AppendixModel.PDFVIP

Appendix Model Consider the following unobserved-component model of real exchange rate (qt) which consists of a stationary component (xt) and a random walk component (yt). qt = yt + xt (1) 2 yt = yt 1 + vt; vt N(0; ?1;t) (2) ? ? 2 xt = á1xt 1 + á2xt 2 + et; et N(0; ?2;t); (3) ? ? ? 2 2 where the variances of the two independent shocks, ?1;t and ?2;t are assumed to be het- eroskedastic. In particular, we assume that they depend on two independent discrete- valued ˉrst-order Markov switching variables, S1t and S2t (S1t = 1; 2; or 3; and S2t = 1; 2; or 3) which evolve independently of vt and et, according to the following transition probabilities: X3 P r[S1t = j S1;t 1 = i] = p1;ij; i; j = 1; 2; 3; p1;ij = 1 (4) ? j j=1 X3 P r[S2t = j S2;t 1 = i] = p2;ij; i; j = 1; 2; 3; p2;ij = 1 (5) ? j j=1 Thus, we can write the variances of the shocks in the following ways: 2 2 2 2 2 2 2 ?1;t = ?v1S1;1t + ?v2S1;2t + ?v3S1;3t; ?v1 ?v2 ?v3; (6) 2 2 2 2 2 2 2 ?2;t = ?e1S2;1t + ?e2S2;2t + ?e3S2;3t; ?e1 ?e2 ?e3; (7) where Sn;kt = 1 if Snt = k and Sn;kt = 0, otherwise (k = 1; 2; 3; n = 1; 2). Writing the above model in a state-space form, we have: