NFL week one spreads we disagree with
Opening week is the worst calibrated week of the season, because every model is running on last year's teams and this year's rosters.
Every NFL model faces the same problem in September: it has been fitted on teams that no longer exist. Rosters turned over, coordinators moved, and the only evidence about the new versions is preseason football played mostly by people who will not be on the roster in October.
Where the disagreements are
Three lines this week are more than two points away from our number, and all three involve a team whose projected strength changed substantially in the offseason. That is not a coincidence: it is where the market and any model are both least anchored, so it is where they diverge most.
The largest is a road favourite giving three. Our number has the game closer to a pick, and most of that comes from an offensive line that lost two starters and a projection that has not caught up.
Why we are staking these smaller
Because the honest error on a week one number is much wider than the error in November, and a wider error means a disagreement of the same size is less likely to be real. This is the second requirement in how to bet on football with a model, applied to the week where it bites hardest.
A three point gap in week twelve, against a number with tight error bars, is a much stronger statement than a three point gap now.
What the market knows that we do not
Week one lines are set by people who watch camp reports, and camp reports are information that does not reach any structured dataset. When a line disagrees with our number in September, the base rate favours the line more than it does in midseason.
What this does not tell you
Nothing about how these finish. Week one is one week, it is the noisiest week, and a model that goes 0-3 here has learned almost nothing about itself.
Common questions
Why is NFL week one hard to price?
Because every model is fitted on last season's rosters and the teams have changed. There is no in-season data yet, so the market is running on projection rather than measurement.
Do spreads move more in week one?
Typically yes. With less information holding a line still, week one lines move further on smaller news than lines in November do.
Should you bet less in week one?
If your edge comes from a model, probably. The uncertainty on every input is at its widest, and a wider error bar means fewer genuine disagreements.