Much of the macroeconomic literature on the `rules versus discretion'
debate proposes that precommitment take the form of a non-contingent or
`open-loop' k percentage money supply rule. Such open-loop policies
prevent governments from responding to unforeseen shocks; and although
feedback rules derived from control theory can provide the flexibility
required, they may prove difficult for a sceptical private sector to
monitor. If private agents are unable to distinguish between a
government's reaction to a contingency and its reneging on a commitment,
such `contingent rules' cannot be credible.
In Discussion Paper No. 515, Research Fellow Paul Levine notes
that a private sector that knows the full nature of the government's
calculations can conduct its own evaluations of the precommitment rule
and deduce the relevant policies for different contingencies. For the
rule to be incentive compatible, there must be a trigger mechanism
whereby the private sector `punishes' a reneging government by believing
only in discretionary policy for some `punishment period'. The recent
literature on credibility has focused on games with less than full
information to assess governments' attempts to eliminate the
inefficiencies of discretionary policy. All this literature assumes,
however, that the private sector knows the general nature of government
behaviour and lacks only some limited information concerning parameters
such as the weight on output in the welfare function. The private sector
can therefore observe policy instruments such as the money supply and
infer the values of these unknown parameters.
Levine assumes instead that a less well-informed public estimates the
policy rule directly by observing the data and applying a recursive
estimation procedure to infer the rule. For an overlapping contracts
model with rational expectations, he shows that the optimal rule with
precommitment takes the rather complicated form of a error-correction
mechanism. Learning a rule of this type proves slower than learning a
simpler rule of lower order. Although the simple rule is sub-optimal in
a complete information setting, its performance becomes significantly
better than its originally optimal counterpart when information is
withdrawn and learning introduced.
Levine maintains that these results support the case for conducting
macroeconomic policy on the basis of a a simple rather than the more
complex `optimal' rule. Other arguments for simple rules include their
intuitive appeal, ease of implementation and relative robustness in the
face of model uncertainty. He concludes that these issues are primarily
empirical, however, and can only be resolved on the basis of results
derived from empirical models.
Should Rules Be Simple?
Discussion Paper No. 515, March 1991 (IM)