I have been experimenting with neural networks to predict future market cap of crypto.
Please dont take the results seriously, they are only the first version without any optimisation to network structure and data input. However, network training is nicely converged and control sample is looking okay.
The network quite robustly (from multiple random initialisations) predicts that the crash will finish at a total market cap around 100 billion, so it expects a further devaluation by about 66% in the next few month. It also predicts that we will rise again starting from roughly june 2018.
If these features depend on the selection of input data or are genuine I honestly cant tell yet.
So this is in no way financial advise and only uses the very FIRST results of a simple neural network. In the coming days, should I have time I will update the analysis and then I will provide a robust analysis.