We estimate the parameters of a dynamic, forward-looking neighborhood choice model in 197 metro areas where households have preferences over the racial composition of neighborhoods. Our inclusion of multiple metro areas in the estimation sample enables us to develop a new, shift-share IV strategy to estimate the impact of the racial composition of neighborhoods on location choice that relies only on across-metro comparisons of similarly situated neighborhoods. For the “shift,” we use national data to determine the probabilities different types of households live in different neighborhoods in a metro when neighborhoods are ranked only by within-metro income quantiles. The “shares” are the metro-level population shares of each household type. Thus, the instrument predicts variation in neighborhood-level racial shares, which for a given within-metro income quantile is attributable exclusively to variation in metrolevel type shares. The overall IV estimate is a weighted average of the contribution from all of the income quantiles. We use the tools of Goldsmith-Pinkham, Sorkin, and Swift (2020) to analyze the comparisons that are weighted most heavily for identification and to derive appropriate balance tests. Our key finding is that many households have very strong preferences to live in same-race neighborhoods. These preferences are so strong that the current demographic composition of neighborhoods is not stable.