Empirical Analysis of Digital Currency Herding and Contagion Effect Using Deep Learning Approaches

In recent years, the use of virtual or cryptocurrency-based money has grown. Blockchain, a digital accounting system, is the foundation of cryptocurrency. The markets for cryptocurrencies are eminent for their high volatility and quick price swings. People tend to be more likely to follow the lead of others at times of notable price fluctuations out of a concern of losing money or missing out on possible gains. In this work, we taking the Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), the hybrid combinations of LSTM-Convolutional Neural Network (CNN), LSTM-Gated Recurrent Unit (GRU) and for Facebook Prophet (Fb-Prophet) model smoothing, triple exponential smoothing (Holt-Winters Method) is utilized, which adds seasonality into the forecasting model. In this work bitcoin exchanges providing real-time information casing the period from January 2012 to July 2020 is utilized. The price of cryptocurrencies fluctuates significantly throughout a range of time periods, from short-term intraday swings to long-term patterns that last for weeks or months. By altering the network layer design and the duration of the input sequences, LSTM-GRU models configured to adapt to various time horizons. In contrast to previous iterations, the hybrid LSTM-GRU performs well. When compared to other models, the model's Mean Absolute Error (MAE) score of 0.056 and Root Mean Square Error (RMSE) score of 0.073 are quite acceptable. Forecasting the value of cryptocurrencies will help traders and investors in making better-informed choices when purchasing, disposing of, and keeping assets.