Editor's note, 15 September 2026. The numbers in this piece are out of date. They come from an earlier version of the model — a three-bucket composite that has since been retired — tested over a three-year window. The current test runs from January 2014 to September 2026, 51 quarterly rebalances, and the figures on the about page are the ones I stand behind. I have also dropped the survivorship haircut described below: it rested on failure rates the data can't verify, and a guess dressed up as a number is the thing this essay argues against. The page now says plainly that companies delisted during the window aren't in the data, and that what that would do to the result can't be measured from it. What hasn't changed is the principle, which is why the piece stays up: publish the figure you can audit, say what it doesn't prove, and let the reader check the work.
I ran the backtest twice. Same strategy, same window, same rebalance schedule, same position count — the configuration files were identical down to the byte. The only thing I changed was where the financial data came from.
Against my data vendor's figures, the strategy returned 54.5% a year.
Against SEC filings as they were actually filed, and dated when they were actually filed, it returned 41.4%.
I published 41.4%.
Why the lower number
The honest reason is simple: the filing-based number is the one that can be checked.
Every figure in that run traces to a document with a date on it, sitting on the SEC's servers, available to anyone who wants to pull it. If someone thinks my result is wrong, they can go and find out. The vendor number can't be audited that way — I'd be asking you to trust a data feed I don't control and can't show you.
Transparency and accuracy matter more to me than a good headline, and this is what that commitment costs when it's real rather than decorative. It's easy to say you value accuracy when accuracy flatters you. The test is what you do when the auditable number is thirteen points worse than the one you could have printed.
I also wanted the work reviewed. Not admired — reviewed. I wanted people who know what they're looking at to be able to cross-reference the data and tell me if I'd got it wrong. That's only possible if the numbers point at something they can independently reach, which rules out the higher figure entirely.
So it was never really a choice. The moment you decide the point of publishing is to be checked, one of those two numbers disqualifies itself.
Then I argued against myself again
Publishing 41.4% against a benchmark of 22.6% would let a reader conclude the strategy beats the market by about 19 points a year. I don't think that's true, and the page says so.
The portfolio ran a beta of 1.29 — it swung roughly 30% harder than the market in both directions — over a test period that was mostly rising. A portfolio that amplifies the market will beat the market in a rising market. That isn't stock selection; that's leverage wearing a disguise.
Adjust for the risk actually taken and the genuine edge is closer to 12 points a year, and about 9 after the survivorship haircut I apply on top. That's the number I believe. It's less than half the gap the raw figures imply.
I could have left that out. Almost nobody includes it — "alpha" gets quoted freely by people who mean nothing more than "we went up more than the index," which is a different claim entirely and a much easier one to make. But a number I'd have to caveat in conversation is a number I should caveat on the page.
The figure that looks worst
The least flattering number I publish isn't the return. It's the hit rate: the strategy beat its benchmark in only 53% of periods.
That sounds like a coin flip, and I understand why it reads badly. But it isn't the weakness it appears to be, and the reason is the shape of the game.
When you buy a stock, your downside is capped at 100% — the worst possible outcome is that it goes to zero and you lose what you put in. Your upside has no such limit. A position can return two hundred percent, or five hundred, or more, and nothing about the arithmetic stops it.
In a game with that asymmetry, you don't need to be right most of the time. You need to be right often enough that the wins you do get are large enough to carry the losses. Being right 53% of the time with capped losses and uncapped gains is a genuine advantage — it just doesn't feel like one, because we're wired to score ourselves on how often we're correct rather than on what being correct is worth.
That's also why the return comes from a few large winners rather than from being consistently right, and why I say so on the performance page. A strategy that produces its return that way demands a different kind of patience than one that grinds out small consistent gains. If you can't sit through the majority of positions doing nothing much, the math never gets a chance to work.
What the honesty is actually for
Here's the part that matters more than any of the disclosure.
If I flatter my own backtest, I'm not primarily deceiving readers. I'm blinding myself. A strategy that looks like it returns 54% doesn't need changing. A strategy that returns 9 points of risk-adjusted edge on a short sample is a strategy I have to keep watching.
If the data showed the approach was ineffective, I'd keep molding it into something that works. That isn't a threat to the method, it's the method. You have to be like water and change as the data tells a clearer story.
You can only do that with honest numbers. Every haircut I apply to my own results — the filing-based data source, the beta adjustment, the survivorship discount — is protecting my ability to notice when something stops working. Optimistic figures don't just mislead the people reading them; they remove the signal you'd need to act on.
That's why the disclosure isn't performance. It's infrastructure.
What I'd still tell you not to trust
The window is 3 January 2023 to 30 April 2026 — a little over three years, rebalanced quarterly, five names per category. That is not a long sample. Three years is one market regime, and this one was mostly favorable.
The deepest drawdown was 31%. Anyone running something like this should expect to watch roughly a third of it disappear at some point and keep going anyway, which is far easier to read than to do.
And a backtest is a reconstruction, not a record. I didn't own these positions across that window; a computer decided what would have been bought using data as it stood at the time. That's the most honest form of the exercise I know how to run, and it's still a simulation.
So: 41.4% is the auditable number, roughly 38% after the survivorship haircut, about 12 points of genuine risk-adjusted edge before that haircut and 9 after, on a three-year sample, with a 31% drawdown, winning barely more than half its periods.
That's a real result. It's also a much smaller claim than the one I could have made, and every step down from 54.5% was a choice I made on purpose.
The Bull Rankings is automated fundamentals research for education, not personalized investment advice and not a recommendation to buy or sell any security. Always do your own research.