Δημιουργία μοντέλου για την αξιολόγηση αποφάσεων επενδύσεων στηριζόμενοι σε τεχνικές sentiment analysis και στατιστικές μεθόδους
Creating a model for evaluating investment decisions based on sentiment analysis and statistics methods

Master Thesis
Author
Τσόπελας, Θωμάς
Tsopelas, Thomas
Date
2022-07-02View/ Open
Keywords
Predictive analytics ; Μετοχές ; Χρονοσειρές ; Προβλέψεις ; Efficient market hypothesis ; Μοντέλα ARIMA ; Python ; NLP ; Neural network ; VADERAbstract
This thesis contains a study in the research field of Predictive Analytics, making use of the capabilities of the Python language and the available libraries as well as an attempt is made to predict the price of a stock as a time series at an initial stage. Regarding the content, some basic concepts related to the share-prices , financial analysis and investments, such as the expected return, efficient market hypothesis, rate of return of an investment, the risk involved in an investment and natural language process as well as sentiment analysis and how these two fields apply to the modern economy and everyday life. The concept of time series and its main characteristics are analyzed in theoretical level, such as stationarity, autocorrelation, periodicity are presented, also the linear models AR(p), MA (q) and ARIMA (p,d,q), which can "fit" the data of a stationary time series and provide forecasts with a relative accuracy.
Finally, the theoretical background of time series and linear models is applied to act in a real share study for the period 17/11/2016 – 16/11/2021. The research concerns the shares of "Home Depot" company which is listed on the American Stock Exchange Dow Jones. The growth of the world wide web has led to the rapid growth of data wherever it is available for processing and analysis in order to draw conclusions, that's why they have build tools for extracting and processing this data. The analysis of personal opinion and belief is a key point where it can be done through the above extracted data. An already trained neural network is then used for correlation publications of authoritative financial sources with the course of the stock through natural language processing and in trying to derive results in the field of sentiment analysis of the financial information site «finviz», using the technique of an already trained dictionary. To develop the time series analysis and sentiment analysis models, the Python programming language was used.