One of the less well understood aspects of the EU’s Markets in Financial Instruments Directive MiFID II is its requirement for trading firms to adopt effective testing frameworks for the algorithms they deploy. While many aspects of MiFID II are aimed at preventing a repetition of the Credit Crisis of 2008, the algo testing provisions are more directed toward avoiding another Flash Crash, the severe market disruption in 2011 that wiped out billions of dollars of market value in milliseconds, before adjusting back to earlier levels.
MiFID II doesn’t represent the first instance of regulators seeking to get a grip on the use of trading algorithms. Individual execution venues and exchanges have their own rules, regulations and market conventions governing the use of automated trading strategies. Since early 2014, for example, traders on Deutsche Boerse’s Eurex derivatives and Xetra cash equities trading systems have been required to provide specific user ID – listing trader or trading programme – with every market order and trade. This allows the exchange to identify the unique decision path leading to any order or quote, including those generated by machines.
But the new requirements under MiFID II take things a step further by insisting firms understand the impact their algorithms will have on the marketplace, including the reaction of other algorithms active in the segment. MiFID II will require all trading firms to certify that their algorithms have been tested to ensure that they do not create or contribute to disorderly trading conditions before being deployed in live markets.
Firms will also be required to understand what constitutes an algorithm, and whether that includes aspects of their smart order router (SOR) functionality, for example. Much of the focus in responding to the new requirements will be on execution venues to provide facilities for members and their clients to test the impact of their aggressive and non-aggressive trading algorithms in realistic market conditions.
How this will translate into practice is yet to be seen. While the venues – exchanges, MTFs, OTFs – will need to make necessary arrangements, possibly with third-party providers, financial institutions will need to commit to and demonstrate use of such facilities. Firms will also need to run their own tests of the impact of their trading strategies and algorithms before they are released to market.
Although no best practices have yet been established, it appears that traditional testing of algos using simple canned replay of market developments would not be enough to meet the criteria under MiFID II, since they require firms to understand the impact of their algorithms in the face of other algorithmic trading activity.
It is, however, unclear how regulators will ascertain whether firms have properly measured and adequately tested the potential impact of their algos. Firms need to play their algos against potentially antagonistic strategies and explore how the algorithm may behave in the real world. Armed with this information – and some idea of the likelihood of an algorithm having a negative impact on the marketplace as a whole – compliance officers will be able to sign off (or not), comfortable that they had done adequate due diligence to meet the MiFID II requirements.
Whether offered by a trading venue or implemented individually by trading firms, any algo testing solution needs to emulate live market environments, including simulation of stressed market conditions, in order to fully test algorithms against the key behaviors that contribute to market disorder events, such as a flash crash. Financial institutions also need to be aware that the testing obligation extends to those firms using third-party trading strategies and models to execute order flow.
Again, how regulators will decide whether firms have tested their (or third parties’) algorithms sufficiently remains to be seen, as does how firms will implement these tests before the regulation deadline without incurring significant costs. By assigning a pass / fail testing result against predetermined indicators of disorderly market behaviors, and supporting testing of algorithms that trade across multiple markets, this kind of approach can help compliance officers ascertain the risk of disruption posed by their firms’ algorithms.