Yield curve modeling and forecasting is important in economics, finance and academia. The thesis at hand proposes a common functional principal component (CFPC) model to reveal the common factors of yield curves of different economic regions. The CFPC model is data-driven and can precisely detect the common dynamic structure of international yield curves. In particular, we extend the autoregression of lag order 1 (AR(1)) forecasting to a local autoregression of lag order 1 (LAR(1)) prediction approach, which does not rely on stationary. In the real data analysis we study the term structure of the U.S. Effective Federal Funds Rate (USEFFR), the Sterling OverNight Index Average (SONIA), the Euro OverNight Index Average (EONIA), and the Tokyo OverNight Average Rate (TONAR), which are overnight indexed swaps (OIS) rates. It is apparent that the CFPC model is the most flexible approach compared to alternative models. For 1-day-ahead and 5-days-ahead out-of-sample forecasts from April 07, 2014, until April 06, 2015, the CFPC model provides a well-suited forecasting approach for district groups of cross-regional yield curves. In contrast, we find that the dynamic Nelson–Siegel (DNS) model is substantially outperformed in low interest rate situations.     «

Yield curve modeling and forecasting is important in economics, finance and academia. The thesis at hand proposes a common functional principal component (CFPC) model to reveal the common factors of yield curves of different economic regions. The CFPC model is data-driven and can precisely detect the common dynamic structure of international yield curves. In particular, we extend the autoregression of lag order 1 (AR(1)) forecasting to a local autoregression of lag order 1 (LAR(1)) prediction ap...     »