What to say in a machine learning interview on the buy-side

If you want to work for hedge fund Brevan Howard's machine learning team, you can expect to encounter Sebastien Guglietta, Brevan's co-head, of 'computational intelligence systematic strategies.' Guglietta joined Brevan from Comac Capital, where he spent 18 years as a prop trading, portfolio manager and senior strategist. These days, he's all about applying 'computational intelligence' (machine learning) to systematic trading and running a team in London that appears to comprise himself and at least another three people.

Guglietta clearly doesn't hire much, but speaking at last week's AI in data science and trading conference in London, he gave a few pointers on what to say if you ever find yourself in front of him. They likely apply equally to interviews with rival funds like Two Sigma.

1. Don't praise simplicity 

A model is about reducing a system's apparent complexity and fighting against its inherent uncertainty, said Guglietta, but a shallow model trying to simulate a macro economic system with a "deep causal structure" isn't going to work. There are huge "interdependencies" in an economy, said Guglietta: each one has 100 macro features alone. "Simplicity is highly overrated.”

2. Don't talk about structural VaR

Guglietta isn't keen on VaR - he says this is one of the model dependencies that produces only "shallow algorithmic subtleties."

3. Do quote the greats

Guglietta loves Alan Turing, the cryptographer who famously broke the Nazi's engima code. He also loves Ray Solomonoff, the inventor of algorithmic probablity. If you can quote either many, you will ingratiate yourself. "Learning is compressing complexity by accepting a given amount of uncertainty," is from Solomonoff.   

Guglietta also likes Ockham's razor ('Models should not be multiplied beyond necesssity') and the principle of multiple explanations which says that if more than one model is consistent with observations, keep them all...

You should also be familiar with Kolmogorov complexity and Bayes probablity theorum. Guglietta is also a fan of the concept of the 'centaur' - a notion introduced by chess chamption Garry Kasparov to describe the combination of human and machine intelligence.

4. Don't speak in favour of data for the sake of data 

Not all data is made the same. "Where we are not sure about the data we don’t use the data," said Guglietta. “You always trade order. You don’t trade noise. If you are trading noise, stop.”

5. Don't presume that, as a data scientist, you can do it all

Guglietta also had a few things to say about the argument made by Marcos Lopez de Prado that data scientists don't necessarily need prior finance knowledge and that they can devise trading models simply based on mathematical patterns if they participate in tournaments. Generally, said Guglietta, finance knowledge is best.

"If you expect to put 50 data scientists in a room and to find a solution then good luck," said Guglietta. "You need traders, economists... you need [finance] expertise. Data scientists on their own will not find a solution." 

Have a confidential story, tip, or comment you’d like to share? Contact: sbutcher@efinancialcareers.com in the first instance. Whatsapp/Signal/Telegram also available.

Bear with us if you leave a comment at the bottom of this article: all our comments are moderated by human beings. Sometimes these humans might be asleep, or away from their desks, so it may take a while for your comment to appear. Eventually it will – unless it’s offensive or libelous (in which case it won’t.)

Photo by Mika Baumeister on Unsplash