How should we police the trading bots?

we police the trading bots

• Please take advantage of the article-sharing options accessible through the share button. The terms of service and copyright policy of FT.com prohibit users from sharing articles that they have copied. More permissions may be purchased by contacting licensing@ft.com. When subscribers use the gift article service, they may send up to ten or twenty articles to friends and family each month.
• One intriguing aspect about the end of flash crashes is that they should no longer occur. This is because well-established, comprehensive, and stringent laws are in place to safeguard markets against algorithmically induced instability.
• It doesn’t seem like those regulations are being followed.
• Sometimes the results of an action are immediately apparent, as was last month when European markets dropped because a Citigroup trader in London allegedly appended a zero to an order. However, the session’s widespread repercussions imply that algos at numerous companies were slow to react to lower-than-usual volume, exacerbating the volatility. This, in turn, begs the thorny issue of whether or whether the same algorithms are causing trouble in less clearly tense circumstances.
• As a general rule, securities laws treat all forms of market abuse equally, whether they are the work of humans or computers. Behavior is what counts. They might expect difficulty whenever a person or company threatens market integrity, destabilizes an order book, sends a deceptive signal, or commits other nebulously defined transgressions. Whether or not this results from a particular process is unimportant.
• In addition, it constitutes market abuse for an algorithm to misbehave when matched against another company’s manipulative or faulty trading approach. No more than a person would be able to use “I was just too stressed to think straight” as an excuse; a robot cannot use this defense.
• Due to this, trading bots need extensive testing before being put into production. In addition, businesses need to ensure they are secure against standard attack methods like momentum igniting and fat finger mistakes. The goal is to prevent a chain reaction of failures like the “hot potato” phenomenon that exacerbated the 2010 flash meltdown by ensuring that algorithms don’t repeat the same mistake.

• MiFID II’s (in effect as of 2018) Voight-Kampff test is comprehensive. Investment businesses utilizing European venues must verify that any algorithm won’t contribute to disorder and will remain to perform efficiently “under stressed market circumstances.” Exchanges should require members to verify that their bots have been tested under “actual market circumstances” before deployment or upgrading as part of their police duties.
• Going into the revised Mifid II RTS is essential for a more in-depth understanding of this.
• If a financial institution wants to ensure that its bots won’t aggravate markets, it may use the fundamental self-assessment approach outlined in RTS 6. However, whether bots would contribute to market disturbance is distinct and different in its successor, RTS 7. To sum up, an RTS 7-compliant company must attest that all systems will not exacerbate any market convulsion and detail the testing procedures that led to this conclusion.
• Although RTS 6 is widely known and used, how many trading businesses are up to speed on RTS 7? Technical director at consultancy TraderServe Nicholas Idelson estimates that less than half of all algo strategies have undergone adequate stress testing. However, given the scope and complexity of the undertaking, even this estimate might need to be higher.
• MiFID II’s definition of an algorithm allows for automated venue routing while catching almost all other forms of automated gambling. Algos are used to generate quotes when “little or no human interaction” is needed. An algorithm is used whenever a set of rules decides the pricing, order size, or time. When a proposal is submitted, algorithms are used if they use any strategy beyond simple implementation. These systems must pass a battery of stress tests to ensure they will function as expected in their final configuration.
• This legislation covers all financial products specified by MiFID II that may be traded on any platform that supports automated trading. Including the “or enables” clause applies to both venues that prohibit and those that do not use auto-matching trading algorithms. (See ESMA’s Q&A about the upcoming year of 2021, specifically Q&A 31.) If one takes a literal reading of the criteria, it becomes almost impossible for any deal to fulfill the best execution requirements without being classified as automated.
• If a company does not comply, it might be fined up to €15 million, or 15% of its annual revenue, and an individual could face up to four years in jail. A similar image can be seen worldwide thanks to IOSCO’s market integrity principles, which may be used as a basis for international regulation.
• However, the approach to algo policing in the UK and Europe has been softly-softly in contrast to the US (where JPMorgan Chase landed a $920mn settlement in 2020 for spoofing futures contracts on precious metals) and Hong Kong (where Instinet Pacific and Bank of America have been penalized for failings in bot management and oversight respectively). As the Financial Conduct Authority (FCA) noted in its May 2021 Market Watch bulletin, our internal surveillance algorithms identified trading by an algorithmic trading firm, which raised potential concerns about the impact of the algorithms responsible for executing the firm’s various trading strategies. In response to our questions, the company revised the relevant algorithm and its control system to mitigate the risk of the firm’s actions disproportionately impacting the market.

• The difficulty in defining the contribution to market disorder stress testing is a problem for regulators.
• Do companies only need to test their bots in a sandbox using historical market data or do they need live data? Or does this method run the danger of missing the feedback loops that form when the fleet engages with receptive markets? Idelson claims that it is hard for an outsider to determine whether any firm’s approach to testing was extensive, superficial, or nonexistent, even though TraderServe has collaborated with regulators on best practices utilizing live market simulations. That’s why it’s helpful to provide some public examples.
• Europe’s nudge approach to bot regulation needs to be revised in light of the vulnerabilities of the Citi flash collapse. Yet, proactive measures of policing are hard to sustain due to the extensive requirements outlined in MiFID II. Show trials may be the most effective method of enforcement available to regulators if noncompliance is as pervasive as it seems.