Quant Library · The machine layer · Football
Stacked Market Ensemble
A classifier that learns, match by match, how much of your model and how much of the market to quote.
Multinomial logistic regression over the model's and the market's log-odds and their disagreement, fitted walk-forward on an expanding window. It closes almost the whole gap between a public model and a sharp market without ever seeing a result before it prices it.
Inputs
4What it takes
| Name | Type | Units | Where it comes from |
|---|---|---|---|
| model_probabilities | dict per match | probability | any match model, including this platform's |
| market_probabilities | dict per match | probability | quoted prices with the overround removed |
| results | list | - | - · only used for matches already played |
| warmup, refit_every, l2 optional | ints and float | - | your choice; 1,500 / 250 / 1.0 are what the validation used |
Outputs
3What it gives back
| Name | Type | Units | Notes |
|---|---|---|---|
| stacked probabilities | dict per match | probability | |
| attention | dict | share of weight | how much of the answer came from the model, the market and their gap |
| walk-forward scores | dict | Brier, nats | beside the model's and the market's on the same matches |
Method
How it works
- Turn each side's probabilities into log-odds, and add their difference as a third feature per outcome: the model, the market, and the disagreement.
- 3 further steps in the licensed specification
Assumptions
- Both inputs are honest probabilities for the same match at the same moment; a stale market price makes the disagreement feature meaningless.
- 1 further assumptions in the licensed specification
Limits
- It does not beat the market, and the comparison that matters is the strict one. Fed only prices that exist before the close, on 7,097 walk-forward matches it scored 0.5872 Brier against our own model's 0.5972 and the closing line's 0.5850: it recovers about 1.7% of the model's error and stays 0.4% behind the close.
- 4 further limits in the licensed specification
In the package
- The full specification: inputs, method, assumptions and limits
- The validation panel as measured on the day of purchase
- A reference implementation extracted from the running source
- Perpetual commercial use for one entity
Reference modules:
- 1 module(s), named in the licensed specification and shipped in the package
Read this before you buy. The platform's own walk-forward tests say the closing line forecasts football better than these models do. What they give you is a coherent price for every market and match, before the market opens, with the band and the working attached. Any model here that has been beaten by something simpler says so in its limits, and the validation panel above is generated from the platform's result files rather than typed in.