How to read a stock story for the fact that changes the business
Earnings releases and the language around them forecast results, and prices often keep moving after the headline.
Most of what changes a stock's intrinsic value shows up first as language: a number in a release, a sentence of guidance, a contract, a loss, a regulatory decision. The price reacts to that language, and the reaction is often incomplete. The research on earnings announcements and on the words inside news stories is unusually consistent on both points. This is a way to read a story so you extract the fact that belongs in a forecast, and leave out the rest.
Earnings news is information, and the price keeps working
Ray Ball and Philip Brown, in the Journal of Accounting Research in 1968, compared the sign of unexpected earnings with the sign of abnormal stock returns. Prices drift in the direction of the eventual earnings news during the year, which means the market is learning before the announcement. Prices still move when the number is released, which means the announcement itself is news. William Beaver's companion paper the same year showed that both trading volume and the variance of returns are higher in the announcement week. The release is not a formality.
Victor Bernard and Jacob Thomas, in the Journal of Accounting Research in 1989, asked whether the later drift was just risk. Their evidence pointed to a delayed response. Stocks with positive earnings surprises keep outperforming, and stocks with negative surprises keep lagging, for weeks to months after the announcement. A reader who stops at the headline print, or at the first-hour return, has left part of the value implication on the table. The drift is also why the text matters after the number is public. The number is one input. The explanation of the number is another, and the market has a record of absorbing it slowly.
Michael Drake, Nicholas Guest, and Brady Twedt, in The Accounting Review in 2014, connect the press to that delay. Broader press coverage of the earnings announcement is associated with less mispricing of the earnings news. Coverage is not a substitute for reading. It is evidence that distribution affects how fast the cash-flow fact gets into the price. If you follow a name the national press covers lightly, the release is doing more of the work.
The words forecast the earnings
Paul Tetlock, Maytal Saar-Tsechansky, and Sofus Macskassy, in the Journal of Finance in 2008, asked whether a crude measure of language predicts fundamentals. It does. Three results from that paper are enough to change how you read:
- The fraction of negative words in a firm-specific news story forecasts lower earnings.
- Prices briefly underreact to the information in those negative words, so the language predicts returns as well as earnings.
- Both the earnings link and the return link are strongest for stories that focus on fundamentals, not for stories that only color the firm's prospects.
Angela Davis, Jeremy Piger, and Lisa Sedor, in Contemporary Accounting Research in 2012, looked at the earnings press release itself rather than at later articles. Net optimistic language is positively related to future return on assets, and the market responds to that language, after you account for the earnings surprise. The release is not a wrapper around the EPS number. The sentences around the number have incremental information about the business.
The measurement has a trap, and Tim Loughran and Bill McDonald documented it in the Journal of Finance in 2011. General-purpose negative dictionaries treat tax, cost, liability, capital, and board as pessimistic. In a 10-K those words are ordinary. A finance-specific negative list, applied to annual reports, lines up with lower returns, higher volatility, and a higher incidence of accounting enforcement. If you or a model scores a filing with a novelist’s dictionary, you will flag the balance sheet as a crisis. Read the sentence the word sits in, and prefer a word list built on filings.
A market mood is a different object
Tetlock's 2007 Journal of Finance paper is easy to mix up with the 2008 paper, and the mix-up leads people to fade every negative story. The 2007 study is about a daily Wall Street Journal column on the market, not about a firm. High pessimism in that column predicts downward pressure on market prices, then a reversal, and unusually high or low pessimism predicts high volume. That is sentiment at the index. It is not a forecast of one firm's next earnings.
Wesley Chan, in the Journal of Financial Economics in 2003, separates stocks that had a headline from stocks that did not. Stocks with news, especially bad news, tend to keep moving in the same direction. Stocks that moved without news tend to reverse. A large return with no story is a weak reason to revise intrinsic value. A large return with a story about fundamentals is a reason to read the story, because the drift evidence says the first move may not be the whole move.
Fang and Peress, Journal of Finance 2009, are the reminder that coverage and information are not the same variable. Stocks with no media coverage earned higher subsequent returns than heavily covered stocks. Absence of a headline is not evidence that nothing happened to the business. It is evidence that the intermediary did not write it up. Brian Bushee, John Core, Wayne Guay, and Sophia Hamm, in the Journal of Accounting Research in 2010, describe the business press as an information intermediary: it packages what the firm disclosed and spreads it past the investors who read the filing. When that intermediary is silent, the filing is still the source.
Try it for 7 days. Signup starts Pro. A card is required. Real-time REST, WebSocket, and MCP, with 1,000 REST requests a day. Paid Pro removes the cap. After that it's $15/month, or Free if you stop.
Count the relevant story
Richard Roll, in the 1988 Journal of Finance, showed that standard factors explain a small share of firm-level daily variance. One reading was that public news is a small part of price movement. Jacob Boudoukh, Ronen Feldman, Shimon Kogan, and Matthew Richardson, Review of Financial Studies 2019, classify news by whether it is actually about the firm and about a real event. Once irrelevant headlines are set aside, identified news days carry a much larger share of volatility. A raw count of articles with the ticker in them will overstate how often the business was revised, and it will understate how much a truly relevant story matters.
A practical screen, before any model runs, is short.
- Prefer the issuer's release and the filing over a rewrite. The rewrite can add reporting, and it can also add a frame the issuer did not state.
- Keep a sentence only if it changes earnings, investment, the life of those earnings, or the risk of the cash flows. Discard the rest, including price targets you did not ask for.
- Mark whether the fact is a level ("revenue was") or a change in the path ("we now expect"). Guidance is a revision to the forecast. A historical recap may not be.
- Note the horizon the text itself names. A quarter, a contract term, and "ongoing" are different persistence assumptions in a residual-income or discounted-cash-flow model.
- If the only content is a mood word and a percentage, you are in the 2007 Tetlock setting, not the 2008 one. Do not capitalize it as a permanent change in the firm's earning power.
Using a feed without turning it into the model
The papers above were written on newswires, the Wall Street Journal, Dow Jones, IBES forecasts, and SEC filings. The conclusion does not depend on which pipe you use. It depends on getting the firm-specific text, knowing when it was published, and not mixing a legal boilerplate item in with a guidance change.
MarketNewsAPI is one way to do that pull. Stock items are typed as news, press releases, and legal notices, and you can ask for one ticker. type=press_release is the issuer's release, which is the document Davis, Piger, and Sedor studied the language of, and the right first read under the screen above. Legal notices stay out of that query. Each story has a published_at time and, once it is written, a short summary built only from the text on hand, so a negative-word story about fundamentals can be separated from a passing mention before you spend a model call on it. The filters are in the API docs, and the same items are on the live desk.
The summary is not a residual-income forecast, and a stream of headlines is not a substitute for the release. The useful thing is narrower. You can put the story that the literature says actually forecasts earnings next to the name, on the day it is published, and ignore the rest.