Before 2017 mainly only IT guys involves in Bitcoin, hold it only for experimental purpose, or technical research interest. The whole market cap is small with high volume, no big firm or government or finanical expert investors involve, and therefore the factor affecting the demand and supply curves underlying is only mining reward changes, which forms a simple, smooth and pure 4-year cycle regression trend. But the increase in popularity of bitcoin draws a lot of non-IT finanial investment and government regulations, bitcoin is no longer a pure technical stuff but a real investment tool, or somehow a real "risk asset". It begins to fluctuate in "stock market" style with high uncertainity. The demand and supply curves underlying already changed, and thus the original regression no longer hold. All the forecasts trying to squeeze all the price moves before and after the finanical firm and governemnt involvement into a single regression, such as some data model like highly overfitted non-linear log(log(p)) without any strong reason or explanation on how log log function related to the underlying demand supply environment, are unreasonable.