Skip to content

Fasteign

05Data explorerIn progress, not publicly launchedAlso known as Fasteignir

A map-first explorer of actual property sales in the Reykjavík capital region.

Illustration, not a screenshotFasteign is not publicly launched yet

The idea

Property prices are usually discussed through asking prices and anecdotes. HMS publishes every registered purchase agreement, which is a far better answer to ‘what do homes here actually sell for’, if you can get at it.

What I built

An Icelandic-language site with a postcode choropleth of the capital region, filters for period, type, size, rooms and build year, market charts, and an address lookup that estimates a value from comparable sales and shows every comparable and its weight.

The interesting part

The simpler model won. A ridge hedonic regression was tested on a time-separated holdout against the comparable-sales baseline and lost clearly, so it stays out of the product. Ingestion keeps source hashes and audit reasons for every row, and the unit assumption in the source data is written down rather than quietly applied.

Decisions

The baseline beat the model

On 5,281 paired predictions from a time-separated holdout, comparable sales beat a ridge hedonic regression: a median absolute percentage error of about 6.3% against 12.4%. The regression is not deployed, and the evaluation says plainly that one split is not a guarantee.

Source: README, Model experiment

Small samples stay quiet

No price is shown for fewer than five sales, and there is a warning under ten. Shared cards apply the same rule as the app.

Source: README, Available now

Assumptions are written down

The official field descriptions omit the monetary unit, so values are scaled by 1,000 based on the observed source scale. The assumption is stated, and it waits for confirmation from the data owner before public launch.

Source: README, Data and methodology

Ingestion leaves an audit trail

Every row keeps its source timestamp, hash, original fields and normalised audit reasons. A daily job tests, ingests and builds, and leaves the existing site untouched if any step fails.

Source: README, Data and methodology; Deploy and automatic refresh