Recency Bias in Investing: Why Ten Headlines Are One Fact

Should you read financial news? Yes, but not as a feed and not every day. Noise versus signal is not a volume problem, it is a selection problem. Say a hundred headlines about one company cross your screen in a month. Zero to two of them change the business. The rest are pieces about a price that moved and about people with a view on the move. I run every item through a sieve of three questions: what role it plays, when it is from, and which assumption in my thesis it challenges. If none, it is noise, even if the stock fell 8% that day. Recency bias in investing is what makes that hard: the headline is fresh, so it feels like it matters.

Two percent and nine hundred urgent words#

A stock moved two percent, so a desk needs a piece on why. Someone writes nine hundred urgent words with an analyst, "investor concerns" and a chart. The honest answer is: because it moved two percent. A two-percent move on an ordinary day needs no reason, and it gets nine hundred words anyway because the page has to be filled, and an empty page collects no clicks.

The point is that the piece runs from the price move to an explanation, not from an event to a consequence. Since the reason is written after the fact, a different one will be found tomorrow, and each will sound plausible, because it was chosen to fit a move already visible on the chart.

Reacting to such a piece is not analysis. It is a response to somebody else's deadline.

Where those moves come from in weeks when nothing changed at the company is a separate piece on why stock prices move between earnings. Here one thing will do: if the company filed nothing on the day of the move, the reason cannot be checked in any document, and a reason that cannot be checked is not a data point. It is a guess with a date on it.

From the 8-K to the panic in four steps#

A real event can reach you distorted too, more than the invented ones. Say a company files an 8-K: a contract with a key supplier slipped by one quarter. An analyst reads it and writes a note that third-quarter margins might see some pressure while the fourth quarter looks solid. A journalist reads the note and files the headline "Analyst warns of margin collapse". You read the headline, not the note, and sell before checking anything.

Every link added something: the analyst a forecast, the desk an adjective, you an emotion. The document at the start of the chain said one quarter of delay, and it still says so. In the United States a company generally has four business days to file the 8-K, as the SEC's investor site explains, and anyone can open it free of charge on EDGAR. In the European Union the same job is done by the issuer's regulatory announcement: inside information goes out as soon as possible because Article 17 of the Market Abuse Regulation says so. Which means that in both places a version of the event exists that predates the headline and that someone is legally accountable for.

Outlets write for the click, filings are written for the rule. That is not a charge against journalists, it is a description of incentives: a newsroom gets nothing when you read the filing and something when you click the headline. How to read the company's documents instead of pieces about the documents is the subject of the guide to primary sources in stock research.

Ten headlines, one fact, and recency bias#

The brain counts headlines, not events. If ten outlets cover the same supplier delay, you feel ten pieces of evidence, though the fact is one, repeated ten times. That is the first distortion. Recency bias is the second: a dramatic headline from today weighs more than a slow structural shift that has been sitting in the filings for two years. Negativity bias is the third: fear holds attention longer than calm, so the bad headline gets more space and more repeats.

The three distortions add up to one. Newsrooms tune headlines for fresh, bad and repeated because they measure clicks, and clicks rise in exactly those three cases. Your sieve has to be tuned the other way: it lets through whatever touches the thesis, however old, calm and singular.

Recency bias has a cousin, the availability heuristic: what comes to mind easily feels more frequent and more important than what has to be found in a filing. A headline comes to you on its own. The footnote in the annual report on customer concentration never will, though it says more about next year's revenue than ten headlines put together.

Hence the first rule: before reading a second piece about the same event, count the events. Usually there is one.

Every headline gets a role and a date#

Since you cannot read everything, every item needs a label before it enters your head. There are four roles.

Without the label an alert turns into a conclusion and an opinion walks in as evidence. "Analyst warns of margin collapse" is an opinion built on an alert, and the whole story above contains exactly one data point: a quarter of delay. The rest is layers somebody added along the way.

You can watch the split happen in a single day. Say a company files in the morning that it lost a customer worth 15% of revenue, which is an alert and one data point. At noon a broker cuts its valuation, which is an opinion where you look for the revenue assumption rather than the number. In the afternoon a news site reports that "investors are fleeing", which explains the price, not the event.

By evening you have four pieces and still the one fact from the morning.

There is nothing wrong with opinions, provided you know the author's angle. A newsletter writer has a favourite argument, an analyst is anchored to a forecast, a social platform rewards whoever sounds most certain fastest. None of these sources lies. Each has a different goal from yours.

The date says more than the headline. A three-year-old article can explain perfectly where a business model came from and say nothing about today's demand. A fresh headline is current and shallow. A competitor's comment from last quarter outweighs an essay on the whole industry from last year. The question is never whether something is current. Current enough for what?

Say you are testing an assumption about this quarter's margin. Then what counts is this quarter's filing and the transcript of the latest earnings call, while last year's article on "structural pricing pressure" is context, not evidence. If you are testing an assumption about a technological lead, a year-old essay may be the best source you have. The same item is signal for one assumption and noise for another.

Which assumption does it challenge#

This one question sieves out the most, which is why it only works when the thesis is written down. An investment thesis is a handful of assumptions, each with a condition you can measure: segment X grows faster than the whole, gross margin holds above a threshold, management does not dilute shareholders. Three or four sentences. An item that touches none of them is noise by definition, however loud the headline.

Say the stock drops 8% today on a report that a large fund trimmed its position.

Which assumption does that challenge?

The fund did not change the margin, the segment or the management. It changed its own portfolio, for reasons you do not know and that may have nothing to do with the company. Which means the 8% is an event in your portfolio, not in the company, and the only thing to do is check whether the company itself published anything that week. If not, the day ends without a decision. What to do when the drop is bigger and has a reason is a separate piece on a stock that fell 20%.

I think this is the hardest rule in the piece. Not because it is complicated, but because it asks you to do nothing on a day when everyone around you is doing something.

Which source ranks higher#

The order of sources does not change from event to event, so it pays to fix it once. The company's own filing is primary: annual report, quarterly report, current report. It is not complete, because the company is writing about itself, but someone is accountable for every sentence. The earnings-call transcript captures management's words and the analysts' pushback, which is what the filing leaves out. A regulator's or a court's document is primary for legal questions. A good industry report adds context. A news article raises an alert. A social post points you to a question.

What happens when the weakest source carries the strongest claim? An anonymous account writes that the company is "losing its biggest customer" and the current reports say nothing. Then the claim waits for confirmation from a source that ranks higher, or it disappears. Between two sources you do not pick the one that sounds more certain. You pick the one that ranks higher.

Read opinions in pairs. One bullish and one sceptical analysis of the same company show which assumptions are contested: one author stresses market size, the other customer acquisition cost. Where one sees pricing power, the other sees promotions. Where one trusts management, the other counts the delays. Instead of choosing the author who sounds nicer, you list the contested assumptions and check which has the better source.

A pair gives you something ten agreeing pieces cannot: a list of what you do not know. Ten agreeing pieces only tell you everyone read the same note.

Should you read financial news every day#

In theory you could do all of this in real time. In practice the daily feed beats every rule, because it was built to win, so the rule has to limit time rather than reading technique. My limit for a company I hold looks like this: the full annual report once a year, the transcript after every set of results, news only when it touches one of the thesis assumptions or when the company itself filed something. Everything else waits for the monthly review. How to set up that rhythm is the subject of monitoring a stock without daily news.

The rule has one exception, and it belongs before the conclusion, not after it. For a small company with one product, an alert can be the whole business: a product recall, a regulator's decision, the loss of the only customer. There, zero to two a month can mean one event changes everything, and a month of waiting is too long. I do not know exactly where the size threshold sits below which a daily alert starts paying for itself. I do know that a company with twenty segments and a hundred thousand employees sits far above it. So the time limit is set per company, not once for the portfolio.

The best case for daily reading goes: whoever reads daily sees the alert first. The one who is actually first, though, is whoever opens the document in that same minute rather than the headline, because the headline says that something happened and the document says what. Everyone else reacts to the headline, which means to a version of the event that already has somebody's adjective attached.

What Monitoring does with this#

The whole procedure is a sieve: items pass through four roles, a date and the thesis question, and zero to two events a month settle at the bottom. In Taufolio that sieve is built into Monitoring. Breakthrough news notifies you only about events that could change the company fundamentally, in our experience usually zero to two a month, and stays silent on a two-percent wobble. The gist of the news sits next to a company's news stream and says what in that noise actually moves the business. The question "which assumption does this challenge" only makes sense when the assumptions are written down, so the Investment thesis is generated from the Full report, broken into measurable conditions, and that is what Monitoring checks the thesis against after every set of results. None of it costs credits. To see what it looks like on a real company, open the example reports.

The sieve does not read for you, it leaves you the two stones you would have had to pick up anyway.