Bitcoin, Sentiment analysis and the Efficient Market Hypothesis, a Machine Learning Approach

Bitcoin, Sentiment analysis and the Efficient Market Hypothesis, a Machine Learning Approach (EN)

Toulias, Georgios (EN)

Gogas, Periklis (EL)

Gogas, Periklis (EN)

Arxontakis, Fragiskos (EN)

masterThesis

2023-06-23

2022-01-30

2023-06-23T11:23:56Z

Cryptocurrency has become extremely popular among investors during the last decade. Scientists around the globe predict that this turn toward crypto is still at its foundations. Investors are intrigued by the extreme volatility that results in extreme returns. When it comes to making money, benefits are followed by disadvantages, and extreme volatility is associated with riskier investments. Since investors are willing to take the extra risk by investing in Cryptos and especially Bitcoin, the least they expect is to invest in a Market that is Efficient. When a Market is efficient, all the available information to any investor, are incorporated into the price of the assets. This mean that the chances of beating the market are eliminated. This paper is testing the Efficient Market Hypothesis in the Crypto Market. The methods we are using on our research are various Machine Learning Models, with a focus on support vector machines. The frequency of the data used is weekly, with a timeline for the test sample from 23/10/2017 until 28/10/2020 which is a total of 167 weeks. The out of sample timeline is from the 04/01/2021 until 02/08/2021 which is a total of 31 weeks. The data we collected are technical, asset based, and sentiment based. We gather our data and create four different data sets. The results of the two first data sets, that do not include sentiment data, seem to back up the efficient market hypothesis. We also perform Sentiment Analysis, which we manage to do with the addition of Google trends as variables to our data set. By adding the sentiment variables, we observe a huge improvement to our model’s accuracy and predictability making them result in satisfactory tests. These results seem to be supporting the claim that machine learning models can be used as a reliable tool for predicting cryptocurrencies in the future. Even though other scientists have tried to predict the Price movement of various cryptos in the past, our diversification is the implementation of various Google Trends as variables and the influence they have on Bitcoins Price Movements. (EL)

Bitcoin (EL)

Cryptocurrency (EN)

Machine Learning (EN)

Αγγλική γλώσσα

School of Economics, Business Administration and Legal Studies, MSc in Banking and Finance

IHU (EL)

Default License

Διεθνές Πανεπιστήμιο της Ελλάδος

Ιδρυματικό Αποθετήριο Διεθνούς Πανεπιστημίου Ελλάδος (ΔΙΠΑΕ)

University Center of International Programmes of Studies (UCIPS)

Master Theses