1# RECORD 122e07c191e40cdc3321c3817fb731f2d55466e12# TITLE: PATH SELECTION OF SPATIAL ECONOMETRIC MODEL FOR MASS APPRAISAL OF REAL ESTATE: EVIDENCE FROM YINCHUAN, CHINA3# YEAR: 2023.0 VENUE: International Journal of Strategic Property Management DOI: 10.3846/ijspm.2023.203764# AUTHORS: Yu Zhao; Xuejia Shen; Jian Ma; Miao Yu5# PAGES: 13 CANDIDATE TABLE PAGES (1-based): [7, 8, 9, 10, 11]6# ABSTRACT: Urbanization, national economic growth, and China’s changing population structure have elevated the importance of real estate assessment in various contexts, including mortgage financing, secondary housing market transactions, and real estate tax reform. To address this need, this study employs a time-spatial double-fixed spatial cross-section data model as a mass appraisal tool to analyze the transaction price data of 429 ordinary residential houses in Xixia District, Yinchuan, China on April 1, 2022. Specifically, this study analyzes 7 spatial cross-section data models, discerning their interconnections. It devises an assignment technique that merges distance and characteristic variable rank into a unified indicator. The results explore spatial lag effects in real estate transaction price generation and assess the descriptive capabilities of different spatial cross-section data models.789===== PAGE 1 (score 8, numeric-paren 1, table-words 0, var-words 7) =====1011Copyright © 2023 The Author(s). Published by Vilnius Gediminas Technical University12This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unre-13stricted use, distribution, and reproduction in any medium, provided the original author and source are credited.14Introduction15The COVID-19 outbreak has affected China’s real estate 16market, leading to reduced new construction and a rise 17in second-hand housing transactions (Song, 2021). This 18has posed a challenge to the implementation of the local 19government’s “land finance” model. On October 23, 2021, 20the “Decision to carry out pilot work on real estate tax 21reform in some areas” was ratified during the 31st meeting 22of the Standing Committee of the National People’s Con-23gress (NPC). Subsequently, since 2022, numerous cities, 24Yinchuan City included, have introduced policies aimed 25at easing property purchase restrictions. In light of the 26demand for property tax collection and the continuous 27growth of the real estate market, the accurate assessment 28of real estate values has gained heightened significance. 29As a result, there has been a growing need for efficient 30and scientific real estate valuation in the financial market, 31tax reform, and the real estate trading market. The mass 32appraisal of real estate is a key area of current real estate 33research, with a focus on providing a value reference and 34pricing basis for buyers and sellers in the secondary hous-35ing market. It also serves as a framework for appropriate 36investment, mortgage, and pricing for developers and fi-37nancial institutions, and a basis for rational and scientific 38International Journal of Strategic Property Management39ISSN: 1648-715X / eISSN: 1648-9179402023 Volume 27 Issue 5: 304–31641https://doi.org/10.3846/ijspm.2023.2037642*Corresponding author. E-mail: shenxuejia6@163.com43PATH SELECTION OF SPATIAL ECONOMETRIC MODEL FOR MASS 44APPRAISAL OF REAL ESTATE: EVIDENCE FROM YINCHUAN, CHINA45Yu ZHAO, Xuejia SHEN*, Jian MA, Miao YU46School of Management, Shenyang Jianzhu University, Shenyang, Liaoning, China47Received 19 May 2023; accepted 1 November 202348Abstract. Urbanization, national economic growth, and China’s changing population structure have elevated the impor-49tance of real estate assessment in various contexts, including mortgage financing, secondary housing market transactions, 50and real estate tax reform. To address this need, this study employs a time-spatial double-fixed spatial cross-section data 51model as a mass appraisal tool to analyze the transaction price data of 429 ordinary residential houses in Xixia District, 52Yinchuan, China on April 1, 2022. Specifically, this study analyzes 7 spatial cross-section data models, discerning their 53interconnections. It devises an assignment technique that merges distance and characteristic variable rank into a unified 54indicator. The results explore spatial lag effects in real estate transaction price generation and assess the descriptive capa-55bilities of different spatial cross-section data models.56Keywords: mass appraisal of real estate, spatial section data model, SDM, empirical analysis.57Online supplementary material: Supporting information for this paper is available as online supplementary material at 58https://doi.org/10.3846/ijspm.2023.2037659urban planning by local governments. Additionally, it pro-60vides a source for evaluating the real estate tax base. Mass 61appraisal, as defined by the International Association of 62Appraisal Officers (IAAO, 2012), has been widely used to 63evaluate a collection of properties at a certain date using 64shared data, standardized methodologies, and statistical 65testing. However, the mass appraisal of real estate faces 66challenges due to inconsistent criteria for data use and the 67inadaptability of the setting and assignment of character-68istic variables. The absence of a unified database and sys-69tem of characteristic variables in the hedonic price model 70and other approaches exacerbates these issues (Chen et al., 712020). Additionally, the geographical characteristics of real 72estate as a commodity are often ignored by the hedonic 73price model, which emphasizes the variety of commodi-74ties (Feng et al., 2019). Therefore, there are concerns about 75the accuracy and flexibility of the estimation of house 76costs using mass appraisal methods. Several studies have 77utilized the average transaction price of residential units 78or the transaction price of a single unit as a sample in 79the empirical examination of real estate mass appraisal. 80However, the data distribution is often unequal due to the 81varied transaction activity of different residential areas 82at a given time, which directly affects the accuracy of re-83gression analysis. In addition, the impact of high-quality 848586===== PAGE 7 (score 18, numeric-paren 0, table-words 2, var-words 12) =====878831089Y. Zhao et al. Path selection of spatial econometric model for mass appraisal of real estate: evidence...90Table 2. Selection and quantification standard of characteristic variables91Feature item92Symbol93Quantization standard94Assignment method95Expected 96impact97Area98Ar99Building area100Actual value101+102Decoration103D104Blank 2, simple 4, medium 6, hardcover 8, luxury 10105Virtual assignment106+107Floor108F109High-rise residence: low 1, middle 3, high 2110Virtual assignment111unknown112Low rise house: low 3, medium 2, high 1113Elevator/househols114EH115Number of elevators on the floor/number of single properties on 116the floor117Actual value118+119Age120Ag121Year the house was built – 2022.4.1122Actual value123–124Greening rate125GR126Green area/residential area127Actual value128+129Parking space ratio130PSR131Total number of households/parking spaces132Actual value133+134Public activity area 135for residents136PAA137Number of areas138Actual value139+140Property fee141PF142Monthly charge per square meter143Actual value144unknown145Orientation146O147Reasonable = 1, unreasonable = 0148Actual situation149+150CBD151CBD152Geographical distance153Anti-geographical 154distance155+156Shopping mall157SM158Service capability: municipal = 3, district 159level = 2, street level = 1160Service capacity/161geographical distance162+163Geographical distance164Main road165MR166Geographical distance167Anti-geographical 168distance169unknown170Bus171B172Bus route number173Quantity/distance174+175The geographical distance to the bus stop176Medical treatment177MT178Qualifications: Grade III Level Special = 10, Grade III Level A = 1799, Grade III Level B = 8, Grade III Level C = 7, Grade II Level 180A = 6, Grade II Level B = 5, Grade II Level C = 4181Qualifications/182geographical distance183+184Geographical distance185Education186E187Qualifications: high quality = 3, good = 2, ordinary = 1188Qualifications/189geographical distance190+191Geographical distance192Developer brand193DB194High quality = 3, good = 2, average = 1195Virtual assignment196+197Development zone198DZ199Rating: national = 3, provincial = 2, municipal = 1200Rating/geographical 201distance202+203Geographical distance204Park205P206Geographical distance207Anti-geographical 208distance209+210Table 3. The description of variable values211Feature item212Value range213Maximum value214Minimum value215Mean value216Area21740–2002181632194422095221Decoration2222–10223822422255226Floor2271–4228422912302231Elevator/households2320–1233123402350.25236Age2371–3523832239224015.5241Greening rate24215–50%24345%24415%24528%246Parking space ratio2470–12480.82490.12500.4251Property fee2520–125312540.252550.65256Orientation2570–1258125902600.8261CBD262–2630.2042640.08332651.01266Shopping mall267–26818.752690.532702.34271Main road272–27314.292740.762753.40276Bus277–278173.912792.0828030.31281Medical treatment282–283902842.7328512.05286Education287–28813202894.57290169291Developer brand2921–6293629412953.5296Park297–298202990.533002.34301302303===== PAGE 8 (score 16, numeric-paren 0, table-words 6, var-words 2) =====304305International Journal of Strategic Property Management, 2023, 27(5): 304–3163063113074. Results and discussion3084.1. The spatial dependency diagnosis of the 309evaluation area310The spatial weight matrix using the K-adjacency matrix is 311constructed using GeoDa1.20. The number of neighbors 312is set to 16 based on the sample’s actual situation, location 313similarity, neighborhood characteristics, and practical ex-314perience of the market method in single-case assessment.3154.1.1. Global spatial autocorrelation test316The global autocorrelation test was conducted using the 317single variable global Moran’s I test tool in GeoDa1.20, 318and the results of the significance test obtained through 319999 permutations are shown in Figure 5. It can be ob-320served that the global Moran’s I index is approximately 3210.4858 under the 0.001 significance level test, which pro-322vides strong evidence for the need of spatial econometric 323analysis.324Figure 5. Global autocorrelation test results3254.1.2. Local spatial autocorrelation326The local autocorrelation test was conducted using the 327univariate local Moran’s I test tool in GeoDa1.20, and the 328resulting Moran’s I scatter plot is displayed in Figure 6. 329The scatter plot reveals the existence of both high-high 330and low-low spatial adjacency effects within the sample 331group.332Figure 6. Local Moran’s I scatter diagram333LISA cluster diagram is shown in Figure 7, and the 334distribution of four spatial clustering phenomena can be 335intuitively observed. 3364.2. Setting and testing of spatial section data model337Before constructing the spatial cross-section data mod-338el, it’s imperative to subject the characteristic variables 339and variable function form to relevant statistical tests 340for validation. These tests encompass evaluating the 341goodness-of-fit for linear, semi-logarithmic, and full-342logarithmic models, examining multicollinearity and 343heteroscedasticity, assessing the significance of variables, 344and computing AIC and log likelihood. Table 4 shows 345the results of these tests reveal: (1) all three forms pass 346the F-statistical test, signifying substantial explanatory 347variable significance; (2) only the full logarithmic model 348exhibits significant multicollinearity and heteroscedastic-349ity; and (3) the semi-logarithmic form outperforms the 350linear model in terms of AIC, log likelihood, and adjust-351ed R-squared. Consequently, the optimal selection is the 352semi-logarithmic function form. Specific variables and 353their associated assignment methods under this form are 354meticulously outlined in Table 5.355Table 4. Inspection and selection results356Test357Semi log358Full log359Linear360Log likelihood361375.449362362.16363–3359.46364AIC365–728.898366–702.323676740.92368Multicollinearity condition 369number37028.4837196.7237228.48373Adjusted R-squared3740.82633750.81523760.8113377Breusch-Pagan test3780.057273790.006473800.06454381F-statistic3820.000003830.000123840.00000385Koenker-Bassett test3860.096933870.002073880.05673389Number of significant 390variables391103921139310394Note: The output results of the Breusch-Pagan test, Koenker-Bassett test 395and F-statistic in the table are the adjoint probabilities of the test results.396Figure 7. LISA cluster diagram397398399===== PAGE 9 (score 83, numeric-paren 48, table-words 8, var-words 1) =====400401312402Y. Zhao et al. Path selection of spatial econometric model for mass appraisal of real estate: evidence...403Table 5. Significant variables and assignment methods404Variable405t-Statistic406Probability407Assignment method408Ar40911.82180 4100.00000 411Actual value412D4138.00813 4140.00000 415Actual value416F4174.94734 4180.00000 419Virtual assignment420EH4214.64670 4220.00000 423Actual value424Ag425–15.09490 4260.00000 427Actual value428CBD42910.10060 4300.00000 431Anti-geographical distance432SM4335.46333 4340.00000 435Service capacity/436geographical distance437MR438–4.24358 4390.00003 440Anti-geographical distance441P4423.17022 4430.00164 444Anti-geographical distance445B4464.59475 4470.00001 448Quantity/distance4494.3. Comparative analysis of model estimates4504.3.1. Summary of model estimation and test results451The results of LM test and Robust LM test of residuals 452based on the regression results of classical linear models 453are shown in Table 6. 454Table 6. LM inspection and robust LM inspection results455Test456Value457Probability458LM (lag)45981.78094600.0000461LM (error)462138.93284630.0000464Robust LM (lag)46530.01954660.0000467Robust LM (error)46887.17144690.0000470The results of R-squared, Log-likelihood, AIC and 471LR test of each model are shown in Table 7. In the 472first test of LR test (LR1), the constraint models cor-473responding to SLX, SLM, and SEM is CLR, the con-474straint models corresponding to SDM, SDEM, SAC, and 475GNS are SLX, SLX, SLM, and SEM respectively, and the 476constraint models corresponding to SDM, SDEM, and 477SAC in the second test (LR2) are SLM, SEM, and SEM 478respectively. 479The regression results of each model are shown in Ta-480ble 8.481Table 7. Summary of model test results482Variable483CLR484SLX485SLM486SEM487SDM488SDEM489SAC490GNS491R24920.82634930.83894940.87284950.90334960.90374970.90414980.90694990.8909500Log-likelihood501375.449502386.575503431,617504467.324505473.693506469.429507467.5302508469.869509AIC510–728.898511–745.149512–839.235513–912.648514–913.386515–906.858516–910.164517–902.126518LR15190.00005200.00005210.00005220.00005230.00005240.00005250.1032526LR25270.00025280.09325290.0740530Table 8. Summary of model estimates531Variable532CLR533SLX534SLM535SEM536SDM537SDEM538SAC539GNS540t(P)541t(P)542t(P)543t(P)544t(P)545t(P)546t(P)547t(P)548Ar54911.8255010.2155112.555213.0655313.0555412.8755512.9155612.84557(0.000)558(0.000)559(0.000)560(0.000)561(0.000)562(0.000)563(0.000)564(0.000)565D56685677.735688.635699.195708.885716.765729.115735.93574(0.000)575(0.000)576(0.000)577(0.000)578(0.000)579(0.000)580(0.000)581(0.000)582F5834.945845.325855.355865.935875.935885.425895.915905.38591(0.000)592(0.000)593(0.000)594(0.000)595(0.000)596(0.000)597(0.000)598(0.000)599EH6004.6460146024.296035.456045.226055.436065.426075.46608(0.000)609(0.000)610(0.000)611(0.000)612(0.000)613(0.000)614(0.000)615(0.000)616Ag617–15.09618–11.75619–12.79620–16.25621–15.74622–15.85623–15.97624–15.87625(0.000)626(0.000)627(0.000)628(0.000)629(0.000)630(0.000)631(0.000)632(0.000)633B6344.596354.876363.786372.66382.996392.516402.666412.29642(0.000)643(0.000)644(0.000)645(–0.009)646(–0.003)647(–0.012)648(–0.008)649(–0.022)650MR651–4.24652–3.88653–2.97654–2.88655–2.68656–2.97657–2.86658–2.88659(0.000)660(0.000)661(–0.003)662(–0.004)663(–0.007)664(–0.003)665(–0.004)666(–0.004)667SM6685.4666956703.586710.76721.946730.566740.816750.17676(0.000)677(0.000)678(0.000)679(–0.485)680(–0.052)681(–0.576)682(–0.417)683(–0.864)684685686===== PAGE 10 (score 47, numeric-paren 17, table-words 11, var-words 0) =====687688International Journal of Strategic Property Management, 2023, 27(5): 304–316689313690Variable691CLR692SLX693SLM694SEM695SDM696SDEM697SAC698GNS699t(P)700t(P)701t(P)702t(P)703t(P)704t(P)705t(P)706t(P)707CBD70810.17099.437105.357113.087123.067133.047143.157152.35716(0.000)717(0.000)718(0.000)719(–0.002)720(–0.002)721(–0.002)722(–0.002)723(–0.019)724P7253.177263.127273.387282.97292.617302.797312.937322.73733(0.002)734(0.002)735(–0.001)736(–0.004)737(–0.009)738(–0.005)739(–0.003)740(–0.006)741W_Ar7421.57743–3.327441.487451.83746(0.116)747(–0.001)748(–0.14)749(–0.067)750W_EH751–0.18752–2.327530.477540.75755(0.861)756(–0.021)757(–0.639)758(–0.451)759W_Ag760–0.097616.077620.94763–0.24764(0.930)765(0.000)766(–0.350)767(–0.813)768W_F7692.42770–0.527711.27721.43773(0.016)774(–0.606)775(–0.231)776(–0.153)777W_D7783.83779–1.41780–0.66781–0.6782(0.000)783(–0.159)784(–0.507)785(–0.546)786ρ78710.9678816.227890.65790–1.13791(0.000)792(0.000)793(–0.513)794(–0.259)795λ79618.8179717.6479811.5679912.04800(0.000)801(0.000)802(0.000)803(0.000)804Note: where: W – spatial lag term of the corresponding variable; ρ – dependent variable spatial regression coefficient; λ – error term spatial regression 805coefficient.806End of Table 88074.3.2. Comparative analysis of appraisal models808The construction path selected based on the optimal mod-809el of ordinary residential real estate (Figure 2) and the test 810results (Tables 6 and 7) is shown in Figure 8. 811Step 1: Based on Tables 7 and 8, the significance of 812the spatial lag variable in SLX is observed, indicating that 813SLX outperforms CLR and cannot be downgraded. Thus, 814including the spatial lag effect of the dependent variable 815in the model setting is necessary.816Step 2: Table 6 shows that the dependent variable and 817the error term have passed the LM test and the Robust 818LM test, and the spatial lag effect of the error term is more 819robust than that of the dependent variable (SEM is better 820than SLM).821Step 3: Table 8 reveals that SDM is superior in terms 822of R-squared, log-likelihood, and AIC test results, and it 823cannot be reduced to a single variable model (SLM, SLX). 824Thus, the model can effectively identify the spatial lag 825effect of independent and dependent variables. Table 9 826Figure 8. Model building path (slt = Spatial lag term; iv = independent variable)827828829===== PAGE 11 (score 40, numeric-paren 0, table-words 15, var-words 3) =====830831314832Y. Zhao et al. Path selection of spatial econometric model for mass appraisal of real estate: evidence...833shows that the two spatial lag variables of SDM are signifi-834cant, while only the spatial lag of the error term is signifi-835cant in SAC, SDEM, and GNS. Therefore, if insignificant 836variables are removed from the model, the three models 837will degenerate into SEM models. SDM is capable of ac-838curately describing the impact of neighboring real estate 839prices and characteristics on real estate prices during data 840generation as there is no significant variable competition 841between the dependent variable spatial lag term and the 842independent variable spatial lag term. However, variable 843competition exists between the error term spatial lag term 844and the dependent variable and independent variable spa-845tial lag term, resulting in the regression analysis of SAC, 846SDEM, and GNS lacking spatial lag essence. Additionally, 847the over-parameterization of GNS significantly reduces its 848overall analytical ability.8494.3.3. Optimization of model based on mass appraisal 850of real estate 851It is not difficult to see from the comprehensive compara-852tive analysis that the comparison between SDM model 853and SAC model, SDEM model, and GNS model can be 854essentially understood as a comparative analysis between 855SDM model and SEM model.856From the perspective of mass real estate appraisal, the 857functional model should exhibit the following character-858istics: (1) The estimated results of the model should be 859close to the actual market value; (2) The variable setting 860should have good interpretability; (3) The model should 861have practical utility.862While the SEM model’s spatial lag term of the error 863can enhance the model estimation’s accuracy, this variable 864is essentially a “black box.” The insignificant influence fac-865tors, hidden influence factors, and exogenous errors in the 866“black box” are difficult to observe, and researchers and 867evaluators cannot explain the economic significance of 868this variable. Consequently, the evaluation results of this 869model will be questioned by both parties, and its guiding 870significance for practical application will be lost.871In contrast, the estimation results of SDM demon-872strate that the regression coefficients of the independent 873variables, the independent variable spatial regression co-874efficients, and the dependent variable spatial regression 875coefficients are significant, and the estimated coefficients 876of each variable’s influence direction are consistent with 877expectations. Hence, it can better elucidate the econom-878ic significance of each variable than SEM. Additionally, 879even without considering the theoretical significance of 880model setting, the SDM test results for Log Likelihood, 881AIC, and R-squared are marginally better than those of 882SEM.883To conclude, SDM has the best evaluation ability. The 884optimal functional form for mass real estate appraisal 885based on ordinary residential real estate data in the study 886area is Equation (8), and the model estimation coefficients 887are shown in Table 9.888= ρ889+β ⋅890⋅θ +β ⋅891+892⋅893⋅θ +894β ⋅895+896⋅897⋅θ +β ⋅898+899⋅900⋅θ +901β ⋅902+903⋅904⋅θ +β ⋅905+906⋅907⋅θ +908β ⋅909+910⋅911⋅θ +β ⋅912+913⋅914⋅θ +915β ⋅916+917⋅918⋅θ +β ⋅919+920⋅921⋅θ922+ ε9231924192529262927392839294930493159325933693469357936793789388939994099411094210943( )944,945Ln Y946WY947Ar948D949W D950F951W F952EH953W EH954Ag955W Ag956B957W B958MR959W MR960SM961W SM962CBD963W CBD964P965W P966where: Y – the vector formed by the single trading price 967of each sample, Yuan/m2; βi – linear regression coefficient; 968θi – spatial autoregressive coefficient of independent vari-969able; Ar, D, Ag, F, EH, B, MR, SM, CBD, P – vectors com-970posed of corresponding eigenvalues of each sample; W – 971K adjacency matrix; ε – error term; ρ – spatial regression 972coefficient of dependent variable.973Table 9. SDM coefficient estimation974Variable975Coefficient Std. Error976t-Statistic977Probability978CONSTANT9792.44599800.34779817.049820.000983Ar9840.00299850.000298613.059870.000988D9890.02199900.00259918.889920.000993F9940.02859950.00489965.939970.000998EH9990.099010000.019010015.2210020.0001003Ag1004–0.015710050.00101006–15.7410070.0001008B10090.000510100.000210112.9910120.0031013MR1014–0.003710150.00141016–2.6810170.0071018SM10190.003510200.001810211.9810220.0481023CBD10240.426910250.139310263.0610270.0021028P10290.004810300.001810312.6110320.0091033W_Ar1034–0.001610350.00051036–3.3210370.0011038W_EH1039–0.075810400.03271041–2.3210420.0211043W_Ag10440.010910450.001810466.0710470.0001048W_F1049–0.005810500.01131051–0.5210520.6061053W_D1054–0.008010550.00571056–1.4110570.1591058ρ10590.690210600.0425106116.2210620.0001063Table 9 shows that for every 1 m2 increase in the area 1064of the surrounding properties compared to the current 1065property (W_Ar), the house price of the current property 1066decreases by 0.0016 Yuan/m2. Similarly, for every unit in-1067crease in the elevator-to-household ratio of the surround-1068ing properties compared to the current property, the house 1069price decreases by 0.0758 Yuan/m2. Conversely, for each ad-1070ditional year in the age of the surrounding properties com-1071pared to this property, the price of the property increases 1072by 0.0109 Yuan/m2. Additionally, when the decoration level 1073of the surrounding properties increases by 1 unit compared 1074to this property, the house price of this property decreases 1075by 0.0080 Yuan/m2. The coefficient of Ar is 0.0029, indicat-1076ing that, in addition to the influence of spatial relations, the 1077housing price increases by 0.0029 Yuan/m2 for every square 1078meter increase in property area.1079The coefficient ρ is 0.6902, meaning a 1 Yuan/m2 1080change in surrounding residential housing price corre-1081sponds to a 0.6902 Yuan/m2 change in real estate price.108210831084===== PAGE 12 (score 9, numeric-paren 0, table-words 2, var-words 5) =====10851086International Journal of Strategic Property Management, 2023, 27(5): 304–31610873151088Furthermore, while the CBD coefficients appear sub-1089stantial, their comparison with coefficients from other spa-1090tial econometric models highlights the SDM’s reasonable 1091estimations. This variable has also successfully cleared the 1092multicollinearity test, and the resulting model outcomes 1093align with the IAAO standard ratio. As such, the model’s 1094applicability is well-supported and can be confidently ac-1095knowledged.10964.4. Appraisal ability measure1097To further assess the practical usability of the optimal 1098model, a single real estate transaction price was randomly 1099selected from the 115 residential districts not included in 1100the training data for testing. The model’s appraisal per-1101formance was then measured against the mass appraisal 1102standard outlined in section 3.2. As depicted in Table 10, 1103the SDM produced three measurement outcomes that align 1104with international standards and exhibit minimal Coeffi-1105cient of Dispersion (COD) values. Moreover, the Mean Ra-1106tio (MR) and Price-Related Differential (PRD) values are 1107closely approximating 1. These outcomes underscore the 1108excellent appraisal capabilities of the SDM within the study 1109area, thus fulfilling international standards.1110Table 10. Summary of measurement results1111Indicator1112MR1113COD1114PRD1115International standard11160.9-1.111175-1511180.98-1.031119Test result11201.029311218.484011221.00971123Conclusions1124The study aims to build and compare the explanatory and 1125appraisal abilities of different spatial cross-section data 1126models using a set of data. Based on the empirical analysis 1127and relevant tests, the following conclusions can be drawn:1128Firstly, the significant high-high and low-low aggrega-1129tion phenomena in the flat transaction price of ordinary 1130residential real estate prove the necessity of spatial econo-1131metric analysis and highlight the bias in the traditional he-1132donic price model’s assumption of sample independence.1133Secondly, the significant spatial regression coefficients 1134of independent and dependent variables in the SLX and 1135SDM regression results explain the influence of neighbor-1136ing real estate characteristics and prices in generating real 1137estate price data.1138Thirdly, the comparative analysis and optimization of 1139the models show that SDM has excellent appraisal ability 1140to describe the spatial spillover effect in the data genera-1141tion process. The measurement results conform to inter-1142national standards, indicating that SDM is superior to oth-1143er spatial cross-section data models in terms of appraisal 1144accuracy and applicability to explain the data generation 1145process.1146In conclusion, the spatial Durbin model based on dou-1147ble fixed effect has good applicability in the field of mass 1148appraisal of real estate, and it can effectively capture the 1149spatial dependence in real estate price data. These find-1150ings can provide a reference for the practical application 1151of mass appraisal of real estate.1152Currently, the model’s scope is restricted to general 1153residential real estate, excluding apartments and com-1154mercial properties. Nevertheless, its usability is con-1155strained by existing limitations such as non-transparent 1156property transactions and the absence of standardized 1157criteria for property characteristics. The following con-1158tents can be further explored in the follow-up research. 1159First of all, with regard to characteristic variables, it is nec-1160essary to select characteristic variables according to the 1161actual situation of each kind of city and to standardize 1162the way of assigning values, that is, to establish a stand-1163ard assignment system for mass appraisal; the second is, 1164in terms of the spatial weighting matrix, when selecting 1165“neighboring” properties, in addition to determining the 1166“neighborhood range” based on the empirical data of 1167house buyers and sellers, a comprehensive survey of the 1168entire study area should be conducted in order to deter-1169mine a universal “neighborhood range” and to verify the 1170impact of different “neighborhood ranges” on the accu-1171racy of the assessment based on actual data.1172Funding1173This study was supported by the Humanities and Social 1174Sciences of the Ministry of Education (22YJA630121), 1175Liaoning Revitalization Talents Program (XLYC2203004), 1176Liaoning Federation of Social Sciences (2024lslyb-1177kt-043), Social Science Foundation of Liaoning Province 1178(L20BJY010), and Basic Scientific Research Project of Col-1179leges and Universities in Liaoning Province (lnms202140).1180Ethical statement1181I testify on behalf of all co-authors that our article submit-1182ted to the International Journal of Strategic Property Man-1183agement: This article does not contain any studies with 1184human participants or animals performed by any of the 1185authors; This manuscript is original and has not been pub-1186lished and will not be submitted elsewhere for publication; 1187No data have been fabricated or manipulated (including 1188images) to support the conclusions; All the authors listed 1189have approved the manuscript that is enclosed and they 1190have no conflict of interest.1191References1192Anderson, S. T., & West, S. E. (2006). Open space, residential 1193property values, and spatial context. Regional Science and Ur-1194ban Economics, 36(6), 773–789. 1195https://doi.org/10.1016/j.regsciurbeco.2006.03.007 1196Antipov, E. A., & Pokryshevskaya, E. B. (2012). Mass appraisal 1197of residential apartments: an application of random forest 1198for valuation and a CART-based approach for model diag-1199nostics. Expert Systems with Applications, 39(2), 1772–1778. 1200https://doi.org/10.1016/j.eswa.2011.08.0771201