The automation of investment is an unstoppable trend. Machines are now the dominant traders in London and New York. But people still run the show, for the moment. That means automated analysis of human actions is helpful in spanning the divide. A new study claims that emotions betrayed in human speech — in this case by central bankers — not only move markets, but can be analysed and acted on.
Algorithms already parse the words of central bankers such as Jay Powell and Andrew Bailey in search of buy and sell cues. The mission of researchers at the universities of Birmingham, Reading and California, Berkeley, was to systematise interpretation of the way those words are spoken.
The academics concluded that when a banker’s remarks in a press conference are delivered in a positive way, they can raise stock market returns by 200 basis points. The more upbeat the tone of voice, the lower inflation expectations tend to be.
Increasing computing power can make quant investing a bewildering subject. But the methodology of the researchers was clear cut. They focused on audio recordings of press conferences held by successive Federal Reserve chairs Ben Bernanke, Janet Yellen and Jay Powell. They built a neural network, or multi-node computing system, to compare segments of each audio recording against a database. This database categorises how emotions are reflected in human speech using recordings of actors delivering the same texts in different ways.
Stock in the giant SPDR S&P 500 exchange traded fund was one security whose reaction was gauged by the researchers: Birmingham’s Oleksandr Talavera, Tho Pham of Reading and Berkeley’s Yuriy Gorodnichenko. They found that a one-unit increase in tonal positivity within their complex scoring system lifted returns by 1 per cent after five days.
Another finding — and I’m paraphrasing here — is that Powell is the gloomiest Fed chair for years. “Ben Bernanke had more positive emotions in his voice than Janet Yellen, who in turn had generally more positive emotions in her voice than Jerome Powell,” the researchers wrote. Perhaps Powell was a bit grumpy because Donald Trump kept yanking his chain.
Of course, there are problems with any quant approach to investment, particularly when dealing with commodities as slippery as human emotion. Astute Lex readers will already be clamouring to point out that correlation does not prove causation. Another issue, frankly highlighted by Talavera, was a small sample size — just 36 press conferences.
A more theoretical stumbling block is that the stock market exhibited worthwhile correlation with voice tone, but bond markets did not. Bond investors follow Fed pronouncements with focused attention, because any hint of inflationary pressure hits prices. Equity investors tend to have a looser approach, with stocks sometimes recommended as hedges against rising prices for goods and services.
What we can say for sure is that the use of data to interpret human behaviour is mushrooming. A year or so ago, Lex met the boss of a huge Chinese financial group who cheerfully told us that first interviews with prospective salespeople were conducted entirely by computer. These analysed the facial expressions of candidates as well as their speech in an attempt to hire honest extroverts. Some quants are now using facial mapping in an effort to deduce the mood of central bankers.
If you are a chief executive or finance director, you can expect your own public appearances to undergo the same kind of scrutiny sooner or later. Talavera and his collaborators have not tried hooking their emotion-detector up to a live audio feed of a company presentation while plugging another cable into an automated stock trading system. It is theoretically possible to do so.
If I was a CEO in such circumstances, I would feel strongly tempted to limit sell-offs by delivering bad news in the happiest possible way. You know that innocent trick dog owners can play on their pets, dishing out criticism in delighted tones? The dogs wag their tails. Perhaps it will eventually work for markets too.
Enjoy the rest of your week.
Head of Lex