Mass Appraisal at the Census Level - Israeli Case Larisa Fleishman Yury Gubman
Valuation of dwelling is carried out with different methods: Price agreed upon in a sale transaction Owners’ self -reported valuation Expert’s (appraiser's) valuation Valuation modeling
In Israel dwelling market values are available only for properties sold during a certain period and reported to the Tax Authority The main goals: To develop a methodology for estimating market values of all dwellings in Israel To estimate property value data for every record in the Dwelling and Building Register
Work Stage (1) Step 2: Step 1: Estimation of Hedonic Model Estimation of Hedonic Model on owners’ dwelling valuations on sale transactions ( Household Expenditure ( Tax Authority, 2011 ) Survey, 2011) Step 3: Comparison of the estimated regression coefficients from the two models
Explanation Model log(Pr ice ) Asset Building CT ijkl 0 1 i 2 j 3 k Locality u 4 l ijkl Dependent variables : (log of) Dwelling Price (log of) Subjective Dwelling Valuation Explanatory variables: – ‘ Asset ’ - dwelling indicators (size of living space, number of rooms, municipal property tax) – ‘Building’ – building indicators (number of apartments in a building, year of construction) – ‘ CT ’ - indicators of the census tract in which the dwelling is located (socio-demographic characteristics of residents) – ‘Locality’ - indicators of locality in which the dwelling is located (demographic, environment and location characteristics
Explanation Model – main results Estimates for (ln of): Variables Owners’ Transactions prices 1 valuations 1 0.018 ( Intercept 12.477* ( 12.485* (0.062) 0.218* (0.003) Number of rooms 0.253* (0.013) Area (square meters) 0.006*(0.0001) 0.006* (0.0005) Interaction: Number of rooms * Area 0.0006*(0.00001) -0.0007* (0.0001) Multistory building (yes / no) -0118* (0.006) -0.094* (0.018) Municipal property tax (NIS 2 per square meter) 0.001* (0.0001) 0.001* (0.0004) Average income from wages and business per 0.007* (0.0001) 0.004* (0.0002) capita in census tract (in 1,000 NIS 2 ) Population in locality (in 1,000) 0.099* (0.001) 0.088** (0/004) Distance from the city center (km) to CT -0.035* (0.0001) -0.022* (0.004) Distance from Tel -Aviv to CT (in 100 km) -0.835* (0.014) -0.889* (0.045) Square distance from Tel -Aviv to CT 0.225* (0.005) 0.239* (0.015) CT’s bordering on the Mediterranean sea shore 0.095* (0.007) 0.045*** (0.025) South District -0.096* (0.008) -0.154* (0.024) North District -0.056* (0.009) -0.068* (0.028) R 2 0.73 0.74 Number of observations 39,244 2,570 1 -S.E. in brackets 2 -New Israeli Shekels *Significant at 1% level; **Significant at 5% level; ***Significant at 10% level
Interim conclusions Sale transaction prices reflect not only the prices of dwellings that are sold, but closely reflect the price level of the entire inventory of dwellings, given the economic situation and the annual amount of sale transactions The model based on transaction data may be used for mass appraisal of the entire housing stock in a given area and at a certain time
Work Stage (2) Step 4: Development of the prediction models Step 5: Testing of models’ ability to estimate market value with Accuracy Indices: Mean/Median Absolute Percentage Error (MAPE / MedAPE) Step 6: Imputation of an assessed market value for every record in the Dwelling Register
Choice of Prediction Model For the full sample of transactions: MAPE 27.73 MedAPE 13.21 MAPE about 30% Benchmark in literature: MedAPE about 20% It is common practice to apply a stratification procedure (division into estimation cells) The distribution of the dependent variable over the officially defined six administrative districts was examined Quantile regression analysis was performed to test the distribution of the estimated coefficients over centiles of the dependent variable
Analysis of model coefficients stability across centiles of distribution of logged price per square meter Logged Mean Income in CT Logged Mean Price per Square Meter in CT Proximity to Tel Aviv Logged Area
Choice of Prediction Model Nine estimation cells: – (1) inexpensive dwellings (lowest decile, all districts); – (2) expensive dwellings (five uppermost centiles, all districts); – (3) seven cells are differentiated by district for dwellings remaining after the removal of those in (1) and (2). Prediction model was estimated independently for each cell
Choice of Prediction Model Accuracy indices received: MAPE 20.12 MedAPE 13.21 Percentile 10 2.21 Percentile 25 5.65 Percentile 75 22.44 Percentile 90 26.86 Division into estimation cells has improved the accuracy of the estimated values The obtained standard deviation was about NIS 16,079 on average, about 1.3 % of the mean value ( 1,262,668 NIS )
Performance of the proposed method in 2011-2013 Year Estimated standard MAPE MedAPE deviation 2011 16,079 NIS 20.12 12.28 2012 14,050 NIS 21.26 12.06 2013 13,616 NIS 19.61 11.58 The proposed method was applied to additional time periods; the accuracy indices continued to be stable during the addressed periods
Final conclusions Estimation of dwelling values at the nationwide level provides new statistical data with a high geographic resolution on a range of topics, e.g., behavior of the housing market, economic profile of residential areas, well-being and inequality, to name only a few. Value data at the individual-record level also facilitate estimations for small geographical units and population groups defined by socio- economic and demographic characteristics
larisaf@cbs.gov.il
Recommend
More recommend
