In previous blogs I have spoken about the fact that 98% of stock pickers lose to markets long-term, mainly due to fees and human error, such as over-confidence after short-term gains, greed, fear and other behavioral issues.
Some technology advocates seem to think they have found the solution; AI powered ETFs and funds.
Increasingly, many companies are using computers to mine huge amounts of data sets, including commentary, social media talk, debit and credit card data and numerous statistics. Using this information, the machine picks certain stocks it thinks it can beat the market.
Blackrock, the largest asset manager with $5 trillion in assets, announced it would be moving some of its actively managed funds to its quants team using AI, lowering fees in the process.
Numerous studies have also looked at the ability of AI, including this one –https://www.gweiss.com/Assets/pdf/Insights/201707-Bike-Ride.pdf. In this work, Visser notes how AI is better at pattern recognition than the human mind.
He also argues that as efficient market hypothesis works best during `normal times`, but works least well during panics like in 2008, the machines may perform particularly well during periods like 1929 and 2008, as certain stocks (like banking stocks in 2008) become particularly undervalued.
Whilst lowering fees should be good for investors, we shouldn’t get too excited just yet. Any discerning readers probably understand something already – some of these machines are using data, such as social media talk, which could be detrimental to the ability to pick stocks.
The whole point of using AI is supposed to be to get rid of the two major problems of active funds, human error and high costs. However, if they are using data that is coming from human interactions, they are compromised.
As per the Charles Mackay book in the reading list, humans tend to follow crowds, as the recent bitcoin craze goes to show.
In addition to that, AI funds are relatively new. There isn’t enough data yet to determine whether the technology can regularly beat the market
There is also the issue that if everybody is using the same data, the same alpha, that will neutralize the results.
Ultimately, investors shouldn’t depend on AI just yet, for a large percentage of their portfolio. It will probably outperform most human investment active managers, but getting `average returns` and tracking the market has produced excellent returns for over 200+ years.
Even if the technology works long-term, there may be significant tax implications of a buy and sell approach, whilst the differences in data sets each AI machine is using makes it hard to determine which AI-powered machine to use, if you aren’t knowledgeable about big data. Choosing the best AI machine leads us back to human error, so we have come back full circle.
We need more time to see how AI works in practice, during crashes and bull markets, before drawing any conclusions.
