# RECORD 9c7b2eaba767769a138e2eee57dcf072ba93d657
# TITLE: From human business to machine learning—methods for automating real estate appraisals and their practical implications
# YEAR: 2022.0  VENUE: Zeitschrift für Immobilienökonomie  DOI: 10.1365/s41056-022-00063-1
# AUTHORS: Moritz Stang; Bastian Krämer; Cathrine Nagl; W. Schäfers
# PAGES: 28  CANDIDATE TABLE PAGES (1-based): [18]
# ABSTRACT: Until recently, in most countries, the use of Automated Valuation Models (AVMs) in the lending process was only allowed for support purposes, and not as the sole value-determining tool. However, this is currently changing, and regulators around the world are actively discussing the approval of AVMs. But the discussion is generally limited to AVMs that are based on already established methods such as an automation of the traditional sales comparison approach or linear regressions. Modern machine learning approaches are almost completely excluded from the debate. Accordingly, this study contributes to the discussion on why AVMs based on machine learning approaches should also be considered. For this purpose, an automation of the sales comparison method by using filters and similarity functions, two hedonic price functions, namely an OLS model and a GAM model, as well as a XGBoost machine learning approach, are applied to a dataset of 1.2 million residential properties across Germany. We find that the machine learning method XGBoost offers the overall best performance regarding the accuracy of estimations. Practical application shows that optimization of the established methods—OLS and GAM—is time-consuming and labor-intensive, and has significant disadvantages when being implemented on a national scale. In addition, our results show that different types of methods perform best in different regions and, thus, regulators should not only focus on one single method, but consider a multitude of them.


===== PAGE 1 (score 3, numeric-paren 0, table-words 0, var-words 3) =====

ORIGINAL PAPER
https://doi.org/10.1365/s41056-022-00063-1
Z Immobilienökonomie (2023) 9:81–108
From human business to machine learning—methods
for automating real estate appraisals and their practical
implications
Moritz Stang
· Bastian Krämer · Cathrine Nagl · Wolfgang Schäfers
Received: 4 December 2021 / Revised: 27 June 2022 / Accepted: 3 July 2022 / Published online: 19 July
2022
© The Author(s) 2022
Abstract Until recently, in most countries, the use of Automated Valuation Models
(AVMs) in the lending process was only allowed for support purposes, and not as
the sole value-determining tool. However, this is currently changing, and regulators
around the world are actively discussing the approval of AVMs. But the discussion
is generally limited to AVMs that are based on already established methods such
as an automation of the traditional sales comparison approach or linear regressions.
Modern machine learning approaches are almost completely excluded from the
debate. Accordingly, this study contributes to the discussion on why AVMs based
on machine learning approaches should also be considered. For this purpose, an
automation of the sales comparison method by using filters and similarity functions,
two hedonic price functions, namely an OLS model and a GAM model, as well
as a XGBoost machine learning approach, are applied to a dataset of 1.2 million
residential properties across Germany. We find that the machine learning method
XGBoost offers the overall best performance regarding the accuracy of estimations.
Practical application shows that optimization of the established methods—OLS and
GAM—is time-consuming and labor-intensive, and has significant disadvantages
when being implemented on a national scale. In addition, our results show that
different types of methods perform best in different regions and, thus, regulators
should not only focus on one single method, but consider a multitude of them.
Keywords Automated Valuation Models · Extreme Gradient Boosting · Housing
Market · Machine Learning · Sales Comparison Method
Moritz Stang () · Bastian Krämer · Cathrine Nagl · Wolfgang Schäfers
International Real Estate Business School, University of Regensburg, Regensburg, Germany
E-Mail: Moritz.Stang@wiwi.uni-regensburg.de
K


===== PAGE 18 (score 9, numeric-paren 0, table-words 2, var-words 3) =====

98
Z Immobilienökonomie (2023) 9:81–108
Table 7
Model prediction errors 2020 throughout Germany
Models
MAPE
MdAPE
PE(10)
PE(20)
R2
EXF
0.2130
0.1624
0.3267
0.5872
0.7735
OLS
0.1736
0.1311
0.3937
0.6940
0.8654
GAM
0.1646
0.1202
0.4273
0.7276
0.8664
XGB
0.1465
0.1084
0.4665
0.7786
0.8995
the performance. The EXF is designed to replicate the practice of traditional real
estate appraisers in an automated manner and is therefore readily understandable.
However, the approach provides the poorest results. Comparing these results with the
performance of the OLS, often used as a baseline model, we can see a performance
improvement. Relatively speaking, the MAPE of the OLS is around 18% lower
and the MdAPE 19%. In addition, using an OLS results in 18% and 20% more
predictions deviating less than 10 and 20% from their actual market value.
Analyzing the results of the GAM, we again see a boost in performance compared
to the OLS. But this time the relative improvement is smaller. The MdAPE of the
OLS is around 9% higher. In addition, the percentage of predictions with a relative
deviation of less than 10 and 20% increased by 9% and 5% respectively. This might
be caused by the ability of the GAM to model more complex non-linearities within
the data, which is extremely difficult to manually reproduce within the OLS, and
practically impossible to implement for 327 districts. This is especially so, since
these manual adaptions have to be done in each of the four quarters.
Overall, the XGBoost yields the best model performance regarding all evaluation
metrics due to its ability to capture and process joint effects, non-linear relationships
and high-dimensional structures within the data with comparably low manual effort.
Comparing the results of the XGBoost with the EXF 43% and 33%, more observation
deviate less than 10 and 20% from their market values.
The chosen extending-window testing approach allows us to further analyze the
performance of all four algorithms over the four quarters of 2020. Confirming the
previous results, the solid line in Fig. 3 shows the trends already mentioned. Addi-
tionally, it is interesting how consistently the models perform over all four quarters.
Moreover, the XGBoost displays better performance the more training data it can
process. The exact numbers can be seen in Appendix III.
One research question of this study is to determine whether modern machine
learning methods are able to outperform traditional hedonic models and the EXF
approach. Analyzing our results at the national level for Germany we can clearly
confirm this. The XGBoost yields a significant performance improvement compared
to the EXF, OLS and GAM. This shows that in the future, regulators should also
discuss the approval of machine learning methods in the field of AVMs. The appli-
cation of machine learning approaches can lead to a reduction in the economic loss
caused by the AVM. Machine learning algorithms are able to better assess possible
risks within the lending process and can thus fulfill the actual purpose of a real
estate valuation in a much more target-oriented manner.
K


===== PAGE 19 (score 2, numeric-paren 0, table-words 0, var-words 2) =====

Z Immobilienökonomie (2023) 9:81–108
99
Fig. 3 Quaterly model error
performance based on MAPE
5.2 Results at the administrative district level
After comparing the models at the national level, we want to examine the model
performance in more detail. Therefore, we focus on the level of the 327 adminis-
trative districts. In Fig. 4, the performance based on the MAPE for the different
methods is shown cartographically. The maps confirm the abovementioned trends.
The EXF again yields the overall poorest performance and again, it can be seen that
the more complex the approach, the better the results. In addition, all four models
are unsatisfactory with respect to estimating the market value in the same admin-
istrative districts. This can also be confirmed by the correlation matrices shown in
Appendix III. Especially in the eastern part of Germany, the MAPE tends to be
higher. This result might be caused by the lower data availability in these regions.
To obtain a better understanding of the model performance at the administrative
district level, we focus on the box plots of the MAPE in Fig. 5. Those confirm
the trend displayed in Fig. 4. The EXF again yields the overall poorest results. It
delivers the largest interquartile range, the longest whiskers and contains the most
outliers. The XGBoost has the lowest median MAPE of all four models, whereas it
has only two extreme outliers. In contrast, the GAM and especially the OLS have
a smaller range of outliers. These results indicate that the XGBoost does not always
Fig. 4
Error comparison at administrative district level
K