Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Optimism among executives at large US companies hit record levels during fourth-quarter earnings season. Yet analyst estimates are deteriorating. What explains the contrast? Part of the blame may lie with AI — and the natural tendency of people to go with the flow.
Bank of America’s measure of S&P 500-wide corporate sentiment — based on analysis of thousands of earnings call transcripts — is at an all-time high.
But as of the end of last week, analysts were expecting index-wide earnings growth of 11.4 per cent for 2025, according to LSEG. That’s decent, but lower than the 14 per cent they were forecasting at the start of January. Predictions for first-quarter growth have come down particularly sharply.
One theory for the discrepancy is that US executives have adapted their communications to respond to investors, who are increasingly deploying algorithms to parse their words for hidden insights.
For a brief window, a few years ago, hedge funds could steal a march on rivals by analysing whether executives were using more positive or negative language during their calls with investors and analysts. The problem was that the so-called “natural language processing” was so effective that everybody quickly cottoned on.
BofA’s sentiment indicator has been trending steadily upward for two decades, with occasional short sharp drops during obvious emergencies like the 2008 financial crisis and the start of the coronavirus pandemic. That is just one bank’s measure rather than a widely-used benchmark, but several academic studies have found similar trends.
Sentiment inflation has forced fund managers to look for ever more advanced ways of measuring companies’ confidence. Some have started analysing audio to pick up hesitations and micro-tremors that wouldn’t show up in a textual transcript. However, companies are already adapting: specialist firm Speech Craft Analytics, for example, set out to provide audio analysis for investors, but soon found it was also drawing interest from investor relations specialists looking to train executives on how to sound good to the algos.
In the US, many large companies already pre-record their results presentations. It doesn’t take much extra effort to run a few versions through an audio analysis to pick the best. At least one company — Nasdaq-listed Cerence — has even delivered an earnings presentation using a totally artificial “voice clone” of its CEO and CFO. There are a number of established consultancies experimenting with similar tech.
Once everyone’s presentation is pitch-perfect, investors will inevitably search for a fresh way to pick out which companies are off-key. In an equity market being pulled by momentum-driven algorithms on one end and social media-obsessed retail traders on the other, a really radical idea would be to pay attention to companies’ actual financial performance too.
https://www.ft.com/content/1d9db48f-d6c7-44dc-9cd6-7d8c752f695c
