Where our numbers come from: SERPAVI, the cadastre, the INE and four more sources
An estimate without a stated error is an opinion with decimal places. These are our sources, how they're combined, and how wrong they are.
Every market figure you see — a city's average price per m², a neighbourhood's average rent — comes from somewhere, and that somewhere determines what the figure means. The useful question isn't “what's the price per m² here?”, it's “what was that computed over, and with what error?”.
The problem with the portal average
When a portal publishes the average price of an area, it almost always computes it over the listings it hosts itself. That has two consequences. First, it measures asking prices, not prices paid — two things that diverge precisely when the market turns. Second, the sample is its own, with its own bias in property type, in agencies, and in the areas where it's strongest. It's a thermometer inside the room it's trying to measure.
We have our own listings too, and we use them too. The difference is that they're one of five signals, and not the heaviest.
Five signals for the same number
For each municipality we estimate two market levels: rent in €/m² per month and sale price in €/m². Every signal is first expressed in that same unit, for a dwelling of that municipality's mean size, so that they're comparable at all. And each one carries the precision we've measured for it against the official series.
| Signal | What it is | Typical error |
|---|---|---|
| SERPAVI index | Spain's state reference system for residential rents, published by the Ministry of Housing. Effectively ground truth. | ≈ 5% |
| Registered deposit contracts | Rents actually signed and registered with the regional deposit bodies. A transacted figure published without floor area, so we divide it by the municipality's mean dwelling size. | ≈ 12% |
| Official appraised value + provincial yield | The ministry's appraised value for that municipality, converted into rent using its province's gross yield. | ≈ 15% |
| Corrected cadastral model | The municipality's residential cadastral value, corrected for its revision year and for its province's systematic residual. | ≈ 17% |
| Our own listings | The median of our rental inventory in the municipality, rescaled from the sample's mean size to the reference dwelling size. | Sample-dependent |
The detail that makes this work is the cadastral correction. A municipality's cadastral value depends enormously on the year its valuation was last revised: two identical municipalities revised in 1998 and 2018 have incomparable cadastral values. By modelling cadastral value together with its revision year, and then adding back each province's mean residual — Basque and Balearic rents sit above what their cadastral values imply, inland Castile below — that signal's median error falls from 12.2% to 10.6% with no artificial territorial cliff.
Combining beats picking the best
The temptation is to keep the best available signal in each municipality and discard the rest. We tested that, and it's worse. What we do is a precision-weighted geometric mean: each signal is weighted by the inverse of its variance, so SERPAVI dominates wherever it exists and the others hold the estimate up where it doesn't.
The validation is the part nobody publishes and the part that matters most. We test against the 333 municipalities that have an official SERPAVI index, leaving each municipality out of the calculation before estimating it, so we're never scoring ourselves on the data point we're using.
- 8.1% Median error when combining (9.0% with the single best signal)
- 88% Estimates within ±20% (82% with the single best signal)
- 333 Control municipalities with an official index
What we do when a data point is missing
Spain has more than 8,000 municipalities and the official series cover nowhere near all of them. The usual way to plug that gap is to fill it with the provincial average and not mention it. We have three explicit rules instead, and we don't break them:
- A missing data point never penalises. The affected factor stays at a neutral 50 and what drops is confidence, not the score. A property isn't a worse investment because its municipality publishes less statistics.
- Confidence measures data coverage, never investment quality. They're two different axes and we show them apart on purpose: a high score with low confidence means “this looks good and we know it on thin information”.
- Every property is ranked against its own class. A dwelling competes with dwellings and a garage with garages. Comparing their yields in one list would be comparing two different businesses.
And when a benchmark isn't from the property's own municipality but from its province or from the country as a whole, the factor using it is explicitly discounted: an estimate borrowed from another market is not evidence about this one.
The sources, one by one
| Source | What it contributes |
|---|---|
| SERPAVI (Ministry of Housing) | The state reference index for residential rents by municipality. |
| MIVAU (Ministry of Housing) | Appraised value, transaction volumes, the rent price index and mortgage foreclosures. |
| Cadastre | Residential cadastral value, revision year, IBI tax rate and mean dwelling area. |
| INE (statistics office) | Population, net external migration, transactions and price series. |
| AEAT (tax agency) | Mean gross household income by municipality. |
| SEPE (employment service) | Registered municipal unemployment, the basis of demand depth. |
| Ministry of the Interior | Crime statistics, as area context and never as an indicator about a tenant. |
| BOE (official gazette) | Announcements of judicial, administrative and notarial auctions. |
| CartoCiudad and IGN | Official postal addresses and administrative geometries. |
Every figure we publish keeps its source, its territorial scope and its period, and that label travels with it all the way to the screen. If a municipality doesn't publish a series, we say so in its place rather than drawing a dash or filling the gap. You can see the data coverage of any area in its territorial market guides.
Frequently asked questions
What is the SERPAVI index and why does it weigh so much?
It's Spain's state reference system for residential rent prices, published by the Ministry of Housing from tax and contract data. It's the most precise official reference available per municipality, so wherever it's published it dominates our estimate: its measured typical error is around 5%, against 12%–17% for the other signals.
How do you estimate the market in a municipality with no official rent data?
By combining the signals that do exist for it: registered deposit contracts, the official appraised value converted with its province's yield, its cadastral value corrected for revision year and provincial residual, and our own listings. Each is weighted by its measured precision, and the published confidence drops to reflect the thinner evidence.
What error does your market estimate carry?
An 8.1% median error, with 88% of estimates within ±20%. It's measured against the 333 municipalities with an official SERPAVI index, excluding each municipality from the calculation before estimating it so we never score ourselves on the data we use.
Does a municipality with no data score worse?
No. A missing data point leaves its factor at a neutral value and reduces published confidence, never the score. Confidence measures data coverage rather than investment quality, and we show them as two separate axes precisely so they can't be confused.
This article is general information, not investment, tax or legal advice. Rent and value estimates are indicative and carry their own stated margin of error.
