Thanks for the answers!

谢谢大家的答案！

No. I try to predict the price.  I am happy when I predict correctly. I don't care if it went up or down.

[ G. T. ] 不，我试图预测价格。我很高兴，当我预测正确。我不在乎，如果它上升或下跌。

[ Sorry for posting in English but I do not speak Chinese and do not trust Google Translate. ]

Is there a "standard" day of the month when salaries are paid in China?  End of every month? Beginning of every month?  Middle of every month?

The reason for asking is: perhaps more money enters the exchanges on certain days of the month than in others.  That would cause the price of BTC to go up.

[ Google translation follows.  I hope that it makes sense. ]
[谷歌翻译如下。我希望这是有道理的。]

有一个月的工资时支付在中国的一个“标准”的一天？每月结束了吗？开始每月？每个月的中间？

究其原因，问的是：也许更多的资金进入交易所在每月的某几天比其他国家小。这将导致BTC的价格上去。

I don't understand your notation, here is mine

Basically, in the log-Brownian model the difference between successive values of Z(i) = log(P(i)) are independent random variables with probability distributions that are symmetric about zero.  Therefore after any number n of steps the probability distribution of Z(i+n) will be symmstric about the starting value Z(i).  That means Z(i+n) wil be less than Z(i) with 50% probability.  Since log is monotonic, it preserves cumulative probabilities, therefore P(i+n) will be less than P(i) with 50% probability.  What is wrong with this argument?  

(Strictly speaking, "Brownian" requires a normal distribution of increments with zero mean and fixed variance.  In practice the variance varies slowly and the distributions have fatter tails than the norma; but by the law of large numbers they become near-normal for large n.)

Er, if it is a joke, I didn get it.  They will be held in Japan, by some Japanese auction house, I imagine.  With remote bidding, perhaps?

IF they decide to sell the coins, I expect that they will divide them into manageable lots, and auction them through some auctioning house.  I believe that would be the easiest way to get the best possible price for them. (The FBI will probably do the same with the SilkRoad coins, except that they will surely use the government auctioning services.) 

But that just pushes the question forward: what wiill the auction winners do with those coins?

Chinese Slumber Method prediction for Friday April 25

Prediction valid for: Friday 2014-04-25, 19:00--19:59 UTC (not before, not after)
Huobi's predicted price: 3067 CNY
Bitstamp's predicted price: 488 USD





[ Plot legend ]

Today's data point was not bad (S = 0.0049 W = 0.614), and above the previous trend line, but not by much (by 37 CNY).  It seems still valid to use a straight-line trend, fitted by weighted least squares to the last six points, Apr/19--24.  Namely, A + B*(d-d0), where d-d0 is the number of days since Apr/19, A = 3138.67 and B = -12.01.

(The older data points should have their weights reduced when fitting the trend, but the effect on the prediction would have been small.)

The Bitstamp prediction, as usual, is the Huobi prediction divided by the currency conversion factor R, which was assumed to be 6.29 CNY/USD. It was 6.30, 6.28, 6.34, 6.27, 6.25, 6.26 at the last five Slumber Times.
 
Checking the previous prediction

Prediction was posted on: Wednesday 2014-04-23, 21:47 UTC
Prediction was valid for: Thursday 2014-04-24, 19:00--19:59 UTC (~21 hours later)

Huobi's predicted price: 3059 CNY
Huobi's actual price (L+H)/2: 3096 CNY
Error: 37 CNY (~6 USD)

Bitstamp's predicted price: 487 USD
Bitstamp's actual price (L+H)/2: 491 USD
Error: 4 USD
 
NOTE: "I never think of the future.  It comes soon enough."  --Albert Einstein

谢谢！

Maybe.  But BTC-China lost almost all its volume after 2013/12.  Perhaps customers migrated to Huobi and OKCoin.
也许吧。但BTC-中国2013/12之后失去了几乎所有它的体积。也许客户迁移到Huobi和OKCoin。

Also, Huobi and OKCoin have no trading fees.  Therefore each customer will make more trade on Huobi and OKCoin than in other exchanges.
此外，Huobi和OKCoin无交易费。因此，每个客户将在Huobi和OKCoin比在其他交易所更多的贸易。

EDIT: fixed wrong quote.

From that paper's abstract:

Aha, I always suspected that 17 was not a prime.  Where do I get in line to buy a PQC (personal quantum computer)?

To be clear: at this point, I belive that the most naive TA (linear or exponential extrapolation) may have some value for very short time periods (a couple of days), in very special circumstances. (That is what I am testing wiith that Slumber stuff.)

Certainly there is some information about the fundamentals encoded in the price, and some of the fundamentals have long-lived trends; so in theory there is some hope that TA could pick up those trends and give more precise and accurate estimates than the log-brownian model.  However, in practice it seems that the useful "signal" is so small compared to the "noise" that the trends detected by TA are mostly derived from the "noise".  Therfore, any perdiction based directly on fundamentals is bound to be infinitely more reliable than one based on price history alone.

Specifically, I do not believe that what the price was or did in 2012 has any bearing on what it will do next month.  Especially when we know that the market and environment have radically changed since then, so there cannot be any significant fundamental factor whose evolution has remained unchanged for these two years.

If the price had been 2000 USD back in 2009, for some reason, and crashed to 0.005 in 2010, why would that reason and that crash have any influence on the the Chinese traders'  behavior tomorrow?

Averaging over long time spans can improve the "signal"/"noise" ratio, but only if there is a "signal" in the fundamental factors that lasts that long.  Even for car manufacturers the fundamental factors -- like steel production, salaries, consumer demand, etc -- can change radically over a couple of years.  What then for bitcoin, whose "fundmental factors" are all virtual, and can go from 0 to 100 and back in a matter or hours?

Even if the past prices had useful information about future prices, what happened in the last two months surely should be much more important than what happened a year ago.  

You surely did not pay attention, but aftter the end-of-March crash I predicted that the price would reboud to the previous levels (~550 USD) if the Caixin article turned out to be false.  I may even have been the most bullish bull at the time.

But I don't know if I will be around for long, whether the price goes up or down,  Everything gets boring eventually...

Yes.  What will be the price on Nov 17, 2014, at 17:23:11 UTC?  The log-Brownian model say that it will be more than 487.15 USD with 50% probability, less than that with 50% probability.  (OK, there is some probability that it will be EXACTLY 487.15 USD,but it is very small.)

The problem with "fundamentals" is that some are unknown (e.g. the true rate of adoption), some unpredictable (e.g. the PBoC decrees) and the rest is hard to translate into price changes.

No wonder TA is so popular.  But of course TA cannot predict unpredictable events either.

My undertanding of the bitcoin fundamentals is that the rally of October-November 2013 was due to the opening of the Mainand China market, and the decline since then is due to the shrinking of that market.  The brief plunge to 340$ on Apr/10, apparently due to rumors in China, tell me that  if the Chinese market were to close today, and absent any other positive development, the price would fall to about 300$ or less.

On the other hand, the market outside China should support at least 120$, as it did in September.

This last estimate assumes that the April 2013 rally was NOT due to an early partial opening of the Chinese market.  Was it?

Salesmen want optimistic estimates; traders should want realistic ones. (But I am not claiming that mine are realistic.)

Those are "fundamentals". Technical Analysis assumes that all the trends of the fundamentals and their influence on price are "encoded" in the price history, so there is no need consider them explictly.

Hey, my goal was just to make some salutary fun of price extrapolations.

Frankly, I have no idea what the price will be next year.  AFAIK, the only model that has some claim to statistical legitimacy is the log-Brownian (or geometric-Brownian) model, which gives a very broad probability distribution, having 50% chance above today's price, 50% below, for any future date.

Indeed, sorry.

He writes that "0.005 USD" is just an arbitrary value.  It seems that there is no data for 2009 -> 2010, and perhaps 300 to 400% 30000 to 40000% for 2010 -> 2011.

I could try to fit a cubic polynomial to those 4 data points, but you would hate the result even more.  

I just used the data that the quoted post used.  What were the percentage changes 2009 -> 2010, 2010 -> 2011?

Don't look at me, obviously that is not mine:
* My plots are nicer.
* My private Technical Analysis method gives 0 USD/BTC by July  27th, 2014 19:23:15 UTC.
(Hey, it is still better than that of those Technical Analists on TV who predicted negative price!)
 

Note that the percent changes (not increases) from year to year were

 2011 -> 2012  300%
 2012 -> 2013 3024%
 2013 -> 2015  314%

Neither constant, not uniformly increasing.

What will be the percent change 2014 -> 2015?

The simplest formula that will fit the three data points above is a parabola. Working in log scale, this is what I get:

The red dots are actual BTC prices in USD on Apr/24 of each year, as per above. 
The first three green dots are the actual percent change from each price to the next, as per above.
The green line is the quadratic A*y^2 + B*y + C that passes through those three points.

According to this quadratic, the percentage price change 2014 -> 2015 (fourth green point) will be 0.34%.
That is, the BTC price on Apr/24 2015 will be 485.00 * 0.0034 = 1.64 USD. 

(Before jumping off the window, check what Mark Twain had to say about extrapolation.  )