Andrew K. Rose, UC Berkeley Summary This paper shows that my estimate of the effect of currency union on trade is significantly lower (and insignificantly different from zero) for the developing countries that unilaterally use the US dollar. It also shows that the coefficient on a dummy for a dollar fix is statistically indistinguishable from that on dollarization. Analysis While I remain sympathetic to the thrust of the paper, it continues to strike me as a small project. This is increasingly true since there have been some recent developments of interest, particularly in Europe where data on EMU is starting to be analyzed. Big I’ve said if before, and I’ll say it again: I’m not sure what to make of your results. On the one hand, it’s clear that one can find cuts of the data for which the CU effect isn’t there. On the other hand, since the effect seems large in the aggregate (as you yourself find), the question is: what makes it stronger here and weaker there? That’s the question of interest (I think), and one that you don’t really address. I think if we really want to understand the sensitivity of the result, that’s at least the question to be asking. As it is, I’m not sure what I learn, other than there are cases where the CU effect is big and somewhere it is small. Another thing that I’d never really noticed before. I think you have to work harder to distinguish between economic and statistical significance. Your most important results are in Table 3. Now column II has a point estimate of .44 with a p-value of .13. Even ignoring other econometric issues (selection bias, pre-filtering, etc.), a point estimate of .44 means that currency union with the US raises trade by exp(.44)-1=55%, which is certainly economically large, even if it’s statistically marginal. There are very few CU observations, so one eliminates them at one’s peril (if they’re the main object of interest). But that’s what you do by eliminating multilateral CUs, developing country CUs, older data, and so forth. Why should one look at trade only between the US and another country? Do we distrust other bilateral trade observations? Are they irrelevant? Do we really learn nothing from currency unions that we see elsewhere? That’s key to your paper, and yet you don’t specify convincingly why your sample restriction is appropriate. In my paper with Glick we use 217 trading partners; but you use our data set and only employ 165. Since the small are more likely to be in CUs, what’s up? Also, why throw away the pre-1974 data? I do not find your motivation compelling. This paper is also starting to feel a little old-fashioned. A few years ago when El Salvador, Guatemala, and Ecuador were all dollarizing, this seemed more relevant. (Alternatively, perhaps you should update the data set and focus more on the newly dollarized.) But now the issues are more interesting, I think, in Europe. There are a few places circulating which look at EMU data (the IADB one is most well-known), and that’s got to be the frontier with 10 smaller central Europeans in the process of accession. One final thing. The thrust of your argument is that the currency union effect on trade is heterogeneous. I continue to think that it would be more compelling if you could parameterize a relationship for that heterogeneity (e.g., dyads with different sizes have a small CU effect). It would be even better if that heterogeneity varied along a dimension of theoretical interest in the optimum currency area literature. Small The references should be updated. There’s no guarantee that sub-samples give more precise estimates (as stated on p 2), and they don’t (judged by confidence intervals) in your analysis. I think you mean more relevant. Some of the paper is hard to understand (such as the abstract!). Again: at the top of p3: the US accounts for 60% of the CUs with industrial countries; Australia accounts for another 25%, but it’s the “only industrial country …” 60% + 25% ≠ 100%, so this should be rewritten. Why should one expect larger CU estimates to result from dollarizers, as stated in the middle of p3? I do not understand that paragraph. You’ve forgotten NZ-UK in note 7. What estimation technique is used in the paper? Column I.b on p8 is very ad-hoc and smacks of data mining. I suggest you eliminate it. To compare fixes with CUs, it’s natural to use the pre-1974 data. You should consider this strongly. Your concluding sentence is way too strong.
