Research on the applicability of time series models for bitcoin price prediction: Taking ARIMA and LSTM as examples

This paper takes the prediction of Bitcoin price as the research goal and comparatively analyzes the applicability of two types of time series models, autoregressive integrated moving average model (ARIMA) and long short-term memory (LSTM). Based on the daily closing price data of Bitcoin from October 2013 to December 2021, the ARIMA model and the LSTM model were constructed respectively, and the performance of the models was verified through stationarity tests, parameter optimization and prediction error analysis. The research results show that the ARIMA model performs stably in the short-term prediction of linear trends, but it is difficult to capture the nonlinear fluctuations of Bitcoin prices. Due to its long-term memory ability, the LSTM model has more advantages in nonlinear feature extraction and long-term dependency processing, and its prediction accuracy is significantly higher than that of the ARIMA model (MAPE is reduced by approximately 60%). Furthermore, this paper explores the potential of the hybrid model and finds that the method combining wavelet decomposition and model fusion can further improve the prediction effect and provide a direction for future research.