Unless you're willing to propose some sort of non-standard regression function that makes sense as applied to the Bitcoin market, complaints about 2081% and -81% are unfounded here. The fact that the Bitcoin market has volatility with respect to the regression trend does not mean that this chart offers nothing worthwhile in terms of representing reality. To get an equation to follow the price more closely, you'd have to throw in something more exotic, like an oscillation component or something, and doing so just to get a better fit is often not the best tactic.
I understand that the idea is to have something very simple and non-complex to approximate the price of Bitcoin, but still I feel uneasy (even very uneasy) when I see the prediction error and the number of data samples on which the trend is based. And I believe that is perfectly founded complain, to which I haven't got any response yet, rather than that one regression is better than the other (linear one). What if both are not closely enough approximating the trend?
Now, you also say that it makes a perfect sense to use such a regression. Can you explain in more detail, what backs that statement?
I don't think more samples would make it more accurate for forecasting. This isn't something where we are lacking data points. What we've got is the best that we can have. You could use intra-day data for more samples, for example, but that wouldn't help anything; the daily weighted price is fundamentally representative of that.
Assumptions have to be made. Trolololo has made some. It's likely that the information necessary to best predict the price may not solely be contained in the past price data. Keep in mind, price growth can't just continue forever. But determining what outside parameters may predict where the model will diverge from a particular pattern is a hard question.
Read through the following for some formal assumptions and interpretation:
https://en.wikipedia.org/wiki/Linear_regression#AssumptionsAlso, the following:
If the guys making the charts could actually predict the market movements, they'd be rich
Edit:
The following is also helpful:
https://en.wikipedia.org/wiki/Overfitting
