Forecasting Bitcoin closing price series using linear regression and neural networks models

This is probably due to at least two reasons: high volatility of the Bitcoin price and market immaturity for cryptocurrencies. This is confirmed by the statistics reported in tables 1 and 2. The results obtained partitioning the dataset into shorter sequences also confirmed the kindness of our hypothesis of identifying time regimes that do not resemble a random walk and that are easier to model, finding that best results are obtained using more than one previous price. It is worth noting that, with this novel approach we obtained the best results for the Bitcoin price series, rather than for the stock market series as happened in the analysis of the series in their totality. As stated before, this is probably 18 due to the high volatility of the Bitcoin price, in fact it is no accident that the best result was found for the time regime identified by a translation step h of 120, where the Bitcoin prices are more distributed around the mean, showing a lower variance. This is confirmed by the standard deviation values shown in table 2. It is important to emphasize that the innovative approach proposed in this paper, namely the identification of short-time regimes within the entire series, allowed us to obtain leading-edge results in the field of financial series forecasting.