The solution of dynamic linear rational expectations models (Q915621)

The problem of multiplicity of equilibria in linear dynamic rational expectations models is by now well-known. A number of techniques have been developed to deal with this problem [see \textit{C. Gourieroux}, \textit{J. Laffont} and \textit{A. Monfort}, Econometrica 50, 409-425 (1982; Zbl 0474.90031); \textit{C. H. Whiteman} [``Linear rational expectations models'', Univ. Minnesota Press (Minneapolis 1983)]; an interesting one of these is the so-called minimum-variance criterion, due to \textit{J. B. Taylor} [Econometrica 45, 1377-1385 (1977; Zbl 0363.90037)]. This criterion selects the solution which has minimum unconditional variance. However, Gourieroux et al. [loc. cit.] have shown that the Taylor criterion is simply a special (and not particularly reasonable) case of a criterion which involves minimizing K-period ahead forecast error variance. In this paper we consider the solution space spanned by the forward and backward solutions and study the optimal linear combination of these solutions (in the sense of Taylor and in the sense of Gourieroux et al.) when the driving process is more complicated than simple white noise. Specifically, we consider the leading example of a first-order moving average driving process.