The Loughran-McDonald Financial Lexicon

Run a general-purpose sentiment dictionary — built for product reviews or news articles — over a 10-K filing, and it will flag routine words like "tax," "liability," "cost," and "depreciation" as negative, simply because those words sound negative in everyday English. But a 10-K is legally required to discuss taxes, liabilities, and costs; their presence says nothing about whether the filing is actually gloomy. Tim Loughran and Bill McDonald built a word list from scratch, using actual 10-K filings and market reactions, to fix exactly this mismatch.

The analogy: a doctor's vocabulary versus a layperson's

If a layperson hears a doctor say "tumor," "chronic," or "invasive," those words sound alarming. But a doctor discussing a routine, benign case uses the same words constantly as part of ordinary clinical vocabulary — their presence alone doesn't signal bad news. A sentiment dictionary built from general English is like a layperson reading the doctor's notes: it flags the technical vocabulary itself as scary. The Loughran-McDonald dictionary is built from the "doctor's" own domain — actual financial filings — so it only flags words that filings themselves use to signal genuinely bad news.

How the word list was built and organized

Loughran and McDonald read a large sample of 10-K filings and manually classified words into categories based on how they're actually used in financial disclosure, then validated those classifications against how stock prices reacted around each filing's release. The resulting lexicon groups words into several categories, most importantly:

• Negative: words that in financial-filing context genuinely signal bad news — "litigation," "restated," "impairment," "default."
• Positive: words signaling good news in context — "improved," "outperform," "strong."
• Uncertainty: words flagging vagueness or unresolved risk — "approximately," "may," "contingent."
• Litigious: legal-process words that flag legal risk without necessarily being "negative" per se — "plaintiff," "pursuant," "claimant."

Words that a general dictionary would flag negative but that appear neutrally in almost every filing — "tax," "cost," "liability" — are deliberately excluded from the negative list.

Worked example: scoring a sentence two ways

Take the sentence: "The Company recorded a tax liability and incurred litigation costs related to a pending lawsuit." A general sentiment dictionary (e.g., Harvard IV) flags "liability," "costs," and "lawsuit" as negative — three hits, sentence classified strongly negative. The Loughran-McDonald dictionary excludes "tax," "liability," and "costs" as routine financial vocabulary, but does flag "litigation," "pending," and "lawsuit" via its litigious and negative categories — giving a more targeted signal that this passage is specifically about legal exposure, not generic financial gloom.

Worked example: computing a document's net tone

For a filing with 5,000 words, suppose 40 words match the Loughran-McDonald negative list and 15 match the positive list. A simple net-tone score is:

$$\text{tone} = \frac{\#\text{positive} - \#\text{negative}}{\#\text{positive} + \#\text{negative}} = \frac{15 - 40}{15 + 40} \approx -0.45$$

In plain English: this filing's word choice leans clearly negative relative to its own positive-versus-negative word count — a score comparable across filings because it's normalized, and built entirely from a dictionary tuned for financial rather than everyday English.

What this means in practice

The Loughran-McDonald word lists are a standard, widely cited starting point for any quantitative sentiment measure built on SEC filings, earnings calls, or analyst reports, precisely because they were validated against real market reactions rather than assembled from generic English usage. They're still a bag-of-words approach, though — no context, no negation handling ("not a material weakness" scores the same as "a material weakness"), so they're best treated as a fast, interpretable baseline rather than a final answer.