Hedonic House Price ModelingBased on Multilevel Structured Additive Regression - Computational Approaches for Urban Environments - page 110

around 1,200 m 2 . It is noticeable that the increasing effect of the plot area for small
properties is more pronounced in the results of the GAMLSS regression. In total,
house prices per sq. m. change by about 480 Euro (mean regression) to 690 Euro
(GAMLSS regression) over the domain of the plot area.
The effect of the age of the building, shown in panel (c), can be considered as
the rate of depreciation of single-family homes. Thus, the initial increase up to an
age of 7 years in the results of the mean regression seems quite unlikely, whereas
the more or less linear depreciation (until an age of about 55 years) in the GAMLSS
regression is in line with our expectations. In both models, the effect stays constant
or even reverses for old buildings. The age of the house covers a range of 560 Euro
(mean regression) to 690 Euro (GAMLSS regression).
The effect of the time index (panel (d)) shows the quality-controlled development
of house prices over time. After a moderate increase from 1997 to 2000, prices
almost stay constant until 2003 and rise afterwards until 2008. In the last year of the
observation period, prices obviously decrease, indicating the effect of the economic
crisis of 2008/2009. In total, the time index accounts for variation in a range of
around 350 Euro.

5.7.1.2 Neighborhood Covariates

In Fig. 5.2 , a selection of the neighborhood effects are displayed, again on the natural
scale of prices per sq. m. In the upper row, the effect of the share of academics
( ln_educ ) is shown in panel (a). Although it enters the equation logarithmically
(see Sect. 5.3.1 ), it is displayed in natural values. The effect is clearly positive, with
a pronounced increase starting at a share of approximately 25 %. The difference
between the mean regression and the GAMLSS regression is negligible. In total, the
share of academics accounts for a variation of up to 1,000 Euro.
The effect of the age index ( age_ind , displayed in panel (b)) is more or less linear
for both the mean regression and the GAMLSS regression. The negative direction
of this effect could be interpreted as a decreasing attractiveness of municipalities
that exhibit an excess of age, which could be expected from our considerations in
Sect. 5.2.2 . The effect of the age index has a bandwidth of up to 535 Euro.
The effect of the population density ln_dens , displayed in natural values in panel
(c), shows a tendency toward higher house prices in more densely populated areas.
Here, we can find a considerable difference of up to 260 Euro between the results
of the mean regression and those of the GAMLSS regression for highly densely
populated areas. In all, this effect accounts for a variation between 630 Euro (mean
regression) and 930 (GAMLSS regression) per sq. m.
Finally, the effect of the house price index wko_ind (the only covariate on the
district level) is shown in panel (d). As expected, this effect is increasing, although
for index values of more than 140, it becomes lightly weaker in the GAMLSS
regression and clearly weaker in the mean regression. Prices per sq. m. increase
by 470 Euro (mean regression) to 560 Euro (GAMLSS regression) from the lowest
to the highest house price index.