Decomposing the disposition effect

We theoretically show that there is a fundamental disconnect be- tween the disposition effect, i.e., investors' tendency to sell winning assets too early and losing assets too late, and its common empirical measure, namely a positive difference between the proportion of gains and losses re- alized. While its common measure cannot identify the disposition effect, it identifies the presence of some systematic bias. We further investigate the measure's comparative statics regarding markets, investors' information level, and their attention. Besides generating novel testable predictions, this analysis reveals that, in contrast to the measure's sign, variations in its magnitude are informative for its cause.