Mifid, initial margin and machine learning

The week on Risk.net, May 11–17, 2019

7 days montage 170519

No more delays on Mifid open access, EU regulator says

European authority confirms July 2020 start date, reigniting argument over market stability versus competition

Initial margin ‘big bang’ could be deferred, says EC’s Pearson

Senior EC official says smaller firms “really struggling” to comply with margin rules

Fund houses get picky over where to use machine learning

Buy-siders limit usage of deep learning techniques due to haziness over their inner workings


COMMENTARY: A way forward for explainability

Fund managers, like most of the rest of the financial sector, have been experiencing problems with machine learning (ML) – in particular, fear of relying on an incomprehensibly complex process and being unable to explain (to superiors, customers, regulators or the police) why it has suddenly gone wrong. In some cases, they’re reacting by using only the simplest linear models. These are easily interpreted and understood, but sacrifice a lot of the potential power of ML to find non-linear patterns in non-traditional data sets.

Other approaches look more promising – constructing decision trees using Bayesian methods rather than choosing from a randomly-generated forest of possible trees saves time on analysis, and also produces a more explicable decision tree. Greater familiarity with the use of ML may help too, as may focusing on areas where ML can augment human analysis rather than replace it, for example by highlighting price-sensitive text in company announcements, analyst reports or other sources.

But the danger of over-reliance on more opaque forms of ML won’t go away – the potential power of sophisticated automated analysis is just too great. The industry needs a constant focus on explainability, including regular exercises to investigate how apparently functional and successful systems are producing such good results. Many frauds, rogue traders and other scandals started because of the financial industry’s unwillingness to investigate the origins of abnormally good profits or unreasonably high growth. ML supervisors need to be more willing to look at their systems before the wheels come off, rather than afterwards.



Australian banks continue to count the cost of misconduct illuminated by the Royal Commission enquiry published in February. The ‘Big Four’ banks collectively subtracted A$1.7 billion ($1.2 billion) pre-tax from first half earnings (September to March), up from A$1.3 billion in the second half of last year. Royal Commission refunds weigh on Aussie banks



“It started up on WhatsApp, and it started propagating. It had a huge, huge impact on the share price of the bank. It can trigger a crisis.” Cosimo Pacciani, from the European Stability Mechanism, on emerging risks from social media after a rogue message claimed Metro Bank was facing closure.

Only users who have a paid subscription or are part of a corporate subscription are able to print or copy content.

To access these options, along with all other subscription benefits, please contact info@risk.net or view our subscription options here: http://subscriptions.risk.net/subscribe

You are currently unable to copy this content. Please contact info@risk.net to find out more.

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an individual account here