How the score works

Most dividend ratings are black boxes. You get a grade, you don't get the recipe. This page is the recipe. Anyone can check it, copy it, or argue with it, and the track record is public too.

The short version

For any stock, we measure 12 things: how much of the company's cash and profit the dividend eats, how much debt sits on top of it, how steady the business is, how the market is pricing it, and the company's own dividend track record. Each factor is compared against 18 years of history of what real dividend cutters looked like before they cut. Each one gets a mark from 0 to 100, the marks are averaged with the weights below, and the result is your score. Higher is safer.

The 12 factors and how much they count

The little chart next to each factor shows how the mark changes as the factor gets worse or better. These curves were learned from history, not set by hand.

No cheating with hindsight

• Companies report their quarterly numbers weeks after the quarter ends. The model respects that: it only ever uses numbers that were actually public on the date of the score.
• If a company has less than two years of usable history, we say "unscorable" instead of guessing.
• Stock splits and one-off special dividends are cleaned out first, so the model sees the real payout trend, not accounting noise.

Banks and REITs are judged differently

Some ratios simply don't describe some businesses. Cash flow numbers mean little for a bank, and REITs look terrible on standard payout ratios by design. The model skips those factors for those companies and lets the rest count more. You'll see them marked "n/a" on the score page.

A detail we like

Nobody told the model that a frozen dividend is a warning sign. It found that on its own: in the history, companies that stopped raising their dividend were far more likely to cut it soon after. Companies freeze before they cut.

What the ratings mean

The line between Borderline and Safe was drawn so that 95% of the dividend cuts in our history happened below it.

What it can't do

No score sees the future. When COVID shut the economy overnight, some rock-solid companies suspended dividends that looked perfectly safe weeks earlier, and the model missed those like everyone else did. A company that collapses very fast between weekly updates can also slip through. And the score says nothing about whether a stock is cheap or expensive. It answers one question only: how likely is the dividend to survive the next 12 months. The full track record, including every miss, is on the performance page.

For the technical reader

The factor curves are isotonic fits on quantile bins of roughly 17,000 point-in-time observations from 2008 to 2019, weights come from logistic regression blended with priors, and everything was frozen before any post-2019 data was scored. The calibration files and reference implementation ship with the project, so the whole thing can be reproduced end to end.