What NFL betting markets price correctly (and the one thing they don't)

The premise

Every NFL broadcast hands you a betting angle dressed as insight. *West-coast team playing an early kickoff — fade them.* *Team's on a short week — fade them.* *Backup quarterback starting — hammer the under.* *Hard to beat a team three times in one season.* They sound like edges. So we tested them — every one — against the closing line, leakage-free, across 4,000+ games from 2006 onward.

The verdict is uncomfortable for the angle-sellers: the market already prices almost all of it. Here's the full ledger, and the single exception that survived.

The angles that are already in the line

Short rest and bye weeks. A team's rest advantage tells you nothing the line hasn't already used. Regressing the spread's error on the rest differential gives a flat, insignificant slope (p=0.42). Teams on short rest cover 49.6% against the spread; teams off a bye cover 51.4% — both indistinguishable from a coin flip. The old "bye-week teams cover" edge was real two decades ago and has been fully arbitraged.

Time zones and the body clock. The most beloved travel angle — a West-coast team dragged into a 1 p.m. Eastern kickoff, body clock reading 10 a.m. — produces a 48.6% cover rate over 395 such games (p=0.61). Traveling two or more time zones east: 49.2% (p=0.70). The market sees the map.

Total travel distance. No monotonic relationship between miles flown and cover rate (slope p=0.86). The one bucket that flashed significance (300–1,000 miles, p=0.04) is an isolated cell among five — multiple-comparison noise, not a signal.

Prime time. Sunday- and Monday-night games are marginally *more* variable than Sunday-afternoon games (residual standard deviation 13.7 vs 13.1, Levene p=0.038) — and it isn't a marquee-blowout effect, since the average spread is identical. But there's no directional edge: home teams, favorites, and overs all land at coin-flip rates under the lights.

Divisional rematches and "the third meeting." When division rivals meet a second time, the game is no calmer than the first — the unpredictable-by-the-line variance is statistically unchanged (Levene p=0.50). The first result repeats only weakly (r=0.24), and the market fully absorbs it: control for the rematch spread and the first game adds nothing (coefficient +0.015). As for "it's hard to beat a team three times in one season" — across 32 playoff third-meetings since 2002, the team that won the regular-season series won the playoff rubber match 64% of the time. The better team just keeps winning.

The trap: backup quarterbacks and the danger of hindsight

This one deserves its own section, because it's where we nearly fooled ourselves.

The on-field effect of a backup quarterback is enormous and real: a team starting a non-primary QB scores 6.3 fewer net points per game, and its straight-up win rate collapses from 53% to 34%. So the natural next step — *fade any team starting a backup* — looks like free money. Our first cut said backup teams covered just 43% of the time, implying their opponents hit a gaudy 57%.

It was an illusion, and the illusion was leakage. That first cut defined a team's "primary" quarterback using the *whole season's* starts — information you wouldn't have on game day. Redo it honestly, defining the established starter only by games *already played*, and the edge evaporates: backup-QB teams cover 49.9% against the spread, profitable in just 5 of 16 seasons. The market prices a known backup correctly. The points are real; the *edge* was hindsight.

The lesson generalizes. A backup-QB indicator absolutely belongs in a model that predicts game *outcomes* — and ours has carried one for exactly this reason. But "improves prediction accuracy" and "beats the closing line" are different claims, and conflating them is how backtests lie.

The one survivor: fade the early-season favorite

After all that efficiency, one angle clears the bar. Through the first quarter of the season, favorites are systematically overpriced. In Weeks 1–4, favorites cover only 46% of the time (n=1,162, p=0.01) — so backing the early-season underdog hits about 54%, clearing the 52.4% break-even against standard −110 juice. Walk-forward and leakage-free, the rule returns +3.97% ROI over 986 bets, profitable in 10 of 16 seasons. By Week 14 the bias is gone entirely (favorites cover a coin-flip 50.6%).

Why does this one persist when everything else gets arbitraged? Because it's an *information* bias, not a *physical* one. Travel, rest, and time zones are known quantities a market clears instantly. But in September the line still leans on last year's reputations and offseason narrative, and the public piles onto the brand-name favorites. It takes a month of real games to wash the recency out. Notably, this is a bet-selection rule, not a model feature — when we folded "early" and the spread into a gradient-boosted model, the signal got diluted into the noise and the ROI went negative. The targeted rule clears +4%; the kitchen-sink model does not. Precision matters.

The meta-lesson

Two things make this study worth more than its individual results. First: the physical angles broadcasters love — travel, rest, body clocks, prime time — are dead money, priced to the decimal. The market's blind spots are informational and temporal, not logistical. Second, and more important: a backtest is only as honest as its hardest leakage check. The most exciting result we found (fade backups, 57%) was the one that didn't survive contact with point-in-time discipline. The least exciting framing of the best result (a plain Weeks 1–4 rule, no model) was the one that did.

If an angle requires hindsight to look profitable, it isn't an angle. It's a mirror.