Market sentiment is an index of investors attitude with respect to the financial market. An estimate of this index can be useful to traders since it is correlated with the trend of stock prices: if the majority of investors has a positive sentiment, stock prices will probably have an increasing trend; on the other hand, if the common feeling of investors is negative, a downward trend is expected. However, it is complex to design a robust estimator of the market sentiment able to predict the market trend, since it depends on many different factors, therefore several estimators are applied in the literature. In the last decade, with the growth of the applications based on Artificial Neural Networks and of Natural Language Processing algorithms, one of the methods introduced to estimate the market sentiment is the application of Sentiment Analysis methods. In particular, they are exploited to extract a sentiment index based on news and documents concerning listed companies or the financial market in general. This project analyses sentiment values extracted from Reuters news and from tweets of Twitter regarding the companies belonging to the S&P 500 index. Specifically, in the first part of this thesis an overview on main Reinforcement Learning models and algorithms and on Natural Language Processing techniques is proposed. Then, after an extensive analysis of the available sentiment features and of the historical series of the S&P 500, the focus of the present work is firstly on the determination of the efficacy of these sentiment features on the prediction of the trend of the S&P 500 index through Supervised Learning methods, with the final purpose to train a Reinforcement Learning agent able to profitably trade on the U.S. Stock Market.
Il sentimento del mercato (market sentiment) è un indice della fiducia media degli investitori sul mercato finanziario. Questo indice viene stimato ed utilizzato in quanto strettamente legato alla predizione dell'andamento dei prezzi degli strumenti finanziari: ci si aspetta che, se la prevalenza degli investitori è fiduciosa, le quotazioni azionarie avranno una tendenza al rialzo; se viceversa la maggioranza degli investitori è pessimista, il trend sarà al ribasso. Tuttavia è in generale complesso estrarre un indice di sentiment che sia robusto ed in linea con quello che accadrà sul mercato, in quanto questo dipende da molti fattori, perciò svariati metodi di calcolo sono presenti in letteratura. Con lo sviluppo delle Reti Neurali e dell'Elaborazione del Linguaggio Naturale, uno dei possibili stimatori del sentimento del mercato è basato su metodi di Sentiment Analysis. Esso consiste nell'estrazione di un indice di sentiment a partire da news o testi riguardanti le compagnie quotate o il mercato finanziario stesso. In questo progetto vengono analizzati indici di sentiment estratti da news di Reuters e tweets provenienti da Twitter riguardanti le compagnie che compongono l'indice S&P 500. In particolare, nella prima parte della tesi si effettua un'ampia analisi delle principali tecniche di Reinforcement Learning e di Natural Language Processing. Successivamente, dopo un'attenta analisi degli indici di sentiment disponibili e delle serie storiche dell'S&P 500, il presente lavoro si focalizza in primo luogo sulla determinazione dell'efficacia di questi indici di sentiment nella predizione del trend dell'S&P 500 mediante l'utilizzo di metodi di Supervised Learning, con l'obiettivo finale di addestrare un agente di Reinforcement Learning in grado di fare trading generando un profitto.
Impact of sentiment analysis on automatic financial trading through reinforcement learning
BONETTI, PAOLO
2018/2019
Abstract
Market sentiment is an index of investors attitude with respect to the financial market. An estimate of this index can be useful to traders since it is correlated with the trend of stock prices: if the majority of investors has a positive sentiment, stock prices will probably have an increasing trend; on the other hand, if the common feeling of investors is negative, a downward trend is expected. However, it is complex to design a robust estimator of the market sentiment able to predict the market trend, since it depends on many different factors, therefore several estimators are applied in the literature. In the last decade, with the growth of the applications based on Artificial Neural Networks and of Natural Language Processing algorithms, one of the methods introduced to estimate the market sentiment is the application of Sentiment Analysis methods. In particular, they are exploited to extract a sentiment index based on news and documents concerning listed companies or the financial market in general. This project analyses sentiment values extracted from Reuters news and from tweets of Twitter regarding the companies belonging to the S&P 500 index. Specifically, in the first part of this thesis an overview on main Reinforcement Learning models and algorithms and on Natural Language Processing techniques is proposed. Then, after an extensive analysis of the available sentiment features and of the historical series of the S&P 500, the focus of the present work is firstly on the determination of the efficacy of these sentiment features on the prediction of the trend of the S&P 500 index through Supervised Learning methods, with the final purpose to train a Reinforcement Learning agent able to profitably trade on the U.S. Stock Market.File | Dimensione | Formato | |
---|---|---|---|
2020_04_Bonetti.pdf
accessibile in internet per tutti
Descrizione: Thesis text
Dimensione
38.16 MB
Formato
Adobe PDF
|
38.16 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/153366