Working Paper: NBER ID: w17882
Authors: Arthur Korteweg; Morten Sorensen
Abstract: We develop and estimate a unified model of house prices, loan-to-value ratios (LTVs), and trade and foreclosure behavior. House prices are only observed for traded properties, and trades are endogenous, creating sample-selection problems for traditional estimators. We develop a Bayesian filtering procedure to recover the price path for each individual property and produce selection-corrected estimates of historical LTVs and foreclosure behavior, both showing large unprecedented changes since 2007. Our model reduces the index revision problem by nearly half, and has applications in economics and finance (e.g., pricing mortgage-backed securities).
Keywords: Real Estate; Loan-to-Value Ratios; Foreclosure Behavior
JEL Codes: C11; C23; C24; C43; R21; R3
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| joint estimation of prices and trade processes (F16) | corrects for selection bias (C24) |
| naive imputation of property prices (C59) | underestimation of the fraction of underwater properties (H82) |
| accounting for idiosyncratic volatility (C58) | increases estimate of underwater properties (C13) |
| joint estimation of price index and trade and foreclosure model (E30) | addresses circularity issue (E10) |
| model (C59) | reduces index revision problem (C43) |
| trade and foreclosure decisions (G21) | depend on current prices and LTVs (G19) |
| LTV distributions deteriorated substantially (G33) | dynamics of trade behavior shifted (F19) |
| properties with higher LTVs (G21) | trading more frequently (F10) |