| Περίληψη: | In this dissertation we tried to examine if the news affect the price of cryptocurrencies.
By saying news, we mean traditional news. Thus, we used articles from
newspapers and for cryptocurrencies the closing price of Bitcoin. We collected
the articles with the scraping method and our training and unknown data sets
included 103 and 70 articles respectively. We used the Sentiment Analysis to extract
the sentiment of each article. Also, to classify the unknown data set we used
the Naive Bayes algorithm since we trained it with the training data set. In the
following, since we got the sentiment results we estimated two regression models.
The first one, was between the closing price of bitcoin and the sentiment of the
articles and in the second model we added the variable of the website. The results
show us that from the first model there is strong correlation between the bitcoin’s
closing price and the sentiment at least in 5% level of significance. Especially, the
positive articles increase the closing price compared to the negative articles. In
addition, from the second model we concluded that when we add the variable of
the website of each article, the variable of sentiment is not statistical significant
but the variable of website is. Thus, the source of the article is more important
rather that the article itself. Finally, we found strong correlation in both models
with the time leads of the bitcoin’s closing price.
|
|---|
