# RECORD doi_10_1002_isaf_287
# TITLE: Hedonic price models and indices based on boosting applied to the Dutch housing market
# YEAR: 2007.0  VENUE: Intelligent Systems in Accounting, Finance and Management  DOI: 10.1002/isaf.287
# AUTHORS: Martijn Kagie; Michiel Van Wezel
# PAGES: 45  CANDIDATE TABLE PAGES (1-based): [11, 14, 15, 16, 25]
# ABSTRACT: Abstract We create a hedonic price model for house prices for six geographical submarkets in the Netherlands. Our model is based on a recent data‐mining technique called boosting. Boosting is an ensemble technique that combines multiple models, in our case decision trees, into a combined prediction. Boosting enables capturing of complex nonlinear relationships and interaction effects between input variables. We report mean relative errors and mean absolute error for all regions and compare our models with a standard linear regression approach. Our model improves prediction performance by up to 39% compared with linear regression and by up to 20% compared with a log‐linear regression model. Next, we interpret the boosted models: we determine the most influential characteristics and graphically depict the relationship between the most important input variables and the house price. We find the size of the house to be the most important input for all but one region, and find some interesting nonlinear relationships between inputs and price. Finally, we construct hedonic price indices and compare these with the mean and median index and find that these indices differ notably in the urban regions of Amsterdam and Rotterdam. Copyright © 2008 John Wiley & Sons, Ltd.


===== PAGE 1 (score 2, numeric-paren 0, table-words 0, var-words 2) =====

Hedonic Price Models and Indices based on Boosting applied
to the Dutch Housing Market
Martijn Kagie∗
Michiel van Wezel
Econometric Institute, Erasmus School of Economics and Business Economics
Erasmus University
P.O. Box 1738, 3000 DR, Rotterdam, The Netherlands
Econometric Institute Report EI 2006-17
April 5, 2006
Abstract
We create a hedonic price model for house prices for six geographical submarkets in the
Netherlands. Our model is based on a recent data mining technique called boosting. Boosting
is an ensemble technique that combines multiple models, in our case decision trees, into
a combined prediction. Boosting enables capturing of complex nonlinear relationships and
interaction effects between input variables.
We report mean relative errors and mean absolute error for all regions and compare our
models with a standard linear regression approach. Our model improves prediction perfor-
mance with up to 40% compared with Linear Regression. Next, we interpret the boosted
models: we determine the most influential characteristics and graphically depict the relation-
ship between the most important input variables and the house price. We find the size of
the house to be the most important input for all but one region, and find some interesting
nonlinear relationships between inputs and price.
Finally, we construct hedonic price indices and compare these to the mean and median
index and find that these indices differ notably in the urban regions of Amsterdam and
Rotterdam.
1
Introduction
Hedonic pricing theory hypothesizes that the price p of a product is determined by a function
p = F ∗(x), where x is a bundle of characteristics that define the product. Hedonic pricing theory
is generally attributed to Court (1939), Lancaster (1966), Griliches (1971b, 1971a) and Rosen
(1974).
In practice the hedonic function F ∗(x) is estimated by a model F(x) which is fitted on a
historical dataset {pi, xi}N
1 . Traditionally, these models are Linear Regression or Box-Cox type
models.
In the context of housing and real estate hedonic models are useful in three ways. In the first
place a hedonic model is a very suitable way to predict house prices. House price prediction can
be used for bulk appraisal for property tax, but can also help real estate brokers by determining
the asking price for a house. However, it is also possible to use a hedonic method for a website
∗We thank Jaap Darwinkel and Dree Op ’t Veld for their helpful comments and suggestions. Corresponding
author: E-mail: kagie@few.eur.nl. Phone: +31 10 4088940. Fax: +31 10 4089031.
1


===== PAGE 11 (score 22, numeric-paren 0, table-words 3, var-words 12) =====

groningen
apeldoorn
eindhoven
amsterdam
rotterdam
zeeland
Size Dataset
3387
2402
3730
8490
7510
2216
NVM Market Share (2003)
77%
91%
51%
76%
56%
33%
Housing Stock
99034
76330
168342
430733
418467
157377
Percentage Houses in Dataset
3.42%
3.15%
2.22%
1.97%
1.79%
1.41%
Total Area Size (m2)
288.10
498.38
308.98
303.33
448.83
2689.38
Size Land (m2)
272.45
496.17
304.20
239.02
337.75
1645.30
Percentage Land
94.57%
99.56%
98.45%
78.80%
75.25%
61.18%
Population Size
217593
189309
392773
855094
899781
349683
Population Density
799
382
1291
3577
2664
213
Apartments
45.73%
22.31%
24.21%
79.41%
61.58%
10.47%
Terraced
34.96%
35.64%
72.65%
18.20%
31.80%
48.29%
Detached
8.92%
25.60%
9.92%
0.94%
2.30%
25.36%
Semi Detached
10.39%
16.44%
17.35%
1.45%
4.31%
15.88%
Table 2: Some characteristics of the 6 regions.
3.1
Regions
Figure 4 shows where the regions are located in the Netherlands and Table 2 summaries some
characteristics of the 6 regions. A brief description of the six regions is given below1:
City of Groningen and environs: The city of Groningen is located in the province Groningen
in the north east of the Netherlands. The region contains the municipalities of Groningen
(179185 inhabitants, 76 km2), Haren (19048, 46 km2), Eelde (6920, 13 km2), Norg (7160,
110 km2) and Peize (5280, 28 km2). Groningen is the largest city in the north west of the
Netherlands and the capital of the province with the same name. It is surrounded by small
rural villages. With 77% the NVM has a relatively high market share in this region.
Apeldoorn and environs: Apeldoorn is located in the province Gelderland in the east of the
Netherlands. The region contains the municipalities of Apeldoorn (156000 inhabitants, 340
km2) and Epe (33309, 156 km2). Like Groningen Apeldoorn is mainly surrounded by small
villages. This region is interesting, because all the house types are almost equally present in
this region and the NVM has a very high market share of 91% in this region.
Eindhoven and environs: Eindhoven is located in the province Noord-Brabant in the south east
of the Netherlands. The region contains the municipalities of Eindhoven (207870 inhabitants,
88 km2), Nuenen (23367, 34 km2), Son en Bruegel (15070, 26 km2), Valkenswaard (31091,
55 km2), Veldhoven (42545, 32 km2), Waalre (16502, 22 km2) and Geldrop (27670, 13 km2).
Eindhoven is the largest city in the south east of the Netherlands and has became this mainly,
because of the factories and head quarters of DAF Trucks and Philips, that were located in
this city. Eindhoven itself is a typical Dutch city with a large amount of terraced houses.
The NVM has a market share of 51% in this region.
Amsterdam: Amsterdam is located in the province Noord-Holland in the west of the Nether-
lands. The region contains the municipalities of Amsterdam (739104 inhabitants, 161 km2),
Amstelveen (78886, 44 km2), Diemen (24049, 12 km2) and Ouder-Amstel (13055, 24 km2).
Amsterdam is the largest city in the Netherlands and the capital of the country. The city
is situated on the borders of the river IJ. Amstelveen, Diemen and Duivendrecht (Ouder-
Amstel) are suburbs of Amsterdam. The NVM has a market share of 76% in this region.
Rotterdam: Rotterdam is located in the province Zuid-Holland in the west of the Netherlands.
The region contains the municipalities of Rotterdam (598923 inhabitants, 206 km2), Capelle
aan den IJssel (65354, 14 km2), Krimpen aan den IJssel (29046, 8 km2),Nieuwekerk aan
den IJssel (22334,17 km2), Vlaardingen (74058,24 km2), Schiedam (75619, 18 km2), Al-
brandswaard (19607, 22 km2) and Nederlek (14831, 28 km2). Rotterdam, the second largest
city in the Netherlands, is one of the largest sea ports in the world and located in delta of
1All statistics that are used in the descriptions and Table 2 are derived from the online StatLine database of the
CBS (de Jonge, Reedijk, and Sluis 2005), except the NVM market shares.
11


===== PAGE 12 (score 16, numeric-paren 0, table-words 3, var-words 10) =====

the rivers Rhine, Maas, Lek and IJssel and near the North Sea coast. The market share of
the NVM real estate brokers in this region is 56%.
Zeeland: The region Zeeland contains almost the complete province of Zeeland, located in the
south west of the Netherlands. The region contains the municipalities of Borsele (22318
inhabitants, 142 km2), Goes (36591 inhabitants, 93 km2), Hulst (27887, 201 km2), Kapelle
(11627, 37 km2), Middelburg (46350, 49 km2), Reimerswaal (20996, 102 km2), Terneuzen
(55412, 251 km2), Veere (22130, 133 km2), Vlissingen (45236, 34 km2), Schouwen-Duiveland
(34416, 231 km2), Noord-Beveland (7124, 86 km2) and Sluis (24596, 280 km2). Zeeland exist
of a number of islands and peninsulas (Schouwen-Duivenland, Noord- and Zuid-Beveland,
Walcheren and Zeeuws-Vlaanderen) and because of that 40% of the total area of Zeeland is
water. Zeeland is one of the less dense populated regions in the Netherlands and also the
NVM has a relative small market share (33%) in this region. Besides the large size of this
region these two reasons would make that Zeeland is expected to be the region that is the
hardest to predict.
3.2
Variables
For these 6 regions all transactions of existing houses (no newly built houses) that were sold by
NVM brokers in the year 2004 are used to create the datasets. The transaction price in euros
is the target variable or response. There are 83 explanatory variables, of which 12 variables are
continuous, 12 variables are categorical and the other 59 are binary. Both the continuous and
the categorical variables have some missing values. Not all categorical variables are compulsory
to fill in by the broker who sold the house and continuous variables contain some odd values like
9999 and 0’s for number of total rooms and 1’s for house sizes. These odd values are replaced by
missing values. Binary values measure whether something is present or not, so missing values are
replaced by 0 in the dataset. A description of the variables for the 6 regions is given in Table 9
up to 14 in the Appendix. The floor specific variables (except rooms per floor) are summarized in
Table 15 for all regions together. When a variable has the same value for all houses in a dataset
the variable is omitted in the table. For the districts only the 5 biggest districts are given in the
tables, but all districts are used in the computations.
3.3
Demographic Data
According to many real estate brokers location is one of the most important characteristics that
determine the house price. In the dataset location is included in the terms of a district code. In
the different regions 3 levels of geographic positioning can be used: On municipality, district and
neighborhood level. Of course we want to include a positioning at the lowest possible level. How-
ever, the disadvantage of the use of neighborhood coding, is its number of possibilities. Because
there are many different neighborhoods and some of them are very small, there are neighborhoods
where during the time of one year only one or two houses are sold and often even none. The use
of neighborhoods coding will thus definitely lead to classification problems (because one tries to
estimate the price of a house positioned in a neighborhood unknown to the model) and overfitting
of the dataset (because many neighborhoods are represented in the dataset by just a couple of
houses). On the other hand the use of district coding has the disadvantage that, especially in
big cities as Amsterdam and Rotterdam, the defined districts are quite large, with up to 80, 000
inhabitants and a surface of up to 4, 000 hectares.
Due to these considerations we use another way represent location, namely with the use of
demographic data. Statistics Netherlands (Centraal Bureau voor Statistiek, CBS) publishes de-
mographic information for all the municipalities, districts and neighborhoods in the Netherlands.
This information is free accessible on the internet via StatLine (de Jonge, Reedijk, and Sluis 2005).
The neighborhood codes of the houses available in the dataset are matched with the CBS neigh-
borhood information. When a characteristic is missing for some neighborhood, this characteristic
is replaced by the value for the district the neighborhood belongs to. When this information is
12


===== PAGE 14 (score 18, numeric-paren 0, table-words 5, var-words 3) =====

Region
Best Model
MAE
Improvement
MRE
Improvement
groningen
LSBoost (5)
€20110.40
26.47%
11.32%
32.01%
apeldoorn
LSBoost (5)
€26643.74
24.38%
9.31%
32.39%
eindhoven
LSBoost (5)
€21856.95
26.71%
8.06%
31.98%
amsterdam
LADBoost (5)
€29320.47
32.84%
9.38%
44.00%
rotterdam
LADBoost (5)
€22637.88
34.60%
10.91%
40.15%
zeeland
LSBoost (5)
€28914.29
31.16%
14.83%
38.03%
Table 3: Out-of-sample performance of the best models per region. Improvement is the improve-
ment with respect to the Linear Regression model. The number between the parentheses represents
the tree depth of the base learner. See text for more comments.
MAE
MRE
Region
LSBoost
LADBoost
Impr. LADB
LSBoost
LADBoost
Impr. LADB
groningen
€20110.40
€20141.72
-0.16%
11.32%
11.62%
-2.65%
apeldoorn
€26643.74
€27824.49
-4.43%
9.31%
9.86%
-5.91%
eindhoven
€21856.95
€21822.29
0.16%
8.06%
8.09%
-0.37%
amsterdam
€29283.76
€29320.47
-0.13%
9.55%
9.38%
1.78%
rotterdam
€22067.10
€22637.88
-2.59%
10.96%
10.91%
0.46%
Table 4: Comparison of the results of LSBoost and LADBoost
of the optimal parameters is a time-consuming task that we considered to be beyond the scope of
our research.
4.2
Experiment 1: Data with Districtcoding
This subsection gives the results of the experiments using districtcoding but without demographic
variables. In Table 3 a summary of the best results is given. A complete overview of all results
can be found in Appendix D, Table 17.
On each dataset 10 boosting experiments are performed (5 LSBoost and 5 LADBoost, with
depths 1 up to 5), except on the Amsterdam and Rotterdam regions, where some less interesting
experiments were skipped, because of their long computation times. All these boosting experiments
outperformed the LR model on their dataset. Adding an extra layer always reduced both absolute
and relative error, but the effect of every extra layer is less than the previous one. The improvement
of a Boosting model with depth 5 with respect to a depth 4 model in terms of absolute error is
around €200. It is therefore expected that the effect of adding a 6th layer is negligible. There
are two possible reasons why extra layers increase the prediction performance. In the first place
because extra layers allow for extra interaction and in the second place, since the number of models
is constant, extra layers just can make more splits on the data and therefore predict more accurate.
The boosting models with depth 5 reduce the absolute error with around 25% – 35% and
relative error with around 30% – 45% with respect to the LR model.
But which of the two
boosting algorithms, LSBoost and LADBoost, performed best differed by dataset.
When we
look at the relative error, LADBoost is the best algorithm in the large cities of Amsterdam and
Rotterdam, where LSBoost performed best in the other 4 regions. However, the difference in
performance between the two algorithms is very small. Only in Apeldoorn and environs LSBoost
is really better than LADBoost. Since LADBoost tries to ignore outliers, it is possible that the
worse performance in Apeldoorn is explained by the bad prediction of expensive houses. In Table
4 a comparison is made between the results of both the boosting algorithms.
4.2.1
Submarkets
It is possible that prediction accuracy is very different among different types of houses or market
segments. In real estate literature market segments are often called submarkets. We analyze the
predictive performance of our models after a very simple market segmentation, in which each of
the six possible house types (detached, apartments etc.) represents a segment.
The prediction error (MAE or MRE) of a submarket is the weighted mean test error of that
submarket, i.e. in every test set the houses belonging to the submarket are collected and the
14


===== PAGE 15 (score 19, numeric-paren 0, table-words 5, var-words 7) =====

Region
Best
Size
LSBoost
LADBoost
Worst
Size
LSBoost
LADBoost
groningen
mid-terraced
901
9.67%
9.24%
detached
302
24.30%
24.98%
apeldoorn
mid-terraced
548
5.08%
5.15%
detached
615
14.39%
14.82%
eindhoven
apartment
903
6.92%
6.93%
detached
370
13.11%
12.26%
amsterdam
apartment
6742
9.06%
8.92%
detached
80
23.68%
21.90%
rotterdam
stepped
61
8.00%
8.04%
detached
173
19.66%
18.57%
zeeland
stepped
34
9.58%
9.83%
detached
562
20.29%
20.82%
Table 5: Best and worst predicted house type per region
mean is taken and multiplied by the number of houses belonging to the submarket in that test
set. Afterward the sum of all these means is taken and divided by the sum of the house counts.
Table 5 gives the best and worst predicted house types per region. More detailed results can
be found in Appendix D, Table 18.
The terraced houses, especially the mid-terraced house, are predicted relatively well in all
regions. There are many terraced houses and when looking at the variables included, we may
conclude that these datasets are especially created for terraced houses (and other single house-
hold residences). Many of the variables are uninteresting for apartments (gardens, floors etc.).
Apartments are also predicted well, except in Zeeland. Detached houses seem to be the biggest
problem, also when there are relative many houses of this type in the region, e.g. in Apeldoorn
and Zeeland. Even then detached houses are quite complex to predict. Detached houses are often
quite unique and besides the included variables also the architectonic style of the house may have a
high influence. Also by the other house types we see when a house becomes more exotic prediction
performance decreases. There are two approaches that may improve the prediction of these exotic
houses, but these approaches create also new problems. Both approaches are based on extending
the datasets. First we can include houses sold in other years in the dataset, but then we have
to correct the prices to compensate for market dynamics. The second approach is to combine
multiple regions. Disadvantage is we get an enormous number of districts, but maybe the use of
demographic data (and preferable other environmental variables) can replace the districts.
It was expected that LADBoost would predict ordinary houses better and LSBoost reached
better results on the larger dwellings.
The submarket results don’t confirm this expectation.
Sometimes LADBoost predicts detached dwellings even better than LSBoost. Overall there does
not seem to be a pattern in which algorithm predicts best on the different house types.
4.3
Experiment 2: Data with Demographic Variables
In Section 3.3 datasets were created where the district codes were replaced by a number of demo-
graphic variables on neighborhood level. The LSBoost and LADBoost, with depth 5, were also
trained on these datasets. In Table 6 the best models per region are listed. A complete overview
of the results is given in Table 19 in the Appendix.
The results of all the boosting models in all 6 regions are better on these new datasets compared
with the results on the old ones, discussed in Section 4.2. The size of the performance increase
differs among regions and models. Overall, the adaptation of the datasets has more influence
on the LSBoost models than on the LADBoost models. In the region of Amsterdam prediction
performance of the models has increased by about 10%, but in the region of Rotterdam this is
only 3% to 4%. It does not come as a surprise that Amsterdam is the region where performance
gain is highest, since the districts in Amsterdam are extremely large. With the use of demographic
information the model can discover a better separation between different locations.
We can conclude that LSBoost or LADBoost models with depth 5 and built with the use of
demographic data are the best choices among the models evaluated so far. These models improve
the Mean Relative Error with 35% up to 49% compared with the original Linear Regression model.
15


===== PAGE 16 (score 13, numeric-paren 0, table-words 2, var-words 7) =====

Region
Best Model
MAE
Improvement
MRE
Improvement
groningen
LSBoost (5)
€18892.04
6.06%
10.41%
8.04%
apeldoorn
LSBoost (5)
€25497.43
4.30%
8.64%
7.20%
eindhoven
LSBoost (5)
€20748.22
5.07%
7.71%
4.34%
amsterdam
LADBoost (5)
€26373.36
10.05%
8.51%
9.28%
rotterdam
LADBoost (5)
€21936.93
3.10%
10.42%
4.49%
zeeland
LSBoost (5)
€27189.76
5.96%
13.90%
6.27%
Table 6: Results of the best models per region with the use of demographic data. Improvement is
the improvement with respect to the model trained on the normal dataset.
5
Interpretation
Parametric techniques often have the advantage that a useful interpretation can be given to the
model parameters, e.g., in linear regression the model parameters can be interpreted as the partial
prices of the product characteristics. Although not parametric, regression trees are also highly
interpretable and can be written as an equivalent set of if-then rules. Boosted trees lack both
these appealing properties. Not all is lost, however, because two useful tools exist that can be
used for the interpretation of these ensemble models:
Relative importance plots, that visualize how important the various independent variables are
relative to one another in predicting the dependent variable. In regression trees, the relative
importance of a variable is measured examining the effect of each split on that variable on the
model outcome. Roughly speaking, this effect is high if the split results in a large difference
in model outcome for the right and the left subtree (with preferably an equal probability to
turn right or left) and/or the split is likely to occur. (The probability that a split occurs
is indicated by the number of patterns that travel through the corresponding node relative
to the number of patterns in the dataset. Generally this probability is higher for nodes
close to the root.) In an ensemble of regression trees such as built by boosting, the relative
importances are simply averaged over all trees in the ensemble.
Partial dependence plots, that visualize the partial dependence of the implemented function
on a subset of the independent variables. For a grid of values for this variable subset, the
‘expected’ output of the model at that point is computed. Ideally, this expectation at a grid
point should be computed with respect to the conditional distribution (given the grid point)
of the variables not in the subset. In practice it is obtained by averaging the model outputs
over all instances in the dataset, keeping the values in the selected variable subset fixed to
the grid point, thus using the dataset to approximate the distribution.
A more detailed description of these plot types is given in Appendix B.
Both relative importance plots and partial dependence plots are made for the six datasets
using an LSBoost model. An LSBoost model, with 300 CART base learners with depth 5 and
learningrate 0.1, is trained on the complete 2004 dataset of a region. First we will use the datasets
with district coding, but in Subsection 5.2 we will also create relative importance plots with the
inclusion of demographic variables and partial dependence plots of some demographic variables.
5.1
Interpretation: Data with Districtcoding
Figure 5 shows the relative importance plots of the 6 regions. Only the 10 most important variables
in a dataset are plotted. Except in Zeeland the most important variable in all regions is a variable
that measures some size: volume, size or lot size. In Zeeland kind is the most important variable,
but kind measures house size implicit. In general there are 6 variables that are in the top tens for
all 6 datasets: volume, size, lot size, kind, construction year and district. It is remarkable that,
although the size of the house is very important in the hedonic model, the number of rooms in a
house hardly has an influence. The total size of these rooms is far more important.
We created partial dependence plots for 5 of the 6 most important variables mentioned in the
previous section. Only for district no plots are created, because of the large number of different
values this variable can have.
16


===== PAGE 17 (score 5, numeric-paren 0, table-words 0, var-words 5) =====

volume
kind
size
district
lot size
constr year
type
rooms GF
garage
garden size
Groningen
0
20
40
60
80
100
lot size
kind
volume
size
district
type
constr year
garage 2+
maint inside
garden maint
Apeldoorn
0
20
40
60
80
100
volume
lot size
type
kind
district
c year
size
maint out
bath 2F
blcny 1F
Eindhoven
0
20
40
60
80
100
size
volume
district
kind
lot size
maint inside
constr year
garden size
garden E−W
heating
Amsterdam
0
20
40
60
80
100
volume
size
lot size
district
kind
constr year
rooms
maint inside
type
garage
Rotterdam
0
20
40
60
80
100
kind
district
volume
lot size
c year
size
type
maint in
garage
heating
Zeeland
0
20
40
60
80
100
Figure 5: Relative Importance of predictors with Districtcoding.
17


===== PAGE 25 (score 16, numeric-paren 0, table-words 3, var-words 5) =====

Lot Size
Volume
Apartments
Region
1985
2004
%
1985
2004
%
1985
2004
%
Groningen
587.39
240.50
-59.06%
356.21
327.93
-7.94%
49.12%
46.06%
-6.22%
Apeldoorn
864.85
654.82
-24.29%
384.59
379.00
-1.45%
13.33%
22.04%
65.32%
Eindhoven
335.85
254.53
-24.21%
425.88
389.39
-8.57%
9.25%
23.48%
153.80%
Amsterdam
170.72
108.14
-36.65%
337.70
269.60
-20.17%
71.24%
79.26%
11.26%
Rotterdam
292.01
138.13
-52.70%
420.17
305.51
-27.29%
49.54%
61.46%
24.06%
Zeeland
675.99
467.29
-30.87%
428.07
377.89
-11.72%
16.20%
10.93%
-32.57%
Table 8: Differences of variable values between 1985 and 2004
index.
In all 6 regions the same pattern is visible only the absolute effects differ. Prices raises in the
first 10 years with 40% in Zeeland up to 90% in Amsterdam, according to the mean and median
indices. The Fisher hedonic shows raises of 58% in Zeeland up to 106% in Amsterdam, which
means that the price raise was accompanied by a quality decline.
In the next 5 years, 1995 up to 2000 these effects became even stronger. In 2000 houses were
almost 4 times so expensive in Amsterdam than in 1985. Corrected for quality this is even 5 times.
Although both effects are the strongest in Amsterdam this pattern is also visible in the other 5
regions. The quality decline is the largest in big cities, such as Amsterdam and Rotterdam, but
also in Eindhoven. In the 3 more rural regions the quality decline is less impressive. Due to
these differences in quality decline, mean and median indices give a wrong notion about in which
regions prices have increased the most. There is no discussion about the fact that Amsterdam is
the region with the highest price increase and Zeeland the region with the lowest. But for the
other 4 regions the mean and median indices give a deceptive view. The most striking example
is Rotterdam. According to the mean and median indices this region is positioned fith out of the
six regions in terms of price increase, but according to the hedonic indices it is positioned second,
since the quality decline of the dwellings in Rotterdam is relatively high.
Table 8 shows some means of important variables in the years of 1985 and 2004. When these
important characteristics have become of less quality (i.e. the average house volume has become
smaller, the average lot size has decreased or more dwellings are apartments) this would be a
decent explanation of the quality decline the hedonic models found between 1985 and 2004. Both
lot size and volume have decreased during these 20 years in all of the 6 regions. Lot size with 24%
in Eindhoven up to 59% in Groningen and volume with 1% in Apeldoorn up to 27% in Rotterdam.
Also the relative amount of apartments has increased in 4 of the 6 regions. The regions with a large
decline in the average house volume are also the 2 regions with a large quality decline according
to the hedonic price index: Amsterdam and Rotterdam. The quality decline during the 90’s can
therefore mainly be explained by the fact that houses have become smaller during this period.
These hedonic indices give only a general overview of all possible dwellings. Since we can see
in Table 8 that in some regions one reason of the decline in quality is the increase of the portion of
apartments, it is possible that the average qualities of the different house types did not decrease.
When one wants to check whether this is the case one could create hedonic indices per house type.
Although the hedonic indices give a good impression of the quality differences during a period,
they are not 100% reliable. There are three reasons for this.
The first reason is a theoretic one and discussed in the previous section. Both the Paasche and
Laspeyres index do not account for substitution effects. The Fisher’s Ideal index only estimates
these effect by using the geometric mean of the other two indices. However, the Laspeyres and
Paasche index are situated very close to each other and thus the Fisher’s index seems to be a good
estimation.
In the second place the boosting models make prediction errors. Since the data quality and
quantity become less going further back in history, this will be a problem. Especially for the
Paasche-like index, because this approach uses only the base period model.
The third reason has to do with the meaning of the variables. There are two variables which
meanings change over time. The first is construction year. This variable measures in fact two
things: the age of the house and the period it is built in, e.g. the architectonic style and average
quality of houses built in that period. This means a construction year of 1980 has a different
25


===== PAGE 26 (score 0, numeric-paren 0, table-words 0, var-words 0) =====

1985
1995
100
200
300
400
500
Groningen
year
index
1985
1995
100
200
300
400
500
Apeldoorn
year
index
1985
1995
100
200
300
400
500
Eindhoven
year
index
1985
1995
100
200
300
400
500
Amsterdam
year
index
1985
1995
100
200
300
400
500
Rotterdam
year
index
1985
1995
100
200
300
400
500
Zeeland
year
index
Figure 10: Hedonic and normal indices of the six regions: Fisher’s Hedonic Index (solid), Mean
Index (dashed) and Median Index (dotted).
26