As above, the maximization of the variance of Sp for given variance of d requires that the time t innovations in d be perfectly correlated(innovations at different times are
necessarily uncorrelated) so that again the dividend process must be forecasted as an
ARIMA process. However, the parameters of the ARIMA process for d which maximize
the variance of lp will be different. One finds, after maximizing the Lagrangean expression (analogous to (9)) an inequality slightly different from (1 1),