Do NFL sportsbooks get sharper through the season?

The intuition, and why it's mostly wrong

It feels obvious that a sportsbook should price games better in December than in September. By Week 14 the book has watched every team for three months; in Week 1 it's working off the offseason, last year's roster, and a preseason no one trusts. More information should mean a sharper line.

So we tested it directly. We pulled every regular-season closing spread and total from 2006 through 2024 — 4,927 games with lines — and compared each one to what actually happened. The answer is more interesting than "yes": by the measure most people mean, books are not sharper later. But the market *does* get more efficient, in one narrow way you could have bet on.

Accuracy is flat all season

The cleanest measure of sharpness is error: how many points off was the line? For the spread, that's the gap between the closing number and the actual margin. For the total, it's the gap between the closing total and the actual points scored.

Both are dead flat from the opener to Week 18:

If anything the error drifts *up* slightly late in the year. Regressing each game's absolute error on its week number gives a slope of +0.013 points per week for the spread (t = 0.53) and +0.022 for the total (t = 0.97) — both statistically indistinguishable from zero. Season by season it's a coin flip: of 19 seasons, only 8 had sharper late-season spreads, and the average difference (+0.31 points) isn't significant (p = 0.37).

The reason is humbling. Most of a spread's error isn't mispricing — it's the irreducible randomness of football. A single turnover or a missed field goal swings a game by a touchdown, and no amount of information prices that out. The residual standard deviation sits near 13 points all season and never shrinks. Week 1 lines are already about as accurate as a line on an NFL game can be.

The line is unbiased the whole way

Sharpness isn't only about error size — it's also about whether the number is *centered*. Regress the actual margin on the closing spread and you want a slope of 1.0 (the book's points map one-for-one onto reality) and an intercept of 0 (no systematic lean). The market clears that bar from the first week:

A favorite laying 7 is, on average, exactly a 7-point favorite — in September just as much as in December.

Where it *does* sharpen: explanatory power

Here's the subtlety. While the *size* of the error holds flat, the share of outcome variance the line explains climbs steadily:

That isn't a contradiction. As the season matures, good and bad teams separate, blowouts get more common, and the spread of actual margins widens. The book widens its spreads to match — so the line keeps pace with reality (R² rises from 0.14 to 0.23) even though the absolute miss stays the same. The market is doing more *work* late in the year; it just isn't getting more *precise*.

The one real edge: early-season favorites are overpriced

Now the part you could have acted on. Track how often the favorite covers the spread, split by early and late season:

In the first quarter of the season, favorites cover only 46% of the time — significantly below the 50% a fair line implies. Translated: blindly backing early-season underdogs would have hit around 54%, clearing the 52.4% break-even against standard -110 juice. By the back stretch the bias has completely evaporated — favorites cover a coin-flip 50.6%, and the edge with it.

This is the recency trap. Early lines lean on last year's reputations and offseason hype, and the public piles onto the brand-name favorites, nudging those numbers a hair too far. It takes a month of real games for the market to wash that out. The home-field and over/under angles show no such early edge — betting the home team (48.6%, p=0.35) or the over (50.6%, p=0.73) was a coin flip in every window.

Bottom line

NFL sportsbooks don't get more *precise* through the season — point error is essentially constant from Week 1, because randomness, not mispricing, sets the floor. What they get is more *efficient*: a genuine early-season tilt toward favorites exists, it's worth more than the vig, and the market arbitrages it away by the back half. "Soft early lines" is true — just not in the way most people assume. The softness isn't sloppy accuracy; it's a predictable, fadeable bias with an expiration date.