On a dark street in Tempe, Arizona, on the night of March 18, 2018, an autonomous car owned by a popular ride hailing service struck and killed a pedestrian, causing what is believed to be the first autonomous vehicle fatality.[1] Although this accident was perhaps no different than the thousands of other fatal car accidents that occur each day that don’t involve “self-driving” cars, it does contribute to a collective concern about the role and implementation of this seemingly inevitable technology into virtually every aspect of our lives.
Physically crashing into people is one of the more extreme and tangible examples of potential harm, but what about when a robo-adviser crashes your portfolio or an algorithmic trading strategy crashes an exchange? Machine learning and artificial intelligence have already progressed to a point where these are no longer hypothetical, science-fiction concerns. These things have already happened and despite our best efforts, to paraphrase a fictional, Hollywood robot that speaks with an Austrian accent from a dystopic future dominated by artificial intelligence, these vast and complicated issues will indeed “be back,” and like some sequels, possibly bigger and more dangerous than before.
Automated Trading
We were given a preview of a potential financial dystopia eight years ago on May 6, 2010, when the so-called “flash crash” occurred. As detailed in a joint report,[2] the flash clash was caused by a confluence of factors. However, one of the primary drivers of the crash was that many of the liquidity providers, who almost exclusively use algorithmic trading strategies, substantially lowered their bids and/or began selling into the decline in prices, further exacerbating the already significant downward movement in the equity-index futures prices. This in turn triggered a response from other algorithms, many of which also started selling. Contemporaneously with the decline in equity-index futures, other products, including exchange-traded funds — several of which hold the stock of blue-chip issuers — were similarly sold causing a similar downward spiral in the cash-equity markets. Although this is one of the most popular examples, it is far from the only one[3] and was likely not the first.
A number of things were done in response to the flash crash and its ilk by regulators, exchanges and issuers themselves, including banning certain trading practices, such as spoofing, layering and front running, as well as implementing modified trading curbs, adding “circuit breakers” and amending existing rules concerning market access. Recently, regulators have refined their rules and brought enforcement actions to give the new regulations some teeth.
For example, in 2014, the U.S. Securities and Exchange Commission brought an action against Athena Capital Research for using an algorithm code-named “Gravy” to engage in a practice known as “marking the close” in which stocks are bought or sold near the close of trading to affect the closing price.[4] The massive volumes of Athena’s last-second trades allowed Athena to overwhelm the market’s available liquidity and artificially push the market price — and therefore the closing price — in Athena’s favor. Athena was acutely aware of the price impact of its algorithmic trading, calling it “owning the game” in internal e-mails.
According to the SEC’s order instituting a settled administrative proceeding, although Athena was a relatively small firm, it dominated the market in the last few seconds of a trading day for stocks that it otherwise traded only slightly using sophisticated, high-speed algorithmic strategies. The manipulative trading described in the SEC’s order made up more than 70 percent of the total Nasdaq trading volume of the affected stocks in the seconds before the market close.
The algorithmic strategies were designed to ensure that the firm was the dominant firm — and sometimes the only one — trading desirable stock imbalances at the end of each trading day. The firm implemented additional algorithms known as “collars” to ensure that Athena’s orders received priority over other orders when trading closing imbalances. These eventually resulted in Athena’s imbalance-on-close orders being at least partially filled more than 98 percent of the time. Athena’s ability to predict that it would get filled on almost every imbalance order allowed the firm to unleash its manipulative Gravy algorithm to trade tens of thousands of stocks right before the close of trading. As a result, these stocks traded at artificial prices that Nasdaq then used to set the closing prices for on-close orders as part of its closing auction. Athena’s high-frequency trading scheme enabled its orders to be executed at more favorable prices.
As recently as Oct. 12, 2018, the U.S. Commodity Futures Trading Commission acted against trader Kamaldeep Gandhi. Gandhi admitted to participating in manipulative and deceptive schemes, along with other individuals, which involved thousands of acts of spoofing (bidding or offering with the intent to cancel the bid or offer before execution) with respect to a variety of futures products traded on the Chicago Mercantile Exchange, Chicago Board of Trade, New York Mercantile Exchange and the Commodity Exchange Inc.[5] The order also said that Gandhi engaged in this unlawful activity while placing orders for, and trading futures contracts through, accounts owned by his former employers.[6] And this was just one of two such actions filed by the CFTC last month for similar conduct.[7]
The CME Group and other exchanges have also brought enforcement actions against firms and individuals for failing to supervise their algorithmic trading systems, or ATSs. For example, the CME Group brought an enforcement action[8] against a member firm for alleged supervisory failures when the firm’s ATS did not operate as intended, resulting in the execution of 17,000 contracts in the various legs of the eurodollar complex, which caused price and volume aberrations, as well as a number of wash trades. The exchange concluded that the firm improperly configured and modified its ATS and did not conduct appropriate testing before deploying the ATS in a live trading environment.
The above example illustrates that without proper testing, ATSs can and do cause market aberrations even when the ATS operator did not intend to engage in any prohibited trading activity. As such, firms using algorithmic trading strategies should adequately test their algorithms before deploying them in the market, and document the results of those tests to establish that the firm fulfilled its supervisory responsibilities.
Regulators also recognized that they needed a better registry of those responsible for the various algorithmic trading strategies and related products. In 2016, the SEC approved the Financial Industry Regulatory Authority’s rule change to include the registration of the “developers of algorithmic trading strategies” and those that are “responsible for day-to-day supervision of such designs or significant modifications” of algorithmic strategies.[9]
This representative registration category now resides in FINRA Rule 1220 regarding “registration categories,” and added the following language requiring registration of securities traders:
In addition, each person associated with a member who is: (i) primarily responsible for the design, development or significant modification of an algorithmic trading strategy relating to equity, preferred or convertible debt securities; or (ii) responsible for the day-to-day supervision or direction of such activities shall be required to register with FINRA as a Securities Trader.
For purposes of paragraph (b)(4) of this rule, an “algorithmic trading strategy” is an automated system that generates or routes orders (or order-related messages) but shall not include an automated system that solely routes orders received in their entirety to a market center.[10]
Foreign Jurisdictions Have Taken Similar Steps
The U.S. is not alone in its increased regulation of algorithmic trading. The European Securities and Markets Authority has also set forth regulations governing algorithmic trading.[11] ESMA requires algorithmic traders to establish pretrade risk controls[12] in all financial instruments. ESMA also requires that trading venues implement certain rules and procedures related to algorithmic trading. Specifically, trading venues are required to: (1) insure that their members conduct conformance testing relating to their algorithmic trading, (2) provide access to a test trading environment, and (3) monitor and test the capacity of their systems, among other things.
Similarly, in 2017, the Japan Diet passed amendments to the Financial Instruments and Exchange Act of Japan, which established a new framework for fund managers and traders engaging in high-frequency trading on Japanese exchanges. While historically, there were no rules or regulations that specifically governed HFT operators in Japan, under the amendments, HFT operators are required to register with the Financial Services Agency of Japan. Further, the Japan Exchange Group, a leading Japanese exchange, also implemented new rules on high-frequency trading intended to make the market more transparent, as well as to reduce price volatility and prevent market manipulation.
Although algorithmic trading rules differ from jurisdiction to jurisdiction, the core principals are generally the same, which is to improve the resiliency of the increasingly algorithmic and interconnected global markets.
The “Robo-Advisers”
In addition to algorithmic trading, the automation of the investment adviser role is another area ripe for regulatory expansion. In February 2017, the SEC’s Division of Investment Management issued regulatory compliance guidance concerning “robo-advisers.”[13] This guidance made clear that the computational autonomy of such systems and products were no less subject to the fiduciary obligations of the Investment Advisers Act than any human operating in the same capacity and set forth the following considerations:
- The substance and presentation of disclosures to clients about the robo-adviser and the investment advisory services it offers;
- The obligation to obtain information from clients to support the robo-adviser’s duty to provide suitable advice; and
- The adoption and implementation of effective compliance programs reasonably designed to address particular concerns relevant to providing automated advice.[14]
With respect to the compliance considerations, the Division of Investment Management suggested the following to advisers offering automated services and products:
- The development, testing and backtesting of the algorithmic code and the post-implementation monitoring of its performance (e.g., to ensure that the code is adequately tested before, and periodically after, it is integrated into the robo-advisers’ platform; the code performs as represented; and any modifications to the code would not adversely affect client accounts);
- The questionnaire eliciting sufficient information to allow the robo-adviser to conclude that its initial recommendations and ongoing investment advice are suitable and appropriate for that client based on his or her financial situation and investment objectives;
- The disclosure to clients of changes to the algorithmic code that may materially affect their portfolios;
- The appropriate oversight of any third party that develops, owns, or manages the algorithmic code or software modules utilized by the robo-adviser;
- The prevention and detection of, and response to, cybersecurity threats;
- The use of social and other forms of electronic media in connection with the marketing of advisory services (e.g., websites, Twitter, compensation of bloggers to publicize services, “refer-a-friend” programs); and
- The protection of client accounts and key advisory systems.
In August of this year, the SEC settled with a registered investment adviser for alleged violations of the Investment Advisers Act’s anti-fraud and compliance policy provisions[15] for quantitative products marketed to customers that improperly characterized the abilities and performance of the models being used as well as the credentials, competence and diligence of those purportedly creating and implementing the product. The SEC’s order explains that customers put billions of dollars into mutual funds and strategies using faulty models developed by the investment adviser and its affiliate investment advisers based upon claims that investment decisions would be based on quantitative models. The SEC’s order finds that the models, which were developed solely by an inexperienced, junior analyst, contained numerous errors, and did not work as promised. The SEC also found that when the investment adviser and its affiliates learned about the errors, they stopped using the models without telling investors or disclosing the errors.
Other investment advisers have previously been sanctioned for defects, poor maintenance and compliance practices when utilizing various quantitative models and products.[16]
For the foreseeable future, the amount of algorithmic trading activity is expected to increase substantially. As such, the actions of one algorithm will almost certainly cause reactions from other algorithms. Given this potentially cascading feedback loop between algorithms, we anticipate that global regulators will continue to pay particularly close attention to market structure and algorithmic trading practices to ensure that the actions of one or a few market participants cannot meaningfully disrupt the prices of any market. Accordingly, we expect regulators to continue to be aggressive in bringing enforcement actions against rogue or manipulative algorithms. Thus, it is important for all market participants using algorithmic trading strategies to appropriately test their algorithms for efficacy and compliance before deploying them in the markets.
This article was originally published at Law360.com.
——————————————————-
[1] Self-Driving Uber Car Kills Pedestrian in Arizona, Where Robots Roam, New York Times, By Daisuke Wakabayashi, March 19, 2018.
[2] On Sept. 30, 2010, after almost five months of investigations led by Gregg E. Berman, the U.S. Securities and Exchange Commission and the U.S. Commodity Futures Trading Commission issued a joint report titled “Findings Regarding the Market Events of May 6, 2010,” identifying the sequence of events leading to the flash crash.
[3] On July 13, 2015, after seven months of investigating the “flash crash” that occurred in the U.S. Treasury markets, the U.S. Department of the Treasury, Board of Governors of the Federal Reserve System, Federal Reserve Bank of New York, U.S. Securities and Exchange Commission and U.S. Commodity Futures Trading Commission issued a joint staff report titled: “The U.S. Treasury Market on October 15, 2014” examining the unusually high level of volatility and very rapid round-trip in prices in market for U.S. Treasury securities, futures and other closely related financial markets.
[4] In the Matter of Athena Capital Research LLC, Respondent. Securities Exchange Act of 1934 Release No. 73369/Oct. 16, 2014, Investment Advisers Act of 1940 Release No. 3950/Oct. 16, 2014, Administrative Proceeding File No. 3-16199.
[5] In the Matter of: Kamaldeep Gandhi, Respondent. CFTC Docket No. 19-01.
[6] The CFTC order finds that from at least September 2012 through March 2014 at Firm A and from at least May 2014 through October 2014 at Firm B, Gandhi, both individually and in coordination with others, placed thousands of orders to buy or sell futures contracts with the intent to cancel those orders prior to execution. In doing so, the order finds that Gandhi intentionally sent false signals of increased supply or demand designed to trick market participants into executing against the orders he wanted filled.
[7] The CFTC issued an order filing and settling charges against the Bank of Nova Scotia for engaging in multiple acts of spoofing in gold and silver futures contracts traded on the Chicago Mercantile Exchange. The order finds that BNS engaged in this activity by and through traders on its precious metals trading desk from at least June 2013 through June 2016. The order requires BNS to pay an $800,000 civil monetary penalty and to cease and desist from violating the Commodity Exchange Act’s prohibition against spoofing. BNS was notified of the misconduct by its Futures Commission merchant, and when BNS became aware of the misconduct, BNS reported it to the CFTC.
[8] See CME Enforcement Action: CME 14-0064-BC.
[9] Securities and Exchange Commission (Release No. 34-77551; File No. SR-FINRA-2016-007), April 7, 2016 Self-Regulatory Organizations; Financial Industry Regulatory Authority Inc.; Order Approving a Proposed Rule Change to Require Registration as Securities Traders of Associated Persons Primarily Responsible for the Design, Development, Significant Modification of Algorithmic Trading Strategies or Responsible for the Day-to-Day Supervision of Such Activities.
[10] FINRA Rule 1210
(4) Securities Trader
(A) Requirement
Each representative as defined in paragraph (b)(1) of this Rule shall be required to register with FINRA as a Securities Trader if, with respect to transactions in equity, preferred or convertible debt securities effected otherwise than on a securities exchange, such person is engaged in proprietary trading, the execution of transactions on an agency basis, or the direct supervision of such activities, other than any person associated with a member whose trading activities are conducted principally on behalf of an investment company that is registered with the SEC pursuant to the Investment Company Act and that controls, is controlled by or is under common control, with the member.
In addition, each person associated with a member who is: (i) primarily responsible for the design, development or significant modification of an algorithmic trading strategy relating to equity, preferred or convertible debt securities; or (ii) responsible for the day-to-day supervision or direction of such activities shall be required to register with FINRA as a Securities Trader.
For purposes of paragraph (b)(4) of this Rule, an “algorithmic trading strategy” is an automated system that generates or routes orders (or order-related messages) but shall not include an automated system that solely routes orders received in their entirety to a market center.
[11] ESMA defines “algorithmic trading” as trading where a computer algorithm automatically determines individual parameters of the orders with “limited or no human intervention,” other than any system that: (1) routes or processes orders without determining such parameters; (2) confirms trades; or (3) performs post-trade processing. See Article 4(1)(39) of Markets in Financial Instruments Directive (Directive 2014/65/EU of the European Parliament and of the Council of 15 May 2014 on markets in financial instruments and amending Directive 2002/92/EC and Directive 2011/61/EU) (“MiFID II”); Article 18 of the Definitions Delegated Regulation
[12] Price collars, max order value, max order volum; and max message limits.
[13] February 2017 | No. 2017-02 Robo-Advisers. This guidance defined the term “robo-adviser” to include both registered investment advisers and any automated investment advisory programs offered.
[14] While this guidance focuses on the obligations of robo-advisers under the Advisers Act, robo-advisers should consider whether the organization and operation of their programs raise any issues under the other federal securities laws, including the Investment Company Act of 1940, and in particular Rule 3a-4 under that act.
[15] Advisors Act Section 206(2), 206(4)-1(a)(5) (false advertising), 206(4)-8 (false or misleading statements by Investment Advisor to a “pooled investment vehicle,” 206(4)-7 (inadequate or nonexistent policies or procedures).
[16] See, e.g., In the Matter of AXA Rosenberg Group LLC et al., Advisers Act Release No. 3149 (Feb. 3, 2011) (settled action) (In a settled administrative proceeding, the commission found that two affiliated investment advisers that used a quantitative investment model in managing client accounts breached their fiduciary obligations to their clients by concealing and delaying to fix a material error in the model. One of the investment advisers was also found to have failed to adopt and implement policies and procedures reasonably designed to ensure that it did not make false and misleading statements to clients and investors, including failing to ensure that the model performed as represented, in violation of anti-fraud provisions of the Advisers Act).
