How Much History Does Persistfolio Need?

Persistfolio looks for companies whose performance relative to the S&P 500 has actually persisted, not just companies that happen to be up over some arbitrary window. That raises an obvious question: how much history do you actually need before “persistent” means anything?

Our shortest standard research period is five years. We wanted to know if that was longer than necessary — if three years could find the same kind of companies sooner, Persistfolio could react faster without losing what makes the persistence idea meaningful in the first place.

So we tested it.

What We Tested

We now have results from three historical Persistfolio portfolio experiments. The earlier one, Testing the Persistfolio Methodology Through Time, used a five-year methodology and a five-stock portfolio. This new one asks a narrower question: does Persistfolio really need five years of history, or can three years get you most of the way there?

These are historical walk-forward tests — at each review date, Persistfolio could only use information that would actually have been available at that time, and the portfolio was then carried forward and measured.

Worth flagging before going further: the 5-year/5-stock test covers a different historical period than the two 10-stock tests, so its return isn't directly comparable to theirs. The clean comparison here is 3-year/10-stock versus 5-year/10-stock — same period, same portfolio size, same testing framework. The only thing that changed was how much history Persistfolio was allowed to use.

Three Years Versus Five Years

Both 10-stock portfolios started March 31, 2014 and ran through September 30, 2024. Neither one held exactly 10 stocks at every point along the way — during a few stretches, particularly 2021 and 2022, fewer than 10 companies actually met the bar, and we didn't lower the bar just to fill seats. “10-stock” here means the target ceiling, not a guaranteed constant count.

The five-year portfolio returned 361.7%, against 215.0% for the reconstructed S&P 500 price benchmark — a 16.53% annualized return versus 12.16% for the benchmark. The three-year portfolio returned 154.6%, or 9.79% annualized.

So five years beat the benchmark by about 4.4 percentage points a year. Three years trailed it by about 2.4 points a year. Shortening the window didn't just fail to help — it made things worse.

The Three-Year Portfolio Wasn't Just a Faster Version of the Same Idea

Going in, one hypothesis seemed plausible: maybe three years would catch the same companies as the five-year methodology, just earlier. There's a bit of evidence for that — among companies that eventually showed up in both portfolios, the three-year version got there first 23 times, versus 12 for the five-year version, with 6 ties.

But zoom out to the whole portfolio and the picture looks different. Average holdings overlap between the two was only 27.9%. Most of the time, they were holding largely different companies, not the same ones on different schedules.

The three-year portfolio also churned much more — 1.95 holding changes per quarterly review, against 1.10 for the five-year version — and its worst drawdown was considerably deeper:

Interestingly, the three-year portfolio actually beat the S&P 500 more often on a quarterly basis — 55% of quarters, versus 52.5% for the five-year version. It just lost bigger when it lost, and that's what shows up in the cumulative number and the drawdown. Frequent small wins don't make up for occasional large losses.

Put together, three years doesn't look like a quicker read of the same signal. It looks like it's measuring something else.

Why Five Years Might Matter Here

Three years can still be dominated by a single market cycle, a sector trend, or one unusually strong stretch for a particular company. That's a real problem if what you're trying to measure is persistence rather than a recent hot streak — an extra two years gives the methodology more room to tell the difference between a durable pattern and a short-lived one.

We can't prove that's the exact mechanism from this experiment alone. But we can see what happened when we took the extra history away: the methodology got less stable, and its results got worse.

What About the Five-Stock Portfolio?

This isn't the first time we've run Persistfolio through history. In our earlier study, Testing the Persistfolio Methodology Through Time, we asked what would have happened if Persistfolio had existed years ago and been applied repeatedly, using only the information available at each point in time.

The portfolio developed through real historical reviews, not a single backward-looking pick. We also reconstructed actual historical S&P 500 membership, so a company could only be considered once it had genuinely joined the index.

That test used the five-year methodology with a five-stock portfolio, and it returned 668.1% against 362.6% for the S&P 500.

This new experiment asks a different question — not whether Persistfolio works through time, but what happens when you change how much history it's allowed to see.

Because the two tests cover different periods, we're not claiming five stocks beats ten, or drawing any conclusion we haven't actually tested. What we can say is narrower: we've now tested the five-year methodology in two different portfolio sizes, and both finished ahead of their benchmarks. When we kept the portfolio size fixed at ten and only shortened the window to three years, it didn't.

Why We're Not Testing Four Years Next

The obvious next question is what happens at four years. Or two. Or six.

We could run all of them and report whichever number looks best. That's not what this is for.

We started with one specific question — can three years replace five without losing the effect — and got an answer. Chasing every nearby variation after seeing this result would turn a hypothesis test into a search for the best-looking backtest, and a good-looking backtest isn't the same thing as a better methodology.

The point of these experiments is to make one reasonable change, test it, and see whether it holds up — not to keep changing things until something does.

What We Learned

We ran this because we suspected Persistfolio might work fine with less history. The data didn't back that up.

The three-year portfolio sometimes found companies earlier, but overall it held a substantially different set of them, turned over holdings almost twice as fast, drew down much harder, and returned far less: 154.6% against the five-year portfolio's 361.7%, versus 215.0% for the S&P 500 over the same stretch.

That's consistent with the separate five-stock test, where the five-year methodology also finished ahead of its benchmark, over a different period.

None of this proves five years is exactly the right number, and it says nothing about what any of these portfolios will do from here. But it's a better reason to keep using five years than simply “that's what we picked originally.”

We asked whether we could shorten it to three.

The evidence says no — five years held up, three years didn't.

---

Historical and backtested results are hypothetical and are not a projection of future performance. The experiments described above cover specific historical periods and use reconstructed historical data. The 5-year/5-stock experiment covers a different period from the 3-year/10-stock and 5-year/10-stock comparison, and their cumulative returns should not be compared directly. The 10-stock study uses a reconstructed S&P 500 price-return benchmark and does not include dividends. Persistfolio model portfolios are standardized quantitative research models and do not represent actual investor accounts. Results do not include individual taxes, transaction costs, or investor-specific circumstances. Persistfolio provides research for educational and informational purposes and does not provide personalized investment advice.