Research into policy coordination has demonstrated that in the presence of well understood policy interactions, coordination leads to better outcomes for all countries. It is also possible to estimate these gains using multi-country econometric models such as those employed by the policy-makers themselves. But if the spillovers are not well understood, and if the policy-makers use an inaccurate model (or fail to agree on the model), welfare gains may turn into welfare losses. Since the true model is unknown, and since econometric models are widely criticized for their sensitivity to misspecification, the key question is: how robust are the gains, or the policy choices, likely to be to moderate misspecifications in the chosen model? If the gains are small, they may be overwhelmed by the uncertainties concerning models and information which are inherent in empirical policy projections. Canzoneri and Minford have argued in Discussion Paper No. 119 that the gains to coordination are highly dependent on the estimated spillover effects: as a result, gains could easily turn into losses with only moderate misspecifications of the 'true' model. Feldstein has argued that this sensitivity is the reason why industrial countries should not attempt to coordinate their policies.

In Discussion Paper No. 190, Gerry Holtham and Research Fellow Andrew Hughes Hallett consider a general framework for analysing the relative benefits of cooperative policies, in which policy- makers recognize model uncertainty as well as the possibility of disagreement over the true model. The framework gives policy- makers the opportunity to select models and policies jointly. This avoids passive dependence on one chosen model with the attendant risk that cooperation may be welfare-reducing if based on the wrong model. The authors compare cooperative and non- cooperative policies for the US and the rest of the OECD (ROECD) in the context of ten prominent econometric models. Model uncertainty means that several cases must be considered: in which both the US and the ROECD agree on a model and the model they choose is the true one; in which they agree on a model (from the set of ten) but some other model is the true one; and in which they disagree about the model, in which case either one or both of the models may be incorrect.

Holtham and Hughes Hallett's simulations differ from previous attempts to estimate the gains from cooperation. The authors include the exchange rate in the welfare function of the two countries involved. This introduces a very direct conflict of interest, since one country's rise is another's fall. The simulations reveal that in the simplest case, in which the two countries agree on a model and this model is the true one, the welfare gains from cooperation are larger than previous research has suggested.

Investigations of policy coordination inevitably encounter the problem of time inconsistency and implementability: if either of the agents involved in the policy bargain would have an incentive to renege in the future, then the bargain is time-inconsistent. The analysis becomes more complicated when there is also disagreement over the appropriate model. In an extreme case each model could indicate that the optimal cooperative policy would yield a welfare gain for the country using the model but a welfare loss for the other. Model uncertainty therefore makes it necessary to distinguish between two kinds of policy bargains: those where all parties expect everyone to benefit from cooperation, and those where at least one party expects one of the other policy-makers suffer from the agreement. The former may be termed a 'strong' bargain (i.e. sustainable) and the latter a 'weak' (or unsustainable) bargain. Holtham and Hughes Hallett find that restricting attention to bargains which are sustainable significantly increases the robustness of cooperative relative to non-cooperative policies.

The empirical results in this exercise show that model variations do affect the policies adopted, the overall gains from cooperation, and the distribution of those gains. These differences, however, were dominated by movements in the exchange rate. This suggests that if exchange rate movements concern policy-makers, then cooperation over that variable will be vital and may secure the lion's share of the potential gains. Even then, the simulations suggested that there would be more consensus across models concerning the expected outcomes of policies (under both cooperation and non-cooperation) than about the policies themselves.