Search academic texts

Information × Registration Number 2124U004713, Article popup.category Препринт Title popup.author Kypybida Roman popup.publication 01-01-2024 popup.source_user Український католицький університет popup.source https://hdl.handle.net/20.500.14570/4851 popup.publisher Description Lots of people want to make a fortune and one can try himself in trading assets like Bitcoin, but how to determine when to sell or when to buy? One could try his chances by analyzing the environment events and news about Bit- coin and try to predict if they have positive or negative effect on the price of this asset. But how can you make an analysis of so much information and make it both fast and quality? Human mind can do that, but it would take too much time and it lies on a single subjective opinion. What if there was something more objective? Sen- timent analysis can allow you to automate the process of evaluating the sentiments and price movements using automation and computational resources of a cold fast machine. Here, we propose using LLMs as they are new popular approach in NLP and are not yet fully researched. The results of the simulation indicate that increasing number of parameters of the model does not necessarily improve the results and that more advanced LLMs can indeed beat primitive approaches like “holding strategy” and less advanced models like FinBERT. Also, we indeed confirm that sentiments are meaningful and have connection to price movements and can be used for trad- ing and that the best approach is using average raw sentiment as price movement prediction without using ML approaches for the prediction task. These findings sug- gest one can reach positive return by predicting correctly only 40% of sentiments by using averages of raw sentiments as price predictors. popup.nrat_date 2025-05-09 Close