Developers: comparison (restricted) and market structure (public)
Script: scripts/modules/developers_internal.py. Source: Dubai Land Department open data on data.dubai only:
transactions (470061), real estate projects (467654), developers (462802). No commercial data is used to
build any figure.
What is published and what is not
Developer and master-developer names are company data (Private Entity Data under the Dubai Open Data Licence).
The developer-by-developer ranking is therefore restricted and is written only to
data/restricted/developers_internal/. Only market-structure statistics are published, with no company
names and no company ids:
| Public file | Content |
|---|---|
data/out/developers_internal/concentration_monthly |
per month and registration type: sales, sales attributed to a developer, coverage, number of active developers, top-1/5/10 shares, HHI |
data/out/developers_internal/attribution_coverage_monthly |
per month and registration type: sales by attribution method, including the reasons a sale cannot be attributed |
data/out/developers_internal/inference_validation |
accuracy of each estimated attribution rule, measured on projects whose developer is known |
site/data/developers_internal.json |
the same series from 2015, compact |
Restricted files (internal only): brand_monthly_ranking, brand_period_ranking, entity_monthly_ranking,
master_developer_monthly_ranking, project_brand_map, entity_brand_groups. inference_validation is public (the
table above, accuracies rounded to 3 decimals); an unrounded working copy is kept with the restricted files.
Scope of sales
- Residential sales from the staging view
sales_residential: DLD group "Sales", usage Residential, segments Apartment, Villa and Townhouse, with valid date, price (>= AED 10,000) and area (100 to 2,000,000 sq ft). See00-staging.md. - Both registrations are counted: Title Deed (ready property, "Existing Properties") and Oqood
(off-plan, "Off-Plan Properties"). Every table has a
registrationcolumn withAll,Title DeedandOqood. - All sales in a developer's projects count: first sales by the developer and resales between owners. DLD open data does not say whether the seller is the developer, so the two cannot be separated.
- Monthly series start in January 2010. The latest month (November 2026, 1st to 5th) is partial and flagged
partial_month.
How a sale is linked to a developer
- Register link. The sale's
project_numberis looked up in the DLD project register, which givesdeveloper_id. That id is then looked up in the DLD developers register to get the legal entity. Caution: in the DLD projects file, the columndeveloper_nameactually contains the master developer's name (it is identical tomaster_developer_nameon every row). This module therefore never uses that column and always goes throughdeveloper_id. The master developer is resolved the same way throughmaster_developer_id. - Brand grouping (editorial, rule-based). Large developers sell through many legal entities (one company per project
is common). Entities are grouped into a brand, in this order:
(a) the legal name starts with a distinctive brand prefix from an explicit list kept in the script;
(b) the entity's webpage domain in the developers register is the same as that of an entity already in a brand, or equals a brand key;
(c) its legal name, after removing legal suffixes (LLC, FZE, PJSC ...), is identical to that of an entity already in a brand;
(d) its register phone number is used only by entities of one brand (applied only to entities with no webpage of their own).
Anything else stays as its own brand (the legal name without suffix). If a registered project's entity has no brand, but the
project's English name contains a listed brand word (for example "<Name> by <Brand>"), the project goes to that brand
(method
register_spv_by_project_name). - Estimated links for projects missing from the register. The project register on data.dubai was last refreshed on
2026-11-01. Its highest project number (4,615) is above every number used in transactions (up to 4,591), so in this build
no project launched after the snapshot has sales; 257 project numbers that are absent from the file (252 of them with sales), most of them older projects and a few recent
launches, appear in transactions but not in the register. Projects launched after a register snapshot are handled the same way. The same rules are also tried for the few registered projects whose
developer id cannot be found in the developers register. In order:
inferred_brand_in_name: the English project name contains a listed brand word (when the name has the form "... by X", only X is looked at, so "... by <other developer>" blocks a brand word elsewhere in the name);inferred_by_developer_in_name: the name has the form "... by X", where X (at least 3 letters, not a generic word such as "The" or "Dubai") is the first word of the legal names of registered developers that all belong to one brand;inferred_sibling_project: the name, without phase, tower and number words, equals that of registered projects that all belong to one brand;inferred_master_community: at least 90% of the registered projects first sold since 2021 in the same master community belong to one brand (minimum 4 such projects);inferred_area: the same test on the DLD area. These are estimates. Each rule's accuracy is measured on registered projects with a known developer, by leave-one-out where relevant, and published ininference_validation. The name rule is tested strictly against the legal entity's brand, so it is a lower bound.
- Not attributed.
unlinked_no_project_number: the sale has no project number (mostly ready villas and older freehold buildings registered outside a DLD project).unlinked_project_not_in_register: the project is missing from the register snapshot and no rule applies.unlinked_developer_not_in_register: the project is registered but its developer id is missing or absent from the developers register.
Measures (per brand, period and registration; restricted tables)
| Field | Definition |
|---|---|
n_txn |
number of sales |
total_value_aed |
sum of actual_worth |
total_sqft, avg_sqft |
sum and mean of the transaction area (DLD procedure_area, converted at 1 m² = 10.7639 sq ft). For villas this can be the plot area, not built-up area |
avg_price_aed |
mean price |
avg_aed_sqft |
mean of each transaction's price per sq ft |
wavg_aed_sqft |
total value divided by total sq ft |
median_aed_sqft |
median of transaction price per sq ft |
n_projects |
distinct projects with sales |
share_inferred |
share of the brand's sales attributed by an estimated rule |
rank_volume, rank_value |
rank by
n_txn and by
total_value_aed within the period and registration (ties share the best rank) |
low_count |
true when n_txn < 5 |
Periods in brand_period_ranking: 2026-11-01..05 (month to date), October 2026, Q4 2026 to date (October), 2026 year
to date, and the last 12 full months (November 2025 to October 2026).
Public market-structure measures
coverage_volume/coverage_value: share of residential sales (count / value) attributed to any developer brand.n_developers_active: brands with at least one attributed sale;n_developers_10plus_sales: brands with at least 10.topK_share_volume/topK_share_value: share of attributed sales (count / value) taken by the K largest brands of that month.hhi_volume/hhi_value: Herfindahl-Hirschman index of brand shares of attributed sales, on a 0 to 10,000 scale;equivalent_n_developers= 10,000 / HHI.- Suppression: any month and registration with fewer than 5 attributed sales has its statistics set to null and its count shown as
<5. In the coverage table, any cell with fewer than 5 sales has its count shown as<5and its value and share set to null. BecauseAll=Title Deed+Oqood, when exactly one of the two is suppressed for a month and method, theAllrow's value is also set to null so the suppressed value cannot be recovered by subtraction. In the site JSON attribution series, a share that represents 1 to 4 sales is set to null.
Caveats
- Concentration figures depend on the brand grouping. Joint ventures and project companies with no shared name, webpage or phone in the register stay separate, so the largest groups' shares are, if anything, understated.
- Off-plan sales of projects missing from the register snapshot are either estimated (see
share_inferredandattribution_coverage_monthly) or left unattributed. With the 2026-11-01 snapshot they are below 1% of off-plan sales in August to October 2026; the share grows as the snapshot ages. - Attribution uses the developer currently recorded for the project, not necessarily the one at the time of the sale.
- Area is the DLD transaction area. It is not always comparable with the "built-up area" quoted in marketing material.