Dear Charles,

Many thanks for your insightful comments. We have looked at them and performed further checks of our analyses and results. Our conclusion is still that an aggregation bias exists, and it is large. Below we explain why, and why we get different results from you. We hope that this clarifies the matters and that you will agree that the aggregation bias is indeed a serious concern. 1. We do bootstrapping in our first experiment because we draw (with replacement) from the estimated cross-section of sectoral rhos and assume some disturbances at the sectoral level. Given these assumptions on the DGP, we have to bootstrap. Your proposed method seems to actually assume that the rhos are deterministic, and perfectly observed. Is that more plausible? 2. You criticize our second experiment on grounds that we have lowered the average rho to 0.8. There are two separate issues here. One is whether to include explosive roots at the sectoral level, the other is a measure of the cross-sectoral heterogeneity in persistence. The paper demonstrates that the bias increases with this heterogeneity. Thus, if the dispersion in cross-sectoral persistence is calibrated so as to make sure no process i