There is yet another way to write the model, which will be useful in the analysis which follows. For this purpose, it is convenient to adopt notation for the innovation in a variable. Let us define the innovation operator t -Et -Et-1 where Et is the conditional expectations operator. Then for any variable Xt the term StXt+k equals EtXt+k - Et IXt+k which is the change in the conditional expectation of Xt+k that is made in response to new information arriving between t- 1 and t. The time subscript t may be dropped so that SXk denotes StXt+k and 8X denotes SX0 or ,X,. Since conditional expectations operators satisfy EjEk = Emin(j k) it follows that E-m,aX,+k =Et-m (Et Xt+ k Et- IXt+ k) = Et-m Xt+ k - Et-mXt+k 0, m > 0. This means that St Xt+k must be uncorrelated for all k with all information known at time t- 1 and must, since lagged innovations are information at time t, be uncorrelated with St,Xt+j t'