REAL EXCHANGE RATE FLUCTUATIONS IN EMERGING MARKETS: THE ROLE OF MONEY, DEMAND, AND SUPPLY SHOCKS

Abstract

This paper uses a structural vector autoregressive model to see which shocks, supply, demand, or nominal/monetary, play the biggest role in the fluctuations of the real exchange rate in five emerging market countries: Brazil, Indonesia, Mexico, South Africa, and Turkey. Using both a classically estimated structural VAR and a Bayesian VAR estimated using a Gibbs sampler algorithm (with an independent Normal-Wishart prior), the results indicate that real shocks, not nominal shocks, explain most of the variance in real exchange rate fluctuations in emerging markets. In particular, supply shocks appear to explain a majority of the variance in the real exchange rate. The paper also investigates whether the system dynamics of the real exchange rate in each country are consistent with the predictions of a stochastic version of a theoretical, two-country rational expectations open macro model with sticky prices. The model exhibits Mundell-Fleming-Dornbusch results in the “short-run”, but also embodies the longer-run properties that characterize macroeconomic equilibrium in the open economy once prices adjust fully to all shocks. We find that the results are not consistent with the theory. Contrary to the model, real exchange rates in emerging markets appreciate in response to a supply shock, and depreciate in response to a demand shock. Finally, the paper explores why the results depart from the theoretical predictions. An ‘over-fitting’ effect, which occurs when there is small sample of data relative to the number of parameters in the VAR, seems to be biasing the results. Moreover, because emerging markets tend to experience more exchange rate (and economic) instability than do advanced economies, the data sample is most likely noisy and volatile. As a result, the noise in the data set is heavily influencing the estimation of the parameters, which only fit the data well because there are a large number of variables (parameters) relative to the number of observations. The fact the Bayesian VAR is not able to produce informative posteriors appears to confirm this fact. A larger and more precise data set, which is not readily available using the usual sources (like the OECD, IMF or World Bank), is needed to be able to determine conclusively which shocks truly drive the results.