How we estimate what a home is worth

What this is, and what it is not

We estimate what a home in Luxembourg would be advertised for today. The answer is a range, not one number.

One model does this. It gives the estimate shown on our listings, and it answers an owner who asks about their own home.

It is not a valuation. Nobody from ChatHome has been inside the home, seen its condition or read its papers. A surveyor, a bank valuer and a notary each do a different job. This replaces none of them.

• It estimates an asking price, not a sale price.
• It covers Luxembourg only.
• Its measured error is published further down this page.

What it learns from

The model learns from the adverts on ChatHome for homes in Luxembourg, for sale and to rent. From each advert it takes the facts of the home and the asking price.

It never learns from sale prices. No recorded sale price goes into the model.

Adverts that have left the market still count. The model is told the month each price was observed, so it knows how old a price is.

Some adverts are left out of the learning: homes outside Luxembourg, plots of land, commercial property, rooms in shared homes, adverts with no price or no living area, and prices per square metre that cannot be right.

What it reads about a home

Every home becomes a fixed list of facts. A fact the advert does not state stays empty. The model does not guess it: an energy class nobody stated is not a bad energy class.

For an owner's home, the facts are the ones the owner gives us. The home is placed in its commune or city quarter, and travel times are those of that area. Facts we only know for an exact address stay empty.

• The home: kind of home, living area, plot, rooms, bedrooms, bathrooms, floor, year built, year renovated, energy class, heating, glazing and parking, and whether it is new, furnished, recently renovated or in need of renovation.
• Its features: lift, balcony, terrace, garden, cellar, fitted kitchen, pool and other features the advert names, plus yearly co-ownership charges when stated.
• The place: commune, quarter, postcode and position on the map. When we do not know the exact address, the home is placed in its street, quarter or commune.
• Getting around: morning travel time by car and by public transport to Luxembourg City and to the nearest work centre, and the time to the nearest stop, station, school, crèche and park.
• The surroundings: shops and services within one kilometre, how often public transport runs, flood and noise zones, and what the commune's land-use plan (PAG) says about the plot.
• What it does not see: the pictures, the description and the agency's name. An earlier step does look at an advert's pictures. A street sign, for example, can help us place the home on the map, and that place is one of the facts the model reads.

Why the answer is a range

The model gives three figures for a home: a low one, a middle one and a high one. Each is a price per square metre, multiplied by the living area.

The low figure is set so that about one home in four is advertised below it. The high figure is set so that about one in four is advertised above it. This is the model's 25% to 75% band, and it is built to hold about half of homes like this one.

After training, we check the range on homes the model did not learn from. We widen or narrow it until it holds about half of them.

So about half of homes are advertised outside their range. A price outside it is normal, not a warning.

A wide range is an honest answer. It means the facts we have leave real doubt.

How we check it

A model checked on the homes it learned from only repeats what it was told. So we never measure it that way.

The adverts are split into five parts. Each home is priced by a model trained on the other four. Adverts that look like the same home always stay in the same part, so one copy cannot give away the price of another.

A kind of home gets a model estimate only when three things hold. The model was tested on enough homes of that kind. It did better than a plain commune figure, which is the living area times the commune's usual asking price per square metre. And its range held close to half of the test homes.

An owner's request is priced by the model trained on all the adverts.

What it covers, and what it cannot do

We would rather publish our limits than let you find them.

• It gives a model estimate for apartments and houses for sale, and for apartments to rent. Studios count as apartments.
• A house to rent and a plot of land get the commune average instead. It is clearly marked as a commune average, not as a model estimate, and it is given only where the commune has at least five such adverts. With fewer, we give no figure.
• Commercial property is not estimated.
• It estimates asking prices. Real sale prices appear only in a separate “sale-level” figure, for apartments for sale in some communes. That figure uses the official average sale price over twelve months from the Observatoire de l'Habitat. It does not train the model.
• That official figure is published for existing apartments only. Houses, rentals and communes with no published figure therefore get no sale-level figure.
• Condition is known only when the advert or the owner states it. The model does not judge condition from pictures, so two homes that differ only in condition can get the same estimate.
• An owner's home is priced as a home known by its commune or city quarter, from the facts the owner gives. Nobody inspects it.

When it changes

The model is retrained by hand, when we decide to. There is no fixed schedule.

A new model comes into use only after the checks described above. The card at the top of this page shows when the current one came into use.

No accuracy figure on this page is typed by hand. The page reads each one from adverts the current model has priced.