Comparison of News Sentiment Effect on Cryptocurrency and Stock Returns

his paper analyses and compares the impact of news sentiment on cryptocurrency and stocks returns. Non-similar structure of the participants of these markets is assumed to result in the different reaction towards the new information. Among the possible reasons for this phenomenon, there are heuristic behavior and informational herding of individual investors which prevail on cryptocurrency markets. Many papers have already been devoted to the discussion of the sentiment effects on each of the markets separately, but none has compared the effects before, especially using the similar models for each. In the current research, this task is accomplished using the modern-day natural language processing techniques and time-series analysis. Empirical estimation is structured in the following way. At first, relevant news articles are chosen from the collected dataset using the created keywords and then marked from 1 to 9 based on their polarity, using the sentiment analysis model, which performed best on the test set. Then, the impact of yesterday's news on today's returns is analyzed using the ARIMAX-GARCH model. Later, the obtained samples of coefficients are compared using graphs and statistical tests. The comparison of sentiment effects is useful not only for further academic research but also has the practical implications in such areas as the design of trading strategies or return forecasting models.