The hedonic model applied to the Luxembourg market
The hedonic model is an econometric approach that decomposes a property's price into the sum of implicit values of its characteristics. Developed by Rosen (1974), it rests on the hypothesis that the market price reflects buyers' preferences for each housing attribute.
In Luxembourg, the Housing Observatory (Ministry of Housing / LISER) regularly publishes hedonic models calibrated on actual transactions recorded by the Registration Administration (AED). These models enable construction of quality-constant price indices, neutralising the effect of changes in the composition of properties sold.
Our implementation uses a log-linear model: ln(Price) = β0 + Σ(βi × Xi) + ε. The dependent variable is the log of price, allowing coefficients to be interpreted directly as elasticities (percentage impact). Explanatory variables cover area, location (municipality/neighbourhood), floor, general condition, CPE energy class, parking availability and outdoor amenities.
Coefficients and explanatory variables
Our model coefficients are calibrated from available Luxembourg data:
- Area: elasticity of -0.35% per additional sqm above 80 sqm. This coefficient reflects the declining marginal price per sqm with increasing area — a well-documented phenomenon in property literature.
- Floor: ground floor suffers a discount of approximately 7%, first floor 3%. Upper floors (4+) benefit from a 3% premium, and penthouses up to 8%. Source: Housing Observatory.
- General condition: a new or fully renovated property earns a premium of approximately 7%. A property needing cosmetic work suffers a 5% discount, and one needing full renovation 15%.
- CPE energy class: classes A-B bring a 5% premium, while F-G carry an 8% discount. Spuerkeess has documented this gap in its Luxembourg green premium studies.
- Parking: an indoor parking space adds approximately 5% to the dwelling price (equivalent to EUR 30,000-45,000 depending on location).
Data sources
The reliability of a hedonic model depends directly on the quality and representativeness of underlying data. In Luxembourg, we use several sources:
- Housing Observatory: hedonic price indices published quarterly, based on notarial deeds transmitted by the AED. Comprehensive coverage of the residential market since 2010.
- STATEC: housing price index (IPIL), average prices by municipality, construction statistics.
- Municipal authorities: PAG/PAP data, building permits, cadastral information.
- Property listings: asking prices serve as a supplementary reference but are generally 5 to 10% above actual transaction prices.
Coefficient updates are performed semi-annually to incorporate market developments.
Model limitations: R² and confidence interval
Every statistical model has limitations that must be understood:
- R² (coefficient of determination): our model shows an R² of approximately 0.75, meaning 75% of price variance is explained by the retained variables. The remaining 25% relates to unobserved factors (view quality, natural light, noise, micro-location, common area condition).
- Confidence interval: the point estimate is always accompanied by an interval. In dense urban areas with many transactions, the interval is ±10%. In rural areas with limited data, it can reach ±18%.
- Atypical properties: the model is calibrated on standard properties. Luxury properties, character houses or properties with major defects may deviate significantly from the estimate.
- Temporal effect: coefficients reflect market conditions at calibration time. During rapid market shifts, a lag may appear.