What the lawyers
knew.

In November 2018, Apple's lawyers added new risk language about China, tariffs, and slowing iPhone sales. Sixty days later Tim Cook cut guidance and the stock lost seventy billion dollars. The warning was already on file. Nobody read it.

→  Good morning

At Citibank, my first task in every quarterly close was the same: line up this period's disclosure next to the prior period's, and find every word that had changed. The work was routine, internal compliance. I did not realize at the time that the same comparison, run across thousands of companies and decades of filings, is one of the most consistently profitable signals anyone has identified in modern finance.

Three economists eventually figured that out. Lauren Cohen and Christopher Malloy of Harvard, working with Quoc Nguyen of DePaul, downloaded every 10-K and 10-Q filed with the SEC between 1995 and 2014, then compared each filing to the same firm's prior filing using four different text-similarity measures. They published the result in the Journal of Finance in 2020. The title was Lazy Prices.

A long-short strategy that bought firms whose disclosure language barely changed and shorted firms whose language changed substantially earned up to one hundred and eighty-eight basis points per month in abnormal returns. Over twenty-two percent annualized. Twenty years of filings. Every public company in the United States.

What Apple's lawyers knew in November.

Here is the example that makes the framework concrete.

On November 5, 2018, Apple filed its annual 10-K. The Risk Factors section had been substantially revised from the prior year. The new language called out specific risks the prior year's filing had not emphasized with the same weight: economic deceleration in greater China, the impact of newly imposed U.S. tariffs, headwinds from a strong dollar, and the possibility that customers were upgrading their phones less frequently than they had historically.

Apple's stock barely moved on the filing. The headline financials were strong. The 10-K is more than two hundred pages.

On the evening of January 2, 2019, Tim Cook released a letter to investors cutting Apple's first-quarter revenue guidance by approximately nine billion dollars. The letter cited China, tariffs, currency, and slower upgrades. Each issue had been previewed in the November filing. The stock fell roughly seven percent in after-hours trading and ten percent the next day. Seventy billion dollars in market value evaporated in a session.

A securities class action followed. In March 2024, Apple settled for four hundred and ninety million dollars without admitting wrongdoing. The plaintiffs' theory was straightforward: the company's senior leadership knew about the deteriorating outlook in China before the disclosure on January 2. The Risk Factors section of the November 10-K is, in retrospect, the public record of what they knew, and when.

Why language changes are not random.

A revision to a 10-K is rarely cosmetic. Two reasons predominate. First, management briefed the legal team about a specific risk that needed to be on the record to satisfy disclosure obligations. Second, an event in the prior year materialized that required updated language. In both cases, the new sentences reflect something the company already knows, something that has not yet broken into the headline numbers.

Apple's November 2018 filing fits the first pattern. The legal team was being protective. By the time iPhone sales actually fell short of guidance two months later, the disclosure was already on the record. The legal review process was working as intended. The language was where the law required it to be, and where investors who took the time to read it could have found it.

The other side of the trade matters at least as much. Firms that do not change their disclosure are firms whose business has been stable enough that no new language was warranted. Cohen, Malloy, and Nguyen found that the non-changers in their sample earned positive abnormal returns of their own. Boring 10-Ks correlate with stable businesses, and the market underprices that stability.

The framework is a way of separating businesses on whether anything worth reporting has changed. The market rewards the firms where nothing has changed, and punishes the firms where something has, on a delay of months.

Apple, Nov 2018 → Jan 2019

~$70B / 60 days

Market value lost on the guidance cut. The risk language was filed sixty days earlier.

Where the signal lives.

The paper found that not every section carries equal predictive content. Three are worth real attention.

Item 7. Management's Discussion and Analysis. The narrative section where management explains what happened. New language describing competitive pressure, slower segment growth, or shifts in capital allocation philosophy is meaningful. Watch for hedging that was not there the year before, particularly around forward-looking guidance.

Item 1A. Risk Factors. The legal team's section. Most of it is boilerplate that varies little year over year. The revisions are almost always informative. Pay particular attention to specific new risks (a country, a customer, a product line) versus general new risks (macroeconomic uncertainty, cybersecurity in general). Specific risks are usually a flag the lawyers were pushed to add for a reason. General risks are often defensive boilerplate updates that carry less signal.

Litigation and executive references. Cohen, Malloy, and Nguyen found these were among the most predictive sub-categories. New language describing a pending lawsuit, or new phrasing referring to the CEO or CFO, each carried independent signal. The litigation-specific subset alone underperformed peers by about seventy-one basis points per month, or eight and a half percent annualized, from one section of one document.

The boilerplate sections (properties, governance, exhibits, signature pages) are mostly noise. Skip them.

A thirty-minute exercise.

For one company, the workflow is straightforward. Pull the most recent 10-K from EDGAR. Pull the prior year's. Open both as PDFs. Use Microsoft Word's built-in Compare Documents feature, or Adobe Acrobat's Compare Files tool, or a free online diff checker for shorter passages.

Most changes will be cosmetic: section number updates, cross-reference tweaks, year-over-year date rolls, formatting adjustments. The substantive changes will stand out. New sentences. New paragraphs. Removed language. The exercise takes about thirty minutes per company once you know where to look. (Now made even easier with things like Google Notebook LM)

Run this on your top one or two holdings every reporting cycle. The first time you find a substantive change you would have missed otherwise, the academic finding will stop feeling academic. The framework is straightforward and the edge comes from doing it consistently.

→  The takeaway

The Apple case is the cleanest illustration, but it is not the only one. Cohen, Malloy, and Nguyen found that the pattern holds across thousands of firms, hundreds of bankruptcies, and decades of disclosures.

The disclosures are always on file. Whether anyone reads them carefully is a separate question.

Have a good Tuesday. Tomorrow we move from the language inside filings to the accounting policies they describe. The same logic applies in a different form: the most economically important changes are often the ones management has the least incentive to highlight.

— Marques

About the author

Marques Blank runs Blank Capital, a fractional CFO and FP&A advisory. Previously Northrop Grumman and Citibank. CMA, MBA, Series 65.

Sources

Cohen, L., Malloy, C.J., & Nguyen, Q. (2020). Lazy Prices. Journal of Finance, 75(3), 1371–1415.  ·  Apple Inc. Form 10-K filed November 5, 2018; letter to investors dated January 2, 2019; In re Apple Inc. Securities Litigation (settlement filed March 2024).  ·  Loughran, T., & McDonald, B. (2011). When is a liability not a liability? Journal of Finance, 66(1).

For informational purposes only. Not investment advice. Specific securities mentioned are case studies, not recommendations. The 188 basis points per month figure represents the upper bound of long-short alpha across multiple specifications in the cited paper. Past performance does not predict future results.