Essays on International Finance and Macroeconomics
- Author(s): Kim, Young Ju
- Advisor(s): Hahn, Jinyong
- Tornell, Aaron
- et al.
This dissertation consists of two studies on International finance and macroeconomics. Each study addresses different topics. The first study exploits Speculators' positions in futures markets to forecast exchange rates. The forecasting model we propose combines Engel and Hamilton's (1990) point that exchange rates follow long swings with Evans and Lyons' (2004) finding that privately available information about market participants' order flow can predict exchange rates over the short-run. We extract speculators' private information by fitting a microfounded autoregressive Markov regime switching model to the speculators'net positions data in the Commitment-of-Traders report and forecasting the speculators' mode of accumulation. We then use this predicted mode to form both directional and point exchange rate forecasts for the six most traded currency pairs. Over forecasting horizons ranging from 6 to 12 months, we evaluate the performance of our forecasts vis-a-vis the random walk. The results indicate that our forecasts are significantly better than those from random walk models for most currencies, except the Swiss Franc.
The second study investigates on the sources of macroeconomic fluctuations in emerging economies. I compare the performance of the Aguiar and Gopinath (2007) model with that of the encompassing model which combines shocks to trend growth with interest rate shocks and financial frictions. Exploiting the recent developments in the theory and implementation of Bayesian methods, I estimate two models using Korea's data over the same sample period as in Aguiar and Gopinath (2007), and Chang and Fernandez (2010). In addition, I explore the role of the transitional dynamics in the estimation results. My findings are contrary to those from existing studies. The magnitude of permanent shocks is much larger than that of transitory shocks and the frictionless stochastic trend model delivers closer a match to the moments calculated from the data.