Three Essays on Time Series Forecasting with Big Data
Type of DegreePhD Dissertation
MetadataShow full item record
In the first chapter, we propose factor-based out of sample forecasting models for US dollar real exchange rates. We estimate latent common factors employing an array of data dimensionality reduction approaches that include the Principal Component Analysis, Partial Least Squares, and the LASSO for a large panel of 125 monthly frequency US macroeconomic time series data. We augment two benchmark models, a stationary autoregressive model and the random walk model, with estimated common factors to formulate out-of-sample forecasts of the real exchange rate. Empirical findings demonstrate that our factor augmented models outperform the benchmark models at longer horizons when factors are extracted from real activity variables excluding financial sector variables. Factors obtained from financial market variables overall play a limited role in forecasting. Our data-driven models tend to perform better than models with international factors that are motivated by exchange rate determination theories. In the second chapter, we propose factor-augmented out of sample forecasting models for the real exchange rate between Korea and the US. We estimate latent common factors by applying an array of data dimensionality reduction methods to a large panel of monthly frequency time series data. We augment benchmark forecasting models with common factor estimates to formulate out-of-sample forecasts of the real exchange rate. Major findings are as follows. First, our factor models outperform conventional forecasting models when combined with factors from the US macroeconomic predictors. Korean factor models perform overall poorly. Second, our factor models perform well at longer horizons when American real activity factors are employed, whereas American nominal/financial market factors help improve short-run prediction accuracy. Third, models with global PLS factors from UIP fundamentals overall perform well, while PPP and RIRP factors play a limited role in forecasting. In the third chapter, we study the characteristics of the real effective exchange rate based on different exchange rate regimes, such as the fixed and flexible regimes. We focus on 17 countries effective exchange rates and we use a long annual time series data spanning from 1870 to 2016. We derive the effective exchange rates using the first common factor (Principal Component) and use the same in different sub-samples to study the mean reversion of the real effective exchange rates. We focus on full sample, pre-Eurozone, fixed and flexible regimes and find that the real effective exchange rate is affected by the regime changes for most of the countries. The unit root test rejects the null of non-stationarity for most countries in the flexible regime but fails to reject the null of non-stationarity in the fixed regime for many countries. This suggests that there is evidence of mean reversion during the flexible regime for most countries in our sample.