##
Real Estate Finance & Investment
Symposium, HK 2023
## Discussion
Combining Algorithmic and
Stochastic Data Models
##
Thies Lindenthal
“In real estate we do not have Big Data—So we need structure.” ~M.F.
# Best of two worlds * Combination of two cultures of statistical modeling - Statistical models using stochastic data models (econometric models, like linear regression) - Algorithmic models (ML) * Complementary strenghts - Econometric models can explicitly control for spatial and temporal correlations, and unobserved heterogeneity - ML algorithms are “easy-to-use”, and can describe complex non-linear relations and interactions
### components * (1) common trend (DGP) * (2) property type trend (DGP) * (3) spatial component (DGP) * (4) property random effect for unobserved heterogeneity (DGP) * (5) property characteristics (ML / DGP)
## Phoenix
Heat maps for single family homes based on neural network
(
Clapp & Lindenthal, JHE, 2022
)
##
Spatial Component
$ s^x = cos(Lat) × cos(Lon), s^y = cos(Lat) × sin(Lon), s^z = sin(Lat)$
##
Cost of Structure
Interaction effects between components? Location and property characteristics are linked