Do Early-Season Lines Actually Have More Value? We Checked.
"Bet the first few weeks — the books haven't figured it out yet." It's close to gospel in betting circles. Before repeating it, we checked it against our own model and our own historical numbers. The answer was more interesting than the conventional wisdom.
What's Definitely True: Week 1 Has No Season Yet
Our player prop model estimates a player's output from their own recent games. In Week 1, that sample is zero for everyone. Without a fix, a model either treats every player at a position as identical or borrows signal from somewhere else. We hit the first problem head-on this week:
Puka Nacua — Receiving Yards, Week 1 line: 90.5
| Version | Model mu (avg) | Model "under" % | Implied edge |
|---|---|---|---|
| Flat league-wide prior (the bug) | 35.0 | 99.9% | "+4,700 bps" — nonsense |
| Fixed: uses Nacua's own 2025 games | 125.1 | ~18% | Real, defensible number |
The flat-prior version wasn't finding an edge — it was blind, assuming a true #1 receiver and a depth piece both average 35 yards. That's a broken model dressed up as a confident one, not a soft market. We fixed it by carrying over each player's own prior-season performance until this season's sample builds up, phasing out naturally as 2026 games accumulate.
So Are Early Weeks Actually More "Volatile"? We Checked.
We pulled our own historical model output — real snapshots from last season, weeks 4 through 14 — plus this year's fixed Week 1 slate, and compared how big our model-vs-market disagreements ("edge") were, week by week.
| Week | Props compared | Median |edge|, bps | Mean |edge|, bps | % with edge > 2,000 bps |
|---|---|---|---|---|
| 1 (2026, post-fix) | 745 | 1,036 | 1,534 | 27.7% |
| 4 | 7,900 | 1,528 | 1,836 | 34.3% |
| 5 | 3,300 | 1,690 | 2,013 | 42.9% |
| 6 | 1,141 | 1,562 | 1,999 | 42.2% |
| 7 | 4,956 | 1,495 | 1,843 | 38.1% |
| 9 | 10,123 | 1,855 | 2,197 | 47.1% |
| 10 | 9,678 | 1,629 | 2,026 | 42.9% |
| 11 | 1,420 | 1,149 | 1,493 | 26.2% |
| 12 / 14 | 7,429 | 1,545 | 1,927 | 39.7% |
Not the clean downward slope we expected. Week 9 had the widest average disagreements in the sample; Week 1, running on the fixed carryover model, was among the tamest. This is one model across parts of one season, not a controlled study — but our own numbers didn't support "Week 1-3 = biggest edges" as a clean rule, so we're not repeating it as one. Our read: edge size mostly reflects how much signal the model has, not how "figured out" the market is — and Week 1 looked tame here specifically because the carryover fix gave it real signal instead of a guess.
What About the Sportsbooks?
This half we can't test directly yet — we don't have historical closing-line data to measure book accuracy week over week. It's a widely held belief among bettors, and the mechanism is plausible (books face the same small-sample problem early), but we're treating it as a hypothesis, not a proven fact. Related but distinct: we've written before about why early-week timing (Monday vs. Sunday) matters within a single game week.
What This Means for You
- Don't assume Week 1-3 props are automatically softer — our numbers didn't show that cleanly. Evaluate each line on its own.
- Be skeptical of huge "edges" on players with no current-season track record, from us or anyone. It more likely means a model is guessing blind than that it found a lock.
- We're tracking this ourselves, in the open, all season — and will publish what we find, either way.
Methodology: edge compares model probability to de-vigged market consensus at the time each snapshot was taken; historical weeks are single snapshots, not full-week averages. Treat this as a first look, not a backtest.
Questions or feedback? Find us on X/Twitter @fourthandvalue.
← Back to Blog