Horse · Model vs Market
Where the model disagrees with the market, who's actually right? · 120513 settled runners
How to read this. Each runner is bucketed by edge = model prob − market prob.
The bottom row (Model ≫ market) is the value test: these are runners the model thinks
are far better than their price. If actual win% there is ≥ the model's predicted %, the
model's optimism is justified and there's an overlay edge. If actual is below it, the model is
just wrong where it's most confident. Sample matters — a result on under ~150 runners is
marked thin/noise; don't trust it until it's deep (the harness 48-runner edge that vanished at
1,128 is why). Analysis only, not betting advice.
Model vs market
behind 0.0013
| Disagreement bucket | Runners | Confidence |
Actual win% | Model% | Market% | Who's right | Flat ROI |
Model ≫ market
model thinks far better than price — the value/overlay test |
494 |
trustworthy |
10.9% |
14.5% |
9.6% |
market ✓ |
-4.0% |
Model > market
model a bit keener than the market |
34527 |
trustworthy |
4.5% |
7.3% |
5.2% |
market ✓ |
-29.7% |
Agree
model and market within 1.5% |
62989 |
trustworthy |
6.6% |
7.7% |
7.1% |
market ✓ |
-26.3% |
Model < market
model a bit cooler than the market |
11389 |
trustworthy |
20.2% |
17.7% |
20.3% |
market ✓ |
-14.9% |
Model ≪ market
model thinks far worse than price — the lay test |
11114 |
trustworthy |
35.9% |
26.4% |
33.6% |
market ✓ |
-9.6% |
Verdict (value bucket):
On the 494 runners the model rated far above market price, actual win rate was
10.9% vs the model's 14.5% prediction.
The model was over-optimistic here (actual below predicted), so these divergences
are not a profit signal — the market is the better guide where the model disagrees.
This is the expected result for a model that matches but doesn't beat the market.