An Empirical Study of Bitcoin Pricing using an Evolutionary Framework

Abstract

Cryptocurrency merchandising is growing as an attractive area of investment. Bitcoin is much preferred over other cryptocurrencies in the world and hence is becoming more popular. But, the bitcoin price is extremely volatile. So, the forecasting of its price is highly desirable. As nature inspired-machine learning is being used extensively for time series analysis and prediction, it can be explored for bitcoin prediction as well. Also, as bitcoin is gradually increasing as a promising virtual asset, its volatility needs to be measured. This paper unveils the consequence of using ChebyShev Ploynomial Neural Networks (CHPNN) for Bitcoin pricing process. The evolutionary algorithms: Particle Swarm Optimization (PSO) and Differential Evolution (DE) are utilized for training the model. This study analyses the performance of the model through three different error measures: Root Mean Square Error (RMSE), RRSE (Relative Root Square Error) and SSE (Sum of Squares Error). It shows that DE-CHPNN predicts better day-ahead price of bitcoin.