and I from the cost function would have allowed these variables to move
to 7. 67 The higher levels of RL are not sufficient to generate a
significant downward movement in I as the strong fiscal action and
high levels of Y are able to keep real I at a desirable level. As
with IEBl, the high levels of RL over the last four periods have no
influence on the system due to the lag structure but a significant
downward shift in I could be expected in the next planning period.
Both monetary and fiscal policy have been strongly applied to
achieve the desired targets. The requirement that both controls be allowed
to move freely has meant that G and DM have been able to bear the
burden of target achievement without excessive adjustment in all other
non-weighted state variables with the possible exception of the total
monetary base and the rates of interest. The variability in the total
function which accounts for the more subdued performance of M in
comparison with these variables. With so many income variables taking on
the role of intermediate variables, comparatively little adjustment to
complement the controls in the achievement of internal balance is
required. The behaviour of the intermediate monetary variables is
reversed however from the income variables with the requirement that they
move excessively to complement the achievement of external balance.
The optimal stochastic solution results in neither internal or
external balance being exactly achieved, which is to be expected. Fifty
monte carlo simulations reveal that the optimal paths for both Y and FR
track very close to their targets for all combinations of additive shock.
An illustrative example is given in Figure 11. The optimal path for FR
fluctuates about its target to a greater degree than real Y but the
shocks are not sufficient to move FR away from the general area of the
target as was the case in IEB1 (remember that identical shocks are used
for the illustrative examples). The additive disturbances are not
sufficient to shift real Y substantially off target with the optimal
stochastic path closely tracking its target and deterministic counterpart.
The stochastic example is so close to its target that it was not possible
to graph a sufficiently distinguishable path from the target in Figure 11.
The behaviour of the other major state variables is also similar to the
deterministic case, allowing of course for the impact of the additive
disturbances. The results clearly illustrate the self-correcting nature
of the optimal control laws in the linear/quadratic framework to additive
uncertainty. The fixed target solution is also self-correcting in relating
to past shocks as x^_ is explicitly included when we solve for the
appropriate values of the controls in each time period. It is worthwhile
it is only in a strongly-Tinbergen world with as many instruments as
targets that the dynamic fixed target approach to stabilisation will be
able to adjust to additive disturbances from period to period.
The optimal stochastic paths of the control variables provide
some interesting results. In the deterministic case strong expansionary
fiscal policy was required to achieve the desired goals. The illustrative
stochastic results (not graphed) indicate that under uncertainty this need
not be the case. The optimal G becomes closer to its target over the
final periods of the planning period than the corresponding deterministic
G. Monetary policy is once again of a stop-go nature but in the stochastic
case it is more severe, particularly in periods fifteen to twenty.
Although the direction of monetary policy has not been reversed, it is
severely contractionary in the last five periods by comparison with the
deterministic results and the target. More importantly, we have a switch
in the optimal mix of monetary and fiscal policy from a situation of an
expansionary fiscal policy in the perfect information case to a less
expansionary fiscal policy and more severe monetary policy in the imperfect
information case. The change in the mix of policy can be illustrated by
the change in the RMSD value for both control variables when we shift from
perfect information to imperfect information.
RMSD RMSD
G DM
D 167 516
S 153 807
The letters D and S refer to deterministic and stochastic solutions
respectively.
The third experiment, IEB3, consisted of attempting to achieve
was emphasised relative to fiscal policy. To achieve this policy mix a
low cost was placed on deviations of DM from its target path while high
costs were placed on G,Y and FR. G,Y and FR were weighted such that
a one percent deviation from the relevant target would have a thousand
times the cost of a similar movement in DM. The relative weights were as
follows
Y FR G DM
1000 1000 1000 1
The computer program accepts the relative weighting specifications and then
generates actual weights which will maintain the desired relativity for all
target values and all time periods. Once again we have a situation in
which the number of targets equals the number of instruments but with the
restriction that only one instrument can adjust in a completely free
manner. This situation is not completely removed from the real world as it
is not uncommon for governments to try and maintain a level of government
spending to satisfy social and political needs and then rely on monetary
policy to carry the burden of economic stabilisation.
The optimal deterministic paths for Y and FR indicate that
given a restricted application of fiscal policy and a free application of
monetary policy, the dual targets of internal and external balance cannot
be achieved simultaneously (see Figures 13 to 15). Initially, real Y is
very close to its target (the first four periods for Y have not been
graphed in Figure 13 to avoid unduly complicating the diagram) but as the
planning period progresses it begins to fall just below target with the net
result being an optimal path for Y which is very similar to the
corresponding path obtained in IEB1 where Y and G were also weighted
R e a l Y ft B il N o m i n a l F R T a r g e t --- I E B 3 F I G U R E 13 I E B 4
ft Bil
ieb
4
--- IEB3