Appendix I Economic Policy Coordination in Context of Uncertainty
- Paul Masson, Morris Goldstein, and Jacob Frenkel
- Published Date:
- June 1991
An important aspect of policy coordination is the lack of knowledge about the effects of policies and hence whether a particular policy choice is likely to have beneficial, or harmful, effects.55 For instance, Feldstein (1988) has argued
Uncertainties about the actual state of the international economy and uncertainties about the effects of one country’s policies on the economies of other countries make it impossible to be confident that coordinated policy shifts would actually be beneficial (p. 10).
Evidence of the lack of knowledge about the effects of policies can be gleaned from a comparison of existing multicountry models that has been made at the Brookings Institution (see Bryant and others (1988)). The second-year multiplier effects on GDP of a standardized increase in government expenditure in the United States ranged from 0.4 to 2.1, while the transmission effects on GDP in the rest of the Organization for Economic Cooperation and Development (OECD) area ranged from slightly negative to 0.7. Moreover, estimates of single parameters—for instance, the interest elasticity of investment or the direct substitution effect of government spending on private consumption—often have very large standard errors relative to estimated coefficients.
Lack of knowledge about the functioning of the world economy—which we will term model uncertainty— should be distinguished from disagreement about the correct view of the world, which may or may not involve the recognition by policymakers that their view of the world may not be correct.56 In an extreme case, each policymaker may be convinced that he has the truth, but that the others do not. In such a case, each may think that he can fool the others into reaching agreements that they think are in their best interest but are not. Disagreement among policymakers is discussed below, but let us first consider the question of model uncertainty.
A natural way to treat model uncertainty is to formulate a general model that includes the various possible models (assuming that they constitute a relatively small set) as special cases with different parameter values—that is, treat model uncertainty as parameter un-certainty. If we can formulate the problem as finding the optimal policies (either coordinated or uncoordinated) in the presence of ranges of possible parameter values, then the analysis of Brainard (1967) applies. He shows that in general there is a trade-off between close attainment of targets and increases in the variance of the target variable. For instance, suppose that, starting from a situation where policy is set to hit a target exactly, an oil price shock threatens to produce a sub-optimal outcome; should an attempt be made to use the policy instrument to counteract fully the effect of the shock? Since the effect of the policy instrument is uncertain, doing so may in fact more than offset the effect of the shock. The main lesson from Brainard (1967) is that policy should be less activist in the presence of model uncertainty, and should not attempt to respond fully to shocks.57 In other words, policymakers in general should not engage in fine-tuning of policy instruments.
What is the lesson for the gains that may result from international coordination of policies? On the surface, policy coordination may seem to be more activist than independent pursuit of policy goals by the countries concerned, but that presumption is not correct. On the contrary, policy coordination may rule out certain types of activist policies, such as the use of the exchange rate in a beggar-thy-neighbor fashion as in competitive depreciation to generate employment or in appreciation to achieve quick disinflation. The question is whether the existence of uncertainty increases the gap between coordinated policies (which, by definition, are fully optimal if problems of time inconsistency are ruled out) and uncoordinated policies.
There is a useful distinction between uncertainty about the effects of policies in the country taking the action (which we will call domestic multiplier uncertainty) and uncertainty about the effects on the home country of policy moves taken abroad (which we will call transmission multiplier uncertainty). In the former, no general results emerge as to whether an increase in uncertainty will increase or decrease gains from policy coordination. In the latter, there is an unambiguous increase in the gap between coordinated and uncoordinated policies, and hence an increase in the gains from policy coordination (Ghosh and Ghosh (1991), Ghosh and Masson (1988)). Uncoordinated policies, because they do not correctly capture the endogenous nature of foreign policy making (that is, the reaction of policy abroad to moves taken at home), do not properly take into account this element of uncertainty, which is larger the greater is the variance of transmission multipliers. Coordinated policies, in contrast, internalize this aspect of uncertainty. Thus, ex ante gains from policy coordination may be larger than is suggested by the simulation of deterministic models that use point estimates of parameter values and ignore uncertainty.
This conclusion emerges from the model simulations performed by Ghosh and Masson (1988), in which a two-country global model of the United States versus the rest of the world was used to quantify gains from policy coordination. Ranges for parameters were established from a survey of empirical work, and three possible models were considered: a midpoint estimate, and the high and the low extremes of the range. Policymakers (and private agents) were all assumed to assign the same probabilities to these possible models, and to set optimal policy on the basis of expected utility maximization. It was shown that uncertainty in most parameters increased the gain from choosing policy in a coordinated fashion relative to independent utility maximization—that is, uncertainty increased expected gains from policy coordination.
A recent instance—the stock market crash of October 1987—may help to solidify the argument and illustrate its real-world relevance. It could be argued that the shock to stock prices also produced greater uncertainty about underlying transmission mechanisms. The central banks were concerned at the time of the crash that liquidity should be increased, to avoid the risk of bankruptcies by investment houses and a crisis of confidence in the real economy. However, a central bank acting alone runs the risk that by increasing the money supply and lowering interest rates, it may provoke a run against the currency, exacerbating financial collapse. In such circumstances, the absence of cooperation among monetary authorities may lead them to increase liquidity by less than the optimal amount; therefore, the uncertainty about effects on exchange markets should be an incentive for enhanced coordination. Of course, the need for coordination depends on the nature of the shocks and the perceived risks. Paradoxically, the fact that the shift out of equities into other assets was generalized across major countries may have minimized the need for coordination in October 1987.
Disagreement About Models
Uncertainty may or may not be associated with disagreement among policymakers about the “correct” representation of reality. If policymakers disagree (and one or both is therefore necessarily wrong), then, as Frankel and Rockett (1988) point out, coordination agreements may lead to losses ex post, rather than to gains, relative to uncoordinated policymaking. They calculate that coordination between the United States and the rest of the OECD is about as likely to worsen welfare as to improve it, when models are chosen from those represented in the comparison of models made at the Brookings conference cited above, and where coordination involves setting policies to maximize joint utility (assumed to depend on both regions’ output and inflation performance).
The significance of this result has been questioned on two grounds. First, coordination is unlikely if one of the partners to an agreement believes that the other is using the wrong model, and that in fact the agreed policies will be demonstrably worse for that country than the alternative, uncoordinated policy (Holtham and Hughes Hallett (1987)). In this case, there is the danger that the agreement might be abandoned by one of the parties. In addition, the perception that one of them had taken advantage of the other might preclude later beneficial cooperation. By ruling out some of the cases considered by Frankel and Rockett (1988), the conclusion that coordination has a good chance of being harmful is considerably weakened. In Frankel (1991), the calculation is redone for only those bargains that improve both countries’ welfare under either of the models the two countries believe in: if the “true” model is chosen from the full set of 10 models, as before, coordination now improves U.S. welfare in 78 percent of the cases, and rest-of-OECD welfare in 76 percent of them.
The second qualification is to suggest that the models probably do not adequately represent the nature of disagreements among the policymakers. Ghosh and Masson (1991) start from alternative estimated variants of a standard two-country open-economy model (Oudiz and Sachs (1984)), which contains about the same degree of reduced-form multiplier uncertainty as the models considered by Frankel and Rockett. They show that if policymakers learn from observations on endogenous variables about the probabilities to be assigned to each of the models, using Bayesian learning, the subjective probabilities converge rather quickly to the objective ones. This result suggests that the experiment performed by Frankel and Rockett (1988) is rather artificial. It may be true that the range of disagreement among the models of the Brookings conference is in fact larger than that among policymakers—some of the models can clearly be ruled out. Alternatively, policymakers’ views of reality may be much more subtle than those represented by the models—they are models after all—and policy setting cannot be represented by such simple optimization exercises.
This appendix draws on Frenkel, Goldstein, and Masson (1991).
It is of course true that a high degree of model uncertainty is likely to be associated with disagreement about the functioning of the world economy.
This conclusion may not apply to general models where there are many targets and instruments, however. We are indebted to David Kendrick for this point.