Predicting the direction, maximum, minimum and closing price of daily/Intra-daily bitcoin exchange rate using batch and online machine learning techniques

Abstract

Bitcoin is the most accepted cryptocurrency in the world, which makes it attractive for investors and traders. However, the great challenge in predicting the Bitcoin exchange rate is its high volatility. Therefore, the prediction of its behavior is of great importance for financial markets. In this way, in recent years, Few studies were proposed based on the use of machine learning techniques to predict the direction of their exchange rate, albeit with low precision. Therefore, as a first contribution of this paper, it can be highlighted the analysis and identification of internal and external variables/attributes considered as relevant for predicting the Bitcoin exchange rate in daily and intra-daily time frequencies. The increased use of machine learning techniques to predict time series and the acceptance of cryptocurrencies as financial instruments motivated the present study to seek more accurate predictions for the Bitcoin exchange rate. For this purpose, it was used different techniques of attribute selection to candidate variables. In relation of internal variables is proposed to use Blockchain information and generate technical indicators commonly used by traders. About external variables is proposed to use international economic indices and social trends extracted from Google and Wikipedia. As a second contribution, a methodology is proposed to predict the direction of the Bitcoin exchange rate against the dollar. In addition, it was explored the possibility of directly predict the maximum, minimum and closing prices, including these information to predict the trend. For this, Artificial Neural Networks, Recurrent Neural Networks, Support Vector Machines and Ensemble models (combining regression and clusterization) were used. As a third contribution for intra-daily time frequency, the data-stream learning methods are explored under the hypothesis that Bitcoin price presents a non-stationary behavior. Thus, it is observed that in long term, Bitcoin behaves more like a traditional instrument and, therefore, is increasingly affected by the international context and economic fundamentals. Likewise, the results showed that the selected attributes and the best machine learning model achieved an improvement of more than 10% in accuracy, for the price direction predictions with respect to the state-of-the-art papers, using the same period of information. In relation to the maximum, minimum and closing Bitcoin prices regressions, it was possible to obtain Mean Absolute Percentage Errors between 1% and 2%. Finally, in the prediction of intra-daily price movement, through the use of data-stream learning techniques, is obtained a result that improves more than 6% in accuracy to other previous studies.