Research
Macroeconomics · Monetary Policy · Household Finance · Housing Markets
Housing Finance and the Dynamic Transmission of Monetary Policy
This paper compares monetary transmission through housing valuation, home-purchase financing, and rate-and-term refinancing. Refinancing activity falls quickly after monetary tightening, but estimated cumulative contributions to the three-year consumption decline are larger for valuation and purchase financing. These findings show why refinancing activity alone is an incomplete gauge of the housing channel.
Data, methods, and findings
I combine mortgage originations by loan purpose with U.S. macroeconomic data for 1998–2023 in an eight-variable Bayesian structural VAR with stochastic volatility. A first-entry path decomposition compares the three margins on a common consumption scale.
In the baseline model, posterior medians indicate larger cumulative contributions from valuation- and purchase-entry paths than from refinancing-entry paths. I assess posterior uncertainty and sensitivity to identifying restrictions, distinguishing the speed of mortgage-market adjustment from its cumulative contribution to spending.
Labor Market Slack, Refinancing Eligibility, and the Distribution of Monetary Transmission
Employment-based refinancing eligibility can exclude unemployed homeowners from monetary easing precisely when liquidity is most valuable to them. In a heterogeneous-household model with illiquid housing and labor-market-dependent unemployment risk, this exclusion delays rather than eliminates gains from a rate cut for unemployed homeowners who keep their homes. The delay hinges on housing illiquidity.
Model and empirical evidence
The model combines long-term mortgages, illiquid housing, and a frictional labor market, calibrated to CPS labor flows, Freddie Mac rate dynamics, homeownership, and first-time homebuying rates. Liquid-asset holdings provide an untargeted check against SCF data. In model simulations, employment verification delays refinancing by one to two quarters for unemployed homeowners.
I examine the eligibility mechanism using SCF and PSID data. Families with unemployment spells were about half as likely to refinance in 2009. This household-level evidence is consistent with the eligibility mechanism.
Separately, HMDA data across 386 metropolitan areas show that the refinancing shortfall in high-unemployment areas was concentrated in applications rather than denials and mostly followed house prices. This regional evidence points to collateral constraints; I do not attribute it to employment-based eligibility.
Work in progress
Mortgage Refinancing Waves and Heterogeneous Interest-Rate Transmission
This project studies why similar mortgage-rate declines generate different refinancing waves. Using a Freddie Mac panel covering 72.8 million loan-months, I distinguish loan-pool composition, refinancing incentives, and borrower responses, and examine whether a common conditional response function can explain activity across episodes.
Current empirical work: mortgage prepayment and refinancing incentives
I construct loan-specific refinancing incentives and benchmark monthly voluntary payoff models on a 2017 origination cohort comprising 1.66 million loan-months. With chronological training, validation, and test splits, XGBoost reduces out-of-time log loss by 7.1% relative to L2-regularized logit on 377,729 test loan-months.
This benchmark predicts broad voluntary payoff, which includes refinancing and home-sale-related repayment. It does not yet isolate pure refinancing. The improvement in individual probability scores does not translate into lower monthly payoff-count errors, a distinction that matters for the project's micro-to-aggregate question.