** You can use EXCEL or any other software. ** You need to include your printout in your submissions. The layout and format of the paper should include the following sections: Title page, Abstract, Introduction, Method, Results, and Discussion. a) Abstract is a very concise summary of the paper. b) The Introduction tells the reader about the topic. Specifically, it should start with the purpose of this paper is to examine.... What the issue is, what is known about it, and the specific focus? Put a business context to it -- write the value added by your work or what businesses can gain from knowledge of the determinants of market profits. c) Empirical Results you should start with a brief discussion of the descriptive statistics for each variable. d) Method you should be able to explain what method you are using for your work. For example, you can start with telling the reader that the ordinary least squares method (OLS) was used to obtain estimated coefficients... (Then write more). You should be able to write the equation from the Excel result. We expect to see you... 1. Using the excel result to generate the equation. 2. Find the standard error of each variable. 3. Find the t-stat of each variable. 4. Find the p-value of each variable. EXAMPLE: = 1.24 + 1.71x1 - 0.83x2 - 2.12x3 (6.79) (1.43) (0.22) (0.85) T-stat [0.18] [1.20] [3.78] [2.48] P-value 0.857 0.247 0.002 0.025 e) The results section should tell what was found from the computed data. Using Chapter 10 and 11 for helping you interpret your result. For instance, you should interpret the Coefficient of Determination (r 2 ), F-test, t-test, p-value and etc. In addition, your result should use the hypothesis equation. Remember that in the results section there should include hypothesis testing for a t-test about each slope coefficient and a F-test for the overall regression model. EXAMPLE: For the T-test, it is written like this: H0 = 1 = 0 Ha = 1 0 Note: You have to decide whether to use one-tail or two-tail test You need to follow chapter 9 materials and declare your alternative hypothesis in each case very well F-test: H0 = 1 = 2 = 3 = 0 Ha = 1 2 3 0, or all three are 0 (Note that: remember to mention what is used in your hypothesis testing). f) The Discussion describes what your findings mean in the light of the information presented in the introduction. It is the interpretive segment of the paper and loops back to answer the issues raised in the introduction. Remember to answer questions 2 and 3. Sample Data for Supermarket Profits. Supermarket Number 1 2 3 4 5 6 7 8 9 10 Food Sales Nonfood Sales (tens of thousands (tens of thousands of dollars) of dollars) 305 130 189 175 101 269 421 195 282 203 35 98 83 76 93 77 44 57 31 92 Store Size (thousands of square feet) 35 22 27 16 28 46 56 12 40 32 Profit (thousands of dollars) Y 20 15 17 9 16 27 35 7 22 23 1. Run a multiple regression using the above data. (with profit as the dependent variable). 2. Compare the result of question (1) with another regression equation obtained without the food sales variable. 3. Which of the models do you prefer? Why? 4. Interpret your results for (1) and (2). 5. In writing your paper, you should start by indicating the purpose of the study. Also, discuss the methodology and conclusions. ** You can use EXCEL or any other software. ** You need to include your printout in your submissions. The layout and format of the paper should include the following sections: Title page, Abstract, Introduction, Method, Results, and Discussion. a) Abstract is a very concise summary of the paper. b) The Introduction tells the reader about the topic. Specifically, it should start with the purpose of this paper is to examine.... What the issue is, what is known about it, and the specific focus? Put a business context to it -- write the value added by your work or what businesses can gain from knowledge of the determinants of market profits. c) Empirical Results you should start with a brief discussion of the descriptive statistics for each variable. d) Method you should be able to explain what method you are using for your work. For example, you can start with telling the reader that the ordinary least squares method (OLS) was used to obtain estimated coefficients... (Then write more). You should be able to write the equation from the Excel result. We expect to see you... 1. Using the excel result to generate the equation. 2. Find the standard error of each variable. 3. Find the t-stat of each variable. 4. Find the p-value of each variable. EXAMPLE: = 1.24 + 1.71x1 - 0.83x2 - 2.12x3 (6.79) (1.43) (0.22) (0.85) T-stat [0.18] [1.20] [3.78] [2.48] P-value 0.857 0.247 0.002 0.025 e) The results section should tell what was found from the computed data. Using Chapter 10 and 11 for helping you interpret your result. For instance, you should interpret the Coefficient of Determination ( ), F-test, t-test, p-value and etc. In addition your result should use the hypothesis equation. EXAMPLE: (Note that: remember to mention what is used in your hypothesis testing). f) The Discussion describes what your findings mean in the light of the information presented in the introduction. It is the interpretive segment of the paper and loops back to answer the issues raised in the introduction. Remember to answer questions 2 and 3. SUMMARY OUTPUT Regression Statistics Multiple R 0.7600866225 R Square 0.5777316736 Adjusted R Square 0.5249481328 Standard Error 5.7155693853 Observations 10 ANOVA df Regression Residual Total SS MS F Significance F 1 357.5581328103 357.55813281 10.945299707 0.0107242534 8 261.3418671897 32.667733399 9 618.9 Intercept Coefficients Standard Error t Stat P-value Lower 95% 3.8804807523 4.9426335188 0.7851038799 0.4550121916 -7.5172525807 Food Sales (tens of thousands of dollars) X1 0.0670463403 0.0202656831 3.3083681336 0.0107242534 0.0203135912 RESIDUAL OUTPUT Observation 1 2 3 4 5 6 7 8 9 10 Predicted Profit (thousands of dollars) y 24.3296145433 12.5965049911 16.5522390687 15.6135903045 10.6521611224 21.9159462925 32.1069900179 16.9545171105 22.7875487164 17.4908878328 Residuals -4.3296145433 2.4034950089 0.4477609313 -6.6135903045 5.3478388776 5.0840537075 2.8930099821 -9.9545171105 -0.7875487164 5.5091121672 Food Sales (tens X1 R 10 5 Residuals 0 -5 50 100 150 200 -10 -15 Food S Upper 95% Lower 95.0% Upper 95.0% 15.278214085 -7.5172525807 15.278214085 Food Sales (tens o X1 L 40 35 30 0.1137790894 0.0203135912 0.1137790894 25 Profit (thousands of dollars) y 20 15 10 5 0 50 100 150 200 Food Sales (tens of Food Sales (tens of thousands of dollars) X1 Residual Plot 100 150 200 250 300 350 400 450 Food Sales (tens of thousands of dollars) X1 Food Sales (tens of thousands of dollars) X1 Line Fit Plot 40 35 30 Profit (thousands of dollars) y Predicted Profit (thousands of dollars) y 25 20 15 10 5 0 50 100 150 200 250 300 Food Sales (tens of thousands of dollars) X1 350 400 450 400 450 rofit (thousands of dollars) redicted Profit (thousands of dollars) Supermarket Number 1 2 3 4 5 6 7 8 9 10 Food Sales (tens of thousands of dollars) X1 305 130 189 175 101 269 421 195 282 203 Nonfood Sales (tens of thousands of dollars) X2 35 98 83 76 93 77 44 57 31 92 Store Size Profit (thousands of (thousands of square feet) dollars) X3 y 35 20 22 15 27 17 16 9 28 16 46 27 56 35 12 7 40 22 32 23 Atl Econ J (2009) 37:159-172 DOI 10.1007/s11293-009-9174-x O R I G I N A L PA P E R Housing Attribute Preferences in a Northern Mexico Metropolitan Economy Karen P. Fierro & Thomas M. Fullerton, Jr. & K. Erika Donjuan-Callejo Published online: 10 March 2009 # International Atlantic Economic Society 2009 Abstract Studies of the valuation of housing attributes have been conducted for numerous metropolitan markets. Empirical analyses of this nature for cities in Latin America and Mexico are less common. This study utilizes data for 175 new houses in Ciudad Juarez, a major metropolitan economy in northern Mexico, to estimate a hedonic pricing model. All units in the sample were completed and sold between November 2006 and April 2007. For each house, a total of 14 characteristics relating to structure and location are employed as explanatory variables. Estimation results indicate that the structural characteristics play bigger roles than the neighborhood amenities. Surprisingly, neighborhood parks are found to lower housing values. Keywords Housing demand . Applied econometrics . Mexico JEL R21 . Housing demand Introduction International studies of housing attribute valuations have been conducted for numerous metropolitan markets. Examples include Africa, Europe, Asia, and North America. Similar research for Latin America and Mexico is less common (Figueroa 1993). To partially fill that gap in the literature, this effort employs a sample of 175 K. P. Fierro : T. M. Fullerton, Jr. (*) Department of Economics & Finance, University of Texas at El Paso, El Paso, TX 79968-0543, USA e-mail: tomf@utep.edu K. P. Fierro e-mail: kpfierro@miners.utep.edu K. E. Donjuan-Callejo Department of Economics, Universidad Autnoma de Ciudad Jurez, Ciudad Jurez, Mexico e-mail: erika@corporacionaem.org 160 K.P. Fierro et al. new houses for Ciudad Juarez, an important metropolitan economy in northern Mexico. All houses in the sample were completed and sold between November 2006 and April 2007. For each house, a total of 14 characteristics, both structural and locational, are utilized to model housing prices in this urban market. As a growing economy, Ciudad Juarez offers an interesting case to study. Throughout Mexico, mortgage banking services are expanding at a rapid rate (Skelton 2006). Some aspects of the expanding housing stock in this city are well understood, such as increased demand for public utility services (Fullerton et al. 2006). To date, however, there have not been any attempts to quantify the contributions of physical attributes to the underlying values for housing structures in this market. To carry out such an effort, hedonic price equations for new dwellings are estimated using a combination of qualitative and numeric explanatory variables. It is expected that 13 of the 14 different characteristics will increase housing values. Several different equation estimates are utilized to analyze the data. Subsequent sections of the material are organized as follows. A brief review of the literature is presented in next section. A description of the data and methodology is then provided. Empirical estimation results are summarized in the subsequent section. Concluding remarks and suggestions for future research are offered in the final portion of the paper. Literature Review Previous work on housing markets has broadly analyzed the demand for structural amenities and attributes. Some of this research utilizes data from developing countries, but the demand for housing in Mexico has generally been analyzed using other approaches. Most of the hedonic attributes pricing articles are based on the methodology proposed by Rosen (1974). It has been applied to data from a fairly large number of different housing markets (Blomquist and Worley 1981; Arimah 1992; Cheshire and Sheppard 1995; Pasha and Butt 1996). These studies have generally established that physical traits and neighborhood amenities play important roles in determining residential real estate values. Can (1992) examines spatial neighborhood characteristics impacts on housing prices. Vicinity or adjacency externalities are added as a third set of explanatory variables. Two sets of estimates are completed. The first includes only structural and neighborhood amenities. The second also includes adjacency externalities. Similar to Case and Mayer (1996), results indicate that the inclusion of spatial factors in equation specifications improves model reliability. Cheshire and Sheppard (1995) reach similar conclusions based on data for towns in the United Kingdom. Empirical evidence in that study indicates that unobserved location amenities are systematically incorporated into housing prices. Just as neighborhood amenities can increase housing prices, location disamenities will reduce housing values. A variety of examples, both observable and unobservable, have been documented in the literature in recent years. Examples include crime, environmental degradation, tax and fee burdens, and traffic congestion (Chay and Greenstone 2005; Cho 1997; Lang and Jian 2004; Pope 2008). Evidence reported in those, and other, studies suggests that hedonic model Housing Attribute Preferences 161 specifications should allow for both positive and negative factors. This can mean including variables that may raise housing values simply because they minimize the impacts of negative factors on that residence. In the context of Latin American housing markets, one example would be the presence of neighborhood guard posts. Relatively few housing market studies have been completed for Mexico (Jones et al. 1993). Among those that have been completed, data constraints are fairly severe. In spite of that, indirect evidence has occasionally been uncovered that documents the demand for structural attributes that match individual household demographics and prospective owner preferences (Gonzalez 1997; Noguchi and HernandezVelasco 2005). Fairly critical housing shortages have also been documented for fast growing regional markets in northern Mexico such as Ciudad Juarez (Pea 2005). Given the international importance of the economy of Mexico, plus the potential construction industry and financial intermediary business opportunities associated with the demand for housing throughout much of the country, better understanding of its real estate markets appear warranted. This study uses a hedonic modeling approach to analyze housing prices in Ciudad Juarez, Mexico. For this purpose, both structural and neighborhood amenities are included as potential explanatory variables. The sample includes a variety of singlefamily units built in different neighborhoods. Because Ciudad Juarez is relatively large, access to roads and social infrastructure is expected to play a crucial role in housing selection. The real-estate market in Juarez offers an opportunity to examine these relationships within the context of a growing urban economy in a middleincome country. To date, very few studies of this nature have been utilized to help quantify the nature of residential real estate demand in Mexico or other Latin American economies. This analysis attempts to at least partially fill that gap in the literature. Data and Methodology The objective of the hedonic modeling approach is to account for how attributes typically associated with residential structures affect housing prices. The methodology is widely employed by numerous organizations, including the United States Census Bureau in its quarterly housing market analysis. The Census Bureau approach incorporates twelve distinct structural characteristics (Rappaport 2007). Of those variables, six are used in this effort. Another eight amenities not used by the Census Bureau are also included in this study. The variables employed are listed in Table 1. Despite widespread interest in the determination of housing prices, hedonic models have yet to be estimated for many metropolitan markets throughout Mexico. Ciudad Juarez has a unique identity as a result of its history, geographic location, and demographic characteristics. Consequently, amenity values will potentially differ what has been measured for other regional economies (Arimah 1992; Cheshire and Sheppard 1995; Pasha and Butt 1996). Similar to other large cities, Ciudad Juarez is laid out in a polycentric manner and residents commute to the various commercial and industrial employment centers. Residential zones have also developed around each of the employment centers in the city (Fuentes Flores 2001). 162 K.P. Fierro et al. Table 1 Mnemonics and description Variable Name Description HP Housing price in pesos LOT Property area in square meters FLS Floor area in square meters BED Number of bedrooms BATH Number of bathrooms PSP Number of parking spaces LEV Number of floors MATNC Brick walls FLOORNC Cement floors GATE Gated neighborhood GUARD Guard post in the neighborhood PARK Park or green areas in the neighborhood SCH School located nearly COMM Commercial area located nearly ML Major avenue or street access The value or price of housing (HP), measured in Mexican pesos, is modeled as a function of the fourteen explanatory variables shown in Table 1. Those right-hand side variables include the size of the property measured in square meters (LOT), the size of the house floor area measured in square meters (FLS), the number of bedrooms (BED), the number of bathrooms (BATH), and the number of stories (LEV). These are all numeric variables and, with the exception of LEV, should increase residential dwelling values. The data are from a residential real estate report that analyzes various aspects of the Ciudad Juarez market (Donjuan-Callejo 2007). In addition, several qualitative variables are also included as structural amenity regressors. For parking spaces (PSP), (0) represents one space with no roof, (1) is one roofed space, (2) represents two spaces without roof, and (3) corresponds to two roofed spaces. For construction materials, (MATNC) is a binary dummy where (1) represents any material other than concrete and (0) is concrete walls. Floor materials (FLOORNC) include anything other than cement (1), while (0) represents the presence of cement as the floor material. Location or neighborhood attributes are also measured using descriptive variables. They include binary indicators for the presence of the dwelling within a gated neighborhood (GATE), whether the neighborhood has a guard post (GUARD), and if the neighborhood has a park or green areas (PARK). Also, metropolitan area location (ML) is represented as location of the house close to a high speed road (2), a main avenue (1), or none of these (0). Proximity to an elementary school (SCH) and proximity to a commercial area (COMM) complete the explanatory variables list. As defined, all of the descriptive dummy variables are expected to increase housing values (Bible and Hsieh 2001; Lang and Nelson 2007). The sample includes 175 observations for new houses completed and sold in Ciudad Juarez between November 2006 and April 2007. Because all of these units Housing Attribute Preferences 163 are less than 12 months old, neither age nor age squared are included as arguments in the hedonic specification (Dehring et al. 2007). Every house in the sample possesses a different bundle of characteristics. In this sense, each of the 175 observations differs from the rest. Summary statistics for the variables that comprise the sample are reported in Tables 2 and 3. The geographic distribution of the housing units in the sample is fairly representative of new residential real estate activity in Ciudad Juarez. There are nine units from the northern sector of the city near the international Bridge of the Americas that crosses into El Paso, Texas. From the northeastern sector, where relatively ample supplies of flat land are available, there are 64 units. Similarly, from the eastern sector of the city, near the international Ysleta-Zaragoza Bridge, the sample contains data for 79 houses. From the southeastern sector, located east of the Ciudad Juarez regional airport where an ample supply of land is also available and public infrastructure is being built, there are 23 units. There are two metropolitan regions for which no units are included. Because of the mountains that lie west of the city, very little infrastructure is available and comparatively few housing construction permits are approved there in any given year. Structures that do get built there without going through the necessary permitting process generally do meet municipal building code requirements and are not sold by licensed realtors. The central region of Ciudad Juarez is largely developed. The absence of undeveloped lots generally precludes new residential construction in this area of the city, also (Fuentes Flores 2001; Esparza et al. 2004). Some residential buildings in this area are occasionally converted to commercial uses such as health care, retail, or office uses, but few houses are sold in any given year. Table 2 reports the arithmetic means and other summary measures for the numeric variables in the sample. As can be seen from the standard deviations and ranges for the different variables included, the sample exhibits relatively good variability. The average price per single-family unit is 896,208 pesos, roughly $82,000 in dollar terms. Most of the units in the sample fall in the middle- to upper-ranges of the houses sold in Ciudad Juarez in 2007 (Donjuan-Callejo 2007). Lot sizes average approximately 160.2 square meters with just over 130.5 square meters of floor space. Reflective of the four person households that prevail throughout the city (INEGI 2005), a typical unit includes three bedrooms and two bathrooms. Most of the houses in the sample are two-story structures. Table 2 Numeric variable summary statistics Variable Name HP Mean 896,208 Standard Deviation Minimum Maximum 397,224 Number of Observations 68,000 4,000,000 LOT 160.189 79.184 103 1,110 175 173 FLS 130.530 44.773 42 440 175 BED 2.85 0.38 2 4 174 BATH 2.23 0.56 1 3.5 174 LEV 1.98 0.13 1 2 174 164 K.P. Fierro et al. Table 3 Qualitative variable summary statistics Series Mode Frequency Minimum Maximum Observations PSP 2 98 0 3 174 MATNC 1 153 0 1 157 FLOORNC 1 150 0 1 174 GATE 1 147 0 1 173 GUARD 1 150 0 1 173 PARK 1 134 0 1 175 SCH 0 109 0 1 169 COMM 1 107 0 1 174 ML 0 75 0 2 175 For the qualitative variables summarized in Table 3, mode and frequency provide a better idea of the sample distribution. Reported minimum and maximum values reflect the qualitative values assigned to each variable, while frequency refers to those values that are observed the most. The majority of the houses in the sample have two parking spaces. Reflective of the general level of insecurity observed in recent years (Cruz 2007; Quintero 2007; Sosa 2008), most of these housing units are located in gated neighborhoods that have guards posted at the entrances. More than half of these dwellings are located near commercial centers and parks, but less than half are near schools or important arteries. For the sample as a whole, only 147 observations include complete information for all of the attributes selected for the econometric analysis conducted below. Although all but one of the variables are expected to increase housing values, these increments are expected to occur at decreasing rates, especially for the numeric variables. To reflect diminishing returns, logarithmic transformations are applied to those variables prior to estimation (DiPasquale and Wheaton 1996; Sirmans et al. 2005). The housing price specification is shown in eq. (1). For the variables PSP, and ML, the coefficients are hypothesized to linearly increase housing prices because the manner in which their respective values are assigned implies better quality or better location. In addition, MATNC, FLOORNC, GATE, GUARD, PARK, SCH, and COMM are binary variables where values of 1 indicate that they are present for that housing unit. Consequently, none of the qualitative variables are logarithmically transformed (Cassel and Medelsohn 1985; Pindyck and Rubinfeld 1998). log HP log b0 b1 log LOT b2 log FLS b3 log BED b4 log BATH b5 log LEV b6 PSP b7 MATNC b8 FLOORNC b9 GATE b10 GUARD b11 PARK b12 SCH b13 COMM b14 ML " 1 Empirical Results Empirical results for eq. (1) are summarized in Table 4. As frequently occurs in cross-sectional data samples, a chi-square test confirms that heteroscdestacity is Housing Attribute Preferences 165 present. Accordingly, the White (1980) procedure is utilized for parameter estimation. All of the coefficients for the numerical explanatory variables exhibit the hypothesized signs, but only those for floor area (FLS) and the number of bathrooms (BATH) exceed, or at least come close to, the 5-percent significance criterion. Among the qualitative discrete variables, only three exhibit the hypothesized positive signs. The only dummy variables that have statistically significant parameter estimates, PARK and MATNC, also have negative signs attached to them. The coefficient of determination for this model is 0.67, adjusted for degrees of freedom it is 0.63. For a relatively small cross-section sample, both values are fairly high and compare favorably to other metropolitan market studies (Anacker and Morrow-Jones 2008). In all, 11 of the 15 coefficients shown in Table 4 fail to satisfy the 5-percent significance criterion. The F-statistic is, however, significant at the 1-percent level, implying that multicollinearity may be present in the sample (Pindyck and Rubinfeld 1998). That possibility is further underscored by a number of correlation matrix values in excess of 0.5. To examine whether multicollinearity is present, several alternative specifications are also considered below. The first step along these lines is the omission from eq. (2) of all nine of the qualitative variables. The variables included in eq. (2) are lot size (LOT), floor area (FLS), number of bedrooms (BED), number of bathrooms (BATH), and number of Table 4 Equation (1) estimation results Variable Coefficient Std. Error t-Statistic Intercept 8.003679 0.562261 14.23480 LOG(LOT) 0.257786 0.184870 1.394421 0.1655 LOG(FLS) 0.937219 0.147390 6.358769 0.0000 LOG(BED) 0.063463 0.243960 0.260138 0.7952 LOG(BATH) 0.250020 0.126459 1.977079 0.0501 0.167879 0.193447 0.867833 0.3871 LOG(LEV) Prob. 0.0000 PSP 0.007249 0.021900 0.331022 0.7412 MATNC 0.220547 0.041998 5.251410 0.0000 FLOORNC 0.025344 0.064789 0.391183 0.6963 GATE 0.063682 0.059223 1.075292 0.2842 GUARD 0.041668 0.054734 0.761287 0.4478 0.148768 0.056935 2.612964 0.0100 SCH 0.028633 0.040831 0.701246 0.4844 COMM 0.036201 0.031080 1.164767 0.2462 0.003810 0.029991 0.127026 0.8991 PARK ML White heteroscedasticity-consistent standard errors & covariance R-squared 0.671592 Mean dep. var. Adjusted R-squared 0.636760 Std. dev. dep. var. Std. Err. Regression 0.288232 Sum squared resid. F-statistic Log likelihood 19.28132 17.80679 Prob. (F-statistic) 13.57692 0.478240 10.96628 0.000000 166 K.P. Fierro et al. stories (LEV). Under this reduced specification, the parameter magnitudes for lot size (LOT), floor area (FLS), and number of bathrooms (BATH) are very close to those reported in Table 4. In contrast, the absolute magnitudes for the BED and LEV coefficients in Table 5 are substantially different from those of their respective counterparts in Table 4. In particular, the negative impact of the number of stories in a dwelling may be less pronounced than what the results in Table 4 indicate. Also, the magnitudes of the adjusted coefficient of determination and the log likelihood function increase in Table 5. log HP log b0 b1 log LOT b2 log FLS b3 log BED b4 log BATH b5 log LEV " 2 Equation (3) excludes BED, but includes four of the discrete variables. As shown in Table 6, the parameter estimates for FLS and BATH are statistically significant. BATH fails to meet the 5% criterion in eq. (1), but is significant in eq. (3). As in previous specifications, the coefficient for LEV is negative, but fails to satisfy the significance criterion. The slope parameter for PSP changes from negative in eq. (1) to positive in this version. It is not, however, statistically different from zero. More notably, the parameter for COMM is almost twice as large as the original estimate in eq. (1) and surpasses the 5-percent significance threshold. That implies that reduced distances to commercial centers increases residential structure values as indicated by other studies (Waddell et al. 1993; Weber et al. 2007). log HP log b0 b1 log LOT b2 log FLS b3 log BATH b4 log LEV b5 log PSP b6 GUARD b7 SCH b8 COMM " 3 Equation (4) represents another departure from eq. (1). This version includes only three continuous variables: lot size (LOT), floor area (FLS), and number of Table 5 Equation (2) estimation results Variable Coefficient Std. Error t-Statistic Prob. 12.91701 0.0000 Intercept 7.822667 0.605610 LOG(LOT) 0.289107 0.186684 1.548643 0.1234 LOG(FLS) 0.855610 0.085702 9.983571 0.0000 LOG(BED) 0.018162 0.210537 0.086265 0.9314 LOG(BATH) 0.299057 0.123765 2.416335 0.0168 -0.058393 0.181697 0.321374 0.7483 LOG(LEV) White heteroscedasticity-consistent standard errors & covariance R-squared 0.651436 Mean dep. var. Adjusted R-squared 0.640938 Std. dev. dep. var. Std. Err. Regression 0.270341 Sum squared resid. F-statistic Log Likelihood 62.04805 16.01588 Prob. (F-statistic) 13.60134 0.451157 12.13202 0.000000 Housing Attribute Preferences 167 Table 6 Equation (3) estimation results Variable Coefficient Std. Error t-Statistic Prob. 13.88380 0.0000 Intercept 8.117161 0.584650 LOG(LOT) 0.204641 0.219496 0.932322 LOG(FLS) 0.861216 0.135064 6.376344 0.0000 LOG(BATH) 0.324610 0.131242 2.473377 0.0145 0.3526 0.073192 0.132012 0.554436 0.5801 PSP 0.006562 0.020640 0.317925 0.7510 GUARD 0.051316 0.039670 1.293583 0.1977 SCH 0.027576 0.039883 0.691414 0.4903 COMM 0.070842 0.029229 2.423650 0.0165 LOG(LEV) White heteroscedasticity-consistent standard errors & covariance R-squared 0.656877 Mean dep. var. Adjusted R-squared 0.639053 Std. dev. dep. var. Std. Err. Regression 0.276301 Sum squared resid. F-statistic Log Likelihood 36.85239 Prob. (F-statistic) 13.59955 0.459896 11.75667 0.000000 16.99664 bathrooms (BATH). It also includes four discrete variables: parking spaces (PSP), guard post (GUARD), nearby school (SCH), and nearby commercial area (COMM). As shown in Table 7, all of the parameters are positive as hypothesized but, once again, only FLS, BATH, and COMM satisfy the 5-percent significance criterion. Coefficient magnitudes for the continuous variables do not differ greatly from those Table 7 Equation (4) estimation results Variable Coefficient Std. Error t-Statistic Prob. 15.25726 0.0000 Intercept 8.079868 0.529575 LOG(LOT) 0.209352 0.211376 0.990427 0.3235 LOG(FLS) 0.854715 0.125521 6.809313 0.0000 LOG(BATH) 0.320452 0.124332 2.577392 0.0109 PSP 0.007062 0.019941 0.354150 0.7237 GUARD 0.046969 0.403988 1.067774 0.2873 SCH 0.029845 0.042923 0.695326 0.4879 COMM 0.071169 0.028705 2.479285 0.0142 White heteroscedasticity-consistent standard errors & covariance R-squared 0.656708 Mean dependent var. Adjusted R-squared 0.641204 Std. dev. dep. var. Std. Err. Regression 0.275476 Sum squared resid. F-statistic Log likelihood 42.35866 17.03686 Prob. (F-statistic) 13.59955 0.459896 11.76248 0.000000 168 K.P. Fierro et al. reported in Table 4. The discrete variable parameter estimates are very close to those shown in Table 6, as are the goodness of fit measures. log HP log b0 b1 log LOT b2 log FLS b3 log BATH b4 PSP b5 GUARD b6 SCH b7 COMM " 4 With a correlation coefficient of 0.755, it is easy to see that lot area (LOT) and floor size (FLS) are highly correlated with each other. Given that, several alternative specifications were also estimated without FLS. As can be seen in Table 8, multicollinearity may be partially masking the true magnitude of the coefficient for LOT in eqs. (1) through (4). Results for eq. (5), and some other simpler specifications, indicate that most of the variables included in the sample are attributes that affect housing values in Ciudad Juarez. A surprising outcome that is associated with these results is that, similar to what is documented in Tables 4 and 8, a negative sign consistently occurs for the PARK qualitative variable. Other studies (Cheshire and Sheppard 1995; Lang and Nelson 2007) have often indicated that such a neighborhood amenity increases housing values. log HP log b0 b1 log LOT b2 log BED b3 log BATH b4 PARK " 5 Crime rate data that would permit formal hypothesis testing are not presently available, but the negative PARK parameter may reflect a level of lawless activities around green areas in Ciudad Juarez that rises above the national threshold for Mexico (Troy and Grove 2008). The results for the alternative specification shown in eq. (6) are reported in Table 9. In it, the parameter estimated for the presence of a neighborhood guard post is positive at the 5-percent level. Space limitations prevent including all of the alternative specification regression results, but the GUARD dummy coefficients generally satisfy the significance criterion and are always greater than zero. A premium seems to be placed upon safety in this housing market. log HP log b0 b1 log LOT b2 log BED b3 log GUARD " 6 Table 8 Equation (5) estimation results Variable Coefficient Std. Error t-Statistic Prob. 11.74308 0.0000 C 8.217414 0.699767 LOG(LOT) 0.911260 0.170068 5.358220 0.0000 LOG(BED) 0.291048 0.218940 1.329354 0.1855 LOG(BATH) 0.741706 0.136453 5.435609 0.0000 0.083919 0.041552 2.019625 0.0450 PARK White heteroscedasticity-consistent standard errors & covariance R-squared 0.574089 Mean dependent var. Adjusted R-squared 0.563887 Std. dev. dep. var. Std. Err. Regression 0.297939 Sum squared resid. F-statistic Log likelihood 56.27515 33.25115 Prob(F-statistic) 13.60134 0.451157 14.82417 0.000000 Housing Attribute Preferences 169 Table 9 Equation 6 estimation results Variable Coefficient Std. Error t-Statistic Prob. C 6.398427 0.602633 10.61746 0.0000 LOG(LOT) 1.324907 0.130168 10.17844 0.0000 LOG(BED) 0.371310 0.280667 1.322958 0.1877 GUARD 0.185221 0.068966 2.685683 0.0080 White heteroscedasticity-consistent standard errors & covariance R-squared 0.437994 Mean dependent var Adjusted R-squared 0.427837 Std. dev. dep. var. Std. Err. Regression 0.342975 Sum squared resid. F-statistic Log likelihood 43.12348 Prob(F-statistic) 13.59931 0.453422 19.52685 0.000000 57.27886 Overall results from the various specifications are largely as hypothesized. As in other studies of this nature, alternate versions that exclude characteristic subsets of the explanatory regressors, help clarify which attributes reliably contribute to increased values (Can 1992). Contrary to what is reported in prior research (Cheshire and Sheppard 1995; Bible and Hsieh 2001; Lang and Nelson 2007), dummy variables such for floor materials, gated neighborhoods, and neighboring schools do not seem to improve housing values in this sample. Experimentation with various alternative specifications indicates that lot size, floor space, and bathrooms increase housing prices in Ciudad Juarez in statistically significant manners. The coefficient for the number of bedrooms is sensitive to the inclusion of additional variables, but also appears to add value to residential structures in this market. As in some earlier empirical analyses, parking spaces, guard posts, and proximity to commercial centers also tend to raise housing prices in this metropolitan market (Gin and Sonstelie 1992; Kockelman 1997; Andersson 2000). Surprisingly, proximity to parks is inversely related to housing prices in this sample. Two possible explanations may exist for this apparent anomaly. One is that parks and open spaces allow unknown persons to have access to the neighborhoods where they are located. That permits illegal narcotic, violent ambushes, and other suspicious activities to occur unexpectedly (Clark and Cosgrove 1990; Andersson 1990; Anacker and Morrow-Jones 2008). A second is that budget constraints frequently prevent municipal agencies from properly maintaining parks and green areas. If those areas then become eyesores, they will reduce housing values. Neighborhood associations sometimes step up to avoid such an outcome and charge fees to cover maintenance costs. Similar to property taxes, those fees can also become amortized into property prices and force them down (Oates 1969; Dehring et al. 2007). At present, crime rate data for the different geographic sectors of Ciudad Juarez are not available. Access to those data would allow examining whether an interaction effect in present between the incidence of crime and the PARK variable, a possibility that cannot be ruled out on the basis of the sample information used in this analysis. Similarly, data on neighborhood association fees are difficult to track 170 K.P. Fierro et al. down for the various regions located within the municipality. Both possibilities offer venues for future research efforts. The sample contains 175 observations. For a city of more than 1.4 million residents, that is a relatively small number of data points. Access to a larger sample would potentially improve upon the results reported above and help reduce multicollinearity. Because relatively few empirical analyses of the values of housing attributes exist for Mexico and other Latin American economies, carrying out similar exercises for other regional markets may prove instructive. As mortgage markets more fully develop throughout the hemisphere, more data sets such as the one utilized for this study will potentially become available. Conclusion This research employs a sample of 175 housing units in Ciudad Juarez to examine the values of physical and locational attributes. Fourteen different explanatory variables are used to estimate hedonic equations for this market. The results indicated that structural characteristics are more influential in housing valuation than locational elements. Moreover, neighborhood parks are consistently found to reduce housing prices. That is in contrast to what has been documented for other metropolitan housing markets and it is unknown if this is related to an interaction effect with the incidence of crime and/or fees that homeowner associations sometimes assess for maintenance purposes. It is unclear whether the results are unique to Ciudad Juarez or are representative of Mexico or Latin America at large. Collective evidence for the other variables indicates that they influence singlefamily housing prices as hypothesized, with the number of bathrooms, lot size, and floor area playing important roles. While some coefficient magnitudes are potentially concealed by sample multicollinearity, experimentation with subsets of the explanatory variable vector helped clarify which attributes contribute to higher housing values. Although those outcomes are useful, additional sampling would undoubtedly prove helpful in obtaining more precise estimates for this specific market. Future real estate construction may benefit, however, by attempting to design floor plans and neighborhood layouts that take into account the various results obtained herein. Despite a relatively large volume of research regarding hedonic housing prices, few studies of this nature have been completed for Mexico. Empirical analyses for housing market data from other cities would also be useful for comparison purposes. The size and diversity of metropolitan economies throughout the country should allow interesting results to emerge from such efforts. An increased number of observations for Ciudad Juarez would also be useful as a means for confirming, or overturning, the results obtained in this initial effort. As regional mortgage markets expand in Mexico, the call for this type of quantitative analysis will likely increase. Acknowledgements Financial support was provided by El Paso Electric Company, Hunt Building Corporation, Hunt Communities, Wells Fargo Bank of El Paso, a UTEP College of Business Administration Faculty Research Grant, and the James Foundation Scholarship Fund. Helpful comments and suggestions were provided by Tony Payn, Doyle Smith, Angel Molina, John Virgo, and an anonymous referee. Econometric research assistance was provided by Emmanuel Villalobos and Enedina Licerio. Housing Attribute Preferences 171 References Anacker, K. B., & Morrow-Jones, H. A. (2008). Mature suburbs, property values, and decline in the Midwest? The case of Cuyahoga County. Housing Policy Debate, 19, 519-552. Andersson, D. E. (2000). Hypothesis testing in hedonic price estimation. The Annals of Regional Science, 34, 293-304. doi:10.1007/s001689900010. Arimah, B. C. (1992). An empirical analysis of the demand for housing attributes in a third world city. Land Economics, 68, 366-379. doi:10.2307/3146694. Bible, D. S., & Hsieh, C. (2001). Gated communities and residential property values. The Appraisal Journal, 69, 140-145. Blomquist, G., & Worley, L. (1981). Hedonic prices, demand for urban housing amenities, and benefit estimates. Journal of Urban Economics, 9, 212-221. doi:10.1016/0094-1190(81) 90041-3. Can, A. (1992). Specification and estimation of hedonic housing price models. Regional Science and Urban Economics, 22, 453-474. doi:10.1016/0166-0462(92) 90039-4. Case, K. E., & Mayer, C. J. (1996). Housing price dynamics within a metropolitan area. Regional Science and Urban Economics, 26, 387-407. doi:10.1016/0166-0462(95) 02121-3. Cassel, E., & Medelsohn, R. (1985). The choice of functional form for hedonic price equations. Journal of Urban Economics, 18, 135-142. doi:10.1016/0094-1190(85) 90012-9. Chay, K. Y., & Greenstone, M. (2005). Does air quality matter? Evidence from the housing market. The Journal of Political Economy, 113, 376-424. doi:10.1086/427462. Cheshire, P., & Sheppard, S. (1995). On the price of land and the value of amenities. Economica, 62, 247- 267. doi:10.2307/2554906. Cho, M. (1997). Congestion effects of spatial growth restrictions: a model and empirical analysis. Real Estate Economics, 25, 409-438. doi:10.1111/1540-6229.00721. Clark, D. E., & Cosgrove, J. C. (1990). Hedonic prices, identification, and the demand for public safety. Journal of Regional Science, 30, 105-121. doi:10.1111/j.1467-9787.1990.tb00083.x. Cruz, J. M. (2007). Asesinan trailero en colonia satlite. El Diario de Ciudad Jurez, (December 12) 6B. Dehring, C., Depken, C., & Ward, M. (2007). The impact of stadium announcements on residential property values: evidence from a natural experiment in Dallas-Fort Worth. Contemporary Economic Policy, 25, 627-638. DiPasquale, D., & Wheaton, W. C. (1996). Urban economics and real estate markets. Englewood Cliffs, NJ: Prentice-Hall. Donjuan-Callejo, K. E. (2007). Monitoreo Semestral de Vivienda Nueva en Ciudad Jurez. Ciudad Jurez, MX: Asesora Econmica & Marketing, SC. Esparza, A. X., Waldorf, B. S., & Chavez, J. (2004). Localized effects of globalization: the case of Ciudad Jurez, Chihuahua, Mexico. Urban Geography, 25, 120-138. doi:10.2747/0272-3638.25.2.120. Figueroa, E. (1993). Hedonic estimations of the market value of public housing programs in Latin America. El Trimestre Economico, 60, 779-805. Fuentes Flores, C. M. (2001). Los cambios en la estructura intraurbana de Ciudad Jurez, Chihuahua, de monocntrica a multicntrica. Frontera Norte, 13, 95-118. Fullerton, T. M,. Jr, Tinajero, R., & Barraza de Anda, M. P. (2006). Short-term water consumption patterns in Ciudad Jurez, Mexico. Atlantic Economic Journal, 34, 467-479. doi:10.1007/s11293-006-9031-0. Gin, A., & Sonstelie, J. (1992). The streetcar and residential location in nineteenth century Philadelphia. Journal of Urban Economics, 32, 92-107. doi:10.1016/0094-1190(92) 90016-E. Gonzlez, L. (1997). Estimacin de la demanda de vivienda: Tenencia y gasto en servicios El mercado metropolitano de Mxico. El Trimestre Economico, 64, 569-598. INEGI. (2005). Censo 2005. Instituto Nacional de Estadstica, Geografa, e Informtica: Mxico, DF. Jones, P., Jimenez, E., & Ward, P. (1993). The land market in Mexico under Salinas: a real estate boom revisited? Environment and Planning, 25, 627-651. doi:10.1068/a250627. Kockelman, K. M. (1997). Effects of location elements on home purchase prices and rents in the San Francisco bay area. Transportation Research Record, 1606, 40-50. doi:10.3141/1606-06. Lang, K., & Jian, T. (2004). Property taxes and property values. Journal of Urban Economics, 55, 439- 457. doi:10.1016/j.jue.2004.01.002. Lang, R. E., & Nelson, A. C. (2007). Boomburb politics and the rise of private government. Housing Policy Debate, 18, 627-634. Noguchi, M., & Hernndez-Velasco, C. R. (2005). A 'mass custom design' approach to upgrading conventional housing development in Mexico. Habitat International, 29, 325-336. doi:10.1016/j. habitatint.2003.11.005. 172 K.P. Fierro et al. Oates, W. E. (1969). The effects of property taxes and local spending on property values. The Journal of Political Economy, 77, 957-971. doi:10.1086/259584. Pasha, H. A., & Butt, M. S. (1996). Demand for housing attributes in developing countries: a study of Pakistan. Urban Studies (Edinburgh, Scotland), 33, 1141-1154. doi:10.1080/00420989650011555. Pea, S. (2005). Recent developments in urban marginality along Mexico's northern border. Habitat International, 29, 285-301. doi:10.1016/j.habitatint.2003.10.002. Pindyck, R. S., & Rubinfeld, D. L. (1998). Econometric models and economic forecasts (Fourthth ed.). Boston, MA: Irwin McGraw-Hill. Pope, J. C. (2008). Fear of crime and housing prices. Journal of Urban Economics, 64, 601-614. doi:10.1016/j.jue.2008.07.001. Quintero, A. (2007). Hallan droga en camioneta. El Diario de Ciudad Jurez, (December 8) 4A. Rappaport, J. (2007). A guide to aggregate house price measures. Federal Reserve Bank of Kansas City Economic Review, 2, 41-71. Rosen, S. (1974). Hedonic prices and implicit markets: product differentiation in pure competition. The Journal of Political Economy, 82, 34-55. doi:10.1086/260169. Sirmans, G. S., Macpherson, D. A., & Zietz, E. N. (2005). The composition of hedonic pricing models. Journal of Real Estate Literature, 13, 3-43. Skelton, E. C. (2006). Laying the foundation for a mortgage industry in Mexico. Federal Reserve Bank of Dallas Economic Letter, 1(October), 1-8. Sosa, L. C. (2008). Detienen a tres personas en interior de auto robado. El Diario de Ciudad Jurez, (July 27) 4B. Troy, A., & Grove, J. M. (2008). Property values, parks, and crime: an hedonic analysis for Baltimore, MD. Landscape and Urban Planning, 87, 233-245. doi:10.1016/j.landurbplan.2008.06.005. Waddell, P. A., Berry, B. J. L., & Hoch, I. (1993). Residential property values in a multinodal urban area new evidence on the implicit price of location. The Journal of Real Estate Finance and Economics, 7, 117-141. doi:10.1007/BF01258322. Weber, R., Bhatta, S. D., & Merriman, D. (2007). Spillovers from tax increment financing districts: implications for housing price appreciation. Regional Science and Urban Economics, 37, 259-281. doi:10.1016/j.regsciurbeco.2006.11.003. White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48, 817-838. doi:10.2307/1912934