Zurich based Marco-Andrea Buchmann uses AI to generate high performing hedge fund and asset management investor returns. Durtti wants to understand exactly how and why the implementation of AI has been key to this success to date.
You worked at CERN as a Scientific Collaborator for your PhD in particle physics. How easy did you find adapting your ML and AI techniques from CERN to the world of hedge funds and asset management that you now operate in, Marco-Andrea?
The two worlds have quite a lot in common, which made it easier for me to adapt.
While the objective is clearly not the same, the two share a lot in terms of techniques and challenges: both have huge datasets and use code along with statistics to find tiny patterns in a vast sea of noise.
After an initial period of familiarization with the new environment, getting to know the terminology, the idiosyncrasies, and so on, I was quickly able to formulate the problems in a way that the methods I was already familiar with could be transferred.
What does your current role as Head of Artificial Intelligence at asset management firm BrightKey Strategies Ltd involve?
I develop algorithmic investment strategies using artificial intelligence.
A single human trader can maybe monitor 10 – 20 stocks, and keep up to date with news for those.
An algorithm, on the other hand, can monitor thousands of stocks without getting tired, read the news for each one, stay up to date with changes in the market, and it can do so around the clock.
Using AI we can take this even further, but we still keep the human element in it: with algorithms proposing investments, the human makes the final decision, manages the risk, and contributes some of his own ideas as well.
Our combination of algorithmic trading, including AI, and human decision making has worked out extremely well so far.
What level of investor returns have you been able to achieve using your methods?
Our returns so far have been rather exceptional – we made over 50%, after fees, for our investors in 2016 and almost 40% in 2015, also after fees.
To compare: the average hedge fund made less than 5% in 2016 and about 2% in 2015, so we didn’t just beat the market, but we also performed very well compared to our competitors.
What is ‘overfitting’ and how does it relate to human behaviour and the commercial success of a business?
Overfitting is the behavioral bias to see patterns where there are none.
This is, to an extent, an evolutionary advantage: our ancestors were safer off assuming that the rustling they heard behind them might be a predator rather than just ignoring it – and potentially be eaten.
Nowadays, however, jumping to conclusions too quickly can be quite expensive.
In this business, for instance, investing according to a pattern that isn’t real can be costly. We would still be accumulating trading costs, and miss out on real opportunities.
That is why we take a range of measures, which we follow very strictly, to reduce the cost of overfitting and deliver better performance for our clients.
We’re all human. How do you try and turn a bad day into a good one if things aren’t going as planned?
I sometimes find it necessary to get some distance to reflect, and to refresh my mind.
In my spare time I love to go for a run or ride my bicycle; these kinds of activities help me clear my mind, gain a new perspective, and then start the next day fully motivated.
Aside from FinTech, which other sectors do you believe stand to gain the most commercial benefit from AI implementation in the next 2-3 years and why?
In general, jobs involving repetitive tasks will see at least some efficiency improvements with AI already at this time scale.
These efficiency improvements aren’t specific to any industry, but some good examples include transportation and manufacturing.
Self-driving cars are in testing, and it is not a stretch to imagine that very soon logistics companies will want to benefit from it.
In this example, AI may serve as assistants at first, and then in a second step, caravans of trucks may only have a human driver in the leading truck, and then the human element could disappear completely… though not in two to three years, of course.
Thinking longer term though I strongly believe that is the direction we’re headed, and the driver’s license will become about as necessary as an elevator operator license is today.
The other example is textile manufacturing, where tasks are very repetitive and automation is already happening today with first factories using “sewbots” – this is a fascinating trend to follow, as it may mean that production lines that went to Asia may come back to the west as labor cost is less of a factor, but having production lines closer to consumers saves on shipping. For example, Adidas now has an automated production line in Germany.
What’s the best piece of business advice you have ever been given?
I think the advice that has resonated with me the most was “innovate or die”.
I love being at the cutting edge, at CERN before and now at BrightKey Strategies, and innovation is key to success.
Finally, Marco-Andrea, what one thing would you tell a classroom of children today to help them prepare for tomorrow?
Learn to code.
There are so many fun ways to start coding, and I am convinced that it will be a crucial skill in the future.
It can help to think differently, approach problems in a different way, and to better understand complex challenges and break them down into manageable pieces.
I myself also started writing code when I was in elementary school, and so have many of my co-workers – it can be lots of fun!
More at www.brightkeystrategies.com
Marco-Andrea is a member of The Artificial Intelligence Group on LinkedIn.