What does Hong Kong horse racing analytics measure?
The analytical unit is the runner within a specific race. Relevant inputs may include recent form, class, distance, course, going, draw, carried weight, pace characteristics, horse attributes and pre-race market information. Each variable needs a clear definition and an availability timestamp.
The aim is not to find one universal rule. It is to compare all runners under the same race conditions, estimate relative chances consistently and retain enough context for the result to be reviewed later.
- Estimated win and place probabilities
- Runner comparisons at the same race and data cut-off
- Market-implied probabilities and odds movement
- Model version, update time and limitations
How is quantitative analysis different from a racing tip?
A conventional tip often presents a single conclusion. Quantitative analysis preserves the estimate, the time it was produced, the comparison benchmark and the uncertainty around it. Two runners can have similar rankings while carrying materially different probabilities and prices.
A model should be judged over a defined sample rather than by a selected winning example. Probability calibration, Brier Score, Log Loss and comparison with the market are more informative than a short sequence of correct selections.
How should readers interpret model probability and odds?
First confirm whether the data is pre-race, live or post-race. Then check the probability definition and timestamp. Decimal odds can be converted into a simplified implied probability by dividing one by the odds, but Hong Kong pool prices include takeout and must be normalised before a fair comparison.
A difference between model probability and market probability is a research signal, not automatic proof of value. It may reflect valid information, delayed data, model error, pool liquidity or a mismatch between timestamps.
| Signal | Question it answers | What it cannot prove alone |
|---|---|---|
| Model probability | Estimated outcome chance at a defined data cut-off | That a runner will win or is attractively priced |
| Market-implied probability | How public prices allocate relative chances at that time | That the market is always correct |
| Difference | Whether model and market estimates disagree at the same timestamp | A persistent or executable advantage |
Conceptual framework only; these rows are not Sigma Quant empirical findings.
What makes an analysis transparent and reviewable?
Reviewable analysis states the race and venue, the information available at prediction time, the model version, the evaluation period and any material exclusions. Historical predictions should remain accessible after results are known so readers can distinguish genuine pre-race output from hindsight.
Sigma Quant is designed around this evidence trail. Public performance claims should separate backtests, out-of-sample evaluation and live predictions, and should always include the sample size and date range when those figures are approved for publication.
What are the limits of Hong Kong racing models?
No model observes every factor. Late withdrawals, changing track conditions, imperfect data, behavioural effects and random race incidents can all affect outcomes. Market structure also changes over time, so relationships learned from one period may weaken in another.
For that reason, Sigma Wagering treats model output as analytical and educational information. It is not a betting operator, does not remove financial risk and does not guarantee a winner or profit.