RDP 8601: New Classical Models and Unobserved Aggregates 7. Conclusions

In the context of a New Classical model with an imperfectly measured monetary aggregate, it has been shown that a flow of partial information can induce persistence into the agents' expectational errors and hence into output. The partial information flow allows agents continually to improve their best guesses about the unobservable true monetary aggregate, but (even over time) never fully reveals the true values. The only requirement is that the unobservable true aggregate depends (directly) on at least one of its own lagged values.

The resulting serial correlation in agents' expectational errors is perfectly consistent with rational expectations. Indeed, there is a sense in which the model's mechanism is a variant of Muth's (1960) result that adaptive expectations may be rational. This was shown by a comparison of the model with that of Brunner, Cukierman and Meltzer (1980).

Given the New Classical assumptions, the use of an imperfectly measured monetary aggregate was shown to not (of itself) lead to the rejection of the (null) hypothesis of neutrality of anticipated monetary disturbances for output. However, the use of these data to test the correlation between output and unanticipated monetary disturbances, was shown to lead to a failure to reject the false (null) hypothesis of zero correlation for some parameterisations of the measurement error.

While these results have been derived under the specific assumption that the true monetary aggregate is unobservable, they may be relevant for other situations. For example, it would not be unrealistic to treat the true aggregate price index as an unobservable variable. Under certain conditions, this assumption could produce similar results to those derived above.

This analysis illustrates the need to examine the restrictive informational assumptions that are made in the rational expectations literature. Further research in this area is particularly important because normally only joint hypotheses of New classical models can be subjected to empirical testing. The controversial nature of such tests can be mitigated to some extent if the effects of different informational assumptions have been clarified.