Meeting of the Motherboards: The Validity of Contracts Created by Algorithmic Traders

“[B]ug, n: An elusive creature living in a program that makes it incorrect. The activity of ‘debugging’, or removing bugs from a program, ends when people get tired of doing it, not when the bugs are removed.”—Datamation

Today’s stock market is dominated by high frequency algorithmic trading. Algorithmic trading refers to the practice of using pre-defined quantitative algorithms to determine when to buy and sell stocks. The algorithms are designed and implemented by so-called “quants,” analytical mathematicians and statisticians who attempt to discover and benefit from patterns in the market. High frequency trading typically involves the transfer of thousands of stocks per second. These exchanges take place automatically and are rarely reviewed. In most cases, the trading algorithms work well, allowing brokers to track prices, propose trades, and influence the market faster than ever before. But what happens when the algorithm fails to function as intended?

On May 6, 2010, “in the space of a few, hair-raising minutes, a breakdown in the market triggered by failures in computer-trading systems across the country caused stocks to plunge about 10%.”[1] A report later issued by the SEC found that the incident, dubbed the “flash crash,” was caused by a single erroneous sale. More recently, on August 1, 2012, Knight Capital, a high frequency trading group, was affected by a bug in its trading algorithms which ultimately caused it to lose over $440 million and 75% of its value in two days. These examples show how software errors in algorithmic trading can lead to a bizarre outcome in which two parties can be subject to an exchange that neither actually wants to happen. Even more problematically, as evidenced by the flash crash, these detrimental and involuntary transactions quickly propagate through the entire market, wreaking havoc on other trading algorithms and making it more likely that more traders will engage in unproductive, unexpected, and undesired exchanges. 

It is a basic tenet of contract law that the formation of a valid contract requires a “meeting of the minds,” an agreement between the parties entering into the contract. It is generally recognized that such a meeting of the minds can exist even when parties to a contract use an automated algorithm to determine the details of the exchange. For example, contracts formed through eBay’s automated bidding system have been upheld by courts. But what are the limits to such automated exchanges? What constitutes a “meeting of the minds” for the purposes of an automated contract? Is it enough to rely on pre-approved trading algorithms, or should we condition a contract’s validity on the conformity of the transaction with a trader’s expectation of how the algorithm should function?

To answer these questions, the National Conference of Commissioners on Uniform State Laws, an organization that seeks to promote uniformity in law, introduced the Uniform Computer Information Transactions Act (UCITA). While only two states have approved UCITA, courts may still look to the act for guidance. Section 206 deals with contracts created by “electronic agents.” Specifically, the act states that contracts entered into by automated systems should be binding unless the contract resulted “from fraud, electronic mistake, or the like.” The UCITA goes on to clarify “electronic mistake,” noting that there might not be a contract if, “because of aberrant programming or through an unexpected interaction of the two agents, operations indicating existence of a contract occur in circumstances that are not within the reasonable contemplation of the parties.”

This places high frequency algorithmic trading in something of a quandary. Because human involvement is needed to determine whether an interaction is “unexpected,” and because millions of transactions can take place before any independent determination can be made, trading algorithms will have no way of determining whether the exchanges produced by other high frequency traders are legitimate or whether they are the products of “unexpected” interactions, subject to future repudiation. That is, algorithms have no way of determining which trades “count” and should be used as criteria to determine future trades. Accordingly, under the UCITA, if traders could exercise their option to repudiate, high frequency trading would become functionally impossible.

If, on the other hand, “unexpected interactions” do not function as a basis for repudiation, if events such as the flash crash are to be expected, or if high frequency traders elect not to repudiate contracts formed from unexpected interactions[2], then, absent any additional oversight or regulation, there would be nothing preventing a sub-optimal trading algorithm from introducing undesired “phantom” trades into the market. High frequency trading would become little more than a game of collective Russian roulette, where each algorithmic exchange is a pull of the trigger, little more than an opportunity for a bug in the code to induce a superfluous algorithmic panic with the potential to destabilize the entire market with millions of undesired, yet legally enforceable, exchanges.

Indeed, given that it is not standard practice to repudiate trades based on software glitches, some think we have already reached the point of collective financial suicide. While the simple solution would be to stop performing these trades, the Federal Reserve could also implement better checks on errant systems to prevent one flawed algorithm from toppling the entire market. Whichever solution is implemented, it is clear that high frequency trading, as it exists now, is unsustainable, irresponsible, and, insofar as traders will be subject to exchanges which harm everyone and help no one, illogical.