In this paper, the problem of aggregation of integrated time series va
riables is contrasted with the case of aggregation of stationary varia
bles. In the integrated case, it is seen that a cointegration conditio
n is sufficient for aggregation bias to vanish asymptotically; no anal
ogous condition is available in the stationary and ergodic case. Thus
tests for aggregation bias, originally designed for stationary data, c
an be misleading when applied to integrated data. This may account for
overrejection of the null hypothesis of no aggregation bias, when the
null is true.