High-Speed Traders Put a Bit Too Much Gravy on Their Meat
One good general rule is that it's harder than you think it is to figure out what's market manipulation and what isn't. Trading a lot, cancelling a lot of orders, putting in orders or doing trades on both sides of the market, trading a lot right before a close or fixing -- all of those things could be signs of nefarious manipulation, or just normal risk management. No single event or pattern proves manipulation. You often need to look for subtle clues to figure out whether a trade is actually manipulative
One subtle clue is, if you name your algorithms "Meat" and "Gravy," there is probably something wrong with you! And your trading, I mean. But also your aesthetic sensibilities. Here is a Securities and Exchange case against Athena Capital Research, which the SEC touts as "the first high frequency trading manipulation case." The SEC caught Athena "placing a large number of aggressive, rapid-fire trades in the final two seconds of almost every trading day during a six-month period to manipulate the closing prices of thousands of Nasdaq-listed stocks." That period was in late 2009, by the way.
Athena settled for $1 million, and while it did so "without admitting or denying the findings," the SEC's order has the usual litany of dumb, so you can tell that Athena was fairly caught. In fact, the SEC is kind enough to put the dumb quotes in boldface, so they're easy to find, though somehow this didn't make it into bold:
Athena referred to its accumulation immediately after the first Imbalance Message as “Meat,” and to its last second trading strategies as “Gravy.”
Heehee that's dumb.
What is going on here? It starts with the fact that Nasdaq basically does two sorts of trading in the late afternoon. One is just its regular continuous order book trading, the kind it does all day. There are bids, there are offers, and there are lots of little trades that are constantly updating the price of every stock. Someone trades 100 shares at $20.01, 100 shares at $20.02, 200 shares at $20.03, 100 more at $20.02 again, etc., all within a fraction of a second.
There is also the closing auction, which is more or less a separate institution. This is an auction that occurs at a single point in time, just after the 4:00 p.m. close. People put in buy orders and sell orders throughout the day, and then they all trade with each other simultaneously just after 4 p.m. at the clearing price of the auction.
These really are two separate markets: a continuous trading market and a discrete-point auction. And those two sorts of trading could get a different price. A stock can be trading at around $20 in the continuous book at 3:59:59 p.m.; meanwhile, the closing auction can have orders to buy 100,000 shares without much price sensitivity, while to get 100,000 sellers you'd need to price the auction at $21. So you could have the weird result of a stock trading at $20 throughout the day, up to the very last second, and then jumping to close at $21.
That's bad. Prices should reflect reality; there should not be two (very) different prices for the same thing at (almost) the same time. So you need a mechanism to align the closing price with the trading price. That mechanism exists, and it has two parts. First, Nasdaq tells people when the prices are likely to be out of line. The way it does this is by sending "imbalance messages " every five seconds throughout the last 10 minutes of trading. An imbalance message says, roughly:
- Based on the orders we already have in the closing auction book, at the current price of $X, the closing auction would have Y more shares to sell than to buy (or vice versa).
- The closing auction would balance at a price of $Z (the "far price").
If Y is small and $X and $Z are pretty close to each other, there's no problem. If they're far apart, then that means that the market price and the closing price may not agree.
The second part of the mechanism is, people fix the imbalance. At 3:50 p.m., Nasdaq tells everyone that, say, there's a buy imbalance of 224,638 shares of EBay at its current trading price of $23.55. That means that, if trading stopped right there and the closing auction was held at 3:50, there would be more buyers than sellers at $23.55, and the closing price would be $23.60 or $23.65 or $24 or something, and the close would look very volatile.
But trading doesn't stop right there. There's still 10 minutes left. And what happens is, people step in to fix the imbalance. They say: OK, if the auction really has 224,638 more buyers than sellers, I will sell those 224,638 shares. (This is called an "Imbalance-Only-On-Close Order.") And then they go out and buy those 224,638 shares in the continuous market over the next 10 minutes. They buy from people who want to sell now, in order to sell to people who want to buy later.
This is a classic market-making function. The people doing this -- and they're not really people, they're algorithmic high-frequency trading firms -- are intermediating across time. There are sellers now, there are buyers later, and the HFT market-makers buy from the sellers and sell to the buyers, giving everyone a smoother and fairer and more informative price.
Of course, the effect of their trading will be to push the price up over the last 10 minutes. That's supposed to happen. There are more people who want to buy at the close than sell. More buyers than sellers pushes the price up. Without Nasdaq's imbalance message and the efforts of the market-makers, the price would be flattish over the last 10 minutes, and then jump up dramatically at the close. Here's the dumbest imaginable chart:
Now if you look at that chart, you'll see that the market maker buys stock over those last 10 minutes at prices ranging from $20 to $20.50. But it sells all its stock at the close at a single price of $20.50. Let's say it pays an average price of $20.25, buying smoothly over time. Then it sells at the close at $20.50. It's made 25 cents a share by intermediating in time. (Obviously, these numbers are at least an order of magnitude too big; I'm just trying to be illustrative. )
I want to dwell for a long, long time on two simple facts. One is: This is good. This is good. There's nothing wrong with this. This is a value-added intermediation function. This buys from people who want to sell and sells to people who want to buy. The people who want to sell can't sell directly to the people who want to buy, because the people who want to buy live 10 minutes in the future. So the market makers intermediate their desires. And the market makers make money doing it. Because they're providing value, and charging for it. All of this is supposed to happen. All of it makes markets better.
The second fact is: This looks manipulative. The market makers put in an order to sell stock at a fixed point in time -- the close, the fix -- and then they buy the same amount of stock in the 10 minutes leading up to that fixed point. As they buy, the price goes up, and then they sell the stock back at the higher price. That looks like classic market manipulation. That looks precisely like the foreign-exchange fixing scandal, except for the scammy chat rooms.
It's a mystery, isn't it? You should rejoice in the mystery. The mystery is real. Stuff that sounds like manipulation may not be manipulation. Or maybe: Stuff that is a normal and essential part of financial markets may be manipulation. Who can say?
Athena isn't a mystery. Athena really was manipulating the market. Athena is a degenerate case of what I described. Basically you'll notice in what I described that the market maker buys at its average price, and then sells at its final price. It has incentives to make those prices as different as possible. One way to make those prices different is to try to buy really efficiently, so you buy at a low average price. Another way to make the prices different is to make the final price really sloppy and inefficient, so you sell at a high final price. That's what Athena did: It bought about half of the shares it was going to buy smoothly over the last nine minutes and 58 seconds, getting a reasonable average price on half of the shares. That's "Meat." Then it would buy the other half sloppily in the last two seconds, pushing up the final price really high and leading to a high sale price. That's "Gravy." Gravy is how it made its money.
That's straightforwardly not good value-adding intermediation. You can tell because they were trying to be inefficient: They wanted to move the price around a lot. If you want to move the price around a lot, you're probably not providing a great service to smooth the market. You're just trying to mess with people to make a profit. That's a useful test of market manipulation: Adding chaos is mostly bad; subtracting it is mostly good. Though it's not an easy test to apply. The dumb e-mail and dumb nickname tests are more straightforward.
There is a lot of puzzlement to the Athena case that is of second-order interest. And there is some basic silliness, as there often is in the manipulation cases that actually get brought. At one point, an Athena analyst e-mailed his manager to gloat that Gravy made $5,300 in a day, moving 33 stocks by as much as 12 cents. The manager wrote back, "going to Vegas tonight." With $5,300! Later, Athena "lost approximately $2-3 million" in a day with this strategy. So it had its flaws.
The lesson here is something like: There are manipulative strategies, and there are good strategies, and it is not easy to tell them apart. You can tell them apart, probably, but you need to understand their purposes first. And dumb e-mails and nicknames can be a big help.
Here they are, with no context:
- "Let’s make sure we don’t kill the golden goose."
- "We have a desired accumulation pattern which includes grabbing stock at the beginning, a period of ‘average price’ accumulation, and a crescendo at the end."
- “We can have some aggressive gravy if we know we have a 100% chan[c]e of getting the fill."
- “dominating the auction” and “owning the game.”
- “gravy [average] move by symbol[.]”
- "biggest dollar move" ... "percentage move" ... "Looks like we have some Mach chips….going to Vegas tonight…."
- "So at 3:59:58, gravy kicked in . . . To try to get a whopping 1000 shares. 1000 shares had 0.0 price impact, but 2000 shares would probably move it a few cents."
- "we should tax a little more"
- "to make sure we always do our gravy with enough size."
- "punching the stock"
- "at home, not here"
- "Protection orders are probably necessary in order to gravy up some of the thinner issues, but since we rank them largely according to volume, we should certainly be able to ramp those guys up."
- "the lack of the blast resulted in extremely poor prices (we essentially gave someone else all the liquidity they wanted with no price impact at all)."
- "Not knowing the flip price going into the gravy phase, then having the gravy push us over the price; thereby changing the direction of the print."
Some are dumber than others. Any time the SEC sees "at home, not here," in a written communication, its interest will be piqued. Also, "Vegas" will always be boldface.
- "Let’s make sure we don’t kill the golden goose."
I guess you could just not have the auction, and just end the continuous-order-book trading at some point, but that's not great either. People want a closing price, and many people want to trade at the closing price. The auction is a way to sort of clear out any residual demand.
Incidentally, the description of the mechanism here, including the specific EBay example, comes largely from the SEC's order.
I'm abstracting away from pretty much everything except the imbalance here. Obviously other stuff can happen in those 10 minutes.
And I'm entirely ignoring risk, changes in the on-close order book, everything that really happens in the real world. This is very very schematic.
Really, you need both. If you do too much Meat and no Gravy, you don't move the price, so you don't get any benefit from manipulation. (The SEC quotes e-mailed complaints about that -- that's the "To try to get a whopping 1000 shares" line in footnote 1.) But if you do too much Gravy and no Meat, you're buying all your shares at 3:59:59 at the same price you're selling them for a second later. Gravy pushes up the final price, Meat pushes down the average price. Your goal is to maximize the difference between the final price and the average price. Intuitively, you should be about half Gravy and half Meat, and that's more or less what Athena did.
Even here: If Athena didn't exist, would the last 10 minutes of Nasdaq trading be more or less chaotic than they actually were? The answer is surely "less chaotic, because someone would have done the market-making strategy I described in a way that was smoother than Athena's manipulation." But that's an empirical point: There actually are a lot of competing market-making strategies that are aware of the closing imbalance and would step in to do the job better than Athena did. But what if there weren't? If the choice is Athena's manipulation or nothing, I'm not entirely sure which is better.
Athena itself takes this view: "While Athena does not deny the Commission's charges, Athena believes that its trading activity helped satisfy market demand for liquidity during a period of unprecedented demand for such liquidity."
Mostly having to do with the fact that they didn't trade in the schematic market I described; they traded in a real competitive ever-changing market. So for instance, when they put in an order to take the entire other side of a closing imbalance, how did they know that they would get filled? The SEC says, "Athena took measures to gain priority over competing limit-on-close orders and Imbalance-Only Orders." My best guess is that they were just the fastest at submitting these orders, and they got filled in time priority. So if Athena was always first in the door with an Imbalance-Only Order, it would always the the one filled on that order, so it could safely buy the other side in the continuous market in the last 10 minutes (or two seconds).
Or there's the fact that the market could move away from them and the imbalance flip the other way. To guard against this, Athena had what it called "protection orders" to limit its losses. But the precise mechanics are not clear from the SEC order, and would be very important to risk-manage this strategy in the actual world.
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 email@example.com
To contact the editor on this story:
Toby Harshaw at firstname.lastname@example.org