indianotes

Every analyst today is using some form of AI to create equity reports. This is evident from the slick arguments, similarity in framing hypotheses, the high-funda language used, and the dramatic pauses.

Beyond the fact that everyone is starting to sound the same – the thought in my head is – Is everybody starting to say the same things?

The concept of “consensus risk” is widely discussed in investment literature. A certain trend or narrative gains traction, and this attracts more investors. Eventually, the price movement itself strengthens the narrative that caused the prices to rise (or fall). In effect, a self-reinforcing feedback loop.

There is nothing wrong with this. In fact, without such periods of consensus, we would not have rallies or dips, nor would the market be as efficient as it is. However, the consensus can (and often does) go too far, and results in price distortions and higher risks.

In fact, extreme cases of over-valuation or under-valuation are the result of such consensus feeding on itself.

Markets have always behaved like this. The question is whether AI may actually be amplifying this behaviour?

If the prevailing view is that a stock’s “premium valuation is justified by superior growth and capital allocation”, then this will already be embedded in available opinion and equity reports. And AI will synthesize from this information ecosystem that already contains consensus.

If a widely held view suggests that a company has an insurmountable moat, then AI will find abundant evidence to support this view.  So the consensus view in fact, becomes the input to the algorithm, which in turn amplifies the view.

And then if a hundred analysts use the same framing and get similar answers, AI will see these correlated opinions as distinct views, and count these as corroborative evidence. And as AI gets better, it will get more convincing.

Am not suggesting we don’t use AI for equity research. Not at all. AI is extremely useful in a variety of tasks; such as forensic analysis, building spreadsheets, gathering industry information, scouring annual reports and transcripts, and so on.

All these fundamentally make the analyst more efficient, and can ensure that all available information is captured and incorporated in the analysis. In fact, AI also helps reduce information asymmetry between large and small investors, which is a good thing.

But trying to get AI to make an investment call is not the same thing, and might only push you toward a consensus trade that is already stretched.

In fact, you may be better off asking AI to try and negate the consensus, or challenge the assumptions and elaborate on the risks. This will make you pause and think.

An interesting way to look at all of this might be as follows.

Investors and analysts traditionally spent most of their time in collecting and collating data, organising information, assessing accounting norms or corporate governance, building spreadsheets, calculating DCFs and so on. AI can now do these tasks much more efficiently, and is getting better day by day.

But the real value is not in information or fancy spreadsheets, but in deciding how to interpret it, and what to do about it. It’s not clear to me why this will change with the use of AI. After all, successful investing has always been about looking beyond the obvious.

What do you think?