About

Ying Ma Tips runs one quantitative ratings model every meeting. It is a gradient-boosted decision tree model (XGBoost) trained on the Hong Kong Jockey Club's published declared fields and race records. For every runner it estimates the probability of finishing in the first three, called p_top3. The top 4 for a race are simply the four highest p_top3 values.

The model only sees what the declared field shows: class, distance, draw, weight, past finishes and speed figures, days since the last run, running style (leader, midfield, closer), and each jockey's and trainer's win and place rates over the last 365 days. It uses no odds at all, which is why ratings can be published as soon as fields are declared, before the market opens.

Signals have three levels: p_top3 of 0.45 or more is Strong, 0.28 to 0.45 is Medium, below that Weak. Pace pressure is the amount of early speed in the field, derived from each runner's historical running style; 0.65 and above is shown as a fast pace.

The race analysis is written by a language model from the structured data above and may only cite facts in the database. Before publication every text is checked mechanically: every horse, jockey, draw, distance and date mentioned must match the declared field and the form records; any mismatch and the text is not published. Analysis is written in Traditional Chinese.

"Payouts hit" after a race are the sum of official dividends per HK$10 unit for every Win, Place, Quinella and Quinella Place combination that fell entirely within the top 4, before stakes.

On a held-out 2025/26 season of 815 races the model's top pick finished in the first three 53% of the time. Backing the market favourite over the same races would have placed 60% of the time. The model is not sharper than the market; what it offers is a ranking and analysis independent of the odds, readable two days before the race. Past performance does not indicate future results.