Market vs Reality [Football Season]

← All reports
2026 Season · Week 8

Week 8: Favorites Roll, but One Big Miss Stands Out

Week 8 delivered 14 completed games, with favorites finishing 11-3. The prediction market posted an average Brier score of 0.1788, better than the 0.2500 baseline. No games ended in ties, so there were no tie settlements or tie wagers to report.

The market got most of the week right, though one upset produced a particularly painful miss.

## Market Accuracy by Week

FOOTBALLBLOGGER / NFL MARKET ACCURACY
===============================================
Average Brier score (LOWER IS BETTER)
Each # represents approximately 0.01 Brier points

WEEK  BRIER   GAMES  CHART
-----------------------------------------------
   3  0.2517     16  #########################
   4  0.2172     16  ######################
   5  0.2236     14  ######################
   6  0.2612     14  #########################
   7  0.2272     14  #######################
   8  0.1788     14  ################## <-- CURRENT
LATEST VS PREVIOUS: IMPROVED (-0.0484)
-----------------------------------------------
50/50 BASELINE: 0.2500
===============================================

Best Call

The sharpest prediction came in TEN @ CIN, where Cincinnati beat Tennessee 43-36.

The market gave Cincinnati an 88.5% chance to win, and the favorite delivered. That produced a Brier score of 0.0132, the best mark among the week’s games.

It was a strong favorite performance, but one successful high-confidence call does not prove perfect calibration. It is simply the week’s cleanest prediction.

Biggest Miss

The largest miss came in ARI @ DAL. Dallas was given a 75.5% chance to win, but Arizona prevailed 33-30.

That result produced a Brier score of 0.5700, easily the worst of the week. The market’s favorite record still finished at 11-3, but this upset shows how much a single unexpected result can affect prediction quality.

A surprising result does not necessarily mean the original probability was unreasonable. A 75.5% forecast still leaves room for the other team to win.

Quarter Kelly

The original fictional Quarter Kelly strategy made no wagers this week.

There were no missing 24-hour snapshots. Instead, all 14 games fell below the strategy’s five-percentage-point entry threshold after the walk-forward model’s historical calibration adjustments. The strategy examined earlier completed weeks only, with Weeks 3 and 4 serving as training-only weeks and Week 5 marking the first test week. It also uses quarter Kelly sizing and a 10% bankroll cap.

The five-point threshold may be cautious, but this week’s data only shows that it was selective—not that making more bets would have been better. The original rules remain fixed and were not changed retroactively.

Quarter Kelly 2.5% (Retrospective Comparison)

The 2.5% version is a RETROSPECTIVE historical comparison introduced later, not a prospectively tracked strategy. It uses the same walk-forward model and quarter Kelly sizing while maintaining a separate bankroll.

It placed two wagers, and both lost:

The bets were on Indianapolis in IND @ JAX and Tennessee in TEN @ CIN. Both teams lost. Twelve other games remained below the alternate threshold. This retrospective result should not be used to claim that the lower threshold is better or worse based on one week.

Bet the House

The fictional Bet the House benchmark divides its entire carried-forward bankroll equally among eligible favorites, using the final permanent snapshot midpoint as the assumed price. It began with $10 per game in Week 3.

The benchmark’s three losses came on Dallas, the New York Jets, and Seattle. The wins included Cincinnati, Jacksonville, Buffalo, and Denver, with Denver’s narrow market edge producing one of the larger individual gains.

What We Learned This Week

All of these wagers are fictional. Midpoint prices may not be executable, fees are excluded, and this experiment is not betting advice or a promise of future profits.

← Week 7 All Reports Week 9 →