On the effects of aggregating cointegrated variables over time

It has long been recognized that aggregating time series introduces correlation between consecutive values of the aggregated observations (see Working (1960)). This paper investigates the effect of aggregation on the relation between variables assuming that the data generating process involves two integrated variables linked by a specific error correction mechanism (cointegration). It will be shown that aggregation does not distort the cointegration relation while some other features of the data generating process will change considerably. Cointegration tests become invalid in a single equation framework but system cointegration analysis seems to be robust against various aggregation strategies.