SAS at Los Angeles County Assessor s Office WUSS 2015 Educational Forum and Conference Anthony Liu, P.E. September 9-11, 2015
Los Angeles County Assessor s Office in 2015 Oversees 4,083 square miles of real estate: 88 incorporated cities and 137 unincorporated areas Processes 2,362,297 real property parcels (SFR, RI, CI) AND ~ 367,593 business and Personal Property assessments Enrolled $1.264 trillion roll value 2 #WUSS15
Assessing real estate using SAS from PROC REG to PROC KRIGE2D Los Angeles County (LAC) Assessor makes use of a computer assisted mass appraisal (LAC-CAMA) model to validate and analyze sale prices of single family residences and condominiums. Ask any real estate appraiser/agent to name the three most important things a home should have, and you will likely hear the three L s - Location, Location, Location! 3 #WUSS15
LAC-CAMA Model plays a vital role in our property valuation system 1,865,934 SFR/CONDO parcels (79% of the total parcels) are routinely evaluated and analyzed by LAC- CAMA model. The values generated by the model are used as a standard against reported sale prices. The values generated may also be used for further analysis of sales that are manually processed by appraisers if the sale prices are missing or questionable. The complete model consists of numerous SAS programs and reports, with many safeguards in place to prevent posting improper values. Using the model, an appraiser can handle an extremely large number of appraisals in a very short time. 4 #WUSS15
Currently, we model Location using over 575 geographical/economic clusters in the County Each cluster has unique characteristics representing its own neighborhood Cluster definition is updated by appraisers as required Clusters are transformed into various mathematical formulae in SAS Closer look of a cluster clusters 5 #WUSS15
LAC-CAMA Model is designed in compliance with Appraisal Principles Approaches to value Cost Approach Market Approach California statutory requirements (Ad Valorem) State Board of Equalization Fair Market Value (FMV) concept Rule 2 (the value concept) Rule 4 (comparative sales approach to value) California Constitutional Provisions, Section 13, Land and Improvements shall be separately assessed Industry best practice the Hedonic Pricing method since 1970 s 6 #WUSS15
Why do we use the Hedonic Method? It is conceptually sound and is still the most powerful analytical tool in mass appraisal Sale price = Regressors of [Time (of sale) + Improvement (building) + Land (area, view, ) + Location (neighborhood)] The Hedonic Method rests on multiple regression analysis via probability theory and calculus to produce the most probable value in a reasonable and fair market The method is also an excellent decision-making tool in real estate market intelligence (trend, forecasting, etc.) It can be easily integrated with geographical information systems (GIS) 7 #WUSS15
Snapshot of LAC-CAMA model y i = β 1 x i1 + β 2 x i2 + + β v x iv + δ i, i = 1,, n This is our basic LAC-CAMA model (via PROC REG, PROC IML) are the regressors; v is the number of regressors, including cluster transformations; β v are coefficients x iv y i is the sale price δ i represents both random error (ε i )and residual ( i ) Today, we will not discuss how x iv are modeled in detail, instead we will explore the options by extracting the residual ( i ) from the δ i and applying an innovated process via GIS to improve our predictions 8 #WUSS15
LAC-CAMA Regressors at a glance General Improvement Land Location Parcel # Living area Lot Size District (148) Class Bath room Frontage Cluster (575) Region Bed room Lot depth Use Code RCN others Ocean view Sale Date Age/depreciation Canyon view Sale Price : : : : : Each District contains a maximum of 9 Clusters 9 #WUSS15
More on the δ i term in the model The δ i component breakdown into deterministic and random parts is difficult because: It may be shown as residuals, due to the effect of sale price from the omitted regressors, if: The combined effect of the omitted variables is independent of each variable included in the basic equation; and The combined effect of the omitted variables has expectation 0. It may also be due to spatial autocorrelation and/or temporal autocorrelation. If detected, autoregressive structure or non-linear relationships should be considered. Finally, some errors (ε i ) will exist and they may be stochastic in nature (beyond our control). 10 #WUSS15
Here are the challenges we face using clusters for location The not-so-bad. Easy to use and explain to novice appraisers and/or tax payers Generally accepted during the old days for the socalled neighborhood Less computer power required Plenty of room for improvement The bad. Rigid and inflexible when neighborhood changes occur Subjective and frequently biased Discontinuities at cluster boundaries - impractical Marginal performance 11 #WUSS15
Options to improve location factor geostatistics approach Geographically Weighted Regression (GWR) Not very practical if used alone in LAC-CAMA model Spatial Kriging Direct approach using GSP (generated sale price) as Kriging variable, via PROC IML, based on β v derived from the LAC-CAMA model Indirect approach Depending upon the δ i, extracting the residuals first 12 #WUSS15
Modeling the Residuals, i i = the residual (the deterministic portion) is extracted first for the following regression model i = the difference between the GSP and the sale price of each property in the sample i is then modeled as Kriging variable y i = β 1 x i1 + β 2 x i2 + + β v x iv + i +ε i, i = 1, n Optimum prediction Kriging algorithms are available in SAS, R, ArcGIS, etc. 13 #WUSS15
What is Kriging Process? A prediction algorithm for geometrically declining sums of future compelling variable(s). It derives the best linear unbiased estimator (BLUE) based on assumptions on covariances making use of Gauss-Markov theorem to prove independence of the estimate and error. Can be done using SAS PROCS and/or ArcGIS packages with some modifications. Our demonstration will also show an integrated SAS + ArcGIS algorithm, taking advantages of the exceptional analytical capability of SAS and the graphic capability of ArcGIS. 14 #WUSS15
Examples of using PROC VAIOGRAM and PROC KRIGE2D 15 #WUSS15
GWR and Spatial Kriging via Direct Approach 16 #WUSS15
Current Cluster Delineation San Marino, California 17 #WUSS15 Micro variances of the neighborhood are not identified using the current cluster delineations.
Integrated SAS and ArcGIS Simulation using the same cluster boundaries Comparing GWR predictions of Lot Size and Improvement Size adjustment factors San Marino, California Two geographical clusters 18 #WUSS15 A more realistic representation of the neighborhood can be identified by using GWR than by using current cluster delineations!
Results of Integrated SAS and ArcGIS Simulation City of Pasadena Fair Market Value Prediction using Kriging Better representation of fair market value by using Kriging than by using current clusters delineation 19 19 #WUSS15 #WUSS14
Contact Information Anthony Liu Los Angeles County Assessor s Office 500 West Temple Street Los Angles, CA 90012 Work Phone: 213-974-3160 E-mail: aliu@assessor.lacounty.gov Web: http://assessor.lacounty.gov 20 20 #WUSS15 #WUSS14