Hong Kong property data does not look like property data anywhere else, and if you model it like a Western market you will get it wrong.
The unit of analysis is not the address. It is the estate: a named residential development, often several towers and thousands of flats, where units are close to fungible. Buyers do not shop for “a two-bedroom in Kowloon”. They shop for a specific estate, then a tower, then a floor, then a unit. Price per saleable square foot within one estate is remarkably tight, and comparisons across estates only make sense with the estate as the key.
The 28Hse API is built around that model. This post covers how the pieces fit together.
Listings, estates, and new developments
Three search endpoints, and the distinction between them is the whole point.
/property-search is unit-level: individual flats for sale or rent, filterable by district, estate, price, floor area and property type. This is what you show in a search interface.
/estate-search is the layer above, and it is the one most people miss. It returns estates with price statistics, transaction volume and rental yield. When an investor compares two buildings rather than two flats, this is the endpoint that answers the question. It is also far more stable than unit-level data, since individual listings churn constantly while estate-level statistics move slowly and meaningfully.
/new-property-search covers first-hand and pre-sale developments. In Hong Kong the primary market runs on a different logic from resale, with developer incentives, staged price lists and mortgage arrangements that do not appear in secondary listings. Keep it separate in your analysis.
Asking prices versus transacted prices
/transactions returns recent registered sale and lease records, optionally scoped to a single estate.
This is the endpoint that makes the rest useful. Listings tell you what sellers want. Transactions tell you what buyers paid. In a market that moves as fast as Hong Kong’s, the gap between the two is itself the signal - a widening spread between asking and transacted prices across an estate is a clearer read on direction than either series alone.
Scoping transactions to one estate and comparing against that estate’s current listings is the single most useful thing you can do with this API.
The endpoints around the edges
/location-autocomplete is the front door. Estate and district names in Hong Kong appear in English, Traditional Chinese, and several romanisations, sometimes within the same dataset. Resolve names to IDs here rather than trying to match strings yourself. Nearly every integration problem in this market traces back to name matching.
/property-filters returns valid property type, sort and price-range values for a search, so you are not guessing at enum values.
/school-net-search returns schools by school net, district, type and gender or religion. School nets are a genuine driver of Hong Kong residential pricing, in a way that is more explicit and more geographically rigid than catchment areas elsewhere. Estates in a desirable net carry a measurable premium. If you are modelling price, this is a real feature, not a nice-to-have.
/mortgage-plans returns current bank mortgage offers with rate, cash rebate and penalty period. Hong Kong mortgage products compete on cash rebates and lock-in penalties as much as headline rate, so affordability calculators built on rate alone will misprice.
/agent-search returns agency companies with per-district listing counts, and /service-apartment-search covers serviced apartments and hotels with room types and price ranges - a distinct segment serving relocation and corporate housing demand.
/news-search returns property market news and video articles, useful as a sentiment or event layer over the numbers.
Fields that behave differently here
Saleable area versus gross floor area. Hong Kong distinguishes these sharply, and the ratio between them varies by building age and design. Since 2013, residential sales material has been required to lead with saleable area. Make sure you know which one you are reading before computing price per square foot, because mixing them produces numbers that look plausible and are wrong.
Floor matters, a lot. Within a single tower, higher floors carry a consistent premium. A price model that ignores floor will have unexplained variance that is entirely explainable.
Tower and unit letter. Orientation and view follow from these, and both are priced.
Age. Hong Kong housing stock is old by international standards and building age interacts with saleable-to-gross ratio, management fees, and redevelopment potential.
What people build
Property search products for the local market, which need estate-first navigation rather than map-first.
Investment screening. Rank estates by rental yield from /estate-search, then validate against /transactions for actual transacted prices rather than asking prices.
Valuation models. Estate, tower, floor, saleable area and school net together explain a large share of price variance in a way that generic property models do not capture.
Agency and market intelligence. Per-district listing counts by agency, tracked over time.
Relocation tools. School nets, serviced apartments and district data combine well for corporate relocation products.
Related APIs
For other Asian property markets, see the SUUMO Japan Real Estate API, which has its own market-specific structure built around railway stations and commute time, or the 99.co Singapore API covering condos, HDB flats and landed homes. Browse the Real Estate Data hub for the full list.
Disclaimer
Happy Endpoint is not affiliated with, endorsed by, or sponsored by 28Hse. All data is collected from publicly available sources.
