Macroeconomic Forecasting under Regime Switching, Structural Breaks and High-dimensional Data

Macroeconomic time series are not constant over time. Recent years have again emphasized that growth and decline periods alternate, with different dynamics characterizing each period. Such instabilities increase the complexity of econometric models and complicate the construction of accurate forecasts. In addition, more and more macroeconomic indicators are available that might, or might not, have an effect on a specific variable of interest.
In current data sets, the number of predictors tends towards the available number of observations. As a consequence, forecast uncertainty increases further. In this thesis, new shrinkage and dimension reduction techniques are developed that reduce the uncertainty that is inherent to modeling unstable time series or using a large number of possibly relevant predictors.