The Limitations of AI in Stock Picking

You all know that Artificial Intelligence is a powerhouse at many tasks. Yet, as you're going to find out, it's not quite adept in the stock-picking arena.

Wall Street Journal reports that AI-driven exchange-traded funds have missed out on capturing returns from the tech-fueled market rally and are lagging the S&P 500.

The problem? These models rely heavily on historical data but struggle to grasp a dynamic fast-changing landscape. Take Meta, for example. AI models believed that Meta was overvalued and hence did not recommend purchasing its stock. But with Meta’s shares skyrocketing over 140% since December, it's clear that AI missed the bullish wave to come.

This underlines why we must scrutinize the drivers behind AI models more intently. Many of these models appear vulnerable to the 'normalcy bias’- the presumption that the future will mirror the past.

Humans often let emotions sway their investment decisions, succumbing to the latest hype or panic. AI models, devoid of emotion, don't overreact - or so we think. Yet, markets are sentiment-fueled; volatility can lead to losses or gains. AI tends to inherit and propagate the beliefs embedded in the data it trains on, coupled with full decision-making power, the risks get even higher.

As Andrew Ng suggests, AI should be seen as a tool, not the boss. It's there to inform our decisions, not to make them entirely. As well, in this perspective, it will probably not create an apocalyptic event as some might argue.

We at Asia SGE have seen the same issues plaguing our clients, primarily startups, when they become overly data centric. We've seen startups steadfastly follow data, oblivious to user feedback. Data should act as a beacon, not the sole compass for decisions.