Democritus University of Thrace (DUTH)
Forecasting financial time series with machine learning techniques
Abstract
dc:descriptionΙn this dissertation I forecast financial time series with machine learning methodologies. During my research I propose various novel forecasting schemes and attack four problems in a machine learning approach: short and long-term exchange rate, housing prices and bank insolvencies forecasting. More specifically, I propose a novel forecasting methodology in short-term exchange rate forecasting that couples a machine learning with a signal processing technique. In the same field I consider machine learning in long-term forecasting, that has never been used before in the relevant literature. When it comes to housing prices forecasting in the U.S., the machine learning model outperforms the econometric ones considered in this dissertation, while it foresees the sudden 2006 downturn in housing prices two years earlier than the actual event. In bank insolvency forecasting, the two-step machine learning models are more accurate than the currently used logistic regression models from the bank supervision authorities in the U.S. The machine learning models outperform all the econometric models examined in this dissertation in terms of forecasting error and directional forecasting accuracy Overall, the empirical findings reveal the superiority of machine learning to econometric models in forecasting the selected financial time series examined in this dissertation.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2015
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Plakandaras, Vasileios
- Πλακανδάρας, Βασίλειος
Subjects
dc:subject × 6Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/36434