2012 Theses Doctoral

Sources of Fluctuations in Emerging Markets: DSGE Estimation with Mixed Frequency Data

Rondeau, Sebastian

In this dissertation, I assess sources of aggregate fluctuations in emerging markets using a small open economy model. I focus on the importance of permanent versus transitory technology shocks. Notably, emerging countries present short quarterly national accounts data, usually since the late eighties, but longer annual series, since 1950. To use this information efficiently, I estimate the model implementing a Bayesian mixed frequency strategy that combines quarterly and annual data for 1950-2010. The mixed strategy allows us to extend the sample period 40 years back with annual data, which helps to identify permanent versus transitory shocks. And at the same time, it keeps the information of shorter quarterly series, addressing potential temporal-aggregation bias of estimation with annual data. In Chapter I, I outline the DSGE Bayesian mixed frequency estimation methodology. Then, I estimate a small open economy model featuring financial frictions for twelve emerging countries under the baseline mixed frequency estimation. I find that transitory technology shocks are the main driver of fluctuations in emerging markets, accounting for 48% of output growth variance on average, while permanent productivity shocks explain 35%. For comparison, I also estimate the model using alternative single frequency estimators based either on quarterly or annual data. Interestingly, these estimators assign a larger role to permanent shocks than the mixed frequency strategy. In Chapter II, I perform a Monte Carlo experiment for a representative emerging economy to assess the relative merits of the mixed frequency strategy. Strikingly, estimations based on short quarterly series exhibit large upward bias for the contribution of permanent technology shocks, yielding an incorrect ranking of shocks importance. Further, I find that the mixed frequency estimation drastically reduces this bias, sorting the shocks in the right order. Finally, the mixed strategy also does a better job than annual estimation along several dimensions. Interestingly, the predictions of the Monte Carlo experiment are in line with the different role assigned to permanent shocks across alternative estimation strategies in Chapter I. Also, I show that the magnitude and sign of these biases are sensitive to the true parameter values in the data generating process, especially with respect to the relative volatility of technology shocks. Overall, the proposed mixed frequency strategy presents large efficiency gains compared with alternative single frequency estimators. In Chapter III, in turn, I analyze the ability of a simpler RBC model driven only by technology shocks to explain emerging markets' business cycles. I find that a frictionless RBC does a poor job at reproducing main business cycle facts. However, the model fit presents a remarkable improvement if I assume a moderate degree of financial frictions by calibrating a larger debt-elasticity of the interest rate. Finally, using artificial data for a representative emerging economy, I find that the mixed frequency estimations deliver significant efficiency gains compared with quarterly estimations, but the gains are not as large as for the financial frictions model of Chapter I and II.