There is a difference between the naive sample size (just your raw n value), and the information content of your sample size. How many crashes has bitcoin had?

There is a difference between the naive sample size (just your raw n value), and the information content of your sample size. How many crashes has bitcoin had? Depending on how you measure it, somewhere between 3-10? And even then, these crashes varied in magnitude, and some occurred at times when bitcoin was fundamentally different than it is now. The point is, you can have the richest data sources you want, at any time granularity, but you can't solve a hard information problem, which is that you only have a small small handful of true points to train on. The rest of the training data is going to be extrapolating or learning from small movements, which should have different dynamics from crashes.