2020/01/04 by Nicola Uras, Uras, Nicola, Lodovica Marchesi +5
Decision Sciences · Economics, Econometrics and Finance · #Stock Market Forecasting Methods #Market Dynamics and Volatility #Complex Systems and Time Series Analysis
paper · pdf · doi:10.48550/arxiv.2001.01127
This paper studies how to forecast daily closing price series of Bitcoin,\nusing data on prices and volumes of prior days. Bitcoin price behaviour is\nstill largely unexplored, presenting new opportunities. We compared our results\nwith two modern works on Bitcoin prices forecasting and with a well-known\nrecent paper that uses Intel, National Bank shares and Microsoft daily NASDAQ\nclosing prices spanning a 3-year interval. We followed different approaches in\nparallel, implementing both statistical techniques and machine learning\nalgorithms. The SLR model for univariate series forecast uses only closing\nprices, whereas the MLR model for multivariate series uses both price and\nvolume data. We applied the ADF -Test to these series, which resulted to be\nindistinguishable from a random walk. We also used two artificial neural\nnetworks: MLP and LSTM. We then partitioned the dataset into shorter sequences,\nrepresenting different price regimes, obtaining best result using more than one\nprevious price, thus confirming our regime hypothesis. All the models were\nevaluated in terms of MAPE and relativeRMSE. They performed well, and were\noverall better than those obtained in the benchmarks. Based on the results, it\nwas possible to demonstrate the efficacy of the proposed methodology and its\ncontribution to the state-of-the-art.\n