I appreciate your combined use of regression analysis and volume analysis
Thanks!...
I believe that it is a consensus that regression does not help predicting future stock prices from past prices alone; and the limited analysis that I have done agrees with that. You may have seen my previous posts on the log-Brownian (or geometric Brownian) model; although there are some more sophisticated models, they do not seem to be useful for predictions.
The model above exploits some features specific to Huobi that hopefully will let us get around that limitation. First, Huobi seems to have little trade by unsupervised robots (which have much more varied behavior than humans, and therefore should be less predictable). Their clients apparently have relatively well-defined demographic profile, and are relatively isolated from the rest of the world in terms of economic news. And, those clients are trading a single item (Bitcoin) and have a relatively narrow channel for injecting money into the system (local banks)
Also, Huobi has a lot more volume and liquidity than the other exchanges; the second largest, OKCoin, has only half as much. So their price is relatively indifferent to the price in those exchanges. (Rather, it is the other exchanges that apepar to track Huobi's price.) It is apparent from the charts that price volatility is correlated with trade volume at Huobi -- but not at other exchanges, like Bitstamp.
As explained earlier, the Chinese Slumber Model assumes that before Huobi's clients go to bed, their "shop closing" actions are such that the price returns to some "natural" value by the "slumber times" (around 19:00 UTC, 3:00 am in China). That "natural" price is suposedly masked during daytime by the large swings that result from intense trading. Moreover, the model assumes that the "natural" price is determined less by trading than by other "physical" factors that vary slowly with time -- such as the amount of money available for trading at the exchange.
The trend analysis is therefore trying to determine the variation of those "physical" factors --- and not of the prices set by trading, which, as observed above, appear to be largely random.
Houbi's artificially inflated trading activity is well known by now
I would not call it "artificially inflated". Since they charge no fees, we should expect that they have many more actively trading clients than other exchanges, and also that each client generates a lot more trade volume than a client would elsewhere.
I have seen many claims that their volume is "fake", but I have yet to see evidence of that. Their volume data seems to be consistent with humans assisted by robots or scripts.
On the other hand, evidence fo fake volume at MtGOX and other sites are quite strong. For example, by the end of January MtGOX's volume was as low as 15 kBTC/day; some dumb robots, that worked steadily day and night, may have accounted for half of that.
It may not be a pleasant conclusion for non-Chinese bitcoiners, but it seems undeniable that bitcoin trading is now largely a Chinese thing; specifically, that the price of bitcoin is now largely determined by the Chinese traders at Huobi and OKCoin., and carried to other exchanges by arbitrage.
I am surprised that sources like Coindesk pay so little attention to Huobi and OKCoin -- or even try to hide their existence. I found only
one interview with Huobi's CEO -- and was on a general finance source, not in a bitcoin news site. We have only vague hints about the number, demographic profile, and motivations of their clients.
Actually, the articles at bitcoin news sources seem to be uncritical copy-paste of press releases, and have very little hard statistics -- such as the number of active users of bitcoin, active traders at each exchange, etc.
Where have all the investigative journalists gone?
Poll
