Accounting for Real Exchange Rates using Micro-Data

There are two stark views of the forces driving aggregate real exchange rates in the short-run. One view is that all of the variance is accounted for by non-traded items in the CPI basket (the classical dichotomy view), the other, due to Engel (1999), claims the opposite, with all of the variance attributable to traded items. We formulate a novel variance decomposition technique to deal with the large covariance of LOP deviations across goods. We find that the facts lie almost exactly between these two views. While the contribution to real exchange rate variability differs across goods, the dichotomous classification into traded and non-traded categories is not a good way to characterize those difference. We argue that the view that all retail goods are composites of traded and non-traded inputs is preferrable as it `convexifies' the two polar views and brings the data much closer to recent theoretical approaches emphasizing a distribution margin or trade in intermediate inputs.