People who ask about Hong Kong horse racing datasets usually want the same three answers: how far back does it go, what exactly is in it, and will it load into what I already use. This page gives all three without a sales deck, using the real tables and the real dates.
If you want the context first, the Hong Kong racing guide covers the two courses and the calendar. Here we stick to the files.
What is in the data
The Hong Kong archive is organised around races and the runners in them. Each of these is a table you can join on a race or horse identifier.
Races. One row per race, 30,112 of them from 22 September 1979 to October 2026. Each carries the date, course (Sha Tin or Happy Valley), race number, meeting number, class, distance in metres, ranking, going, race type, surface, number of runners, prize and each-way terms. The cumulative times and section times for up to six sections sit on the same row.
Records. One row per runner per race, 365,326 in all. This is the results table: finishing place, horse number, weight, declared weight, draw, the distance behind the runner ahead, win odds, place odds, the position, margin and time at each section, and the finishing time. Jockey and trainer identifiers are on each row.
Racecards. What was known before the race: rating and the change in it, best time, days since the last run, gear, priority, owner, sire, dam, import category and the horse’s recent runs. The gear field is explained in Hong Kong gear codes.
Horses, jockeys and trainers. Profiles with career totals and a split between Happy Valley and Sha Tin: runs, wins, places and strike rates for each venue. Horse profiles add colour, sex, sire, dam and dam sire, prize money and the last rating. There are 16,541 horses, 606 jockeys and 505 trainers.
Veterinary records. 53,966 notes from the Jockey Club, dated from September 2016. Each has the horse, the race, a date, the details and, where there is one, a “passed on” date. The veterinary records post explains how to read them.
Dividends. About 300,000 rows from September 2000, giving the betting pool, the winning combination and the payout in Hong Kong dollars. There are more than a dozen pools, from win and place to quinella, tierce and the multi-race doubles.
How far back each part goes
Not everything starts in 1979, and it is worth being exact about the exceptions.
| Part | Starts |
|---|---|
| Results and race-level section times | 1979 |
| Runner-level section times | 2008 (about 97 to 98% of runners from 2009) |
| Veterinary records | September 2016, mostly 2019 onwards |
| Dividends | September 2000 |
That means a model that uses each runner’s position at the three-quarter mark has 18 seasons to learn from, not 47. A model that only needs finishing order and time has all of them.
Formats
The data is delivered as CSV, SQL, JSON and as MySQL and PostgreSQL dumps, so it fits a spreadsheet, a notebook or a warehouse without conversion. The SQL and dump versions keep the relationships between tables.
There are two ways to buy it. A historical package gives you the archive as it stands. A daily updated service adds each race day’s results and racecards as they are published. Most teams do both: one load for the history and then the daily feed to stay current. The Hong Kong daily updated service page has the details, and the historical exports page covers bulk packages.
Whichever you pick, ask for a sample first and test one join before anything else: take a single race identifier and follow it through races, records and the racecard. If that works in your tool, the rest will.
Dataset or API?
The API serves the same entities over HTTP, with 30 Hong Kong endpoints for races, records, racecards, horses, jockeys, trainers, courses and veterinary records. Choose it when you want a day’s results or one horse’s profile on demand. Choose the files when you want to run a join across thirty years of races, which is slow and expensive one request at a time.
That is the whole archive, without the adjectives. If a column you need is not on the list, say so before you buy. The datasets overview lists every package, and the API documentation shows the Hong Kong plans.
Comments are closed