Neural Networks and Deep Learning have gained increasing importance in the research on diverse subjects and topics. Particularly, Neural Networks' ability to extract, learn and represent non-linear relationships between variables as well as to reproduce such relationships given new data has attracted great attention to financial time series modeling with Neural Networks. We focus the master thesis on the applicability, practicality and potentials of financial crisis forecast with an emphasis on time series classification of economic states with Neural Networks. In the training and validation of Neural Networks, we observe stability problems in modelbuilding and testing performances, which are common due to over-parameterization and inadequate utilization of data in Neural Networks. Motivated by this observation, we develop two mechanisms, which are the Ensemble Sensitivity Analysis and the modified
Walk-Forward Validation. Upon developing the two mechanisms, we demonstrate Neural Networks in classification as well as the proposed improved mechanisms in an example focusing on the forecast of the economic states with Neural Networks in classification. Theoretically, the demonstration provides practical evidence to the two improved mechanisms we develop and offers distinctive perspectives in the investigation of a time series classification problem. Practically, we build a classification model that effectively and efficiently makes forecasts of financial crises in the future. Last but not least, we also acquire and present pragmatic ideas regarding potential directions for further research based on the development and finding in this master thesis.
«
Neural Networks and Deep Learning have gained increasing importance in the research on diverse subjects and topics. Particularly, Neural Networks' ability to extract, learn and represent non-linear relationships between variables as well as to reproduce such relationships given new data has attracted great attention to financial time series modeling with Neural Networks. We focus the master thesis on the applicability, practicality and potentials of financial crisis forecast with an emphasis o...
»