Forecasting German Government Bond Development by (Deep) Neural Networks on Technical and Economic Data

Abstract

Artificial Intelligence is on its way to change many aspects of every-day-life. One often underestimated industry, where this change happens, is the financial industry. Much work in the area of Artificial Intelligence and Finance is concerned with time series forecasting. One specific and economically important type of time series is the development of government bonds prices over time. This thesis presents an overview of state-of-the art forecasting techniques on government bond prices and compares the established techniques with a newly developed, long short term memory recurrent neural network based technique for bond price forecasting. Initial results show that neural network based approaches can outperform other established techniques. However, further research in this direction needs to be conducted.