Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 75

Application of a Hedonic Pricing Model for Assessment of Apartments in Tirana,

Albania

Marsela Thanasi (Boe)

University of Tirana, Faculty of Economy
Albania
marsithanasi@yahoo.com

Abstract

Through the results generated by a hedonic pricing model for apartments in Tirana, it was found
that, besides residential area location, there are a number of structural features of the
apartments, which affect their value, as the flat surface related to the number of rooms, view,
the opportunity for parking and furniture. Other features such as residential floor, partial
furniture, the age of the apartment, the presence of more than one toilet, number of balconies,
the presence of central heating and orientation are estimated to have insignificant impact jointly
on the price of apartment. These results were recommended to the government of Albania on
reassessment of legal reference prices of apartments in Tirana. The government uses the results
in addition to the review of the property evaluation methodology in order to increase the
transparency and availability of information on real property value.

Key words: real property, mass appraisal, reference price, hedonic model

Aim of the study

The approved methodology of property valuation in Albania evaluates the reference property
prices based on the average contractual price of properties sold and take into account only
factors such as price, square metres of living and location of property (Thanasi & Hysi, 2013).
Simple method is the one used on evaluating the reference price for group of properties with
similar characteristics. The objective of this study is that, by building a hedonic prices model
for apartments in Tirana, to be proved empirically that in addition to location, there are other
characteristics that affect the value of apartments. These variables should be taken into account
in the valuation model specification. In confirmation of this hypothesis, the capital city is
chosen as a case study for the construction of hedonic pricing model. From the database
registered on Immovable Property Registration Office it is found that apartments sales take 84
 Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 76

percent of property sales transactions in Tirana. Also the complexity of the issues and
apartment characteristics variation in Tirana is high.

Literature review on hedonic pricing model

Hedonic models are displayed for the first time with the assessment of agricultural land by
Haas (1922) and Wallace (1926), but the term "hedonic" was mentioned for the first time by
Court, (1939), to determine the hedonic price index of automobiles. Lancaster (1966) defined
the concept of utility of the products through the value of their special characteristics,
considering that a linear relationship. His study is based on the theory of consumption,
according to which the demand for a heterogeneous product such as real estate, depends on its
features, which create the consumer benefit. Rosen (1974) estimated that properties are
heterogeneous and the total value is built by the sum of each property characteristic value,
estimating that hedonic pricing models should be nonlinear. According to Rosen, the benefits
consist of a set of attributes, which cannot be separated and are selected according to a
combination of consumer preferences.

Specification of the model

Database

In the database were included 1421 apartments and for each of them are collected data as: price
of the property, square metres of living, price per square metres, the number of rooms, ehe
number of balconies, the number of toilets, residential floor, elevator, parking possibility,
furniture, orientation directions and age of the building. In addition to the variables in linear
form, the model included the square of the surfice and number of rooms and an interaction
between them. Data collection has been very careful and had no missing values for any of the
variables under consideration.

Location of the property is found by one of the 32 legal reference areas where the property is
located. These are classiffied into three main groups, which are specified as three independent
variables that identify the location of the apartments. The first group area includes the most
favorable areas located in the center of Tirana, near Lake and National Park of Tirana; The
second group area includes properties that are located within a small circle of Tirana and the
third group area includes properties located outside the small circle of Tirana.
 Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 77

Specification of variables

The analysis of descriptive characteristics of the dependent variable (price) on the completion or not the
assumption of normality confirmed that the price of houses did not have normal distribution, thus it was concluded
that error term had no normal distribution, too. For this reason, we applied what literature review recommended
in similar cases, transform the price variable to its natural logarithm. This solution eliminated a very large amount
of variable deviation from its normal distribution. Semi-logarithmic functional form was found more reliable and
provided better results in the hedonic model, as the literature suggested. The test of Ramsey and Link to the
functional form provided clear evidence that the null hypothesis of well-specified functional form could not be
rejected.

Analyses of model specification

The control for existence of multicollinearity in the model was conducted by analyzing the
correlation between variables (Table 1). Variables that were identified to show a very strong
correlation between them were excluded from the model, and this led to the improvement of
the model. T-test was performed to compare the apartments average price, according to
different variables. Based on the results, we found that variables contained no information in
explaining the differences in price categories, except three variables related with furnishing,
for which the average difference was statistically insignificant at 5 percent, although there was
evidence for apartments average price, related with full furnishing variable to be statistically
significant at 10 percent level. For variables with more than two categories, we conducted one-
way ANOVA analyses. The results showed statistically significant difference between groups.
At the end of this section, we confirmed that the selection of explanatory variables in the model
has been optimized, as they were discriminating about the price and this feature creates high
expectations on explanatory of the model.
 Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 78

Table 1 Coefficients of correlation between independent and dependent variables of the model
price lnprice sqrm rooms balcon0 balcon1 balcon2 bathd Floor floord lift parking aged heat Fully_f Part_f not_f orient pamja zone1 zone2
Rooms 0.682 0.644 0.697 1
0.000 0.000 0.000
balcony0 0.610 0.645 -0.160 -0.212 1
0.000 0.000 0.000 0.000
balcony1 -0.108 -0.157 -0.229 -0.173 -0.290 1
0.000 0.000 0.000 0.000 0.000
balcony2 -0.192 -0.168 0.294 0.252 -0.035 -0.946 1
0.000 0.000 0.000 0.000 0.190 0.000
Bathd 0.237 0.228 0.536 0.448 -0.056 -0.325 0.358 1
0.000 0.000 0.000 0.000 0.036 0.000 0.000
Floor 0.438 0.428 0.103 0.041 -0.061 -0.049 0.072 0.087 1
0.000 0.000 0.000 0.120 0.021 0.066 0.007 0.001
Floord 0.104 0.092 0.069 0.019 -0.054 -0.020 0.039 0.071 0.784 1
0.000 0.001 0.010 0.466 0.041 0.451 0.140 0.007 0.000
Lift 0.063 0.054 0.277 0.181 -0.081 -0.054 0.084 0.238 0.197 0.129 1
0.019 0.042 0.000 0.000 0.002 0.042 0.002 0.000 0.000 0.000
parking 0.151 0.156 0.157 0.130 0.012 -0.066 0.065 0.132 -0.033 -0.047 0.077 1
0.000 0.000 0.000 0.000 0.661 0.013 0.015 0.000 0.215 0.079 0.004
Aged 0.135 0.129 0.278 0.181 -0.082 -0.053 0.083 0.238 0.195 0.124 0.994 0.077 1
0.000 0.000 0.000 0.000 0.002 0.044 0.002 0.000 0.000 0.000 0.000 0.004
Heat 0.154 0.167 0.119 0.047 -0.017 -0.079 0.088 0.130 0.074 0.040 0.043 0.029 0.043 1
0.000 0.000 0.000 0.075 0.519 0.003 0.001 0.000 0.006 0.129 0.103 0.280 0.105
fully_f 0.084 0.092 -0.031 0.000 0.034 -0.054 0.045 0.003 0.039 0.028 0.003 0.013 0.006 0.047 1
0.002 0.001 0.247 0.998 0.200 0.040 0.088 0.925 0.139 0.286 0.925 0.622 0.823 0.074
Part_f 0.048 0.060 0.001 0.018 -0.026 -0.025 0.035 0.024 0.013 -0.019 0.056 0.079 0.055 0.015 -0.125 1
0.069 0.025 0.979 0.490 0.333 0.343 0.187 0.371 0.616 0.481 0.036 0.003 0.037 0.585 0.000
not_f -0.005 0.009 0.028 -0.010 -0.017 0.063 -0.060 -0.015 -0.043 -0.016 -0.032 -0.054 -0.035 -0.051 -0.848 -0.421 1
0.848 0.734 0.296 0.711 0.513 0.017 0.023 0.572 0.105 0.549 0.227 0.041 0.187 0.054 0.000 0.000
Orient -0.041 -0.059 0.111 0.046 -0.031 -0.123 0.139 0.137 0.080 0.016 0.116 0.050 0.116 0.111 -0.020 0.025 0.005 1
0.119 0.026 0.000 0.084 0.248 0.000 0.000 0.000 0.003 0.551 0.000 0.058 0.000 0.000 0.446 0.344 0.848
Pamja 0.062 0.086 -0.005 -0.026 -0.022 0.022 -0.015 0.042 0.069 0.024 0.060 0.022 0.060 0.050 0.013 -0.024 0.001 0.050 1
0.019 0.001 0.856 0.330 0.406 0.416 0.570 0.117 0.010 0.377 0.023 0.411 0.024 0.060 0.622 0.370 0.976 0.058
zone1 0.096 0.080 0.041 0.055 -0.003 0.001 0.001 0.057 0.039 0.032 -0.047 -0.049 -0.043 0.006 -0.006 0.009 0.001 0.025 0.079 1
0.000 0.003 0.123 0.038 0.904 0.984 0.984 0.032 0.145 0.230 0.077 0.068 0.107 0.809 0.813 0.741 0.968 0.346 0.003
zone2 0.369 0.359 0.085 0.127 -0.023 -0.037 0.046 0.063 -0.010 0.000 -0.015 -0.009 -0.016 0.034 0.039 -0.019 -0.025 -0.006 0.034 -0.318 1
0.000 0.000 0.001 0.000 0.384 0.167 0.082 0.018 0.701 0.993 0.583 0.737 0.548 0.204 0.143 0.465 0.343 0.820 0.203 0.000
zone3 0.181 0.249 -0.111 -0.162 0.025 0.035 -0.045 -0.101 -0.017 -0.022 0.047 0.043 0.046 -0.037 -0.033 0.013 0.024 -0.012 -0.088 -0.395 -0.745
0.000 0.000 0.000 0.000 0.353 0.185 0.089 0.000 0.514 0.403 0.076 0.107 0.086 0.161 0.210 0.634 0.373 0.658 0.001 0.000 0.000
 Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 79

Method

Multiple regression analysis was chosen in this model as a method to test hypotheses about the
causal relationship between house value Y and two or more independent variables XS, representing
the characteristics of the property. Population equation parameters 0, 1, 2 ... n., were evaluated
by Ordinary Least Squares (OLS) method. Selection of -standard is 5 percent, taking into account
the significant number of observations and the degree of regression freedom. The procedure
followed is sequential.

Results of the full model evaluation with OLS method showed that the model is statistically
significant, as 70 percent of the logarithm of the price variation is explained by the factors taken
into consideration. F test for the global significance of regression with 18 and 1402 degrees of
freedom was worth almost 183, and the value of the probability to observe such figure for a case
variable with these characteristics was 0.

While performing the test for heteroskedasticity Breuch-Pagan, the null hypothesis of
homoskedasticity was rejected with complete certainty, suggesting that the standard errors of
coefficients obtained from our evaluation were not available. It was decided that heteroskedasticity
correction to be made according to different segments, as a better way than correcting for unknown
forms. For this reason, errors were robust to heteroskedasticity segments according to official
reference areas. Some of the variables changed statistical significance after correction for standard
errors. Square metres of living variable was not significant in the full model, but became important,
to 5 percent, at the second model. Likewise, statistical significance of complete furniture and view
was improved.

Skewness test/Kurtosis for normality of error term brought as a result that distribution of our
assessors were not normal. Due to the large selection of observations, and based on the Central
Limit theorem and the Law of Large Numbers, we did not take any corrective step for the lack of
normality.

Given that one of the reasons why the normality tests cannot give proper results is the presence of
extreme observations, it was checked and tested for the presence of extreme observations in the
model (Figure 1).
 Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 80

1
0
Residuals

-1
-2
-3

15 16 17 18
Fitted values

Figure 1 Residuals versus fitted values

Consequently, we found the presence of four extreme observations, as shown in Table 2. One of
these observations has a residue less than -2, while the other has the estimated value greater than
17.5. The exclusion of these four extreme observations improved the outcome of regression, as the
coefficient of determination increased to 71.15 percent. F test for the global significance of
regression with 18 and 1398 degrees of freedom was 192 (increased from 183 to 192).

Table 2 Results of extreme observations test

dfits lnprice yhat uhat

9. -.7118553 13.21767 15.70104 -2.483365
307. 7.650204 17.7355 17.69501 .0404963
394. 2.608342 17.6912 16.86493 .826271
729. -.6653056 14.64842 15.54236 -.8939373

Variables such as the square of the number of rooms, dummy variables for the number of balconies,
a dummy variable for the presence of more than one toilet, floor apartment, age, presence of central
heating, partial furniture and orientation resulted jointly insignificant, so were excluded from the
model (F (9, 31) = 0:38, Prob> F = 0.9348).

In the final model (Table 3), the functional form was significantly improved. Although the model
removed a number of explanatory variables, its coefficient of determination still remained about
70 percent. F test for the global significance of regression with 9 and 31 degrees of freedom was
rated almost 375.
 Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 81

Table 3 Final hedonic model

Linear regression Number of obs = 1421

F( 9, 31) = 374.66
Prob > F = 0.0000
R-squared = 0.7004
Root MSE = .27797
(Std. Err. adjusted for 32 clusters in zf)
------------------------------------------------------------------------------
| Robust
lnprice | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
sqrm | .0116118 .0017784 6.53 0.000 .0079847 .0152388
sqrmsq | -8.32e-06 1.70e-06 -4.89 0.000 -.0000118 -4.85e-06
rooms | .2634096 .0410389 6.42 0.000 .1797101 .347109
sqrxrms | -.0011195 .0005304 -2.11 0.043 -.0022013 -.0000376
parking | .1254876 .0300995 4.17 0.000 .0640993 .1868758
fully_furn | .0740034 .0211437 3.50 0.001 .0308805 .1171263
pamja | .1140003 .047017 2.42 0.021 .0181084 .2098922
zone2 | -.2822059 .0407601 -6.92 0.000 -.3653368 -.1990751
zone3 | -.6003058 .0655384 -9.16 0.000 -.7339723 -.4666393
_cons | 15.12883 .1090786 138.70 0.000 14.90636 15.3513
Ramsey RESET test using powers of the fitted values of lnprice Prob > F = 0.5767
Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Prob > chi2 = 0.0000
Skewness/Kurtosis tests for Normality: Prob>chi2 = 0.0000

Findings

Estimated regression presented in Table 3 is written in equation form, with robust standard errors,
as follows:

= . + . . + . .

[. ] [. ] [. ] [. ] [. ]

The equation shows that all variables included in the model are statistically significant at the 5
percent level, and with the exception of two of them, the others are all significant at 1 percent.
Although the model removed a number of explanatory variables, the coefficient of determination
is still about 70 percent.

Regarding the effect of the square metres of living on the apartments price, the regression
recommends that this effect depends on the level of the square metres, since the relationship is
 Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 82

nonlinear and depends more on the number of rooms. Referring to the average square metres of
living (about 90 m2), if an apartment has a room, an increase of its surface to 10 square meters
would cause a price increase by about 10 percent, if all other factors are held constant. Marginal
effect for an additional 10 square meters in a house that has two rooms is 7.8 percent while for a
house with 3 rooms this effect sits at 6.7 percent, ceteris paribus. Although the effect decreases
with increasing the number of rooms, as expected, it is statistically significant in all three cases.

In the same logic, the effect of an extra room depends on the surface of the house. For an apartment
with an average surface, the addition of a room within the same space would affect a house price
rise 16.4 percent on average, if all other factors are held constant. Meanwhile, the addition of a
room with a surface of 10 square meters is estimated to increase the price of the apartment with
40.8 percent, when all other factors are held constant.

In the same conditions, an apartment with parking opportunities is estimated 12.6 percent more
than an apartment with the same features that has no parking opportunity; a fully furnished
apartment costs approximately 7.4 percent more than an unfurnished one; a favourite apartment
view is estimated about 11.4 percent more than the apartments that do not have such view. An
apartment located in the second zone is estimated 24.6 percent less than the apartments located in
the first zone, while an apartment located in the third zone is worth 45 percent less than an
apartment in the first zone.

Conclusion

Based on this study, it was proved empirically that the characteristics of the apartments as square
metres of living, number of rooms, access to parking, furniture, view and surface of living affect
their price. As literature review showed, location is the most important variable that affects the
value. Results showed that the marginal effect that the number of rooms has in the apartment value
depends on the living square metres of the apartment. In the same line, the effect of the square
metres of living on value depends on the square metres, since the relationship between these
variables is nonlinear and depends more on the number of rooms.
 Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 83

Limitations of the study

According to researchers, hedonic model is built based on data generated from sales (Gloudemans
and Almy, 2011). Immovable Property Registration Office, Albania does not have a digital
database on sales property contracts with real characteristics of the properties sold (Thanasi, 2012).
It was impossible for us to find the information on sales contracts. That is why we were oriented
towards the market and took into account the properties offered for sale at one of the biggest Real
Estate Agency elsi. In this way, we could get more accurate information regarding properties
characteristics. Of course, this information was limited and depended on the best apartment
charasteristics that potential sellers wanted to advertise, but a more complete information would
have brought more complete results from the model. On the other side, in the context of the study
while contractual sales price is oriented by the reference price (Thanasi & Hysi, 2013), the applied
methodology has resulted in better evaluations of real estate prices, which reflect the market price.

Housing market segments can be modelled and not imposed as official reference zones, bringing
more objective assessment results (Goodma & Thibodeau, 1998). Variables included in the
hedonic model belong to the category of structural variables and the location, while due to lack of
data, the model could not include variables that represent the characteristics of the neighborhood,
as well as other variables that have been identified by the empirical literature to affect the value of
property. The inclusion in the model of other useful data such as distance of the property from
workplaces, schools and shopping centers, the level of crime, quality of education and hospital
services, local tax rate property, information on the days that property remains on the market
without being sold, would help improve the performance of hedonic pricing model (Thanasi,
2014).

Identificition of the reference area where the property was located is carried out manually for each
property and this may be associated with human errors. This problem can be solved through the
use of geografic informations systems that enable clear geographical identification of the property
according to specific neighborhoods within the areas of assessment.

In the future, it would be of high interest that in addition to apartments, similar studies be
expanded, involving other types of properties.
 Journal of Economic Development, Management, IT, Finance and Marketing, 7(1), 75-84, March 2015 84

References

Court, A.T. (1939) Hedonic Price Indices - With Automotive Examples. The Dynamics of
Automobile Demand. New York: General Motors Corporation, 99-117
Gloudemans, R. J. and Almy, R.. (2011) Foundamentals of Mass Appraisal, pq. 180-193.
Goodman, A. C., and Thibodeau, T. T.. (1998) Housing market segmentation. Journal of Housing
Economics, 7, 121143.
Haas, G.C. (1922) Sales Prices as a Basis for Farm Land Appraisal. Technical Bulletin 9, St.
Paul, The University of Minnesota Agricultural Experiment Station.
Lancaster, K. J. (1966) A new approach to consumer theory. Journal of Political Economy, 74(2),
132157. http://dx.doi.org/10.1086/259131
Rosen, S. (1974) Hedonic prices and implicit markets: Product differentiation in pure competition.
Journal of Political Economy, 82 (1), 34-55. http://dx.doi.org/10.1086/260169
Thanasi, Marsela. (2012) Improvement of service-delievery in Immovable Property Registration
Organization in Albania. Paper presented in ABSCR Conference, Venice, Italy.
Thanasi, Marsela. (2014) Hedonic Pricing Model Literature Review. Paper presented at the 2d
Scientific Conference, organized by FEUT, Tirana, Albania
Thanasi, Marsela. (2014) Property Mass Evaluation in Albania. Journal of Marketing and
Management, 5 (2), 1-11, Nov 2014 1 1
Thanasi, Marsela. Hysi, Vjollca. (2013) Impact of the Model of Taxation on Transfering
Properties in Immovable Property Market in Albania. Journal of Finance, Accounting and
Management, 4(2), 1-10.
Wallace, H.A. (1926) Comparative farmland values in Iowa. Journal of Land and Public Utility
Economics, 2:385-392.