Persist
Does the leadership hold?
Updated June 3, 2026
What it actually measures. Every week, all ~500 stocks are ranked against each other and sorted into ten buckets. This metric asks: over the last year, how much of the time did this stock stay in its most-common bucket? A high score means it kept its place in the line. A low score means it rotated through many buckets — chop.
How to read it
| If you see… | It means… |
|---|---|
| High Persist + High Perform | A durable leader. The cleanest single read on the card. (Note: because everyone can see it, forward returns are only average — the durability is real, the bargain is not.) |
| High Persist + Low Perform | Durably weak. Stuck at the back of the pack on purpose. Rarely worth buying without a specific thesis for why that changes. |
| High Persist + Middle Perform | Often a utility or staple whose weekly returns barely move — the stability is low drama, not genuine leadership. |
| Low Persist | Either a recent breakout that hasn't earned its position yet, or directionless chop. Check the shorter-window reads to tell which. |
Using it in an IRA
- This is the cleanest sorter of the four and it rewards durability over spikes — a natural fit for long-horizon retirement money that you don't want to churn.
- Use it as a quality-of-evidence filter: a high-Perform stock with low Persist has "run but not proven it"; a high-Perform stock with high Persist has earned the leadership. For an IRA you can afford to wait for the second kind.
- The 13-week and 26-week supporting reads move faster than the 1-year. A rising 13-week with a still-low 52-week often means a name that just stabilized into a new position — sometimes early to a longer trend.
The card tells you about the stock. Your account tells you how much to own.
The fine print — formula, evidence, and where it fails
rankPersistence52W = the share of the last 52 weeks the stock spent in its modal decile of the universe's weekly excess-return-vs-SPY ranking, then percentile-ranked across the universe. Modal (most-common) decile, not mean, so a leader that took a few large hits still reads as a leader.
Stability. 45% stay-rate in the same decile month-to-month, 66% within ±1. Slightly less stable than Perform, because the trailing window itself slides.
Independence. The most orthogonal tile on the card by a wide margin — average absolute correlation with the others is +0.01, essentially zero. It also has the most positive marginal information contribution (+0.004): it carries something the other three genuinely don't.
Tier-1 evidence
A single tile's job is to describe a stable, distinct dimension of stock character — not to produce alpha on its own. We test that with three questions: does the metric stay where it puts a stock month-over-month (stability), is it independent of the other tiles on the card (orthogonality), and does it carry information the others don't (marginal contribution).
Correlation with the other panel headlines
Diagnostics regenerated 2026-06-01
The decile evidence. The cleanest gradient of the four: a near-monotonic climb from ≈ −0.07pp/month at D1 to ≈ +0.60pp/month at D10 across ~397 anchors. The best sort key on the card — though still a screen, not a strategy; no single decile produces a tradeable Sharpe alone.
Evidence — decile screening
Mean forward excess return vs SPY for each decile of the metric, averaged across 397 monthly anchors from 1996-01-05 to 2026-05-15. Cohort: S&P 500 (point-in-time membership via fmp PIT view).
Failure modes
- Cleanest decile gradient among the four Tier-1 tiles — the decile mean climbs almost monotonically from roughly -0.07 pp/month at D1 to +0.60 pp/month at D10. Persistence as a sorter beats persistence as a buy-signal: the cross-section ranks meaningfully, even though no decile produces a tradeable Sharpe on its own.
- Tier-blind by construction — the metric scores a name persistently in the bottom decile exactly as high as a name persistently in the top. Splitting by direction-of-mode under PIT, the winner-tilted half earns +4.2% CAGR excess and the loser-tilted half earns +4.6% — virtually identical. The metric is genuinely about position stability, not direction.
- Sparse middle deciles — the persistence score concentrates at fractional values (k/52 weeks at the mode), so cross-sectional deciles are unevenly populated. Deciles 3-6 typically have a quarter to a third the support of the tails (see the n column above). The monotone gradient survives this, but per-decile precision varies.
- Slow signal — a 52-week window is by definition slow to respond. Stocks that just broke into the top decile score low on persistence even when the recent momentum is strong. The supporting 13W / 26W tiles provide faster reads of the same structural-position question.
- PIT-correct but not fully survivorship-free — the input excess-return panel doesn't carry names that left the S&P 500 before the panel was built (Lehman, Bear, WaMu, Sears). The PIT filter removes the late-additions lift; it doesn't add back the dropped losers. True ex-ante decile means would be marginally lower than what shows above.
Signal = share of last 52 weeks the ticker spent in its modal decile of weekly excess-return rank vs SPY. Forward = next 4-week cumulative excess return. Top-decile basket = top 20 by persistence score, equal-weighted; stats are computed on the resulting per-anchor excess-return series (so Sharpe = Information Ratio and CAGR = annualized excess vs SPY).
Evidence regenerated 2026-06-01