DOJ’s Flawed Spoofing Theory: Are High Frequency Traders Really the “Victim”?

The Commodities and Exchange Act (CEA) defines “spoofing” as “bidding or offering with the intent to cancel the bid or offer before execution.” ¹ Put into plain language, the law makes it illegal for traders to enter orders onto securities exchanges that they do not actually want to be filled.  

 In the only two criminal spoofing cases that have so far gone to trial, United States v. Coscia² and United States v. Flotron³, the Department of Justice presented high-frequency traders (HFTs) as the defendants’ intended victims.  In both cases, DOJ contended that “trick” orders—i.e., orders intended to be cancelled—were placed on one side of the order book so that HFTs would be deceptively induced into filling “real” orders lying in wait on the opposite side.  To substantiate this claim, DOJ called a series of HFT executives to testify as to how the defendants’ orders supposedly affected the decisions of their trading algorithms in a way that caused trading losses.  DOJ’s strategy of casting HFTs as the innocent targets of defendants’ spoofing was no doubt designed, at least in part, to give the jury a reason to care about the crime it was trying to prove—it was a way of emphasizing, “What the defendants did was wrong and someone lost money because of it.”  

In a filing on September 2, 2020, DOJ signaled its intention to deploy this strategy once again in the upcoming Vorley trial by calling “two algorithmic traders,” Anand Twells and Travis Varner, to “testify as victims.” ⁴ Since the “HFT as victim” playbook looks to be one that we will continue to see in spoofing cases, it seems appropriate to ask whether it fairly characterizes the market activity at issue.  ⁵

Based on an analysis of how Twells testified in previous spoofing cases, in the author’s opinion the answer is no.

Twells’ Prior Testimony

Twells is a trading supervisor at Citadel L.L.C., the $35 billion Chicago-based hedge fund, who testified for the government in Coscia and Flotron.  On both occasions, he walked the jury through instances where trade data showed that Citadel’s algorithm filled orders placed opposite to the defendants’ alleged spoof orders.  ⁶

In the Coscia case, for example, Twells examined a one second slice of data from the gold futures order book on September 2, 2011 in which Citadel’s algorithm filled buy orders that were placed by Coscia’s firm (Panther Energy Trading) immediately after Panther placed a series of sell orders. ⁷ Twells stated that Panther’s sell orders “were a factor influencing [Citadel’s]” decision to place its own sell orders, which resulted in Citadel selling gold futures to Panther. ⁸ Just 400 milliseconds later, Citadel bought back the same amount for $0.20 more, resulting in a loss of $480.⁹

Because Panther’s sell orders were left open for only a brief time period, DOJ used Twells’ testimony to convince the jury that they were illicit spoof orders designed to dupe Citadel into filling the buy orders that Panther supposedly really wanted to trade.

What Twells Didn’t Say

The questions posed to Twells on direct examination in Coscia framed Citadel’s brand of high-frequency trading as a relatively simple exercise: its algorithm processes various bits of information (including the aggregate number of open buy and sell orders in the order book for a particular security), estimates the security’s “fair value,” and trades accordingly.   ¹⁰

In this rendition, Panther’s spoof orders illegitimately skewed the algorithm’s fair value assessment by making it believe there was more supply and demand than was actually the case.  As a result, Citadel got tricked into engaging in transactions that it would not have had the spoof orders not been entered—or as DOJ puts it, spoofers cause HFTs like Citadel to trade “futures contracts at prices, quantities, and times that they would not have otherwise.”¹¹

But what Twells did not say is that, irrespective of its algorithms’ estimate of gold futures’ “fair value,” the reason Citadel placed orders in the same direction as Panther’s alleged spoof orders is that it was almost certainly trying to profit off Panther by trading ahead of them.¹² After all, this is one of the primary ways that HFTs make money.  

As explained by Professor Irene Aldridge, “The central value proposition of HFT is enabled by tick-by-tick data processing and high capital turnover.  The technology behind identifying small changes in the quote stream is what differentiates this business and enables a trader to send rapid-fire signals to open and close positions.”¹³ In simple terms: HFTs profit by quickly reacting to orders placed by other traders; while profits from each individual HFT transaction may be no more than a single tick, huge sums can be generated because algorithms are able to enter thousands or even millions of trades every day.¹⁴

While this overarching business model encompasses a variety of specific strategies, Twells’ testimony suggests Citadel’s algorithm was quote-matching, a tactic that involves “mimic[king] the limit orders of another trader,” “rid[ing] the market impact the original orders generate,” and then quickly taking “advantage of the move, reversing positions and capturing the profit.”¹⁵ Essentially a legal form of “electronic front running,” this strategy seeks to detect large order entries and then profit by trading in front of them at supra-human speeds. ¹⁶

In the data Twells discussed in Coscia, at 9:39:34.533 AM Panther entered a “real” order to buy 28 lots at¹⁷ $1,880.70.  At 9:39:34.558 AM, Panther placed a “spoof” order to sell 201 lots at $1,881.00.¹⁸ Citadel’s algorithm likely interpreted Panther’s large volume sell orders as a signal that someone in the marketplace needed to dump a big position.  To profit from this need, 0.221 seconds after Panther’s sell order was entered, Citadel placed its own sell orders in front of Panther’s and thereby incidentally filled Panther’s 24 lot buy order (i.e., the order that Panther actually wanted to execute). ¹⁹

Presumably, Citadel’s strategy was that by selling in front of the 201 lot sell order, it could ride a downward price movement and then buy back its position for cheaper.  But to Citadel’s chagrin, the anticipated price drop did not occur because Panther cancelled its sell immediately after its buy orders were filled.²⁰ As a result, prices bounced up and Citadel bought back the 24 lots it had just sold at $1,880.70 for $1,880.90 more, handing it a small loss.

So what really happened here?  The data indicates that in the competitive “contact sport” of trading,²¹ both Citadel and Panther were trying to profit off the orders entered by the other.  In this particular instance, Citadel’s strategy lost out.  If the market had moved another way, Panther could have easily lost out to Citadel (and, indeed, this result is what Citadel’s algorithm was likely vying to produce).  To present Citadel as a victim in such circumstances is disingenuous: it was no more a victim than a losing football team is a victim of the winner.²²

Did Citadel Really Not Trade at “Fair Value”?

To hammer home the “HFT as victim” narrative in the Coscia trial, DOJ elicited testimony from Twells that implied Citadel had traded gold futures at a price other than their “fair value.”²³ What exactly Twells meant by this term was never explained at trial, but the impression it likely gave the jury was that Citadel had been fooled into selling its gold futures for less than they were really worth. ²⁴

While at first blush this seems sensible (more traders wanting to sell = more supply = lower value), it looks more problematic when the trade data is examined in detail.  Just before Panther’s supposed “spoof” sell orders were entered, the highest bid for gold futures was $1,880.70 and the lowest offer was $1,880.90.  This means that at the exact moment in question (specifically, 9:39:34.523 AM on September 2, 2011), the market, which was as yet untainted by Panther’s spoofs, valued gold futures at roughly $1,880.80 (the midpoint between the buyers and sellers).  After Panther’s spoof orders were entered, Citadel sold 24 lots for $1,880.70 at 9:39:34.867 AM.²⁵

Juxtaposing this data against Twells’ testimony, it appears that DOJ’s theory was that the “real” fair value for gold futures at the time of Citadel’s sale was the “untainted” midpoint price of $1,880.80 shown at 9:39:34.523 AM (i.e., the midpoint that would have continued to exist but for Panther’s spoofs).   The $0.10 difference between the purportedly “real” and “tainted” fair values is ~0.005% of the total contract price.  Following this logic, instead of Citadel selling its 24 lots for the “right” price of $4,513,920, it was only able to get a measly $4,513,680 because of Panther’s deceptive tactics—hardly a discernable difference.

The point here is that trying to determine a futures contract’s exact, objective “fair value” on a millisecond by millisecond basis is extremely difficult, if not impossible.  The reality is that the “fair value” is never exactly one price or another: rather, it constantly fluctuates within a range over any given time period.²⁶ To implicitly claim that Citadel was cheated out of 0.005% of its gold futures’ “real” value is misleading. ²⁷

Which is not to say that Twells intentionally gave inaccurate testimony.  Rather, he was never pressed for details on what exactly the concept of “fair value” means, and the impression left by his high-level responses to DOJ’s queries was left to stand unchallenged.   

Would Citadel Have a Viable Fraudulent Misrepresentation Claim?

Another way to assess the “HFT as victim” theory is to consider whether Citadel would have a viable claim of fraudulent misrepresentation against Panther in the transactions cited above.  At a basic level, gold futures contracts are no different than ordinary contracts: they are agreements to exchange a quantity of goods (in this case, gold) at a particular time for a particular price.  As such, they can, in theory, be rendered void when one party has “fraudulently induced” the other party into contracting based on a material misrepresentation. ²⁸

According to the DOJ, a spoofer fraudulently induces a counterparty into filling its “real orders” using the false pretense that its “spoof orders” were placed with the “intent to trade them.”²⁹ Applying this theory to the Citadel example, DOJ’s position is that Citadel entered into a contract to sell 24 lots to Panther based on the false representation that the 201 lots Panther also had on offer would remain in the order book until they were filled (or at least for some time period longer than they actually were).   ³⁰

In other words, DOJ’s perspective is that Citadel was tricked into entering one transaction—a sale to Panther—based on the false representation that another proposed transaction—a buy from Panther—would remain available for some indeterminate time after the sale transaction went through.  

This confused idea is akin to a person complaining that they sold a gold ring (Ring A) under the false pretense that a different, but identical, gold ring (Ring B) would remain available to buy afterwards.  The only way this could conceivably serve as grounds for the seller to unwind the Ring A transaction is if the party who it sold to represented that Ring B would continue to be left available for sale until someone bought it.  But even then, it is not apparent how Ring B’s continued availability for sale could constitute a material condition of the transaction for Ring A.    

In any event, in futures markets this type of representation is never made—orders are not required to be left open for any minimum time period and every order can be pulled from the order book at any moment.  Going back to the Citadel sale to Panther at 9:39:34.867 AM, to the extent its algorithm was influenced by the number of lots then available for sale, its decision-making was based on completely accurate information.  For if we imagine a freeze frame of the order book at the precise moment Citadel transacted and count up the number of lots bid and offered, Panther’s sell orders should rightly be included in that number (regardless of whether Panther intended to cancel those orders moments later).  Even though Citadel’s assumption that Panther’s sell orders would remain open until executed turned out to be wrong, the responsibility for that error lies with whoever programmed Citadel’s algorithm, not Panther.  

At bottom, Citadel contracted to sell a specific quantity of gold to Panther for a particular price.  The terms of that transaction were fully and accurately disclosed, and Citadel got exactly what it contracted for.  If Citadel sued Panther for damages on a fraudulent representation theory in such circumstances, it would almost certainly lose (since there would be no evidence that Panther made any false representations to Citadel about how long its sell orders would be left open).   ³¹

If HFT firms would be unable to prove that they were fraudulently induced into contracting by alleged “spoofers” in a civil context, it is unclear why DOJ’s “HFT as victim” theory should be given any more credence in the criminal one.  

ENDNOTES

1. 7 U.S.C. § 6c(a)(5).
2. 14-cr-00551 (N.D. Ill.).
3. 17-cr-00220 (D. Conn.).
4. United States v. Vorley, 18-cr-00035 (N.D. Ill.), Dkt. #297 at 1.
5. Based on public filings, DOJ’s case in the upcoming United States v. Bases, 17-cr-00048 (N.D. Ill.) and United States v. Smith, 18-cr-00669 (N.D. Ill.) trials appears to include the theory that HFTs are victimized by spoofers.
6. While Varner has not yet appeared for the government in a spoofing case, one of his colleagues at Quantlab Financial, John Huth, testified in Flotron. The substance of Huth’s testimony was essentially identical to Twells’ in the Coscia and Flotron cases. See Vorley, Dkt. #297 at 1-2. DOJ expects Varner to testify consistently with Huth in the Vorley trial. See id. at 4.
7. See Coscia, Trial Tr. 635-638 (Oct. 28, 2015).
8. See id., Gov’t Ex. 11, Dkt. #177-11.
9. Id.
10. See Coscia, Trial Tr. 635-638 (Oct. 28, 2015).
11. See Vorley, Dkt. #12 at 4; see also Bases, Dkt. #301 at 11-12 (court determining that “the indictment sufficiently pleads that Defendants induced market participants into transactions that they otherwise would not have executed, under the false pretense that supply and demand were at a certain level when, in fact, they were not”).
12. To be clear, this article does not suggest that Twells made any intentional omissions in his testimony. Rather, he simply was never asked any questions that explored the details of how Citadel’s algorithm operated.
13. Irene Aldridge, High Frequency Trading (Hoboken: John Wiley & Sons 2013), p. 117.
14. See, e.g., Matthew Leising, “Virtu Never Loses (Well, Almost Never),” Bloomberg News (August 11, 2016) (describing how HFT firm earned $723 million in 2014 by making up to 5 million trades per day).
15. Aldridge, supra, p. 199.
16. See, e.g., Merrit B. Fox, et al., High-Frequency Trading and the New Stock Market: Sense and Nonsense, 65 Duke Law Journal 191 (2015).
17. Gold futures are traded in “lots” comprised of 100 contracts each.
18. See Coscia, Gov’t Ex. 12, Dkt. #177-12. Note that the actual data showed that Panther entered a 124 lot sell order at $1,881.00 and a 77 lot sell order at $1,880.90. To avoid unnecessary complication, these orders will be considered as if they were a single 201 lot order at $1,881.00.
19. Since the order book is anonymous, Citadel would have no way of knowing who it was selling to in real time. It only became apparent that Citadel was selling to the same party who had placed the 201 lot sell order after the fact (through trade data provided by the government). See Coscia, Trial Tr. at 634.
20. See Coscia, Gov’t Ex. 12, Dkt. #177-12.
21. Thomas Hieronymous, Economics of Futures Trading for Commercial and Personal Profit 327-28 (1977).
22. Coscia, Trial Tr. at 635. According to Twells, Citadel lost $0.20 per contract it sold to Panther (2,400 x $0.20 = $480).
23. See Coscia, Trial Tr. at 637 (“Q: And why were [Panther’s sell orders a factor in Citadel’s trading decisions?] A: . . . [O]rders in the in the order book can be used . . . as a sign of supply and demand, and so they’re one factor that we use . . . to estimate fair value.”).
24. DOJ intends to make this same claim in the Vorley case. See Vorley, Dkt. #297 at 2 (“The significance of [the] anticipated testimony [from Citadel and Quantlab] is that automated trading strategies . . . necessarily rely on futures market orders as expressions of genuine supply and demand to calculate a commodity’s fair value.”).
25. See Coscia, Gov’t Ex. 11, Dkt. #177-11.
26. Indeed, Twells’ testimony conflated “fair value” with “market value.” As the CME Group itself explains on its website, futures contracts generally do not trade at their intrinsic “fair value” because “short-term supply and demand will cause price to fluctuate around fair value.” See https://www.cmegroup.com/trading/ equity-index/fairvalue.html. Fair value (i.e., the contract’s theoretical value) is therefore distinct from market value (i.e., whatever price at which the market is willing to buy or sell it). Calculating the fair value of a commodities futures contract involves an exceedingly complex assessment of a variety of inputs (e.g., the commodity’s spot price, time until the contract expires, and cost to store the physical commodity), none of which turn on millisecond by millisecond changes in the order book. See, e.g., John C. Hull, Options, Futures, and Other Derivatives (Toronto: Pearson 2015), at p. 779-784.
27. See Hieronymous, supra (“A futures market is not a scholarly seminar in which learned men debate what is, and arrive at, an equilibrium price; it is a game in which businessmen . . . compete to establish a market price that works. Competition is sometimes a vicious business but it works well.”).
28. See Ralph C. Anzivino, The Fraud in the Inducement Exception to the Economic Loss Doctrine, 90 Marquette Law Review 4 (Summer 2007), at 931.
29. See, e.g., Bases, Dkt. #301 at 7 (“[T]he primary victims of the fraudulent scheme were not the counterparties to Defendants’ [spoof orders] but the counterparties to the [spoofer’s ‘real’ orders], who—along with the rest of the market—reasonably believed that the [spoof orders] were posted to the exchange . . . with the intent to fill them.”).
30. See id. at 11-12.
31. Note that class actions have recently been filed by investors against certain banks who have reached settlements with the government with respect to their traders’ purported spoofing. See, e.g., Robert Charles Class A, L.P. v. J.P. Morgan Chase & Co. et al., 20-cv-03206 (N.D. Ill.). Whether these actions have any merit remains to be seen. But in any event, they seek damages under a market manipulation theory, not one based in contract, and therefore do not have to prove any material misrepresentations were made. See id., Dkt. #1. Rather, they need only prove that the orders in question were placed with an illicit intent. See United States v. Coscia, 866 F.3d 782, 796-97 (7th Cir. 2017).