RegTech Intelligence

Know thy algo: how to define it, prove it, tame it. Part 1

JWG analysis.

Regulators across the globe appear divided on the question of whether tighter control of algorithmic trading is necessary.

The Australians are pretty laid back about it, the Germans are ahead of the game, whilst political debate rages in the US.  Regardless, while the value of algo trading to global markets is generally considered to be substantial, the idea that it compromises market safety is becoming more widely accepted.

Under the auspices of MiFID II, the EU looks set to build on the German HFT act and go further than any other jurisdiction in attempting to tighten controls on algorithmic trading.  Pushing aside the doubters down under, ESMA appears to believe that prescribing safety measures will give Europe an edge in an unforgiving global financial system.

The MiFID II hearings were clear: if you want to get your views in, start writing now – you have 22 days left to submit responses.  They are moving into uncharted waters that could have grave commercial consequences, so the importance of consultation is high.  This is especially true since, looking across the 860 consultation and discussion questions, there is a huge number of issues which will affect trading models – many of which are open for debate.

We have been looking at two main groups of questions that MiFID II raises for algo traders, which will exist regardless of the direction of travel of some of the more political questions about benchmarks, DEA (DMA to some) and the commercial basis for pricing.

Here, we focus on the questions of how you define algorithms and the extent to which you look after them.  As noted above, it’s not too late to get your responses in.  Given the size, scope and potential depth of the remediation exercise required, it makes sense to get started on making the unknowns known.

Question 1: what is an algo?

Algorithmic trading has been defined fairly broadly and will include any trade in which a computer algorithm automatically determines individual parameters of an order, for instance, timing, price or quantity with no, or limited, human intervention.

Leaving aside the issues of what rules apply when a machine sits outside the EU, this means that, if you have a spreadsheet summing a few columns, checking a reference table and doing a VBA lookup on a server, you might well now have multiple algorithms that you think of and control as one system, with one set of data licences and operational risk controls.

No doubt exchanges and market data providers will be quite interested in reviewing firms’ statements for, not only the number of algos that they have, but how much of their ingredients they consume.  Worried about the impact of this on the cost/income ratio?  Get your views on questions 167-171 in the consultation paper in to ESMA.

Question 2: how and when is it tested?

Under article 17 of MiFID II, investment firms engaged in algorithmic trading will be subject to rigorous testing requirements, the cornerstone of which is mandatory use of testing in a non-live environment.  This also includes ongoing testing of systems, procedures and controls, targeted at ensuring that algos will not destabilise the market – the criteria for which have yet to be defined.  It seems likely that there will be serious difficulties here in terms of segregation, as well as ensuring the testing environment sufficiently reflects the complexities of the live environment.  These obligations go much further than the German HFT Act, which does not attempt to control the operation of algos (being much more focused on monitoring).

The depth of testing that will be enforced in practice remains unclear, but it will concern firms to the extent that such testing requirements, particularly rules around cautious roll-out and non-live initial testing, have the potential to constrict the adaptability that makes such trading practices so effective.  Key to this debate will be how significant a change is.  If a ‘self-learning’ algo decides to upgrade its behaviours, is that meant to trigger a comprehensive set of testing procedures?  Views on questions 199-204 in the discussion paper, please.

Question 3: how will operational risks be mitigated?

According to the Level one text, firms engaging in algorithmic trading will be required to have in place effective systems and risk controls to ensure that their systems are resilient.  Key provisions are stress testing and circuit breakers designed to ensure that HFT trading does not result in disorderly markets.  Only firms that can prove that they have such systems and controls in place will be authorised to engage in algorithmic trading.  ESMA has stipulated that all such controls should be up for a twice yearly review.

At this stage of the process, the precise controls that firms will need to have in place remain unknown, and it is possible that firms may be able to rely on, as yet unspecified, exemptions and therefore not be required to get authorisation.  Clearly, this will not be the first time operational risk teams have looked at trading activity.  It could well be, however, that the prescriptive nature of the controls cuts across the current RCSAs – both as to timing and level of detail – and new op risk policies, controls, reporting and governance may well be required.  Comments on questions 212-218 in the discussion paper quickly, please.


We have until 1 August to help shape this debate.  Do not miss this chance to get your views in.  Your friendly trade associations are there and ready to help.

Watch this space for our follow-up piece on the issues of proof and transparency.

To promote global dialogue on how to deliver regulatory change JWG post hundreds of focused articles a year to thousands of subscribers. Get involved and join the mail list.

By hitting the subscribe button you agree to our Privacy Policy