New to these reports? Start here
- Dotted-underlined words have a plain-English definition — hover or tap them. Every term is also on the glossary page.
- "Null" means we found nothing, not that something broke. Most reports here are negative results, on purpose — knowing an idea doesn't work is the point.
- Two questions get asked separately. First, is the effect real? Second, is it already priced into the betting odds? An effect can be completely real and still useless to bet on.
- A "calibration" row is a self-check. It runs the same method on something already known to be true. If that fails, the whole report is unreliable — so it's reported alongside the findings.
- If a confidence interval includes zero, the real effect might be nothing at all, so no claim gets made.
$2,500 across 24 small confidence pools (2026-10-24)
portfolio_sim.py → portfolio_sim_results.json. Always-favorite (ordered by
market probability) in every league; each league has its own rivals; payout
top 3 = 50/30/20 of the pot; $104 a league; 2014-2025 seasons.
Rival field calibrated to our Yahoo pool: favorite-pick rate median 0.815 (real 0.81), 10th-90th 0.74-0.88 (real 0.72-0.89), gap to always-favorite 4.1 points of max (real 4.3). On 2025 weeks 1-12 the sim reproduces the pool's spread (sd 2.9% vs real 2.6%) and baseline-minus-best (-0.3 vs +0.4).
| Scenario | Mean profit | Median | P(down) | 5th pct | Cash rate / league | 1st / league |
|---|---|---|---|---|---|---|
| 12-person leagues | +$10,100 | +$10,900 | 0.3% | +$5,100 | 97% | 68% |
| 12-person, real outcomes 2014-25 | +$10,200 | +$10,800 | 0.1% | +$5,700 | 97% | 68% |
| 30-person leagues | +$22,200 | +$25,000 | 3.3% | +$1,600 | 84% | 45% |
| 12-person, 10% site fee | +$8,900 | +$9,500 | 0.5% | +$4,500 | 97% | 68% |
| 12-person, sharper field (2x skill, half the ordering noise) | +$5,700 | +$6,400 | 17% | -$2,250 | 74% | 34% |
| 6 leagues instead of 24 | +$10,100 | +$10,800 | 0.4% | +$4,500 | 97% | 68% |
Worst real season: 2016 (30-person: mean +$6,700, P(down) 21%).
Why to discount these
- The empirical check is less rosy than the sim. Resampling the real 2025 members' weekly scores gave always-favorite P(1st of 30) = 30% over 12 weeks; the sim gives 41% on the same weeks. Treat the sim as ~25%+ optimistic.
- 2026 so far cuts the other way: always-favorite is 7th of 33 after 6 weeks (64.2% of max vs leader 66.7%). Six weeks is noisy, but it is a live miss.
- The field is one pool. Leagues of strangers may be sharper; the sharper row is the realistic downside, and it is still clearly positive.
- Diversifying across leagues barely reduces risk (6 vs 24 leagues: same mean, same P(down)): your score is identical everywhere, so the risk is your season, not any one league's rivals.
- Not modeled: weekly prizes, missed entries, collecting from strangers.
On this page
Related
NFL survivor pools: which strategy earns the most pot equity? (2026-10-24) Survivor strategies vs the REAL ESPN field, 2020-2025 (2026-10-24) Do pick'em players get gutsier late, or when behind? (2026-10-19) How well does the public pick? — field baseline, 2014–2025 Winning a 30-Person Confidence Pool Testing "always pick the favorite" — and why a Monte Carlo can't do it Season Prize vs Weekly Prize: Shoot for the Season Best Pick'em Strategy for 2026: Pick Every FavoriteSource
backtests/pickem/FINDINGS_PORTFOLIO.md
updated 2026-10-24 16:31