Counting Gloomy Words: Tetlock Turns a Newspaper Column Into a Trading Signal

"Investor sentiment" is one of the most-used and least-defined phrases in finance. Markets are described as fearful, greedy, complacent, panicked. Everyone knows what is meant, and nobody can measure it.

The measurement problem is real. You cannot survey the market's mood in a way that is timely, unbiased and daily. The proxies people used, closed-end fund discounts, IPO volume, survey data, were slow, indirect and contested.

Paul Tetlock's contribution, which won the Journal of Finance's prize for the best paper that year, was to notice that there is a daily, timestamped record of financial mood sitting in plain sight: the newspaper.

The problem: sentiment had no thermometer

The title of the paper is doing real work. "Giving content to investor sentiment" means: taking a vague, hand-waving concept and attaching an actual measurement to it, so that claims about it can be tested rather than merely asserted.

The obstacle was that markets are mostly numbers, and sentiment is mostly words. Financial economics had spent a century building tools for numbers. It had essentially no tools for words. Text was regarded as unstructured, subjective, and unusable, which meant that the enormous volume of financial language produced every day, news, analyst reports, company filings, was simply invisible to quantitative research.

The key idea via analogy: reading the room by counting the frowns

Suppose you want to measure the mood of a room but you cannot ask anyone. You could count faces: how many are frowning, how many are smiling. It is crude. It is also objective, repeatable, and it captures something real.

Tetlock did this to the Wall Street Journal.

He took the "Abreast of the Market" column, a daily piece commenting on the state of the market, and processed every day's text through a word-counting dictionary. The dictionary, borrowed from psychology and content analysis (the Harvard psychosocial dictionary), classifies words into categories, and the category that mattered was negative or pessimistic words.

The procedure is almost aggressively simple. Take today's column. Count how many words fall into the pessimism category. That count, suitably scaled, is your daily reading of media pessimism.

No sentiment model. No understanding of the sentences. No parsing of meaning. Just: how many gloomy words did the paper use today.

Then he asked what that number predicts. The findings were subtle and, importantly, not what a naive story would suggest.

High media pessimism predicts downward pressure on prices, followed by a reversion. The word "reversion" is the crucial part. If the newspaper's gloom were conveying genuine information about fundamentals, prices would fall and stay down, because the market would have learned something true. That is not what happened. Prices fell, and then they came back.

That pattern is the signature of sentiment, not information. Something in the pessimistic coverage was pushing prices temporarily away from fundamental value, and then fundamentals reasserted themselves. Media gloom was moving prices without carrying news.

Unusually high or unusually low pessimism predicts high trading volume. Note that this effect is symmetric: extremes in either direction stir up trading. That is consistent with media coverage energising investors who trade on noise rather than information, which is exactly what behavioural models of markets predict.

Put together, the picture is that the media is not simply a neutral pipe carrying news to investors. It is a participant. Its tone influences behaviour, that behaviour moves prices, and the price move partially unwinds. The market is, at least temporarily, reacting to how things are said rather than only to what is said.

Why it mattered

• It opened text as a data source for finance. This is the paper that made "we can quantify words" respectable. Every sentiment signal, every earnings-call-tone model, every news analytics product, every language model chewing through filings today sits downstream of the moment Tetlock counted negative words in a newspaper column and found they mattered.
• It gave sentiment a number. After this, "investor sentiment" stopped being an appeal to intuition and became a variable with a time series, a mean and a standard deviation. That is what allows a concept to be argued about scientifically.
• The reversal is the point. The finding is not "bad news makes prices fall," which would be trivial. It is that negative tone moves prices temporarily, in a way that reverses, which is direct evidence that something other than information is at work. That is a genuine contribution to the debate about market efficiency, made with a clean and clever test.
• It showed the media as an actor, not a mirror. Journalism does not just report on markets. It affects them. That has real implications for how we think about news, about attention, and about the fragility of prices when coverage turns.
• It made a very simple method respectable. Word counting is a laughably crude way to read text. It works well enough to find real effects. That lesson, that a simple, transparent, reproducible method beats a sophisticated one you cannot audit, has served textual finance well.

The honest limitations

• The dictionary was borrowed from the wrong field. This is the big one, and it is the criticism that produced the next great paper in the area. The Harvard dictionary was built for general psychological content, not finance. It classifies words like "liability," "cost," "capital," "tax" and "vice" as negative, and in a financial document those are just neutral vocabulary. Loughran and McDonald showed a few years later that a large majority of the words the Harvard dictionary flags as negative in financial text are not negative in a financial sense at all, and they built a finance-specific replacement. Tetlock's finding survived, but his tool was blunt.
• Word counting cannot read. It has no idea about negation, sarcasm, context or comparison. "The company denied rumours of a loss" and "the company reported a loss" contain the same gloomy words.
• One column, one paper, one era. The data source is a single daily column in a single newspaper over a specific historical window. That is a narrow slice of the information environment, and it was a much bigger slice in 1990 than it is now.
• Causation is genuinely hard. Journalists write gloomy columns because the market has been gloomy. Disentangling "the column moved prices" from "the column and the price move share a common cause" is very difficult, and while Tetlock addresses it carefully, the identification is not airtight.
• It is not obviously tradeable. The effect is a short-lived push and reversal at the market level. Whether it survives transaction costs, and whether it survived publication, are different questions from whether it exists.

The one-line takeaway

Tetlock counted the pessimistic words in a single daily newspaper column and showed that media gloom pushes prices down temporarily and then they bounce back, which is evidence that tone moves markets even when it carries no news, and which single-handedly opened up text as a legitimate data source for quantitative finance.