Over the years, repeated sales models have come to wide and even commercial use. However, considering the subset of dwellings sold twice entail several challenges. Small sample problems constitute a special concern in repeated sales models, since sample sizes tend to be smaller than hedonic methods based on all transactions in a given period of time. Moreover, a cluster of observations in one time period does not only influence the index corresponding to that particular time period, but all other estimated indexes. A simulation approach is used to study the interplay between sample size and temporal aggregation. The analysis shows that serious mis-measurements may occur even in cases where the statistical diagnostic tools like R <Superscript>2</Superscript> and t-values and empiric standard deviations indicate good explanatory power. However, the risk of serious biases driven by sparse data sets tends to be small, even if the actual estimated price curve show signs of under-smoothing. Mis-measured curves have unstable estimates with respect to temporal aggregation. Two repeated models, one with a slightly finer time partition achieved by adding one more time dummy, used on the same sample can alter the index estimate at a given time with as much at 10–15%. The simulations reveal that varying temporal aggregation is a powerful diagnostic tool and should be employed routinely. The last part of the paper shows that choosing an appropriate temporal aggregation involves more than merely a balance between under-smoothing and over-smoothing. Efficiency questions tend to be better addressed by a higher temporal aggregation, than a good overall estimation of the price curve alone calls for. Copyright Springer Science + Business Media, LLC 2006
