Week 6 Report: Market Accuracy Slips, Quarter Kelly Sits Out, and Bet the House Loses Again
Week 6 included 14 completed games. There were no ties, so there are zero tied-game scores to report. No tie wagers were made. Ties are excluded from binary Brier scoring and from the favorite win/loss record; a tied contract would settle at $0.50, while a fictional tie wager would be reported separately rather than as a win or loss. Historical binary calibration training also excludes ties.
The market’s average Brier score was 0.2612. The 50/50 baseline was 0.2500, so the market was 0.0112 higher than that baseline. Lower Brier scores are better, meaning this week finished worse than the baseline.
The market favorite won 8 games and lost 6.
The best prediction was KC over LAC. KC was given a 0.7750 probability and won 22–6. That game’s Brier score was 0.0484.
The biggest miss was LA against ARI. LA was given a 0.8650 probability, but ARI won 18–16. The Brier score was 0.7482.
Other strong predictions included NE over NYJ, with a Brier score of 0.0600, and SF over WAS, at 0.0552. BUF over LV also graded well at 0.1024. Several close calls landed on the wrong side, including SEA over DEN, JAX over HOU, BAL over CLE, IND over TEN, and DAL over GB.
Fictional Walk-Forward Quarter Kelly
Quarter Kelly was in WALK_FORWARD_HISTORICAL status this week. Its supplied starting bankroll was $153.44, and its ending bankroll was also $153.44. It placed 0 bets, with 0 wins, 0 losses, and 0 ties. The weekly profit/loss was $0.00, and the return was 0.0%.
Weeks 3 and 4 were training only. No bets were intended during those weeks, and the bankroll remained $160.00. Week 5 was the first test week. This is a small-sample, fictional walk-forward evaluation, not a retrospective simulation for the training weeks.
For Week 5 onward, the model uses only earlier completed weeks. It never uses later market prices or game outcomes for the game being evaluated. The fixed model takes each earlier game’s 24-hour snapshots, groups the two team probabilities into ten-percentage-point bins, computes the sum of historical outcome-minus-probability residuals in each bin, divides that adjustment by 40 plus the number of observations, and adds the resulting shrunk adjustment to the current game’s 24-hour midpoint probability.
A team must have at least a five-percentage-point edge to qualify. Eligible wagers use quarter Kelly with a maximum wager of 10% of the bankroll. Missing 24-hour snapshots cause a game to be skipped.
This week, all 14 examined games were below the edge threshold. There were 0 missing 24-hour snapshots, 0 games that passed the edge threshold, and 0 zero-sized wagers. As a result, no bets were placed because the games were below the threshold, not because of missing data.
The training set contained 90 observations. The evaluation remains limited by its small sample. Midpoint prices may not be executable, and fees and slippage are excluded. No real money was wagered, and the results do not establish predictive skill or profitability.
Fictional Bet the House Benchmark
Bet the House is a deliberately naive fictional benchmark. The experiment began with $10 per game in Week 3. Each week, the entire carried-forward bankroll is divided equally across all eligible market favorites, using the final permanent snapshot midpoint as the assumed purchase price. Fees are not included.
This week’s supplied starting bankroll was $98.49. The benchmark made 14 hypothetical wagers and finished with 8 wins and 6 losses. The amount wagered was $98.49. Weekly profit/loss was -$8.63, ending bankroll was $89.86, and the return was -8.76%.
Notable wins included CAR over PHI for $4.99, CHI over ATL for $6.62, and NO over NYG for $6.89 in fictional profit. The largest listed losses were DAL over GB at -$7.10, along with several -$7.03 losses, including the favorites in HOU @ JAX, BAL @ CLE, TEN @ IND, ARI @ LA, and SEA @ DEN.
This benchmark is fictional. No real money is wagered, it is not betting advice, and results are reported whether they are good or bad. The rules are not changed retroactively.
What We Learned This Week
- The 14-game average Brier score was 0.2612, which was 0.0112 higher than the 0.2500 50/50 baseline.
- The market favorites went 8–6.
- KC over LAC was the best prediction at 0.0484.
- LA over ARI was the biggest miss at 0.7482.
- Quarter Kelly placed no bets because all 14 games were below the required edge threshold.
- The fictional Bet the House benchmark lost $8.63 this week, ending at $89.86.