LAND VALUATION MODEL FOR LAND BANKS Guy Thigpen, MUSA, MPhil Director of Analytics, Philadelphia Land Bank Doctoral Student, Philadelphia University
Land Banks What are they? Land Banks are public or community-owned entities created for a single purpose: to acquire, manage, maintain, and repurpose vacant, abandoned, and foreclosed properties the worst abandoned houses, forgotten buildings, and empty lots. U.S. Department of Housing and Urban Development https://www.hudexchange.info/resources/documents/landbankingbasics.pdf
Why Land Banks are needed
Philadelphia Land Bank The Philadelphia Land Bank was formed by the Passage of State and Local legislation in 2012 and 2013 respectively.
Typical Land Bank Properties Inventory (6,500) from existing land holding agencies Potential tax delinquent and vacant properties (25,000)
Challenges Facing the Land Bank Tax collection versus strategic reuse of Acquisition Targets Quality of existing land holding inventory What is the highest and best use? How much is the value of the property?
Portfolio Management Valuable versus Valueless properties Should we have a single policy or should different types of assets be handled differently? Developers versus Community What is the best use? Affordable housing? Sideyard? Market rate development? Revenue generation versus Development subsidy Should the property be auctioned? Sold? Gifted for development?
Proposed Solutions: Decision Trees
Decision Trees - continued
Decision Trees - continued Community Garden Commercial Development
Data Based Decision Making Public Record Tax Status Ownership Planning Priorities Analysis Property Value Market Conditions Vacancy Status Policy Adjustment Highest and Best Use
Determining Property Value The Land Bank basis much of it s disposition policy on the value of a property. Three valuation methods exist to value property in Philadelphia: 1. Appraisal slow and expensive 2. Tax Assessment comprehensive but market insensitive 3. Upfront Pricing Model
Evolution of the Upfront Pricing Model Simple Hedonic Pricing Model Iteratively Weighted Least Squares (IWLS) with Spatial Adjustments Machine Learning (Random Forest) Hybrid Model with Geographic Weighting and Time Weighted Regression
Simple Hedonic Model - OLS A method of identifying price of something that is not observable, but known contributing factors exist. Also, a method of breaking down the total price of a product into the value of it s individual attributes. EQUATION: - Total PRICE = (Prices accumulated by individual attributes) In our case, - LAND PRICE = f(property characteristics) More flexibility compared to other valuation methods (income, cost), presents an opportunity for hypothesis testing and develop pricing strategy. P = β 0 +β 1 X 1 + +β n X n +Ɛ
IWLS Regression Variables Final IWLS Regression of Land Value Dep. Var.=Ln(Price/SqFt), N=1,517, Adj. R-Sq.=0.75 Variable Est. Coeff. Std. Error t-value Pr> t Intercept 2.31332 0.44544 5.19 <.0001 art_assmb -0.09003 0.04449-2.02 0.0432 dist_cbd_parcel -0.00012 6.37E-06-19.3 <.0001 dist_cbd_sq 1.61E-09 9.76E-11 16.54 <.0001 Lot_sqft1-0.00022 4.24E-05-5.21 <.0001 lot_sqft1_sq 3.45E-09 1.57E-09 2.19 0.0285 frontage 0.00679 0.00203 3.35 0.0008 depth -0.00561 0.00119-4.7 <.0001 ratio_frt_sqft -22.7815 5.6085-4.06 <.0001 view_dum -0.61181 0.05924-10.33 <.0001 dist_park -0.00024 2.75E-05-8.76 <.0001 dist_vaclot 0.000486 1.77E-05 27.46 <.0001 dist_park_sq 3.84E-08 3.18E-09 12.09 <.0001 gma_d 0.70623 0.12977 5.44 <.0001 gma_e 0.71918 0.09679 7.43 <.0001 gma_f 0.31926 0.05055 6.32 <.0001 gma_g -0.22136 0.04425-5 <.0001 gma_h -0.32153 0.04848-6.63 <.0001 gma_k -0.26384 0.04157-6.35 <.0001 gma_l 0.63188 0.114 5.54 <.0001 gma_m 0.7446 0.12908 5.77 <.0001 gma_n 0.97441 0.091 10.71 <.0001 (Continued from left) Variable Est. Coeff. Std. Error t-value Pr> t avg_psf_eighth 0.00438 0.000217 20.24 <.0001 numsales_eighth 0.00276 0.00101 2.73 0.0064 sd_psf_qtr 0.0019 0.000359 5.29 <.0001 ASSMB_dumm 0.28046 0.0296 9.47 <.0001 lg_2006-0.0909 0.01935-4.7 <.0001 lg_bachlr 0.17856 0.01665 10.72 <.0001 lg_eoi -0.11387 0.01849-6.16 <.0001 lg_rstr -0.11327 0.02871-3.94 <.0001 lg_dry 0.24648 0.03167 7.78 <.0001 lg_vcnt 0.09282 0.01759 5.28 <.0001 lg_bridge -0.22534 0.03073-7.33 <.0001 lg_crime 0.24287 0.02863 8.48 <.0001 lg_hwy_ext 0.22765 0.02516 9.05 <.0001 lg_permit -0.14233 0.01847-7.7 <.0001 lg_rail 0.15878 0.02141 7.42 <.0001 big_ratio -0.15547 0.06706-2.32 0.0206 big_sd -0.13127 0.06628-1.98 0.0478 small_sd -0.1539 0.04598-3.35 0.0008 far_cbd 0.20879 0.08529 2.45 0.0145 big_lot 0.25899 0.09901 2.62 0.009 isolated 0.05239 0.04174 1.26 0.2096 close_park -0.0676 0.08424-0.8 0.4224 high_eoi -0.36911 0.10137-3.64 0.0003 low_eoi -0.08278 0.06574-1.26 0.2082
Important Characteristics Most Powerful Locational Variables in Philadelphia, Ranked by Contribution to the Model's R-Squared Proximity to Recent High-Priced Sale Distance to City Hall Distance to Highway On/Off Ramp # of Nearby EOIs Assemblage Opportunity Distance to Park Amount of Frontage (Feet) # of Nearby Bldg Permits Depth of Lot (Feet) Distance to Septa Rail Station 3.2% 2.3% 2.2% 2.0% 1.2% 6.8% 6.6% 6.0% 10.0% 25.9% 0.0% 5.0% 10.0% 15.0% 20.0% 25.0% 30.0%
Value of Various Characteristics $Value of Different Locational Attributes to a Square Foot of Vacant Land in Philadelphia Distance to City Hall Total Square Feet of Lot Nearest Dry Cleaner # of Nearby Septa Rail Stops Assemblage Opportunity Nearest Restaurant Distance to Highway On/Off Ramp Distance to University Campus Site Has a Nice View # of Nearby EOIs Depth of Lot (Feet) Distance to Park Site is on Major Commerical Corridor Amount of Frontage (Feet) # of Nearby Bldg Permits Distance to Park Near to River Property has a (vacant) structure ($0.36) Property is a Government ($0.55) Sale ($2.95) Property is a Sheriff Sale $1.65 $1.59 $1.01 $0.77 $0.56 $0.55 $0.53 $0.45 $0.13 $2.47 $2.46 $2.38 $2.94 $2.79 $2.72 $3.59 $7.75 This chart shows the average change in the $Price/SqFt of vacant land in response to a unit increase in the value of the locational attributes of land in Philadelphia. ($4.00) ($2.00) $0.00 $2.00 $4.00 $6.00 $8.00 $10.00
Machine Learning Uses various algorithms to find the mean of values in different spatial areas. The same variables are used and it uses a random forest algorithm to calculate value. A tight clustering of observations around the 45-degree line indicates accurate predictions. Chart courtesy of Ken Steif, Ph.D.
Limitations of OLS Regression Spatial phenomena will vary across a landscape which affects property prices including that of land. OLS regression can capture the location component of observations only as much by fixed effects with underlying assumptions that modeled relationships are constant across time and observations are independent, stationary.
Introduction of GWR/TWR Geographic Weighted Regression (GWR) local spatial relationships to be tracked across space, while Time Weighted Regression (TWR) allows the same to be tracked across time. What if instead of having a single, global, regression (OLS), we have separate, local, regressions for each location (point, block group, tract, zip code, etc.) For instance, higher home values might be correlated with closer transit stops in some locations and more distance from public transit at others. We improve the performance of all techniques by transforming the data before running the model.
The GWR Formula The GWR model extends the traditional regression framework by allowing parameters to be estimated locally so that the model can be expressed as where (u i,v i ) denotes the coordinates of the point i in space, β 0 (u i,v i ) represents the intercept value, and β k (u i,v i ) is a set of values of parameters at point i.
Visualization of GWR
The TWR Dimension
Validation Process
Land Value Index 8,000.0 7,000.0 6,000.0 5,000.0 4,000.0 Philadelphia Land Price Indices by Submarket: 1980-2015 Central Delaware Central Philadelphia North Philadelphia-East North Philadelphia-West Northeast Philadelphia Northwest Philadelphia South Delaware South Philadelphia West Philadelphia-North West Philadelphia-South 3,000.0 2,000.0 1,000.0 0.0 1980 1980 1981 1982 1983 1983 1984 1985 1986 1986 1987 1988 1989 1989 1990 1991 1992 1992 1993 1994 1995 1995 1996 1997 1998 1998 1999 2000 2001 2001 2002 2003 2004 2004 2005 2006 2007 2007 2008 2009 2010 2010 2011 2012 2013 2013 2014 2015 Notice grouping of indices: high v. medium v. low appreciation rates.
Broader Uses for Pricing Models Policy Changes The Official Disposition Policy is based on value levels generated from the model Project Prioritization Property acquisition is based on neighborhoods with appreciating values
Future Considerations Dynamic Updates www.phillylandworks.org Agent Based Modeling http://runthemodel.com/models/391/ htttp://runthemodel.com/models/198/