26th August 2020

Algorithms don’t do a good job of detecting their own flaws

An algorithm is a clearly defined step-by-step set of operations to be performed by a computer. 

Let’s say you are asked to create an algorithm for making a gin and tonic to drink from a specified ‘gin’ glass. The algorithm or set of actions for that will be — to get the gin, in my case at the moment Sipsmith (other gins are available in BBC speak),open the bottle, pour the defined measure of gin into the correctly selected gin glass, add a wedge of lime then premium tonic, add lots of ice and place the glass on a coaster in just the right place, at the right time, for me to drink.  

Deep joy, Stanley Unwin would have said!

Similarly, in a computer system, when you need a machine to do something for you, you explain the job clearly by setting instructions for it to execute. And that process is also called programming.

Algorithms are the thing in financial services at the moment.

Previously it was called mechanical, systematic, black box or rule-based trading. Sometimes even high frequency trading. Now most people refer to it as algorithmic or algo trading, but the idea has not changed. The core philosophy is that all the rules for buying and selling (the “trading system” or “trading strategy”) are 100% defined, and strictly followed.

Fast forward to August 2020 and the exam results fiasco. 

As we all know, students in the UK did not sit exams this year because schools were closed following the coronavirus lockdown.

In England, A-level students were given algorithm determined grades by the official exam regulator, Ofqual. 

“The algorithm was fed a few different strands of raw data. The first was the teacher’s predicted grade for each student based on their performance in class and the mock exams. But this was deemed insufficient on its own, so teachers were also asked to rank each student from highest to lowest in terms of their expected grade”.

And, with the benefit of hindsight, some may even say foresight, the programming went wrong.

I have to say that if the ‘algo’ had been put together by the same Ofqual people to do trades for a fund manager I suspect the programmer would have been fired because of the huge losses incurred, the FCA would have asked some serious questions and investors would be demanding compensation.

Now to Robo advice.

An advisor’s experience and judgment is replaced with an automated low-fee investment platform offering asset allocation and portfolio re- balancing. 

Ultimately, fully automated advice algorithms may come to play an important role in the financial advice landscape for many products, both as stand- alone models and as tools for advisors. But so far it is not clear whether these firms can move beyond simple investment solutions, capture non-millennial investors at scale, or replicate the trust that can come from a human advisor. The potential exists, but the jury is still out.

I worry greatly that although this heralds a new era, the danger in ‘algo-failure’ still places the blame on the adviser and not where it should be with the programmer.

So, apart from getting the ‘algo’ right, the FCA should licence said ‘algo’ as fit for the clearly defined and designed purpose and that any adviser who uses ‘algo’ should not carry the blame or fund the compensation for the failure of the ‘math’. 

That is down to those ‘nerds in flipflops, vests and shorts, ‘clever b******s’, as Ian Dury’s song went, to fund that with their own pool of money and PI.

FCA approved robo advice platforms operating on purpose based algorithms may make ‘robo’ viable by absolving the adviser that recommended their use, from responsibility of ‘math failure’ that was totally outside their control?

Just a thought?

Panacea Comment