In this thesis we evaluate three different techniques to predict changes in the zero-rate curve. The arbitrage-free Nelson-Siegel model is purely based on zero rates, the macroeconomic vector autoregressive model adds inflation and real activity as macroeconomic data to a latent model, and feedforward neural networks trained on the same data. Our analysis shows a positive influence of macroeconomic data on the directional forecasts for some neural networks and the principal component analysis as a tool to reduce the dimension of the zero-rate curve data. The lowest root mean squared error is obtained by the arbitrage-free Nelson-Siegel model and the best directional forecasts by the neural network using macroeconomic factors and the zero rates decoded in three principal components. In the end, we use our predictions to implement investment strategies to simulate their application for practical applications. A duration management strategy based on the afore- mentioned neural network shows the highest risk-adjusted returns resulting in a substantially higher Sharpe ratio than the benchmark.      «

In this thesis we evaluate three different techniques to predict changes in the zero-rate curve. The arbitrage-free Nelson-Siegel model is purely based on zero rates, the macroeconomic vector autoregressive model adds inflation and real activity as macroeconomic data to a latent model, and feedforward neural networks trained on the same data. Our analysis shows a positive influence of macroeconomic data on the directional forecasts for some neural networks and the principal component analysis as...     »