(Not for publication) Appendix to: Testing long-horizon predictive ability with high persistence, and the Meese-Rogoff puzzle

j=1 bρ(j) 11 L is a generic lag polynomial that captures the serial correlation in y1,t − y1 − β0y2,t − d1t. After these three steps, the researcher repeats regression (b) – call it step (d) –, then repeats regression (c) – a fifth step – and continues to zigzag back and forth between them until convergence is reached.2 The Cochrane-Orcutt procedure is the estimation method used in Meese and Rogoff (1983a,b and 1988). However, the conditions under which the Cochrane-Orcutt procedure consistently estimates the parameters of the model are quite stringent. They include the absence of lagged endogenous 1 In this theoretical section, we assume that OLS gives consistent estimates of the parameters of interest. However, in the empirical section we use GMM estimation step by step to ensure consistency of the estimates. 2 In what follows, it will be much easier to work with the canonical regression in order to derive the order of convergence of the estimates.