# RECORD 38bde6e6f3fb947d8b6029d64eef6268c93759cc # TITLE: Marginal prices of improvements made to blocks of flats: empirical evidence from Romania # YEAR: 2016.0 VENUE: nan DOI: 10.3846/1648715x.2015.1121415 # AUTHORS: Adela Deaconu; D. Lazar; Anuța Buiga; Gheorghe Fatacean # PAGES: 16 CANDIDATE TABLE PAGES (1-based): [8, 10, 12] # ABSTRACT: This paper examines a specific segment of the housing markets from Central and Eastern European countries, namely the stock of flats in the urban area. With focus on Romania, we emphasize the improvements made by owners and authorities, aimed to enhance the quality of life and the maintenance of old blocks of flats; are highlighted some aspects regarding the cost of these improvements. In addition, the empirical section investigates whether and to what extent the improvements made to the flats and buildings have a significant effect on sale prices and rental prices. The marginal prices of these attributes are provided by hedonic pricing models, estimated in a generalized linear model framework; gamma and log-linear regression models perform in a similar way in terms of fit and prediction accuracy. The physical condition of the building and finishing works inside the flats, as variables of interest in our models, have a significant effect on sale and rental prices, but with different intensity. A common set of significant attributes, in the hedonic models for both sale and rental price, is suggested. ===== PAGE 1 (score 8, numeric-paren 0, table-words 0, var-words 8) ===== Copyright © 2016 Vilnius Gediminas Technical University (VGTU) Press http://www.tandfonline.com/tspm * Corresponding author. E-mail: dorina.lazar@econ.ubbcluj.ro MARGINAL PRICES OF IMPROVEMENTS MADE TO BLOCKS OF FLATS: EMPIRICAL EVIDENCE FROM ROMANIA Adela DEACONU a, Dorina LAZAR b,*, Anuta BUIGA b, Gheorghe FATACEAN a a Department of Accounting and Audit, Faculty of Economics and Business Administration, Babes- Bolyai University, Cluj-Napoca, Romania b Department of Statistics, Forecasting, Mathematics, Faculty of Economics and Business Administra- tion, Babes-Bolyai University, Teodor Mihali 58-60, 400591 Cluj-Napoca, Romania Received 7 April 2014; accepted 3 July 2015 Abstract. This paper examines a specific segment of the housing markets from Central and Eastern European countries, namely the stock of flats in the urban area. With focus on Romania, we empha- size the improvements made by owners and authorities, aimed to enhance the quality of life and the maintenance of old blocks of flats; are highlighted some aspects regarding the cost of these improve- ments. In addition, the empirical section investigates whether and to what extent the improvements made to the flats and buildings have a significant effect on sale prices and rental prices. The marginal prices of these attributes are provided by hedonic pricing models, estimated in a generalized linear model framework; gamma and log-linear regression models perform in a similar way in terms of fit and prediction accuracy. The physical condition of the building and finishing works inside the flats, as variables of interest in our models, have a significant effect on sale and rental prices, but with differ- ent intensity. A common set of significant attributes, in the hedonic models for both sale and rental price, is suggested. Keywords: Hedonic model; Flat; Sale price; Rental price; Improvements attributes; Generalized linear model 1. Introduction Residential property accounts for a significant amount of homeowners’ wealth, representing an important asset for most of them. In Europe, the value of properties ranges from 0.4 times the GDP in Poland to 1.78 times in France, as reported by Colleta (2013). On the other hand, there is a strong correlation between the housing market, in par- ticular the real estate prices index, and macroeco- nomics (Hepsen and Vatansever 2012). Goodhart and Hofmann (2008) highlight the interdepend- ence between housing cost dynamics and economic activity, bank credit, inflation, and the implica- tions of house price fluctuations for monetary and public policies. Following the pattern of other European econo- mies in the last decade, the rapid development of real estate markets in Central and Eastern Euro- pean (CEE) countries was accompanied by an up- ward trend in the housing prices, and the expan- sion of bank loans. Besides fundamentals, these movements in housing prices are related to fac- tors specific to the transition towards a functioning market economy. Huynh-Olesen et al. (2013) found that the movements in residential property prices between 1999 and 2011 were driven by economic fundamentals such as disposable income and inter- est rates, transition-specific factors like fast credit growth and funding through remittances, and by construction costs. In the context of real estate market growth, an accurate valuation of residential properties be- came more critical, taking account of some mar- ket specific features, such as: heterogeneity of properties, highly segmented market, infrequent trading of properties, depreciation and renovation of housing over time, durability, and high trans- action costs. An accurate estimate of the value of a given property avoids arbitrary measurements and is relevant for valuation companies, valua- tors and public authorities. This is critical even International Journal of Strategic Property Management ISSN 1648-715X / eISSN 1648-9179 2016  Volume  20(2): 156–171 doi:10.3846/1648715X.2015.1121415 ===== PAGE 8 (score 60, numeric-paren 33, table-words 3, var-words 18) ===== Marginal prices of improvements made to blocks of flats: empirical evidence from Romania 163 Table 1. Variables for sales price respectively for rental price models: descriptive statistics and ANOVA Name Variable Sales price Rental price ANOVA F (p-value) % Mean price ANOVA F (p-value) % Mean price Price Attributes Thermal Heating Finishing Finishing2 PhCond PhCond2 Sales price of the flat (euro) Rental price of the flat (euro) Thermal insulation of the block: Yes No (reference category) Heating system: Individual boilers Central heating (reference category) Finishing work into flat: Super finishing Medium finishing Demi-finishing (reference category) Finishing work into flat: Super finishing or medium finishing Demi-finishing (reference category) Overall physical condition: Very good Good Unsatisfactory (reference category) Overall physical condition: Very good Good and unsatisfactory (ref. cat.) 8.4 (0.01) 28.5 (0.00) 17.6 (0.00) 35.2 (0.00) 16.9 (0.00) 59.1% 50.9% 62.7% 37.3% 23.4% 56.3% 20.3% 79.7% 20.3% 40.2% 46.4% 13.5% 40590 42295 38948 42927 36655 42454 42208 33948 42281 33948 44117 39319 34441 31.6 (0.00) 22.7 (0.00) 8.7 (0.00) 18.7 (0.00) 36.1 (0.00) 31.2% 68.8% 45.5% 54.5% 26.5% 56.8% 16.7% 26.0% 62.0% 12.0% 26.1% 73.9% 220 256 204 245 214 241 220 187 261 204 218 260 206 City Location Rooms Floor Floor2 Area Year Garage NoFloors Type Elevator Balcony Year­_tr The city: Arad Iasi Timisoara Cluj-Napoca (reference category) Location in the city: Central and intermediate district Urban periphery district areas Suburban areas (reference category) Number of rooms: Two rooms One room (reference category) Floor level: Ground floor Intermediate Last floor (reference category) Floor level: Intermediate Ground or last floor (ref. cat.) Floor area (m2) Year of construction: fc a 1990 <1990 (reference category) Existence of a garage: Yes No (reference category) Number of floors of the block: >5 floors cc b 5 (reference category) Type of flat: Detached Not-detached (reference category) Existence of an elevator: Yes No (reference category) Existence of balcony: Yes No (reference category) Year of transactions 2011 2012 (reference category) 8.9 (0.00) 11.7 (0.00) 137.5 (0.0) 2.2 (0.11) 3.4 (0.07) 0.9 (0.34) 20.6 (0.00) 2.5 (0.10) 20.1 (0.00) 2.6 (0.11) 12.2 (0.00) 0.9 (0.33) 15.5% 25.3% 31.7% 27.5% 44.5% 32.5% 23% 71.2% 28.8% 15.5% 67.3% 17.2% 67.3% 32.7% 36% 64% 10% 90% 31.9% 68.1% 70.6% 29.4% 36.9% 63.1% 65.4% 34.6% 53.6% 46.4% 34383 39596 42522 42778 42904 40798 35813 44433 31078 40162 41337 38052 41337 39054 41372 40144 48515 39736 41934 39961 42251 36602 41822 39871 42056 37851 41115 39983 30.1 (0.00) 30.3 (0.00) 49.0 (0.00) 1.63 (0.20) 27.0 (0.00) 113.7 (0.00) 19.5 (0.00) 1.8 (0.18) 19.2 (0.00) 20.1 (0.00) 3.5 (0.07) 8.0% 25.6% 39.3% 21.1% 21.6% 33.1% 45.3% 67.5% 32.5% 16.5% 67.9% 15.6% 16.7% 83.3% 9.0% 91.0% 27.1% 72.9% 66.8% 33.2% 27.4% 76.2% 63.7% 36.3% 50.9% 49.1% 152 242 191 262 275 217 196 239 180 204 223 228 271 211 346 209 245 208 230 218 249 211 234 196 227 213 Note: The sample size N = 483 for sale transactions, respectively N = 468 for rent transactions. ===== PAGE 9 (score 29, numeric-paren 0, table-words 9, var-words 11) ===== A. Deaconu et al. 164 category and also the results from the analysis of variance (ANOVA) method. For the entire sam- ple, the average sale price is of 40590 Euros, with a standard deviation of 12834 Euros. The distri- bution of prices has a positive skew and is plat- ykurtic. Two-room flats (71.2%), with an average floor area of 52 m2, prevail; for one-room flats the average floor area is about 36 m2. Most transac- tions seemed to occur in the three largest cities, respectively Cluj (27.5%), Iasi (25.3%) and Tim- isoara (31.7%). A large part of the flats are located in blocks with less than 5 floors, built before 1990, and without elevators. The attributes measuring the improvements made to the flat or building are encoded as Ther- mal, Heating, Finishing, and physical conditions as PhCond. A large part of the blocks were built before 1990, but their average price does not seem to be significantly lower than that of newer flats, revealing potential improvements to the old blocks. About 80% of the flats have undergone interior fin- ishing works, 23.4% being superior-finished and 56.3% medium-finished. Examining the moderni- sation features of the blocks, namely physical con- dition and thermal insulation, it can be noted that a large number of the flats are in good or very good state in terms of overall physical condition; 59% of the blocks were included in thermal rehabilita- tion programs. The values for the variable heating system confirm our preliminary descriptions con- cerning the prevalence of flat owners who decided to quit the district heating in favour of individual boilers. We mention that these statistics are drawn from our sample, without being generalised to the entire market. The data indicates, however, that some efforts were made during the last two dec- ades on the housing market to improve the physi- cal state and comfort of blocks of flats. As a preliminary analysis, we examined each attribute separately and computed the F-test from ANOVA, presenting the results in Table 1. The null hypothesis that the average prices are equal across the groups is rejected for all improvement attributes, except for thermal in- sulation, which is borderline rejected (p-value = 0.011). Therefore the average price seems to vary across the categories, for the physical condition of the building, degree of finishing, and heating system. As regards the control attributes, the null hypothesis is rejected for the age of the building (before/after 1990) and year of transaction, and borderline rejected for the number of floors in the building, flat floor, existence of an elevator. The transaction dates cover the period from late 2011 to 2012, during which the market prices remained relatively stable, therefore we expected this con- clusion for the year of transaction. Upon exami- nation of the F statistic, the city does not seem to hold a very high explanatory power for the price; as the descriptive statistics suggest, except for Arad where the average price is significantly lower, the average prices for the other three large cities are comparable to each other. The entire sample (sample size is 483 transac- tions) will be used to estimate the hedonic pricing models. Table 2 contains the variables with sig- nificant effect on the sale price, for each of the five models, which were estimated using the maximum likelihood method. A backward selection approach has been applied. All attributes were included in the model, then the attribute with the highest p- value was removed, and the new specification was estimated, and these steps were repeated until all coefficients had a p-value lower than 0.111. For the Finishing variable, the coefficient of dummy associated to medium finishing is not significant, therefore this variable has been redefined as a dummy with two categories Finishing2, as shown in Table 1. In a similar manner, the variable Floor was redefined as Floor2; there is no significant dif- ference between the prices of flats located on the top floor compared to the ground floor. Also, due to few flats being sold in the central area, for some cities, the central district and intermediate district locations are compressed into a single category. An overall goodness-of-fit is indicated by the likelihood ratio test. The Wald chi-square test is applied to test the statistical significance of each coefficient. The empirical results seem to be consistent with different specifications for GLM; generally the same explanatory variables were retained as having significant coefficients. Esti- mates obtained for marginal effects in Gamma-log M2 and inverse Gaussian-log M3 are close to those provided by the traditional hedonic log-linear M1 model. In the framework of maximum likelihood method, the coefficient 2 1 0 1 / R D D = − , where 1 D is the residual deviance for the estimated model and 0 D the residual deviance for the model includ- ing only the intercept, measures the proportion of the null deviance accounted for by the model (Fox 2008). These coefficients suggest a medium goodness-of-fit for our models; a similar value of this coefficient is obtained for traditional log-linear 11 The empirical results for models including all attrib- utes can be provided upon request. ===== PAGE 10 (score 181, numeric-paren 130, table-words 15, var-words 14) ===== Marginal prices of improvements made to blocks of flats: empirical evidence from Romania 165 model12. To compare the (in-sample) performance of the three models, we also considered the differ- ences between observed and predicted prices, and computed the mean absolute error MAE13. This measure suggests that Gamma-log model M2 pro- vides a slightly better agreement between observed and predicted prices than M1 and M3. The Akaike information criterion (AIC)14, R-square and MAE usually suggest that the Gamma-log model performs better than the inverse Gaussian-log model, in terms of goodness-of-fit and prediction accuracy. Examining the estimated coefficients from Table 2, where are indicated exp( ) β , the attrib- utes have the expected signs. The percentage effect of an attribute over the price is given by [(exp( ) 1] 100 β − ⋅ . As regards our primary goal, to measure the effect of the improvement attributes 12 We mention that the traditional R-square coefficient, computed for the log-linear model estimated by OLS has the same values, respectively R2 = 0,54 for M1 and R2 = 0.61 for M4, when the same regressors, as in Table 3, are included. 13 Willmott and Matsuura (2005) suggest that MAE is better suited than the Root Mean Squared Error to measure the model performance. 14 The AIC cannot be used to compare M2 and M3 mod- els with the traditional log-linear model M1, as long as in the log-linear model the prices were transformed, being thus on different scales. on pricing, the physical condition of the building and the finishing variables have a significant effect on the price in all specifications. As the log-linear model seems to perform slightly better, in terms of R-square coefficient, we look at the implicit mar- ginal prices provided by M1. Thus, the prices of flats located in buildings that are classified as good and very good are higher by 7.8% and by 15.4% respectively than the prices of flats from buildings with unsatisfactory technical conditions. The fin- ishing works inside the flat determines price in- creases of 10%. These findings are in line with the empirical results from other studies. The market prices of flats and houses in which changes and renovations were undertaken appear to be higher (Portnov et  al. 2005; Kryvobokov, Wilhelmsson 2007; Sipos, Crivii 2008; Herath et al. 2015). The marginal effect of the heating system is borderline significant. There is a willingness to pay a price approximately 6% higher for the exist- ence of an individual boiler heating system. The thermal insulation dummy does not have a sig- nificant coefficient, but this characteristic may be viewed as embedded into the physical condition of the building. If the physical condition of the build- ing is not included in the model, the Thermal vari- able has a significant effect on pricing, at 5% sig- nificance level (Wald test = 4.15; p-value = 0.04). Table 2. Summary of hedonic models for sales price Model M1 M2 M3 M4 M5 Variable Log-linear Gamma-log In.Gauss-log Log-linear Gamma-log Finishing2 PhCond Very good Good Heating City Arad Iasi Timisoara Location Central and interm. Urban periphery district Rooms Area Garage NrFloors Type Balcony Elevator Floor2 Intercept 1.099*(9.4) 1.154* (12.9) 1.078* (4.4) 1.062** (4.6) 0.750* (58.0) 0.941 (3.4) 0.963 (1.3) 1.270* (56.9) 1.191* (27.8) 1.44* (218.4) – 1.118* (8.1) 1.061** (4.9) 1.121* (20.1) 1.090* (12.2) – – 18688.2* LR = 367.0* AIC = 13.6 MAE = 6329.3 R2 = 0.54 1.084* (7.4) 1.149* (13.3) 1.057 (2.9) 1.062 ** (4.9) 0.752* (58.9) 0.962 (1.4) 0.951 (2.2) 1.250* (52.4) 1.177* (24.8) 1.419* (213.5) – 1.115* (8.1) 1.046 (3.06) 1.120* (21.1) 1.080* (10.3) – – 20216.7* LR = 345.89* AIC = 10193.9 MAE = 6319.5 R2 = 0.50 1.096* (8.3) 1.183* (16.3) 1.081** (4.5) 1.067** (4.6) 0.736* (54.3) 0.959 (1.2) 0.941 (2.4) 1.260*(44.8) 1.199*(23.6) 1.424* (184.9) – 1.114* (5.1) 1.058 (3.5) 1.120* (16.8) 1.081* (8.4) – – 19382.8* LR = 285.8* AIC = 10322.5 MAE = 6431.2 R2 = 0.45 1.095* (10.5) 1.150* (15.7) 1.091* (7.0) – 0.729* (74.4) 0.972 (0.9) 0.955 (2.4) 1.217* (45.7) 1.138* (17.6) – 1.017* (375.1) – – 1.041** (2.8) 1.067* (8.1) 1.062** (6.1) 1.066* (8.1) 12381.2* LR = 448.4* AIC =–69.1 MAE = 5472.1 R2 = 0.61 1.077* (8.4) 1.127* (41.7) 1.056* (15.4) – 0.738* (93.6) 0.994 (0.3) 0.944 (3.0) 1.192* (41.9) 1.117* (14.5) – 1.018* (438.6) – – – 1.061* (7.3) 1.057**(6.2) 1.061*(8.1) 13017.0* LR = 466.53 AIC = 10061.4 MAE = 5463.2 R2 = 0.62 Notes: For each model are reported Exp(β ) where β are the coefficients of explanatory variables, and the Wald test (in parentheses). *Significant at 1%, **significant at 5%. For inverse Gaussian-log model including the Area are obtained: LR = 398.4, AIC = 10205.9, MAE = 5936.2, R2 = 0.56. ===== PAGE 11 (score 35, numeric-paren 0, table-words 9, var-words 16) ===== A. Deaconu et al. 166 The chi-square test supports this remark, a signifi- cant degree of association being observed between the PhCond and Thermal variables (chi-square statistic = 168.0 and p-value = 0.00; the contin- gency coefficient = 0.51). On the other hand, as expected, the number of rooms has the highest explanatory power. The price of two-room flats is 44 % higher than the price of one-room flats. Location inside the city also has a significant effect on pricing. The flats from blocks located in the central district, inter- mediate district and urban periphery district areas have a significantly higher price than those from suburban areas, by 19% and 27% respectively. If we compare the prices in the four cities, we find, as expected, significantly lower prices in Arad (by about 25%) than in Cluj-Napoca. For Iasi and Timisoara, the conclusion is not so clear, as there seems to be no significant differences in prices compared to Cluj-Napoca; in these two cities the price does not appear to be significantly different, as also suggested by the Student t-test for equality of two means (t = 0.18; p-value = 0.85). The coef- ficients of corresponding dummy variables seem to be insignificant in some specifications; therefore, they must be interpreted with caution. Generally the coefficients for dummies for cities should be only interpreted as indicative, due to the small sample of transactions from each city. The other structural characteristics, namely the existence of a balcony and a garage, detached flats, have a positive effect on the price. Flats located in high blocks (more than 5 floors) seem to be more ex- pensive than those from blocks with fewer floors. The four-floor blocks are generally old buildings; therefore, this attribute may incorporate partial information about the age of the building. We conducted a similar empirical analysis for hedonic regressions, considering the floor area instead of the number of rooms. A backward se- lection approach was applied for the same speci- fications as above (S1-normal distribution with identity link for logarithm of prices, S2-Gamma distribution with log link, and S3-inverse Gauss- ian distribution with log link). Table 2 reports the results for the Log-linear model denoted by M4 and for the Gamma-log denoted by M5. For inverse Gaussian specifications, not reported here, we ob- tained a lower value for R-square and a higher value for MAE, than for M4 and M5. In terms of fit and prediction accuracy the Gamma-log model (M5) performs slightly better than the log-linear model (M4). We can see that the floor area has a positive and high significant effect on pricing; one additional square meter raises the flat’s price by 1.8%, as long as the other attributes are the same. The physical condition of the building and finishing variable have a significant impact on the sale price, and the estimated coefficients are close to those obtained from models which include the number of rooms as an explanatory variable. If the physical condition of the building is removed from the models, the thermal insulation becomes a significant attribute for the sale price (Wald test = 9.07; p-value = 0.00), therefore the overall physi- cal condition variable seems to capture the effect of thermal insulation. The other structural charac- teristics, with positive effect on pricing are: exist- ence of a balcony, floor number, detached flats, and existence of an elevator. Therefore, the physical condition of building and the finishing works into the flat have a significant effect on sale prices for all distributions considered for sale price (log-normal, Gamma, inverse Gaus- sian); our results are robust to the distribution of sale price. On the other hand, the multicollinearity is often an issue in regression models. The Variance Inflation Scores (VIF), a measure usually used for detecting the presence of multicollinearity, not ex- ceeds 3.28 for any explanatory variables, suggesting that the multicollinearity is not a major problem. Although no definitive rules exist for the threshold of VIF, the general approach is that multicollinea- rity is not a serious problem if the VIF not exceed 5 (Troy, Grove 2008). A likelihood ratio (LR) test for omitted variables is also conducted. According to LR statistic, the physical state of the building (LR = 9.36, p-value = 0.00 for model M1, and LR = 13.2, p- value = 0.00 for model M4) and the finishing works (LR = 10.5, p-value = 0.01 for model M1, and LR = 15.6, p-value = 0.00 for model M4) have a signifi- cant contribution to explaining the variation in the sale price. Regarding the heteroscedasticity issue, in order to avoid a potential bias in the variance of estimated parameters, has been used the White’s heteroskedasticity consistent covariance matrix es- timator (White 1980). The provided results were si- milar to the reported ones; our variables of interest have significant coefficients. Empirical results for rental price For our sample of 468 rental transactions, the av- erage monthly rental price is of 220 Euros, with a standard deviation of 90 Euros. Table 1 indicates the percentage of transactions and average rental price determined for each attribute and category, and also the results from the analysis of variance method. The distribution of rental prices has a positive skew ===== PAGE 12 (score 148, numeric-paren 107, table-words 11, var-words 13) ===== Marginal prices of improvements made to blocks of flats: empirical evidence from Romania 167 and is strongly platykurtic, therefore into the GLM framework we look at the distributions that take ac- count of this behaviour. The rented flats from the three large cities, located mainly in the intermedi- ate district and urban periphery district areas, and in blocks with less than 5 floors, are predominant. The descriptive statistics computed for improve- ment attributes, Thermal, Heating, Finishing, and overall physical conditions PhCond, indicate low percentages for unfinished flats, and for flats lo- cated in buildings with inadequate technical state; the majority of flats are equipped with individual boilers. Examining each attribute separately, the F-test from ANOVA rejects the equality of average rental prices among the categories, for all improve- ment attributes. The rental prices seem not to vary across categories for the following variables: year of transaction, type of the flat (detached/ non-de- tached) and floor of the flat. The descriptive statis- tics suggest, as expected, that the rental prices are significantly lower in Arad than in the other three large cities, and that they are significantly higher in the central area. Following a careful analysis of the significance of coefficients for dummy variables, in the models including all attributes, the variable PhCond was redefined, as indicated in Table 1. Also, the urban periphery district and suburban areas are consid- ered a single category. The backward selection ap- proach provides the empirical results reported in Table 3; almost the same variables were selected regardless of specification, as the attributes with significant effect on rental prices. The set of signifi- cant attributes, in addition to the physical state of the building and finishing, includes: the city, loca- tion inside the city, number of rooms/ area, exist- ence of a garage, existence of a balcony, generally included also in the hedonic models for sale prices. We first examine the M1, M2, and M3 models, considering the number of rooms. According to the AIC information criteria the Gamma-log model performs better than the inverse Gaussian-log model. The in-sample performance of these models is almost the same, as suggested by the MAE, and by the R-square coefficient measuring the propor- tion of the null deviance accounted for by the mod- el. The physical state of the building and finishing variables are significant in all specifications, but have lower effect than on sale prices. According to the log-linear model, for a very good physical condition of the building there is a willingness to pay a 10.5% higher price, as long as the other attributes are the same. As in the case of the sale price, this attribute seems to incorpo- rate the effect of thermal insulation; when it is re- moved from the model, the thermal insulation has a significant effect on rental prices (Wald test = 6.55; p-value = 0.01). The finishing level of the flat has a significant but not strong effect on rental prices, as long as this variable remains borderline significant. A superior level of finishing in the flat has a marginal effect on rental prices, triggering Table 3. Summary of hedonic models for rental price Model M1 M2 M3 M4 M5 Variable Log-linear Gamma-log Inv.Gauss.-log Log-linear Gamma-log Finishing Superior Medium PhCond2 City Arad Iasi Timisoara Location Central Intermediate Rooms Area Garage Balcony Intercept 1.106** (5.7) 1.066 1.103* (9.4) 0.628* (83.7) 1.052 (1.7) 0.879* (13.7) 1.263* (45.6) 1.064** (4.6) 1.298* (94.9) – 1.290* (30.5) 1.080* (8.8) 147.2* LR = 320.7 AIC = 95.3 MAE = 42.85 R2 = 0.50 1.096**(4.8) 1.065 (3.3) 1.131* (15.7) 0.660* (69.3) 1.082** (4.8) 0.893* (10.2) 1.266* (45.1) 1.051* (3.7) 1.302*(100.5) – 1.300* (33.1) 1.077* (8.1) 147.2* LR = 349.1 AIC = 5072.9 MAE = 43.25 R2 = 0.51 1.100* (5.5) 1.059 (2.8) 1.140* (16.7) 0.634* (98.7) 1.080 (3.6) 0.905* (7.3) 1.245* (36.8) 1.041 (1.9) 1.299* (100.3) – 1.274* (18.2) 1.074* (7.4) 147.6* LR = 324.1 AIC = 5086.1 MAE = 43.38 R2 = 0.49 1.118* (7.9) 1.066** (3.9) 1.084** (5.6) 0.575* (129.5) 1.065 (3.0) 0.904* (9.6) 1.199* (27.8) 1.061** (4.7) – 1.010* (159.6) 1.182* (14.1) 1.067* (6.9) 109.7* LR = 371.7 AIC = 44.6 MAE = 38.56 R2 = 0.55 1.114* (7.6) 1.060* (3.2) 1.090* (8.2) 0.612* (104.5) 1.100* (7.2) 0.909* (8.7) 1.192* (29.8) 1.061* (4.8) – 1.010* (149.8) 1.183* (14.8) 1.072* (8.2) 110.9* LR = 398.6 AIC = 5021.5 MAE = 38.74 R2 = 0.56 Notes: For each model are reported Exp(β) where β are the coefficients of explanatory variables, and the Wald test (in parentheses). *Significant at 1%, **significant at 5%. For inverse Gaussian-log model including the Area, are obtained: LR = 351.0, AIC = 5057.3, MAE = 39.6.2, R2 = 0.53. ===== PAGE 13 (score 28, numeric-paren 0, table-words 6, var-words 15) ===== A. Deaconu et al. 168 a price surge of about 10%. Bala et al. (2014) also conclude that renovated dwellings carry higher monthly rents. Owners are motivated to improve the physical characteristics of properties, in order to rent at higher prices. Tenants are looking for finished flats, to improve their living standards, but are also interested in some improvements that could reduce the cost of utilities. The number of rooms, existence of a balcony, garage, and location have a significant effect on rental prices. The price of two-room flats is higher roughly by 30% than the price of one-room flats. Flats from blocks located in the central area carry a significantly higher rent than those from the ur- ban periphery district and suburban areas, rough- ly by 26%. The lowest rental rates are in Arad, and the highest in Iasi and Cluj-Napoca; in these two cities the rent does not appear to be signifi- cantly different, as also suggested by the Student t-test for equality of two means (t = 1.37; p-value = 0.17). These two cities are important university centres, thus a large part of the demand for rental flats comes from students. Due to the small num- ber of rent transactions in each city these statistics should be viewed as indicative. When the floor area is considered instead of the number of rooms, usually the MAE and R- square indicators show better fit; the continuous variables are richer in information than categori- cal variables. The empirical results from the log- linear model and Gamma-log model are indicated in Table 3 (model M4 and M5). As the estimated hedonic regressions suggest, one additional square meter of floor area triggers a rental price increase of 1%. The overall physical state of the building and finishing variables still remain significant, at 5% significance level, but the effect of the first variable is slightly lower than in previous specifi- cations. Overall, both variables, the physical state of the building and finishing works, remain significant regardless the distribution of rental price. The multicollinearity is not an issue as long as no VIF exceeded 3.03. The LR test for omitted variables indicates that the physical state of building (LR = 9.30, p-value = 0.00 for M1, LR = 5.56, p-value = 0.02 for M4), respectively the finishing works (LR = 5.73, p-value = 0.00 for model M1, and LR = 7.85, p-value = 0.00 for model M4) are variables that significantly influence the rental price. Using the White’ heteroskedasticity robust standard errors, similar results are provided; the variables of inter- est are still significant. 5. Conclusions This paper examines a specific segment of the Ro- manian housing market, namely blocks of flats in the urban area. Similarly to other markets from Central and Eastern Europe, the Romanian real estate market inherits some stock of low-quality blocks of flats, with a small living surface and low comfort, undergoing a process of deterioration. We highlight the steps taken by owners and authori- ties to improve the housing maintenance, and the living standards in these dwellings. This analysis may contribute to a better understanding of the housing market from Romania and from the CEE countries. The empirical section investigates whether and to what extent the improvements made to the stock of flats have a significant effect on sale and rental prices. As described in Section 2, the total cost of building modernization and finishing works inside the flats, is about 179 Euro/m2, without VAT, as an average for flats with one or two rooms. On the other hand, in our sample, the average sale price is 906 Euro/m2 for flats with very good physical conditions of the building and super-finishing level inside the flats, respectively 717 Euro/m2 for flats with unsatisfactory physical state of the building and demi-finishing level inside the flats; the dif- ference between these two average prices is 189 Euro/m2. Overall, the marginal price and the cost of modernization can be considered as being com- parable, taking account of the difficulty of accu- rately estimating the cost and that average values are involved. Two hedonic analyses were conducted, one for sale transactions and another one for rental trans- actions, based on samples of flats from blocks lo- cated in four large towns from Romania. Hedonic models are estimated in a GLM framework, which considers the following specifications: the log- linear model, Gamma distribution with log link, and inverse Gaussian distribution with log link. Generally, for our data, the empirical results re- garding the marginal prices of the improvements attributes are consistent with these specifications. The traditional log-linear and Gamma-log models provided close results in terms of fit and prediction accuracy. The inverse Gaussian-log models lead to a weaker agreement between the actual and pre- dicted prices. The robustness of empirical results was also investigated from other points of view, namely multicollinearity, omitted variables and heteroscedasticity.