after disclosure.

There are two basic ways to think about high-frequency trading -- or about anything in financial markets, really. 12 One approach is: This is a sensible economic phenomenon, and also sometimes people cheat. If that is your assumption, then you explain the retail execution business in terms of market microstructure and adverse selection. And then you hear that sometimes Citadel did this SIP latency arbitrage thing, and you think, "hahaha, you cheeky little imps!" And you have a good laugh at their chutzpah, and fine them a few million dollars.

The other approach is: This is a pure creation of cheating and corruption, and any economic benefit is a minor accident. If that is your assumption, then latency arbitrage must be at the heart of the retail execution business; it's how high-frequency traders can make a risk-free profit by front-running retail investors. And if that's what you think, then the SEC's order must be a disappointment. It finds SIP latency arbitrage, yes, but so little of it, and for so little money. The problem is real, or was real, but it's so small. It hardly explains anything.

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

  1. I mean, it's not always Citadel. It's some "internalizer" or "wholesale market maker," of which Citadel is a big one, one that "accounts on average for approximately 35% of the average daily volume of retail equity shares traded in the U.S. markets." 

  2. Those illustrative numbers come from this Citadel comment letter to the Securities and Exchange Commission. (Technically Citadel is describing the average price at "a leading wholesale market maker," which may or may not be Citadel itself.)

  3. We have talked about it occasionally. (Here, for instance.) The basic idea is that if you are an electronic market maker, posting bids and offers to buy and sell stock on the public stock exchange, then every so often some big mutual fund will come to you looking to buy 1,000 shares. And you will sell them the 1,000 shares, at $40.02, and then they will go buy 100,000 more shares, and the price will go up to $41. And so you lost 98 cents per share, because you sold to a big buyer just as the stock was about to start going up. This is called "adverse selection," or "getting run over." 

    But if you are an electronic market maker trading only with retail orders -- because you have paid a retail broker for its order flow -- then you know that will never happen. (Or, very rarely.) Retail investors tend to trade much smaller size (and, often, be less informed), so if you see a 1,000 share order, there's unlikely to be another 100,000 behind it. So your chances of getting run over are much lower.

    Getting run over costs you money. If you can eliminate that cost, you can operate more efficiently. And you can pass those savings on to your customers. The way market makers charge customers is with the spread: A market maker will bid to buy for $40.00, and offer to sell at $40.02, hoping to make 2 cents per share most of the time (and have enough profit to cover any losses on adverse selection). A market maker without adverse selection can charge a lower spread: It can bid to buy for $40.005, and offer to sell at $40.015, for instance, a spread of 1 cent instead of 2 cents. And customers -- the retail investors buying and selling the stock -- get those savings.

    (That's not exactly what happens -- Citadel isn't posting those sub-penny prices anywhere -- but it's effectively what happens.)

    The other way to pass on the savings is by paying for order flow. Citadel could charge the same $40.00 / $40.02 spread, but pay E*Trade or Charles Schwab or whoever 1 penny per share for the privilege of interacting with its retail orders. (E*Trade or Schwab could then pass those savings on to customers in the form of lower commissions, free research, whatever. Or not!) Or you could do some combination of both, say charging a 1.2-cent spread ($40.004 / $40.016) and also paying 0.2 cents per share to the broker. Or whatever. 

    This all oversimplifies: In particular, wholesalers seem to avoid even some retail flow, and their economics may involve optimizing order flow in ways that go beyond just "take only retail orders." But as a simple microstructure explanation, it more or less works.

    There is also something to be said here about fees. Stock exchanges charge fees for executing orders. If you are a market maker buying and selling on the exchange, those fees are a cost for you. If you are a market maker buying and selling from captive retail orders without ever going to the exchange, you avoid those fees, and can make tighter markets. (This is complicated by the fact that stock exchanges often pay market makers for providing liquidity, though.)

  4. CES is Citadel Execution Services, the wholesale market making unit of Citadel. "NBBO" stands for "national best bid or offer," the best price available in the public stock markets. "Depth of book feeds" is the SEC's phrase for "direct feeds." "Internalize" just means that Citadel traded with the order itself -- sold stock to a buy order, etc. -- rather than routing it to a market.

  5. From the SEC order:

    The second strategy, known as SmartProvide, was triggered when the SIP NBB or NBO, as applicable, was better than the best price from at least one of the depth of book feeds. SmartProvide did not internalize at the SIP price, nor did it seek to obtain an execution at that price by sending an order to the market. Instead, assuming all other conditions for order handling by SmartProvide were met, SmartProvide would route a non-marketable order to the market.

    For example, if CES was handling a marketable order to buy shares, and the SIP NBO was $10.01, and the best offer from one or more of the depth of book feeds was $10.02, SmartProvide would send a buy order to be displayed in the market at a price less than $10.01, such as $10.00. This order would be displayed for up to one to five seconds, depending on the size of the order. If this order received an execution, the customer order would benefit from the execution at the better price (i.e., the shares purchased by the customer would be at a price at least one penny better than the NBO). This occurred for approximately 18% of the shares handled by SmartProvide. If the order did not receive a full execution from this routing, CES’s algorithms reassessed the handling of the remaining shares, and could either internalize or seek to obtain an execution in the market.

  6. I mean, arguably it isn't, if Citadel could also make symmetric profits by trading at the stale SIP price when it was in Citadel's favor. But the SEC objected to that too!

    To be fair, the SEC actually says:

    Some of the orders CES internalized after SmartProvide displayed an order in the market on their behalf received a price that was worse than they would have received had CES immediately internalized at the SIP NBB or NBO, as applicable, or sent an order to the market at that price.

    Presumably sending a buy order to the market at the $10.01 SIP best offer, when the actual market best offer was $10.02, wouldn't have worked that often, but it would have worked sometimes, and Citadel instead sent the order with an even more unmarketable $10 price.

  7. But sometimes it did!

    In a narrow set of circumstances, FastFill internalized the order even though CES did not have a principal interest in the order or internalization of the order would cause CES to exceed its position limit in the relevant security. In these situations, contemporaneous with determining to internalize the order at the SIP NBB or NBO, as applicable, FastFill sent a proprietary order to the market in an effort to execute for itself at a price better than the SIP NBB or NBO, as relevant. This routing was known as a “trade out attempt” and occurred with respect to less than 6.9% of the shares internalized by FastFill.

  8. I'm just looking at the "500 - 1,999" share order size range on these charts, which show price improvement of $5.41 per order on S&P 500 stocks and $7.22 on other stocks. The SEC's statistics -- 1.2 million orders per day, for 2.3 billion shares -- suggest that Citadel's average order size is around 1,900 shares. And 1.2 million orders times $5 or so is $6 million per day in savings. This is not particularly scientific math but there you go.

  9. As Matt Hurd puts it:

    This is basically saying that if the SIP was behind the direct feed, the customer got the SIP and CES would do its best to get something better for itself. That may not have always worked out for CES, but it probably did. This is the direct feed (DF) versus SIP feed, a so called latency arb that isn't, but is frowned upon at an ATS or exchange. I'm not sure if this was improper for an internaliser under rules at the time. I don't think it was. Newer rulings, subsequent to January 2010, may have an impact in current interpretations. The SEC did not take the view it was improper, they just wanted proper disclosure. For example, say that you used some smart ML to determine the price was about to change and thus filled your clients over the spread with the expectation you would do better as you hedge, but with less certainty, is that wrong? At what price is innovation? It quickly gets cloudy.

  10. Sal Arnuk asks:

    If Citadel said "we will trade in front of your order, and give you a worse price", there would have been no fine?

    I mean, probably, right? In general, it's not illegal to charge people what you tell them you'll charge them for a service, though brokerage is different.

  11. Though in November 2015 the Financial Industry Regulatory Authority said in a footnote that "a firm that regularly accesses proprietary data feeds, in addition to the consolidated SIP feed, for its proprietary trading, would be expected to also be using these data feeds to determine the best market under prevailing market conditions when handling customer orders to meet its best execution obligations."

    In any case, the best execution requirement really belongs to the retail broker -- E*Trade or Schwab or whoever -- rather than to internalizers like Citadel.

  12. This draws on Scott Alexander's discussion of "figure-ground inversion" in "House of God."

To contact the author of this story:
Matt Levine at mlevine51@bloomberg.net

To contact the editor responsible for this story:
James Greiff at jgreiff@bloomberg.net

Before it's here, it's on the Bloomberg Terminal.