There are 492 companies on this site that the Bull Rankings score refuses to touch. Banks. Insurers. REITs. Search for JPMorgan and you'll get a page, and metrics, and a number — but not the number. Not the 0–100 quality-growth score the rest of the site is ranked by.
That's deliberate. It's also not what I wanted.
What breaks
The model rests on three pillars: quality, growth, and a fair price. The quality pillar is built from return on invested capital, free cash flow, and the debt on the balance sheet. Those are the right questions to ask an industrial company, or a software company, or a retailer.
They are the wrong questions to ask a bank.
For a bank, debt isn't a risk to be measured against equity — debt is the product. Deposits are how the business funds itself. A debt-to-equity ratio that would be alarming at a manufacturer is ordinary at a lender, and reading one the way you'd read the other tells you nothing. Free cash flow has the same problem. When the balance sheet is the business, cash moving across it isn't the surplus that a factory throws off after paying for itself — it's the inventory.
Banks are valued differently, and they always have been. They trade on price-to-book. Insurers trade on book value and underwriting. REITs trade on the value of what they own and what it pays out. None of that flows the way my algorithm runs, and none of it is a flaw in the companies.
I did not accept this quickly
I tried, and tried, to fit them in.
The most serious attempt was a bank-appropriate quality bonus. The logic was reasonable: financials were shut out because the compounder test asked for cash flow and leverage figures they don't produce in a comparable form, so give them a route in on the terms that do fit — reward a high return on equity bought at a cheap multiple. That is close to how a bank investor actually thinks. I built it, and I walk-forward tested it the same way I test everything else.
It made the model worse.
- −2.9 percentage points of annual return across the full window
- −4.0 points in-sample
- Deeper drawdown than without it
And the detail that settled it: at the margin, the bonus slipped banks into the top five picks — and those picks underperformed the names they pushed out. It wasn't neutral. The financials it let in were actively worse than the compounders they displaced.
I got creative, and it wouldn't work. That's the whole story of it.
The result stands on the years I tested
I want to be clear about the limit of that evidence, because it cuts against me.
My test window was 2023 to 2026. That was one rate environment, not all of them. Banks can lead in a different one — a value regime, a steeply sloped yield curve — and my window doesn't contain that. So the honest version of my finding isn't "banks don't belong in a quality-growth screen." It's "on the period I could test, letting them in made the model worse, and I have no evidence about the periods I couldn't."
That distinction matters enough to keep writing down, even though nobody would notice if I quietly dropped it and kept the result that went my way.
For now they stay separate. I don't see a way I'd combine them any time soon. But never say never — and if the rate environment moves somewhere genuinely different, that's a reason to re-run the test, not a reason to assume the old answer still holds.
What I built instead
The easy path here is a blank space. The page loads, the metrics show, and where the score should be there's nothing — no explanation, just an absence that reads like something is broken.
I didn't want that either, so financials get their own scale: a Financial Strength score, 0–100, built from the things that actually describe these businesses. Price-to-book. Return on equity. Dividend yield and whether the payout is covered.
It's aware of what it's looking at. A REIT's return on equity reads nothing like a bank's, because depreciation flattens it. A REIT's payout ratio measured against net income is structurally above 100% and is supposed to be — penalizing that would be scoring the accounting rather than the company. Capital-light asset managers, where book value means very little, get anchored on earnings instead of book. Three sub-types, because treating them as one would repeat the exact mistake that got financials excluded in the first place.
What it gives you is an easily digestible read on the strength of the book — the book value, and the metrics behind it — for a company my main model can't judge.
What it is not
It is a coarse strength read, and I'd rather say so than let it be mistaken for more.
There is no funds-from-operations figure behind it, no net interest margin, no combined ratio, no non-performing loan data. That information isn't in the data I have. A real credit model would need all of it. This is a structural read built from what can be sourced reliably, and it's labeled that way everywhere it appears.
Why the two scores never share a column
This is the part I feel most strongly about.
A 70 on the Financial Strength scale means a strong financial for its type. It does not mean as good as a 70 on the quality-growth score. They are different measurements of different things, and the moment they sit next to each other in one ranked list, a reader will compare them — because that's what a column of numbers is for.
I could have normalized both to a single 0–100 and put everything in one table. It would look tidier. Nobody would know. And it would be comparing apples to oranges with a straight face.
This site is already dense. There are a lot of numbers on it. I think we do a good job of explaining what each one is and where it comes from — that's most of the point of the whole project — but density is exactly why a second scale has to be labeled as a second scale rather than blended into the first. Adding one more number is fine. Adding one more number that quietly means something different from the number beside it is not.
So they stay designated as different. The site says so, out loud, on every page where it matters.
The names I actually like
I like JPMorgan. I like Goldman Sachs. I don't own either of them. The one I do own is Realty Income — I think their book is strong.
On the scale I built for exactly these businesses: JPMorgan reads 60.5, Goldman 59, Realty Income 71.9.
My instinct about Realty Income and my model's read of it agree, on the one part of this site my main score refuses to touch. That's a small thing, and I don't want to oversell it. But it's the reason the second scale exists rather than a blank space — I'd rather measure what I can measure, say plainly what it does and doesn't cover, and keep it in its own lane.
The alternative was to force a number I didn't believe in. On the page, it would have looked exactly like the ones I do — same scale, same styling, same air of precision. You would have had no way to tell them apart.
That is precisely why it isn't there.