UC Berkeley

This dissertation consists of three chapters dealing with different topics in time series econometrics including generalized method of moments (GMM) estimation and vector autoregressions (VAR). These econometric models have revolutionized empirical research in macroeconomics. Previous work by Hansen and Singleton (1982) showed that the GMM method can be applied to estimate nonlinear rational expectations models in a simple way that the models need not even be solved. The seminal work of Sims (1980) has demonstrated how VAR models can be used for macroeconomic forecasting and policy analysis. The objective of this dissertation is to provide some new econometric tools for applied research in macroeconomics using time series data.

The first chapter develops an asymptotic theory for the GMM estimator in nonlinear econometric models with integrated regressors and instruments. We establish consistency and derive the limiting distribution of the GMM estimator for asymptotically homogeneous regression functions. The estimator is consistent under fairly general conditions, and the convergence rates are determined by the degree of the asymptotic homogeneity of regression functions. Similar to linear regressions, we find that the limiting distribution is generally biased and non-Gaussian, and that instruments themselves cannot eliminate the bias even when they are strictly exogenous. Therefore, GMM yields inefficient estimates and invalid $t$- and chi-square test statistics in general. By implementing the fully modified method developed by Phillips and Hansen (1990), we obtain an efficient GMM estimator which has an unbiased and mixed normal limiting distribution.

In the second chapter, we develop a novel shock identification strategy in the context of two-country/block structural vector autoregressive (SVAR) models to identify the transmission of credit shocks. Specifically, we investigate how credit shocks originating in the U.S. or euro area affect domestic economic activity in emerging Asia. Shocks within each block are identified using sign restrictions, whereas shocks across the two blocks are identified using a recursive structure (block Cholesky decomposition). This strategy not only enables us to distinguish the external credit shock from the other structural shocks, but also captures the responses of the domestic country. The main findings include that the transmission of credit shocks across countries through the channel of credit contagion is fast and protracted. The adverse effects of external credit tightening are mitigated by domestic credit policy easing in China, but lead to significant decreases in credit and GDP growth in the other emerging Asian countries. We also find that the external credit shocks play a non-negligible role in driving economic fluctuations in emerging Asia, although the role is smaller in China.

In the last chapter, we use a global vector autoregressive (GVAR) model to forecast the principal macroeconomic indicators of the original five ASEAN member countries (i.e. Indonesia, Malaysia, Philippines, Singapore, and Thailand). The GVAR model is a compact model of the world economy designed to explicitly model the economic and financial interdependencies at national and international levels. Our GVAR model covers twenty countries which are grouped into nine countries/regions. After applying vector error correction model (VECM) to estimate parameters in the GVAR, we generate twelve one-quarter-ahead forecasts of real GDP growth, inflation, short-term interest rates, real exchange rates, real equity prices, and world commodity prices over the period 2009Q1-2011Q4, with four out-of-sample forecasts during 2009Q1-2009Q4. Forecast evaluation based on the panel Diebold-Mariano (DM) tests shows that the forecasts of our GVAR model tend to outperform those of country-specific VAR models, especially for short-term interest rates and real equity prices. These results suggest that the interdependencies among countries in the global financial market play an important role in macroeconomic forecasting.