Friday 14 August 2020

The trouble with Algo's

Algo's have been all the rage in the financial markets now for the best part of three decades. As far back as the Long-Term Capital debacle of the mid-1990's, through to the flash crashes and algo driven 2009 crashes and thence on to "flash boys", it has been all about the algo's. 

But in recent years, they have topped out a little - admittedly because in part they are the only show in town. Still, Hedge Fund launches have been tailing off for a few years and this year there is no real change. Perhaps some optimism that a recession will see an uptick, but nothing as yet. The main driver of this has been that a generally rising market has meant the much cheaper buy and hold strategies have paid better than expensive hedge fund algo trading - which is prone ot making racy bets and trying to corner markets against the trend. 

So it is not surprising that now Government's are following the lead and using Algo's - in the UK Dominic Cummings is a big fan. if it works for high finance, then surely it is a good idea for Big Government. So when a challenging situation appears like the A-Level results all needing to be modelled out of the blue, here is an Algo solution which has found its problem. 

Yet it has not worked out to the Government's advantage, overall the results are likely very fair, but individually there are thousands and thousands of errors. Each error is a person with hotline to social media and the wider world. Each one is aggreived and a victim, all have parents behind them saying little Jenny or Jack worked all the hours god sends, is very clever and deserves at least a place at Oxford and not Salford.

So basically today ends up being an Algo catastrophe, much like in the trading world, where getting 99% of your bets right but shipping all your profits on the 1% higher risk postion that cratered ruined your year. Here most results are likley fair, but the numerous injustices willl be impossible to justify. 

Here begin the opportunities and pitfalls of real technocratic governance. 

12 comments:

Don Cox said...

It was a crazy thing to attempt. However you guess what a student's exam results would have been, there are going to be a huge number of angry students.

If the universities can't be bothered to organise some kind of entrance filter for themselves, such as getting applicants to physically go to the university, sit down and write a couple of essays, and then have a safely spaced interview, then they should draw lots.

From my experience of University students, the range of ability in a single class is so great that they might just as well have been chosen by lot.

Anyway, which universities other than Oxford and Cambridge really have more applicants than they can take ?

Don Cox

Anonymous said...

And let's not get started on the damage that algorithms and modelling have had on predicting covid....

Anonymous said...

Don Cox +1
Entirely agree. Exam results reveal very little about a person apart from having a good memory.
M.

andrew said...

It is not just variation in ability.
Maybe things have changed but the first 2 terms of maths were getting everyone to the same place. People with further maths cruised and so did scottish highers. From memory the london board ? people struggled.

dearieme said...

It's all said as if there were never any injustices with conventional A-level exams. Whatever you use can deliver at best rough justice. Which is just as well since the people taking A-level English must know that if we all got our just deserts "Who should 'scape whipping?"


(If I may say so, that comment is well worth an A*, don'cha think?)

estwdjhn said...

I mostly blame the teachers. They predicted grades in such an "all must have prizes" style that the government couldn't possibly have just run with it.

There is also the crying about good institutions not getting marked down as much as bad ones. This is obvious when you think about it - there is a hard cap at the good end of the grade predictions - no matter how good your pupil, they can't do better than an A/A*/9 (depending on the result system).

If 80% of your kids get the top marks every year, then that's 80% who you can't over-predict.

Meanwhile at some grotty dump where getting a C puts you well above average, it's perfectly possible for your teachers to predict pretty much everyone in the school up a grade on reality, which then get moderated back down again when the system compares the school's predictions to its previous results.

Elby the Beserk said...

Anonymous said...
And let's not get started on the damage that algorithms and modelling have had on predicting covid....

8:42 pm
=========================================================================

Not to mention climate models...

E-K said...

Yet another ramification that hasn't been graphed out properly.

Add teenage suicides/drug addiction to death by lockdown, not CV19.

One clearly highly intelligent, articulate, composed and mature young lady predicted A A A was on TV having lost ALL her university offers yesterday having been awarded much lower grades.

Don Cox said...

She may get a place through the clearing house.

If she is really "highly intelligent, articulate, composed and mature young lady" then she should be able to find a job for a year while she keeps up her studies and retakes the A levels, especially if they run normally next year.

A year between sixth form (or Further Education college) and university is a good thing, especially at a time when she will not get proper face to face teaching. Nobody with any sense will go to a university this year.

Don Cox

E-K said...

WTF ?

Anonymous said...

@E-K

As parliament will be opening in a few weeks, perhaps you can ask your MP to pose that exact question to Williamson.

CityUnslicker said...

The Government do not appear to be getting a handle on this as yet at all. Shame, surely they can do something competently at some point, the law of averages must catch up with them?