Licence 3 MIASHS · semester 5 · R, Python

What is a Grenoble flat worth?

A hedonic pricing study on DVF open data: what property characteristics are implicitly priced, and what happens when you stop insisting the model be linear.

Results

Explained variance by specification. The log transform is the whole story: prices are right-skewed and multiplicative in their characteristics, which is exactly what the log-linear hedonic form assumes. Source: ProjetEconometrie/results.json.

Prediction error, euro-denominated models only. The two log-price specifications are deliberately absent: their RMSE is in log units and does not belong on this axis. Source: ProjetEconometrie/results.json.

What the random forest costs

It divides error by about 4.5, and it does so by giving up exactly the thing a hedonic model exists to produce: an interpretable implicit price per characteristic. The two models answer different questions, so the write-up reports both rather than declaring a winner.

The study

Group project Full write-up Data preparation, literature review, OLS with diagnostics, and the random forest extension. Source Model training The regression output, printed in full.

Data: DVF, the French property transaction register (DGFiP), first half of 2025, published as open data.