The Effect of cohesion frequency and qualitative cohesion score on Prediction Performance of the Bitcoin Price Jae Hyun Byun

Abstract

The Effect of cohesion frequency and qualitative cohesion score on Prediction Performance of the Bitcoin Price Jae Hyun Byun Department of Business Informatics Graduate School of Hanyang University The study focuses on enhancing financial market prediction models by integrating investor opinions, with a specific emphasis on the use of advanced sentiment analysis technologies like RoBERTa and DistilBERT. Traditional financial models often overlook the crucial role of investors' opinions in shaping market dynamics, a gap this research aims to fill. This research introduces a novel approach where investor opinions are quantified and incorporated into market analysis models. The key innovation lies in the use of RoBERTa and DistilBERT, two state-of-the- art sentiment analysis tools, to capture and interpret the nuanced opinions of investors from online discussion platforms. These tools are integrated into Long Short-Term Memory (LSTM) models, widely recognized for their effectiveness in time series prediction. A detailed case study on Bitcoin price prediction highlights the impact of this approach. The study compares the performance of LSTM models with different cohesion indicators: a general model, a model informed by RoBERTa, and a model using DistilBERT-informed cohesion. The results show that the DistilBERT-informed model significantly outperforms the others, indicating a more precise understanding of market sentiments leads to better predictive accuracy. The model with RoBERTa-informed cohesion also shows improved results over the general model, though not as pronounced as with DistilBERT. Statistical analysis through Paired Sample T-Tests further validates these findings. The tests reveal a statistically significant improvement in the performance of models using RoBERTa and, more notably, DistilBERT- informed cohesion, compared to the general model. These results emphasize the effectiveness of incorporating advanced sentiment analysis tools in financial prediction models. In conclusion, this study demonstrates the substantial benefits of integrating sophisticated sentiment analysis technologies like RoBERTa and DistilBERT into financial market prediction models. By capturing the often-overlooked investor opinions more accurately, these models offer a more comprehensive and precise tool for understanding and forecasting market trends, particularly in volatile sectors like cryptocurrency. This approach not only enhances current market analysis practices but also paves the way for future advancements in the field of financial analytics