| Abstrakt: |
A number of financial factors have been investigated for the cryptocurrencies asset pricing. However, cryptocurrencies have other features such as data-driven, corporate-alike, fiat-alike features and decentralized algorithm-oriented. These factors are nonstandard distributed, skewed and long-tailed. We combine a number of blockchain factors with financial factors to explain the cryptocurrencies returns by deep learning methods. Empirically, compared to linear regressions and neural networks, our results from deep learning models such as convolutional neural network (CNN) and recurrent neural network (RNN) perform extremely well in categorization. These models successfully use the new blockchain factors’ characteristics and financial factors. However, when try to predict the exact point of the return values, deep learning models do not work so well. It still has some performance in pricing. In other words, using factors to explain accurate returns. But the prediction power is not so good, in a similar level of random forest model. |
Full text from SpringerLink