How we predict dividend cuts 18 days early

A dividend cut is rarely a surprise to the company. It's a surprise to the shareholder — usually on the morning the press release drops and the stock gaps down 20%. Our reliability model exists to close that gap. On average, it flags at-risk payers about 18 days before the cut is announced.

Here's how the model works, what signals actually matter, and — just as important — what it can't do.

The cost of being late

When a high-yield name cuts, two things happen at once: the income you were counting on drops, and the price falls because income investors leave. You get hit on both sides. The investors who do best aren't the ones who react fastest on cut day — they're the ones who trimmed or hedged in the weeks before.

What actually precedes a cut

Cuts don't come from nowhere. In the historical record, the same fingerprints show up again and again:

• Coverage erosion — the payout ratio creeps above what free cash flow can sustain, quarter after quarter.
• Rising leverage — debt grows faster than earnings, and interest starts competing with the dividend for the same dollars.
• Falling cash flow — operating cash flow turns down while the declared dividend stays flat.
• Yield blowouts — the market pushes the yield to double digits not because the payout rose, but because the price is pricing in the cut.
• History — a company that has cut before is far more likely to cut again.

Building the reliability model

We trained a gradient-boosted model on years of payout decisions — every increase, hold, and cut we could attribute to a clean set of fundamentals at the time. The label is simple: did this company cut its dividend in the next two quarters? The features are the fingerprints above, computed point-in-time so the model never gets to peek at data that wouldn't have existed yet.

That last detail matters more than the algorithm. The most common way to build a model that looks brilliant in a backtest and useless in production is to accidentally feed it future information. Our pipeline timestamps every input and reconstructs what was knowable on each historical date.

The features that matter most

When we rank feature importance, cash-flow coverage and the trend in coverage dominate. A single bad quarter is noise; a steady slide over four quarters is signal. Leverage and prior cut history follow. Raw yield, on its own, is a weak predictor — but yield combined with deteriorating coverage is one of the strongest.

A 9% yield on a company with rising coverage is fine. A 9% yield on a company whose coverage has fallen four quarters running is a countdown.

The data pipeline

The model is only as good as what feeds it. Every night, our pipeline ingests dividends, prices, and fundamentals from multiple providers, reconciles them against a canonical symbol registry, and recomputes scores. Disagreements between sources are flagged rather than silently averaged, so a bad data point doesn't quietly move a score.

How to read the score

In the app, every dividend payer gets a reliability score. A high score means the fundamentals support the payout today. A falling score is the part to watch — direction matters more than the absolute number. When a holding's score drops into the danger zone, that's your cue to look closer, not to panic-sell.

What it can't do

No model predicts a board's decision with certainty, and it can't see a sudden one-off — fraud, a surprise acquisition, a regulatory shock. The score is a probability, not a prophecy. We publish it as one input into your judgment, never as a buy/sell command.

The point isn't to be right every time. It's to make sure you're rarely the last to know.