In light of equation (32), unconditional loss is given by
I_____
stpst +
tr [n1 PN∑] p (st) dst,
tr[pω] + ∖tr [nipnς] ,
(35)
1 — p L -I
where p (st) denotes the density function for st and Ω represents the unconditional variance
of st. The performances shown in Figure IA were calculated using (1 — β)Lt, while those in
Figure IB were calculated using (1 — β)L.
Rather than integrate with respect to the entire state, st, as equation (35) does, I propose
to integrate with respect to qt conditional on xt, and to evaluate timeless perspective policies
where p (qt∖xt) denotes the density function for qt conditional on xt. To evaluate this integral,
partition Ω (and subsequently Mss and P) conformably with xt and qt, then the mean and
variance of qt conditional on xt, are given by
according to
stPst + ɪtr ∣Ni PN∑]
1 — p L J
p (qt\xt) dqt,
(36)
qt = ΩqxΩx3^xt, (37)
^qt ∣χt = Ωqq-ΩqχΩχ∙XΩχq, (38)
and equation (36) is equivalent to
Lt = xt (Pxx + PχqΩqχΩχX + ΩχXlΩqχPqχ) xt + tr [PqqΩqt∣χ] +ɪtr [NPN∑] . (39)
Equation (39) contains three terms. The first and third terms represent the penalties
attributable to the known initial state and to the stochastic shocks, respectively. The second
term represents the penalty associated with the conditional variance of the auxiliary states
that are introduced by timeless perspective policymaking. By integrating out the auxiliary
state variables, equation (39) measures average loss for a given state, xt. In the absence of
any auxiliary states, equation (39) is equivalent to equation (32). Further, in the limit as
P ↑ 1, equation (39) converges to equation (35).
Before leaving this section, it is worth noting that policy performance, as assessed by equa-
tion (39), is invariant to how the timeless perspective equilibrium is represented, unaffected
by the particular choice of dt or by the fact that Gχl may not be unique. To understand
why, note that the substantive difference between the optimal commitment equilibrium and
14