Equilibrium Real Exchange Rate Estimates Across Time and Space

Abstract

Equilibrium real exchange rate and corresponding misalignment estimates differ tremendously depending on the panel estimation method used to derive them. Essentially, these methods differ in their treatment of the time-series (time) and the cross-section (space) variation in the panel. The study shows that conventional panel estimation methods (pooled OLS, fixed, random and between effects) can be interpreted as restricted versions of a correlated random effects (CRE) model. It formally derives the distortion that arises if these restrictions are violated and uses two empirical applications from the literature to show that the distortion is generally very large. This suggests the use of the CRE model for the panel estimation of equilibrium real exchange rates and misalignments.

How frequent a BEER? Assessing the impact of data frequency on real exchange rate misalignment estimation

“…As in Fischer andHossfeld (2014) andMancini Griffolo et al (2015), the dependent variable of our BEER model is the bilateral RER, as opposed to the multilateral REER. This choice, which is discussed more thoroughly in Annex A, presents the advantage that the RER captures relative prices in a cleaner fashion, since it is unaffected by changes in trade weights, which in turn can be endogenous to exchange-rate variations as they modify the relative value of trade flows across partners.…”

“…Fidora, Giordano and Schmitz (2018) and Giordano (2018) adopt various price and cost indicators to construct the REERs, as well as employing alternative Balassa-Samuelson proxies, carefully selecting the set of relevant explanatory variables and using different estimation techniques. Finally, Fischer (2019) tests alternative estimation methods.…”

“…The inclusion of fixed effects removes any cross-sectional variation in the average level of the series in the panel(Fischer, 2019). Since price or cost index-based RERs do not contain any meaningful between-group information, it is quite natural in this context to employ a fixed-effects estimation, which eliminates the meaningless relative level information from the estimation process; the use of fixed-effects residuals in turn eliminates this information from the computation of the RER misalignment.14 The CCEMG procedure provides consistent estimates of the averages of long-run coefficients, but the latter are inefficient if homogeneity is present, as discussed inDe V. Cavalcanti, Mohaddes and Raissi (2015) andCouharde et al (2017).…”