Blog post

More on Quants and the Risks with AI

By Andrew White | August 29, 2017 | 0 Comments

Quant (Quantitative analyst)Artificial Intelligence


I never even knew what a “quant quake” was, until Sunday last.  I was reading my US print edition of the Financial Times and there was an article titled, “A Decade on, the quant quake still has much to teach us“.  It turns out that I did know of the event to which the article was associated; I just did not know that it was called by the specialists of the day a quant quake.

Apparently two years ago the Goldman Sachs then finance chief complained that his bank had been subject to events or movements in the market that were 25-standard deviations from the mean, and several days in a row.  Clearly something of that magnitude is in no way normal.  The fact that this repeated for several days just doesn’t make sense.  It seems the result was a loss of a third of Goldman’s quantitative funds value.

As you read the article you realize what took place.  It seems the quants in the article developed sets of AI code that monitor the market, to predict ahead of others an action that would lead to a financial advantage.  It seems that a number of quants (more precisely the number of AI algorithms) were following each other along.  The article, by John Authors, mentions words such as cascading and crowding.  The AI’s behaved like people – we all thought we had a good bit of news and we all followed along.  It seems this worked so effectively that a huge black swan event took place – and took place several times.

There clearly is risk here in that quants may react in a certain way to fewer human signals and increasing quant signals. What happens if the perturbation of the market start to decline?  How do we create an environment where we can prevent the quants from conning each other?  Do we need super-quants or quant-supervisor algorithms to monitor the behavior of the quants?  Maybe we need ‘quantarents’ – senior parents of junior quants.

I had written about quants and AI a few times now.  Back in April I blogged about how we might indeed learn ways to compete with quants, rather than against them.  Just the other day I noted that AI could run amok by doing too much.  I was calling out a very similar phenomena as Mr. Authors (of the FT) in that AI engines could end up acting as a heard – following each other – as the number of signals they monitor decline in numbers as humans leave a market.  All told the whole AI space is a fascinating battle ground.  And we get to come up with cute, interesting names for hugely black swan-like events such as “quant quake”.

Such a name just happens to mention a word that I happen to love since it refers to a favorite PC game  of mine, “Quake Champions“, by id Software.  I have to admit my geek-credentials that I am an old Doom player.  Looking forward to unwinding with my Quake Champions soon.

Comments are closed