top of page
  • Julean Albidone

AI is not the Midas Touch, But Why You Should Probably Still Use It

Machine learning seems to be quite polarizing in the asset management space. People either see it as a godsend or a complete waste of time. I can very confidently refute both of those claims.

Addressing the overly optimistic side: If we look at simple base rates in the field of asset management, we should know by now that there is no free lunches anymore, there isn't one single factor or type of algorithm that is going to conquer all. We are talking about improvements on the order of basis points, not tens of percentage points.

On the overly pessimistic side: I feel most of this comes from either one of two areas. Either people have lost momentum or the drive to innovate, they are simply stuck in their ways for better or worse (perhaps lack of access to talent or just sheer laziness to innovate) and want to detract from a genuine innovation and new source of alpha. The second area I see is a genuine misunderstanding of how the algorithms work and how to properly deploy them in a financial forecasting setting.

Let me tell you that the reality is undoubtedly somewhere in the middle of these two extremes.

Machine learning is certainly not the Midas touch. It will not take a terrible investment strategy of Sharpe zero, and magically turn it into a strategy if Sharpe 2. It does not work like that. In fact I would go as far as to say if you do that, then it will be purely overfitting, and that the same person who is likely to do this is likely the same person who does not know how to properly apply machine learning toolsets.

I tend to view machine learning through the Pareto principle, i.e. the 80/20 rule. A core investment thesis with conventional quantitative analytical techniques should take you 80% of the way there, it is machine learning that can extract that final 20% to take you over the edge.

Framed another way, machine learning can be viewed as a multiplier. Apply machine learning to a strategy with zero alpha, and you will still end up with zero alpha. Apply it to a strategy with great alpha, and - applied properly - you will likely end up with more alpha.

So should you use it or not? I'm of the camp that you should always seek to innovate and push yourself to find more alpha. The asset management space is already competitive as it is. You should not be leaving any alpha on the table, you should look to exploit every advantage that you have.

Does that mean you should slap machine learning and AI onto every investment problem that you have? The answer is obviously no, but on balance it adds value. I will leave it for another blog post to discuss when and when you should not use machine learning.


bottom of page