We investigate the rapid growth in the dispersion of housing prices across metropolitan statistical areas (MSAs) in the United States during 1975-2007. We first examine several explanations for this pattern, and find that it is difficult to fully explain it. Our econometric analyses show that the Log of price-to-rent ratios follows a random walk process. We then set up a parsimonious asset-pricing island model. We find that the dispersion of fundamental housing prices grows too slow relative to that in data. Incorporating rational bubble solutions, our calibrated model can match the rapid growth in the dispersion of housing prices.