The second component is a measurement equation for the nonpredetermined variables and the
decision variables
dt = Sdxlxt + Sdv,χ. ■ + Sdddt-1. (25)
It should be clear from equations (24) and (25) that system estimation of timeless perspec-
tive models involves the very same techniques that are used to estimate rational expectations
models. Thus, the timeless perspective raises no hurdles for estimation other than those al-
ready encountered for rational expectations models (c.f. Juillard and Pelgrin, 2007). Specifi-
cally, after introducing any necessary measurement error terms, the likelihood function can be
evaluated directly (Fuhrer and Moore, 1995; Dennis, 2004; Schmitt-Grohé and Uribe, 2007)
or be built up recursively using the Kalman filter (Hansen and Sargent, 1980), depending on
the model, and then either maximized or combined with a prior for Bayesian estimation.
4 Evaluating timeless perspective policies
An obvious alternative to timeless perspective policymaking is for the policymaker to conduct
policy with discretion. Discretion is an obvious alternative because discretionary policies are
rule-based, time-consistent, and, critically, they do not require a commitment mechanism. In
the absence of a commitment mechanism, the discretionary policymaker simply reoptimizes
period by period. Since neither policy is optimal, an essential question is whether timeless
perspective policymaking dominates discretion. This is the question I now address.
However, in order to address this question I need a method for evaluating the performance
of timeless perspective policies, an exercise that is complicated by the presence of the auxiliary
state variables. I begin by considering two possible methods. The first method is to evaluate
loss conditional on the entire (including the auxiliary) initial state vector; the second method
is to evaluate performance using unconditional loss. Using the simple new Keynesian model
to illustrate, I show that neither of these approaches is entirely satisfactory and use their
deficiencies to motivate and derive an alternative measure of policy performance.
4.1 The simple example continued
To illustrate the central issues, it is useful to return to the simple model introduced earlier.
Accordingly, I state without proof that the explicit targeting rule associated with discretionary
policymaking in that model is given by
^t + ~yt--it = θ∙ (26)
к σκ
10