R and Shiny to support real estate appraisers: An expert algorithm
implementation for Automated Valuation Models (AVM)
Beatriz Larraz-Iribas1, José-Luis Alfaro-Navarro2, Emilio L. Cano1, Esteban Alfaro-
Cortés1, Noelia García-Rubio2 and Matías GÔmez-Martínez1
1Quantitative Methods and Socio-economic development Group, Institute for Regional Development (IDR);
2Faculty of Economics and Business Administration. University of Castilla-La Mancha (UCLM) – Spain
With the support of:
Traditional Approach
1. Visit house and neighborhood
2. Look for comparables
3. Estimate value (mean)
4. Correction factors
The problem
The banking sector needs to assess their real
estate portfolio, not only for accounting
purposes, but also for regulations
compliance, e.g., The Basel II International
Banking Agreement.
AVM Expert algorithm
1. Properties to be assessed are known and in a database
2. Collect dwelling characteristics and prices from Real
Estate portals and clean database
3. Look for comparables as a human appraiser would do,
by steps:
1. Close and similar (surface, characteristics)
2. Relax constraints (distance, surface, characteristics)
3. Iteratively, until at least 6 comparables are found
4. Remove ā€œlocalā€ outliers at each step
Multiple combination for all these rules
Estimator:
Modified Inverse Distance Weighting
(IDW)
! , where !
! is the typical inverse distance for the
weighting, and ! is a coefficient that depends
on the quality of the comparable, labeled as
green, yellow, orange or red by the algorithm,
based on the stage the house is accepted as
comparable.
n
āˆ‘
i=1
γiyi γi =
αiβi
āˆ‘
i
αiβi
αi
βi
Shiny App and R programs
Ongoing work
Machine Learning and Geostatistics
models to improve the precision of the
AVM. Presumably meta-modeling by
clusters, which are also under review.
Related work
4th prize!
EnvyRState: App to explore and model
open data related to Real Estate.
Challenge: new insights into economics
and finance.
Configuration
Config files manage the rules at each
stage, and their consequences.
Different versions have been tested.
Attributes and scalars can also modify
the expert algorithm behavior.
Selection
Filters are made until
a target is identified.
Different datasets for
search comparables
can be used. Previous
in-batch estimations
can be also explored.
Estimation
The algorithm looks
comparables and show the
result (table and map).

R and Shiny to support real estate appraisers: An expert algorithm implementation for Automated Valuation Models

  • 1.
    R and Shinyto support real estate appraisers: An expert algorithm implementation for Automated Valuation Models (AVM) BeatrizĀ Larraz-Iribas1, JosĆ©-LuisĀ Alfaro-Navarro2, Emilio L.Ā Cano1, EstebanĀ Alfaro- CortĆ©s1, NoeliaĀ GarcĆ­a-Rubio2 and MatĆ­asĀ GĆ”mez-MartĆ­nez1 1Quantitative Methods and Socio-economic development Group, Institute for Regional Development (IDR); 2Faculty of Economics and Business Administration. University of Castilla-La Mancha (UCLM) – Spain With the support of: Traditional Approach 1. Visit house and neighborhood 2. Look for comparables 3. Estimate value (mean) 4. Correction factors The problem The banking sector needs to assess their real estate portfolio, not only for accounting purposes, but also for regulations compliance, e.g., The Basel II International Banking Agreement. AVM Expert algorithm 1. Properties to be assessed are known and in a database 2. Collect dwelling characteristics and prices from Real Estate portals and clean database 3. Look for comparables as a human appraiser would do, by steps: 1. Close and similar (surface, characteristics) 2. Relax constraints (distance, surface, characteristics) 3. Iteratively, until at least 6 comparables are found 4. Remove ā€œlocalā€ outliers at each step Multiple combination for all these rules Estimator: Modified Inverse Distance Weighting (IDW) ! , where ! ! is the typical inverse distance for the weighting, and ! is a coefficient that depends on the quality of the comparable, labeled as green, yellow, orange or red by the algorithm, based on the stage the house is accepted as comparable. n āˆ‘ i=1 γiyi γi = αiβi āˆ‘ i αiβi αi βi Shiny App and R programs Ongoing work Machine Learning and Geostatistics models to improve the precision of the AVM. Presumably meta-modeling by clusters, which are also under review. Related work 4th prize! EnvyRState: App to explore and model open data related to Real Estate. Challenge: new insights into economics and finance. Configuration Config files manage the rules at each stage, and their consequences. Different versions have been tested. Attributes and scalars can also modify the expert algorithm behavior. Selection Filters are made until a target is identified. Different datasets for search comparables can be used. Previous in-batch estimations can be also explored. Estimation The algorithm looks comparables and show the result (table and map).