Year Published
2005
Abstract
Property market dynamics depend on changes in long run equilibrium and on
impediments to adjustment towards equilibrium. One of the most important aspects of property
market dynamics is often attributed to the activities in the mortgage market. However, many
impediment factors, such as changes in family status, education, neighborhood effects and job
relocation, are often unobservable from micro household-level data. Since these omitted
variables contribute to moving decisions and therefore to sale and default decisions, utility
functions for sale and default are correlated through these unobservable variables; thus, the IIA
assumption of the widely used Multinomial Logit Model (MNL) is violated. Under such
circumstances, econometric theory suggests that the Nested Logit Model (NMNL) is a better
choice, which obviates the limitation of MNL by allowing correlation in unobserved factors
across alternatives.
This paper empirically investigates the omitted household mobility characteristics
problem in mortgage termination, and tests NMNL against MNL. Using loan level micro data,
we find significant correlation between sale and default due to omitted borrower mobility
characteristics. Our simulations find that NMNL out performs MNL in out-of-sample prediction.
impediments to adjustment towards equilibrium. One of the most important aspects of property
market dynamics is often attributed to the activities in the mortgage market. However, many
impediment factors, such as changes in family status, education, neighborhood effects and job
relocation, are often unobservable from micro household-level data. Since these omitted
variables contribute to moving decisions and therefore to sale and default decisions, utility
functions for sale and default are correlated through these unobservable variables; thus, the IIA
assumption of the widely used Multinomial Logit Model (MNL) is violated. Under such
circumstances, econometric theory suggests that the Nested Logit Model (NMNL) is a better
choice, which obviates the limitation of MNL by allowing correlation in unobserved factors
across alternatives.
This paper empirically investigates the omitted household mobility characteristics
problem in mortgage termination, and tests NMNL against MNL. Using loan level micro data,
we find significant correlation between sale and default due to omitted borrower mobility
characteristics. Our simulations find that NMNL out performs MNL in out-of-sample prediction.
Research Category