Πρόβλεψη χρηματοοικονομικών χρονολογικών σειρών
KeywordsΧρηματοοικονομικά ; Χρονολογικές σειρές ; Μηχανική μάθηση ; Χρονοσειρές ; Time-series analysis
The prediction, analysis and the study of financial time series variables, mainly stocks, as well as the development of a financial strategy for the stocks market and generally, as subject matters are very popular subjects not only in the academic sector but also for everyone who has ambitions on making profit through this practice. The reason this subject matter is so popular is more than obvious to everyone who wants to build a model which will manage to "beat" the market, forecast when a stock rises or falls, when someone should buy or sell, which stock will earn him more in the long-term period, etc. The time series analysis has plenty of techniques the purpose of which is to predict the future either concerning a day, an hour, a week, etc. However, the time series analysis cannot take a lot of other factors which have been proven to affect the market into consideration, for example the daily news. Thus, with the extensive use of machine learning techniques, a lot of new methods have been proposed and plenty of publications with relevant results have been made. Taking a step further, the predictions are now being made with combination of techniques such as neural networks with sentiment analysis and the use of historical data, with the results being increasingly promising. This thesis is about the analysis, study and prediction of stocks with the use of techniques, such as time series analysis and machine learning. The use of machine learning techniques for this subject shows that the accuracy of predictions can exceed 70% in some models while for the rest their performance is just satisfactory. In either case though predictions exceed 50%. The predictions through time series analysis can also be considered as satisfactory.