FundFire: Beware of False Patterns in Big Data InvestingFundFire:BewareofFalsePatternsinBigDataInvesting

Read the transcript

• –

  RICHARD HENDERSON, ASSOCIATE EDITOR, FUNDFIRE: Hi, I’m Richard Henderson, the associate editor of Fundfire. I’m sitting here with Andrew Dyson, the CEO of QMA.

  Andrew, I wanted to ask you, in this time of Big Data, of AI, how quant firms such as yourselves and some of your competitors are approaching Big Data.

  ANDREW DYSON, CEO, QMA: Long-term quants, those people are generally looking for factors or signals to drive their performance, which are robust, have an underlying theory behind them and they can rely on over long periods of time.

  Now, for the long-term players, Big Data becomes another area you can look at for signals. So, for somebody like ourselves, we will incorporate where we think it adds value, a Big Data signal or more than one, alongside the more traditional signals that we’ve used for a long time. So it becomes another arrow in your quiver, but it’s not driving the decision process, per se.

  Very important, as I say, to recognize in all this noise that these things are not identical, these three different uses. Fundamental managers using Big Data as a technique; for long-term quants such as ourselves, who also use it as a technique, but inside a quantitive process; and then the shorter-term players who are much more allowing it to drive the portfolios, but with the attendant risk of data-mining that that brings.

  The third group, which I would characterize as the more short-term group. Now, that’s where you are seeing this machine-learning AI driving the actual portfolio decision process. And that I think is a more challenging way to go about building portfolios for clients because inherently in our industry there has always been a data-mining challenge. So I can look back and find a hundred patterns that might succeed, looking at the historic data. But the reality is only one of those hundred will work going forwards, 99 will fail.

  And so I think the challenge when you implement Big Data in that more shorter-term way to actually drive the selection process is how can I avoid data-mining, if you like, false pattern recognition and how can I be confident that this will be robust in the forward-looking process?

  Now there are some fantastic players out there who are very skillful at that, but I think it’s a tough game to win at and, as more and more people chase it, the half-life shrinks those strategies and the costs of pursuing it grow.

  RICHARD HENDERSON: Thank you very much for your time.

  ANDREW DYSON: Great, thank you