M. R. Hoque
Resume

Go to a page

← All projects

RESEARCH TOOLS · VERSION 1.0

Disclosure Gap Calculator

Two disclosures. A clearer view of the differences.

Compare an earnings call with a filing, or two versions of the same report. See exactly which words drive the results.

Runs in your browserNo text uploads or storageHow it works ↓Animated walkthrough →Methods & reproducibility

01 Add your disclosures

For example: earnings-call remarks

0 / 200,000 characters

For example: 10-K MD&A

0 / 200,000 characters
Word lists Illustrative starter lists · not LM

The starter lists demonstrate the method and are not a validated financial sentiment dictionary. Edit them below or import the official Loughran–McDonald master CSV for research use.

Get the LM dictionary ↗

Exact word matches; no stemming or negation adjustment. “Not strong” still counts “strong.”

All gaps are A minus B.

TRANSPARENT BY DESIGN

What the numbers mean

Tone & uncertainty

Each category counts matching word occurrences, divided by all word tokens and multiplied by 100. Net tone is the positive rate minus the negative rate. Tone and uncertainty gaps are percentage-point differences: A − B.

These are descriptive text measures, not evidence of deception, a forecast, or an investment recommendation. Compare similar sections and periods; quotations, negation, context, and document length can change interpretation.

Readability

We report average words per sentence and the Automated Readability Index (ARI): 4.71 × characters/words + 0.5 × words/sentences − 21.43. A higher ARI indicates greater estimated reading difficulty; it does not measure disclosure quality. We retain raw scores, including negative values.

ARI formula reference (NIST) ↗

Counting rules, limits, and reproducibility

English text only. Text is Unicode-normalized (NFKC) and lowercased. Tokens are English letter sequences (internal apostrophes retained) or numbers (decimal/comma groups retained); hyphens split words. Numbers enter the denominator. ARI characters are token letters and digits. English Intl.Segmenter detects sentences, so abbreviations and browser versions may affect counts. No stop words are removed. Inputs are limited to 200,000 characters each; short passages below 100 words are flagged.

LM imports use category values strictly greater than zero; negative removal-year flags are excluded. Categories may overlap. Exports contain exact word lists, all calculated measurements, version, and input SHA-256 fingerprints, but exclude the original texts. Keep your original texts to reproduce a comparison.

Text and imported dictionaries stay in memory in this browser tab. Editing an input hides old results until you calculate again. Nothing is saved to local storage. The theme preference is stored separately by the portfolio.

Built by M. R. Hoque · Methods & reproducibility · Send feedback