Keen watchers of the ever-developing exchange-traded product space may have noticed an intriguing development last week, as the first purely “artificial intelligence”-based stock-picking ETF launched. Powered by IBM’s “Watson” platform, the fund sponsors claim to use a proprietary quantitative model to select stocks that will outperform, based on machine learning applied to vast data sets.
One cannot help wondering if they have missed a trick: as far as I can tell, their algorithm does not explicitly allow for the possibility that – rather than trying to pick stocks – a truly intelligent option might be to invest their entire portfolio in a low cost index fund, or otherwise replicate the market portfolio. Certainly, buying such funds is nowadays as easy as buying stocks, while the data would suggest that this is more than a viable option.
Perhaps, one day, another sponsor will create a fund including this option. Perhaps, one day, our machines will be so advanced that they can draw conclusions from the entire range of academic and practitioner studies that examine the performance of stock-picking compared to low-cost passive investing. Perhaps, if it helps, they can check their conclusions 10,000 times a second.
Such a fund may never exist, but if it one day does, I hope they call it “Holmes”.