This study estimates a structural version of the Genesove and Mayer (2001)loss aversion model. Our model diers in several ways: instead of using ahedonic regression to estimate the current expected selling price of a home, weuse MSA-specic housing indices and include a property-specic structuralerror term that depends on the time from the date of purchase. We estimatethe model by simulated maximum likelihood using data from the AmericanHousing Survey (AHS). The AHS is an ideal dataset for studying loss aversionas it is large, and the homes surveyed span over several boom and bustperiods. We nd strong support for loss aversion behavior. On average,homeowners over-value their homes 21% for a 10% increase in losses.