How prices are calculated: Prayk publishes its methodology with live figures

Every market price on Prayk, the Good deal badge, the Deal Score and calculators such as What's my home worth? all follow the same rules. Those rules are now published in one place, with live figures, on the How prices are calculated page.

The data: how much and from where

The calculations rest on real asking-price listings published on the market, Prayk's own catalogue included. Each listing is an observation: price, details, location, when it appeared and when it disappeared. The indicators at the top of the page come straight from the live database. At the update on the morning of 27 September:

• 188,563 observations over the last 90 days;
• 57,873 price baskets;
• 19,675 of them ready to show, meaning they hold at least 8 listings.

The calculation in five steps

1. Observations are split into tiered baskets. For cars: make → model → year (±1) → body type → engine, gearbox, mileage. For property: deal → type → district → rooms → floor-area band → building type.
2. In each basket, outliers are removed with the IQR × 1.5 rule, then the median and the P25–P75 range are calculated. The median rather than the average, so a few inflated listings cannot drag the result.
3. Reading starts from the most specific basket and stops at the first one with at least 8 listings. Baskets are never added together — one basket is always read.
4. For a property listing with coordinates, a separate geo-basket is built from neighbouring listings within 500 metres, which is more precise than district statistics.
5. Baskets are rebuilt every night; the gap to market shown on listing cards is refreshed every 6 hours.

Honesty rules

• A basket with fewer than 8 listings is not shown to users — a “market price” from a tiny sample is misleading.
• A price 45% or more below market is not a “great opportunity” but a risk signal.
• If at least 3 similar listings in the same building or complex (within 150 m) do not back up the neighbourhood comparison, the listing is not called cheap or pricey.
• The Good deal threshold is ±10%: a smaller gap counts as market level.
• All prices are asking prices, not the prices deals closed at.

The page makes a point of saying that prices do not come from a chat model. Statistical baskets produce them, and a hedonic model does so in the value calculators. Artificial intelligence checks a listing's text and photos and explains risks, but it does not change the number.

Accuracy: what is ready and what is still being collected

The value calculator's model is tested on listings it has not seen: median absolute percentage error (MdAPE) and the share of estimates that land within ±10% / ±20%. To be clear: the weekly live backtest table has not filled up yet, so for now the page shows values from the initial measurement and says so. By those values, the median error is 8.3% at high confidence, 10% at medium and 15% at low. As the live table fills, the figures will update automatically.

Where to find it

You can reach the page from the Prices section and from the “Read the full methodology” link in each tool's method block. Tools that work by these rules include What's my home worth?, the Check a listing report on each listing page, the market comparison and the Good deal badge. For what the badge means, see our earlier piece.