1# RECORD 154f55871a6811fb1dd94f5c020e97dabe79c8c32# TITLE: Efficient property value estimation for single-family homes in central Florida3# YEAR: 2024.0 VENUE: International Transactions in Operational Research DOI: 10.1111/itor.135654# AUTHORS: S. Lozano; E. Gutiérrez; Kotryna Klizentyte; A. Susaeta5# PAGES: 29 CANDIDATE TABLE PAGES (1-based): [10, 16, 17, 18, 19]6# ABSTRACT: This paper proposes the use of a metafrontier efficiency analysis approach to estimate residential property values. The objective is to mitigate inefficiencies in these valuations, thereby enhancing tax revenue and fairness in tax burdens. Property features (such as acreage, living area, number of bedrooms, number of bathrooms, age, and pool presence) along with contextual variables (such as county and median household income) are utilized to estimate just market value. Observations can be grouped based on some of these variables, allowing a metafrontier analysis to be conducted. This analysis not only assesses efficiency scores but also identifies technology gaps for each group. Furthermore, the proposed approach can be employed to estimate the value of any property based on its characteristics. To illustrate the proposed approach, we utilize data on over 30,000 properties in Orange and Seminole Counties in central Florida. Additionally, we present the results of a robust regression approach for comparison purposes. While our approach can serve as a valuable tool across various sectors, its primary aim is to promote best practices in governance regarding land use and real estate valuation. This includes balancing multiple objectives such as tax revenue, economic growth, environmental sustainability, and social considerations.789===== PAGE 1 (score 20, numeric-paren 0, table-words 1, var-words 18) =====1011Intl. Trans. in Op. Res. 0 (2024) 1–3012DOI: 10.1111/itor.1356513INTERNATIONAL14TRANSACTIONS15IN OPERATIONAL16RESEARCH17Efficient property value estimation for single-family homes in18central Florida19Sebastián Lozanoa20, Ester Gutiérreza,∗21, Kotryna Klizentyteb22and Andrés Susaetac23aDepartamento de Organización Industrial y Gestión de Empresas I, Universidad de Sevilla, Sevilla, Spain24bNatural Resource Economics and Policy, School of Forest, Fisheries, and Geomatics Sciences, University of Florida,25Gainesville, FL, USA26cDepartment of Forest Engineering, Resources and Management, Oregon State University, Corvallis, OR, USA27E-mail: slozano@us.es[Lozano]; egm@us.es [Gutiérrez]; kklizentyte@ufl.edu[Klizentyte];28andres.susaeta@oregonstate.edu[Susaeta]29Received 20 February 2024; received in revised form 17 July 2024; accepted 30 September 202430Abstract31This paper proposes the use of a metafrontier efficiency analysis approach to estimate residential property val-32ues. The objective is to mitigate inefficiencies in these valuations, thereby enhancing tax revenue and fairness33in tax burdens. Property features (such as acreage, living area, number of bedrooms, number of bathrooms,34age, and pool presence) along with contextual variables (such as county and median household income)35are utilized to estimate just market value. Observations can be grouped based on some of these variables,36allowing a metafrontier analysis to be conducted. This analysis not only assesses efficiency scores but also37identifies technology gaps for each group. Furthermore, the proposed approach can be employed to estimate38the value of any property based on its characteristics. To illustrate the proposed approach, we utilize data39on over 30,000 properties in Orange and Seminole Counties in central Florida. Additionally, we present the40results of a robust regression approach for comparison purposes. While our approach can serve as a valuable41tool across various sectors, its primary aim is to promote best practices in governance regarding land use42and real estate valuation. This includes balancing multiple objectives such as tax revenue, economic growth,43environmental sustainability, and social considerations.44Keywords: residential property value; local tax revenue; tax burden fairness; non-parametric metafrontier efficiency anal-45ysis; Free Disposal Hull (FDH)461. Introduction47Housing prices, representing the value at which a property is sold in the open market, serve as48crucial economic indicators (Gabrielli and French, 2021). These prices not only reflect the overall49∗Corresponding author.50© 2024 The Author(s).51International Transactions in Operational Research published by John Wiley & Sons Ltd on behalf of International Federation52of Operational Research Societies.53This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which54permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no55modifications or adaptations are made.565758===== PAGE 10 (score 25, numeric-paren 0, table-words 3, var-words 16) =====59601061S. Lozano et al. / Intl. Trans. in Op. Res. 0 (2024) 1–3062Table 163Descriptive statistics for the clean dataset (n = 30,555).64Min.65Max6625th67percentile68Median69Mean7075th71percentile72IQR73Std. dev74JMV7558,373761,477,34477217,94478276,35579312,51680361,97781144,03382149,651.883LIVARE8467285929886175787222488239489283190107491931.1392AGE (yrs)93494121952196349733.019843992210013.43101ACREAGE1020.0201035.9901040.2101050.2881060.4021070.4261080.2161090.401110MHHI11151,38611297,18511356,20911479,25211575,94311695,23911739,03011815,644119NBEDS1201.001217.001222.001233.001243.001254.001262.001270.92128NBATHS1291.001306.001313.001323.001333.361344.001351.001360.84137County138n139%140Orange141704314223%143Seminole14423,51214577%146Pool147n148%149Yes15015,57715151%152No15314,97815449%155Note: IQR (interquartile range) = 75th −25th percentile.156of those observations. Figure 3 illustrates the results, showing the dispersion plot of observed JMV157versus the deviation of observed JMV from estimated JMV values (in percentage). The visual158comparison between Figs. 3a,b reveals the improvement in the residuals of the regression model159after the removal of outliers, with the average absolute deviation decreasing from 11.49% to 9.43%.1603.2. Non-convex deterministic property valuation model161In this section, an FDH property valuation model is presented. Due to the discrete nature of162some variables, a non-convex DEA technology is assumed. The model assumes VRS to account163for scale size effects and considers the JMV of the property as the primary output variable,164while the property features, such as acreage, living area, number of bedrooms, and number of165bathrooms, are modeled as non-discretionary inputs. The median household income of the area166is also considered as a non-discretionary input as is the age of the property. This classification167of the non-discretionary variables as input or outputs is based on their expected effect on the168JMV output. Thus, it is assumed that, ceteris paribus, the larger the input variables the larger the169corresponding JMV should be and that, ceteris paribus, the older the property, the lower its JMV170should be. Following the approach outlined by Banker and Morey (1984), a property is deemed171inefficiently valued if another property with less acreage, less living area, and lower number of172bedrooms and bathrooms (and less median household income and a greater age) has a higher173JMV. The model computes, for a certain property and given its characteristics, the maximum JMV174it could be assigned based on the JMV of comparable properties in the sample. Furthermore, this175ability to compute the property value using as benchmarks specific existing observations (and not176a fictitious target computed as a convex combination of observed properties) is one of the strengths177of FDH.178© 2024 The Author(s).179International Transactions in Operational Research published by John Wiley & Sons Ltd on behalf of International Federation180of Operational Research Societies.181 14753995, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/itor.13565 by Readcube (Labtiva Inc.), Wiley Online Library on [17/10/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License182183184===== PAGE 11 (score 14, numeric-paren 0, table-words 0, var-words 14) =====185186S. Lozano et al. / Intl. Trans. in Op. Res. 0 (2024) 1–3018711188Before formulating the model, let us introduce the notation:189Data190J191Set of properties192ACREAGE j193Acreage of property j ∈J194LIVAREAj195Living area (air-conditioned) of property j ∈J196NBEDS j197Number of bedrooms of property j ∈J198NBATHS j199Number of bathrooms of property j ∈J200MHHIj201Median household income of the area of property j ∈J202AGE j203Age of property j ∈J204JMVj205Just Market Value of property j ∈J2060207Index of property whose JMV efficiency is being assessed208Decision variables209(λ1, λ2, . . . , λn)210Binary variables used to compute the target JMV of Property 0211TARGETJMV0212Target JMV of Property 0213ξ0214JMV efficiency of Property 0215FDH model216Maximize217TARGETJMV0218s.t.219220j∈J221λ j · ACREAGE j ≤ACREAGE0222223j∈J224λ j · LIVAREAj ≤LIVAREA0225226j∈J227λ j · NBEDSj ≤NBEDS0228229j∈J230λ j · NBATHSj ≤NBATHS0231232j∈J233λ j · MHHIj ≤MHHI0234235j∈J236λ j · AGE j ≥AGE0237238j∈J239λ j · JMVj ≥TARGETJMV0240241j∈J242λ j = 1243λj ∈{0, 1} ∀j ∈J.244(1)245Once the target JMV of DMU 0 is determined, the corresponding JMV efficiency score can be246computed as247ξ0 =248JMV0249TARGETJMV ∗2500251.252(2)253Since the solution λ0 = 1, λ j = 0 ∀j ̸= 0, TARGETJMV0 = JMV0 is feasible in (1), it follows254that the optimal solution TARGETJMV ∗2550 ≥JMV0, and, hence, ξ0 ≤1. The JMV undervaluation256© 2024 The Author(s).257International Transactions in Operational Research published by John Wiley & Sons Ltd on behalf of International Federation258of Operational Research Societies.259 14753995, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/itor.13565 by Readcube (Labtiva Inc.), Wiley Online Library on [17/10/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License260261262===== PAGE 16 (score 14, numeric-paren 0, table-words 4, var-words 3) =====26326416265S. Lozano et al. / Intl. Trans. in Op. Res. 0 (2024) 1–30266Table 2267Average group and metafrontier just market value (JMV) undervaluation for each group and for the whole sample.268County269Pool270No. of271properties272Avg. JMV0273Avg. JMV SLACKg02740275US$ (also in %)276Avg. JMV SLACKM2770278US$ (also in %)279Seminole280No28110,643282282,79928343,249 (16.2%)28449,116 (18.2%)285Seminole286Yes28712,869288371,20828980,753 (22.4%)29097,620 (27.4%)291Orange292No2934335294223,14529528,803 (12.8%)29683,220 (35.6%)297Orange298Yes2992708300293,45630143,369 (15.0%)302104,303 (34.6%)303ALL304No30514,978306265,53430739,068 (15.2%)30858,986 (23.2%)309ALL310Yes31115,577312357,69131374,254 (21.1%)31498,782 (28.6%)315Seminole316ALL31723,512318331,18931963,776 (19.6%)32075,664 (23.2%)321Orange322ALL3237043324250,18032534,403 (13.6%)32691,326 (35.2%)327ALL328ALL32930,555330312,51633157,006 (18.2%)33279,274 (26.0%)333Table 3334Average technology gap ratio (TGR) and group and metafrontier JMV efficiency for each group and for the whole sample.335County336Pool337ξ g338ξ g,M339TGRg340Seminole341No3420.8783430.8663440.982345Seminole346Yes3470.8423480.8103490.964350Orange351No3520.8953530.7543540.822355Orange356Yes3570.8783580.7603590.847360ALL361No3620.8833630.8333640.939365ALL366Yes3670.8483680.8013690.946370Seminole371ALL3720.8583730.8353740.971375Orange376ALL3770.8883780.7563790.833380ALL381ALL3820.8653830.8173840.943385instead of a simple arithmetic average, the single-output case of the TGR aggregation method pro-386posed in Walheer (2018) has been used. As expected, the group efficiency scores are always larger387than the corresponding metafrontier efficiencies (0.865 vs. 0.817, for the whole sample), and, simi-388larly, the average TGR are consistently less than (or at most equal to) unity. These patterns persist389throughout the table. Consequently, on average, both in the group and metafrontier projections,390properties with pools demonstrate lower JMV efficiency, compared to properties without pools391(0.848 vs. 0.883 in one case, and 0.801 vs. 0.833 in the other). However, while the group frontier392projection indicates lower JMV efficiency for properties with pools in both counties, the metafron-393tier projection reveals this trend only in Seminole County, not in Orange County. Additionally, it394is evident that the average group efficiency of all properties (with and without pools) in Seminole395County is lower than in Orange County, while the opposite holds true for the average metafrontier396efficiency.397Regarding TGR, this measures the undervaluation of the best-valued properties within a given398group relative to the best-valued comparable properties across all groups. A TGR value closer to399unity signifies relatively better-valued properties within that group. As per Table 3, the distance400between the group efficient frontier and the metafrontier, aggregated for all properties, yields a401TGR value of 0.943. This value aligns with the aggregate TGR of both with-pool and without-pool402© 2024 The Author(s).403International Transactions in Operational Research published by John Wiley & Sons Ltd on behalf of International Federation404of Operational Research Societies.405 14753995, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/itor.13565 by Readcube (Labtiva Inc.), Wiley Online Library on [17/10/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License406407408===== PAGE 17 (score 63, numeric-paren 21, table-words 13, var-words 11) =====409410S. Lozano et al. / Intl. Trans. in Op. Res. 0 (2024) 1–3041117412Table 4413Results of the stochastic frontier analysis.414Variables415Coefficient416Standard error417Variables418Coefficient419Standard error420Production frontier421Inefficiency422log(LIVAREA)4230.8086***4240.0032425County4260.3097***427(0.0033)428log(NBEDS)4290.0480***4300.0032431Pool432−0.0575***433(0.0032)434log(NBATHS)435−0.0615***436(0.0038)437Constant4380.0262***439(0.0034)440log(MHHI)4410.2214***442(0.0037)443σ u4440.0116**4450.0059446log(ACREAGE)4470.0750***448(0.0013)449σ v4500.1246***4510.0059452log(AGE)453−8***454(0.0024)455λ = σ u/σ v4560.0929***4570.0063458Constant4594.2852***460(0.0440)461Observations46230,555463Wald chi2(6) = 253,536.27***464Log-likelihood = 20,211.15465∗∗Significant at 0.05;466∗∗∗Significant at 0.01 level467Table 5468Technical efficiency (TE) scores.469Technical efficiency470Mean471SD472Max473Min474TE4750.92744760.10844770.99694780.7117479TE (Seminole)4800.98624810.01134820.99694830.9686484TE (Orange)4850.73104860.02064870.76024880.7117489TE (with pool)4900.89884910.11754920.97864930.7117494TE (without pool)4950.95494960.09074970.99694980.7540499groups, indicating no significant overall differences between the two groups when within-group un-500dervaluation is factored out. Although the with-pool group marginally outperforms the without-501pool group in Orange County, this is offset by the reverse situation in Seminole County. Nonethe-502less, the impact of County location on property undervaluation appears to outweigh the presence of503pools. Thus, across all properties, including those with and without pools, the best-valued proper-504ties in Seminole County appear to be superior to the best-valued comparable properties in Orange505County.506The results of the SFA estimation are shown in Table 4. All the regressors are significant at507the 0.01 level. The signs of the coefficients are as expected, except in the case of the number of508bathrooms, which seems to have a small negative effect on the JMV.509The efficiency scores computed by SFA, shown in Table 5, indicate an average value of 0.927510for the whole sample. Contrary to the DEA results, properties without pools show higher efficiency511scores than those properties with pools (0.955 vs. 0.898). It can be also seen that the efficiency scores512in Seminole County are higher than those in Orange County (0.986 vs. 0.7310), which is consistent513with the TGR results mentioned above.514© 2024 The Author(s).515International Transactions in Operational Research published by John Wiley & Sons Ltd on behalf of International Federation516of Operational Research Societies.517 14753995, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/itor.13565 by Readcube (Labtiva Inc.), Wiley Online Library on [17/10/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License518519520===== PAGE 18 (score 53, numeric-paren 20, table-words 11, var-words 7) =====52152218523S. Lozano et al. / Intl. Trans. in Op. Res. 0 (2024) 1–30524Table 6525Ordinary least squares coefficient estimates with robust standard estimators and performance accuracy metrics.526Coefficients527Estimate528(std. error)a529CI lower530CI upper531INTERCEPT5324.641****533(0.049)5344.5445354.737536log(LIVAREA)5370.803***538(0.004)5390.7965400.810541log(ACREAGE)5420.078***543(0.001)5440.0755450.080546log(MHHI)5470.222***548(0.004)5490.2145500.230551log(NBEDS)5520.042***553(0.003)5540.0345550.049556log(NBATHS)557−0.060***558(0.004)559−0.067560−0.052561log(AGE)562−0.173***563(0.002)564−0.177565−0.168566County567−0.308***568(0.003)569−0.314570−0.301571Pool5720.024***573(0.002)5740.0205750.028576County*Pool5770.035***578(0.003)5790.0295800.042581Goodness-of-fit statistics582Adjusted R2 = 0.911583AIC = −40553584BIC = −40441585RMSE = 0.124586Note: Dependent variable JMV; number of observations = 30,555.587aHeteroskedasticity-consistent covariance matrix estimator (type H2C).5884.2. Comparison with regression analysis589Before presenting the results of the log–log regression, let us mention that outliers/extreme values590were removed because they increase predictive deviations between the observed and fitted JMV.591Also, to decide the specification of the regression, that is, a linear model (linear function of re-592pressors) versus a log-linear model (linear function of logs repressors), the test proposed in Koop593(2008) was implemented. The results confirmed the rejection of the hypothesis that the linear and594log–log model perform equally well. Moreover, since the residual sum of squared (RSS), RSSlinear595= 1036.88>RSSlog-log = 471.75, the log–log model seems preferable to the linear model at 1% sig-596nificance level (lambda statistic = 12,031.5; p < 0.01).597To address the challenges posed by heavy-tailed error distribution, heteroskedasticity in error598variance, and extreme values, a robust estimation procedure was employed. This procedure utilizes599robust standard estimators within the OLS estimation framework (Blair et al., 2022). Additionally,600robust regression via M-estimation, employing the iterative re-weighted least squares estimation601method based on the Huber function, was applied (Venables and Ripley, 2002). Both approaches602yielded similar results. The OLS robust regression coefficients for the log–log model are presented603in Table 6. Note that not only are all the regression coefficients individually highly statistically604© 2024 The Author(s).605International Transactions in Operational Research published by John Wiley & Sons Ltd on behalf of International Federation606of Operational Research Societies.607 14753995, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/itor.13565 by Readcube (Labtiva Inc.), Wiley Online Library on [17/10/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License608609610===== PAGE 19 (score 25, numeric-paren 1, table-words 5, var-words 12) =====611612S. Lozano et al. / Intl. Trans. in Op. Res. 0 (2024) 1–3061319614significant (based on the F-statistic F_value = 2.937e+04; p < 0.01), but it can also be concluded615that jointly the nine variables are highly statistically significant. The interpretation of the estimated616coefficients reveals that the most influential variable affecting the JMV is the LIVAREA so that the617coefficient of Log(LIVAREAi) is 0.803, that is, an increase of 1% in LIVAREA increases the JMV,618on average, ceteris paribus, about 0.8%. The second most prominent variable is determined by the619structural characteristic of the county location so that the coefficient of the County is −0.308, that620is, the average Orange County property prices are 26.5% lower (e−0.308 −1 = −0.265), compared621to Seminole County property prices, if all other variables are kept constant. The presence of a pool622in the property seems to be the least influential variable, that is, the average JMV increases by just6232.4% (e0.024 −1 = 0.024) if the property has a pool.624Other explanatory variables with an impact on house prices are MHHI (an increase of 1% in625MHHI increases, on average, ceteris paribus, the JMV by 0.22%. Also, as expected, the age of the626property has a negative effect on JMV (−0.173% for an 1% increase in AGE), and the size of the627property has a positive effect on JMV (0.081% for a 1% increase in ACREAGE). On the other hand,628the lower the number of bathrooms and the higher the number of bedrooms, the higher the JMV,629with a slightly higher effect of the NBATHS variable (−0.06% vs. 0.04%). The interaction dummy630variable also has a highly significant effect on JMV. Thus, on average, a property located in Orange631County has a lower average JMV (−26.5%), and, on average, the presence of a pool increases JMV632by about 2.4%. And when both circumstances concur, the overall effect on the JMV is a reduction633in the JMV of 22.0% (e−0.308+0.024+0.035 −1 = e0.249 −1 = −0.220). Finally, the model’s accuracy634metrics (Adjusted R2, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC),635Root Mean Square Error (RMSE)) support its validity.636Note that the results from the SFA estimation (Table 4) show similar outcomes in terms of statis-637tical significance and the magnitude of the common coefficients. For example, a 1% increase in the638age of the property reduces the JMV by 0.196%, while a 1% increase in the number of bathrooms639decreases the JMV by 0.0615%. In the SFA model, we use the location of houses and the presence640of pools as factors that explain inefficiency, with the results indicating that properties with pools641lead to an increase in the technical efficiency of the property appraisal, while properties located in642Seminole County reduce the technical efficiency of the property appraisal.643Table 7 shows, for each group pf observations and for the whole sample, the comparison between644the JMV fitted by the regression model (labeled FITTEDJMV reg) and: (a) the observed JMV,645(b) the group frontier target JMV, and (c) the metafrontier target JMV. Note that, since for each646Property 0 it holds that JMV0 ≤TARGETJMV g06470648≤TARGETJMV M6490 , it follows that these650three ratios are decreasing, that is,651FITTEDJMV reg6520653JMV0654≥655FITTEDJMV reg6560657TARGETJMV658g06590660≥661FITTEDJMV reg6620663TARGETJMV M6640 . Moreover, since665regression captures the central tendency, that is, it estimates average values instead of best values666as DEA does, the ratios667FITTEDJMV reg6680669TARGETJMV670g06710672and more so673FITTEDJMV reg6740675TARGETJMV M6760677are below unity, with overall678averages of 0.866 and 0.826, respectively. The average ratio679FITTEDJMV reg6800681JMV0682is, understandably, close683to unity (1.010), with very small differences between the different groups. The relative differences in684the average value of685FITTEDJMV reg6860687TARGETJMV688g06890690and691FITTEDJMV reg6920693TARGETJMV M6940695between the different groups are a reflection696of the corresponding group and metafrontier target JMV mentioned above. For example, since697on average, in both the group and metafrontier projections, those properties with a pool are more698© 2024 The Author(s).699International Transactions in Operational Research published by John Wiley & Sons Ltd on behalf of International Federation700of Operational Research Societies.701 14753995, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/itor.13565 by Readcube (Labtiva Inc.), Wiley Online Library on [17/10/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License702703704===== PAGE 20 (score 16, numeric-paren 0, table-words 1, var-words 12) =====70570620707S. Lozano et al. / Intl. Trans. in Op. Res. 0 (2024) 1–30708Table 7709Comparison of regression JMV estimates and group and metafrontier targets.710County711Pool712Avg.713FITTEDJMV reg7140715JMV0716Avg.717FITTEDJMV reg7180719TARGETJMV g07200721Avg.722FITTEDJMV reg7230724TARGETJMV M7250726Seminole727No7281.0137290.8817300.869731Seminole732Yes7331.0077340.8407350.807736Orange737No7381.0127390.8987400.754741Orange742Yes7431.0097440.8797450.758746ALL747No7481.0137490.8867500.835751ALL752Yes7531.0077540.8477550.798756Seminole757ALL7581.0107590.8587600.835761Orange762ALL7631.0117640.8917650.755766ALL767ALL7681.0107690.8667700.816771undervalued than properties without one, both ratios772FITTEDJMV reg7730774TARGETJMV775g07760777and778FITTEDJMV reg7790780TARGETJMV M7810782are lower for783properties without a pool than for properties with one. Similarly, in the group frontier projection,784properties (including with and without pool) are, on average, more undervalued in Seminole County785than in Orange County, while the opposite happens in the metafrontier projection. Therefore, the786ratio787FITTEDJMV reg7880789TARGETJMV790g07910792is lower for properties in Seminole County than in Orange County, while, on the793contrary, the ratio794FITTEDJMV reg7950796TARGETJMV M7970 is lower for properties in Orange County than in Seminole County.7984.3. Using the DEA-based property value estimation function799As discussed in Section 3.4, the proposed FDH model serves as an estimation function for the JMV800of a property based on its characteristics and group affiliation. Furthermore, it can also estimate801and compare the values of similar properties across different groups. To demonstrate the utility802and effectiveness of this capability, we examine how the JMV of a property varies according to803its characteristics, county of location, and the presence or absence of a pool. Consider a property804with median acreage (0.288 acres), median living area (2224 sq ft), median number of bedrooms805(3), median number of bathrooms (3), median MHHI value ($79,252), and median age (between80630 and 40 years). Figures 4–6 depict the corresponding JMV of this property, estimated for the807four groups considered, while varying one property characteristic at a time and keeping the others808at their median values. It is important to note that, apart from the qualitative trends discussed809below, the key strength of using the proposed DEA-based estimation function lies in its ability to810provide quantitative estimates derived from the best-valued similar properties within its group. By811utilizing the estimation function for a different group, it is also possible to estimate and compare812the maximum JMV value of a comparable property in that group.813Note, first of all, the monotonic, non-concave character of the curves, typical of FDH technology.814In each plot, the metafrontier estimation JMV M is represented by the non-concave curve (not815labeled in the figures) that envelops the curves of all four groups from above. Regarding the living816area of the property, it is evident that an increase in the living area leads to a higher estimated JMV817© 2024 The Author(s).818International Transactions in Operational Research published by John Wiley & Sons Ltd on behalf of International Federation819of Operational Research Societies.820 14753995, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/itor.13565 by Readcube (Labtiva Inc.), Wiley Online Library on [17/10/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License821