Hedge-word density
Counting the words a company reaches for when it is not certain. No model involved.
What it measures
A count of hedging terms — may, might, could, approximately, substantially, potentially, believe, estimate and their relatives — expressed per 1,000 words of narrative text.
The figure is then compared against a baseline drawn from comparable filings, so the output is a position relative to normal rather than a raw count.
This is pure counting. There is no model in this metric at all, which is why it carries the computed badge and never the AI one.
It is most useful on the sections where management has discretion over wording: the MD&A and the risk factors, rather than the financial statements.
How to read it
There is no good or bad level in the abstract. Filings in regulated and litigious industries hedge more as a matter of course. What is informative is movement against the same company’s own prior filings.
| Range | What it suggests |
|---|---|
| Near the baseline | Typical for a filing of this type |
| Well above, and rising | Management is qualifying more than it used to |
| Falling sharply | Also worth a look — unusual confidence is a change too |
What it does not tell you
The part most tools leave out. A score you over-trust is worse than no score.
- Hedging is often required. Forward-looking statements carry legally mandated qualifiers, and counting them tells you about the law, not the company.
- A change of counsel or a new disclosure template can move the figure without anything in the business changing.
- It cannot distinguish a hedge about a trivial matter from a hedge about a material one. It is a pointer toward where to read, not a finding.
- Document length affects it. A filing that grew by adding boilerplate will dilute its own density.