This paper examines whether the rational jumpiness/stubbornness hypothesis can explain forecast biases. Using a dataset of professional GDP forecasts for the G7 countries over the period 1989-2010, we find evidence supporting the rational stubbornness hypothesis. Specifically, forecasters underreact more when large forecast revisions are highly indicative of low forecast ability. Underreaction is less likely when the size of forecast revisions is unrelated to ability. These findings are consistent with the hypothesis that forecasters choose to smooth GDP forecasts to maximize their perceived ability.