So put simply, Chinese bed time = coin dump because they only want to hold fiat overnight?
Not quite...
It is a fact that Huobi's volume is very low around 03:00 am local time. (OKCoin has a steady background traffic that does not disappear even at those times, may be fake volume.)
My impression (not yet verified quantitatively) is that the prices sampled daily
at that time of the day are more likely to follow steady trends over successive days than prices sampled at other hours, or even mean daily prices.
It is my impression, also, that the price at that hour often deviates from the current trend whenever the trade volume is large at that hour, compared to the daily volume (i.e., when Huobi's clients stay up trading through the night). However, the price sometimes returns to that trend on the next night, if Huobi's clients go to bed at the usual time.
These impressions are the basis of the "Chinese Slumber Method". I look the price and volume around 03:00 am. If the volume too high relative to the day's total volume, I ignore that data point (W near zero). Then I try to fit a simple trend formula to the most recent good points (W near 1), either a straight line or a shifted exponential, and extrapolate it to predict the value at the next "slumber time".
At each day I must decide whether to continue with the previous trend or assume a trend break and start a new trend, Bad "slumber points" are preferred trend break points. Apart from this general principle, the choice of breaks is still subjective (but I hope to automate it, with dynamic programming, if Huobi does not collapse on April 15).
I do not believe that past prices by themselves are useful to predict future prices. However, I believe that the price is influenced by certain "concrete" factors, such as the amount of money and coins in the exchange, or the general mood of the traders; and that these factors tend to change gradually over several days -- but suddenly at times, in response to news or new exchange policies. The "Chinese Slumber" method tries to determine the price trends determined by those "concrete" factors.
I don't know why the "slumber prices" should follow the trend more faithfully than prices sampled at other times (if they indeed do). One guess is that most traders return to their "base positions" (their preferred ratios of CNY:BTC in their accounts) before going to bed; and those moves somehow cause the price to drift back to the "ideal" price determined by the "concrete" factors. Whereas, during the day they may deviate considerably from those positions, thus adding "noise" to the price. But that is only a guess.
My brain is not as big as a lot of people on this forum and I can't claim to fully understand your method. But i can grasp the concept (I think).
I like to use more, visual prediction methods. Looking at your diagrams there is obviously some truth to it. Are you saying even though the volume is low at these times on Huobi, the overall sentiment is like an amplification of feeling causing a chain reaction, through the collective "final trade" of the night?
You made predictions last night. Do you still hold true to those predictions? where do you see the price going in the next 2-4 hours?