[..] from what I understand, the derivative of price does not follow a normal distribution.  There are far too many data points with high standard deviations.  Instead you need a distribution with so-called fat tails.
[/quote]
Yes, that is what I mean.  I will try to make some plots later today.

The "voucher" mechanism adopted by BTC-China (which of late had only 1/10 the volume of Huobi, by the way) apparently uses e-commerce sites as money transfer channels.  Will China's central bank be fooled by that trick?  Anyway its volume seems to have fallen quite a bit in the last 24 hours.

It usually does at this time of the day.

Indeed, in other time intervals and other scales it is quite visible that the amplitude of the "random" component (the price changes) varies over a scale of tens of hours, so that there are "steadier days" and "wilder days".  That should account for "clustering of high standard deviation data points".

In that interval, in particular, the red curve has larger deviations in the first half, smaller in the middle (from ~120 to ~150) and then intermediate ones near the end.

However, that effect seems to be rather subtle.  Perhaps it was the non-normal distribution of the changes that made the red curve look different?

Good. Can you tell what looked "wrong" with that one?

Not at all! "Random" means "due to arcane reasons that we mortals are not given to understand". 

The red line (Y) is Bitstamp's hourly closing price from 2014-01-05 to 2014-01-13.

The blue line (X) is generated by starting with

  X(0) = 840
  Z(0) = log_10(X(0)),

then, for i = 1,2,...

  Z(i) = Z(i-1) + 0.058*RND(),
  X(i)  = 10^Z(i)

where RND() is a random number with normal distribution, mean 0 and standard deviation 1.

In simple words, the simulated price X(i) increases or decreases at each step by a random percentage, without regard for the past history.

I had to try twice, with different random generator's seeds; on my first attempt the simulated price dropped too low at one point, it would have been a dead giveaway.

One of these plots is actual hourly closing price (USD/BTC) from one exchange, the other is a series of fake prices generated by a simple "log-Brownian" model.

Can you tell which is which, without looking at other charts?
If you think you can, what difference do you see?

Oops, thanks!

I saw several messages claiming that Huobi's volume is "fake", but I have yet to see evidence of that.

Certainly the 0% fee makes it very attractive for day traders, both human and robots.  One would expect a higher volume of legitimate transactions just for that reason (and because China is so big).

As others have noted, "robots are people too". A trasaction by robots or humans is not "fake", unless the two accounts involved belong to the same person.  Is there reason to believe that is happening?

Huobi's volume changes with the hour of the day as expected for a market centered on China's latitude.  In particular, it falls regularly to near zero between 19:00 and 23:00 UTC, which I believe is 04:00 am to 08:00 am in China.  (There are a few days when that pattern was broken, usually during large crashes or rallies.)  Legitimate human or robot speculation is expected to behave like that; but would one think of putting to sleep a robot that is trading with himself only to build up volume?

On the other hand, MtGOX has been caught posting "fake" price data during the Dec/01 crash. It and other exchanges do seem to have robots making tiny transactions in the bid/ask spread, which may add up to a significant fraction of their volume.

Methinks Huobi deserves as much (os as little) trust as the other exchanges...

(I sense a strong dislike in this forum to the idea that Chinese speculators may be dominating the Bitcoin market.  Is that so?)

Quite possibly. But even so it makes sense to know what it is doing...


I have a suggestion for that text: instead of "float" say "exponential format, such as '2.3e-2'".  For many programmers, "2.07" is still a float value.

I noticed that your chart squeezes out "empty" intervals, which seem to be common in some sites. For example, this is an excerpt of the 1-minute data for BTC-China provided by bitcoincharts.com:


2014-01-14 18:29:00   5009.61   5009.61   5009.61   5009.61   0.43   2154.13   5009.61
2014-01-14 18:30:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:31:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:32:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:33:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:34:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:35:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:36:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:37:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:38:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:39:00   1.7e+308   1.7e+308   1.7e+308   1.7e+308   Infinity   Infinity   Infinity
2014-01-14 18:40:00   5009.61   5009.61   5009.6   5009.6   0.41   2043.92   5009.6


(those are UTC times)

On your chart the 18:29 datum is immediately followed by the 18:40 datum.  That makes the horizontal scale non-linear.

While one can get used to it, wouldn't it be clearer and simpler to leave blank spaces for empty intervals (i.e make the horizontal axis linear, with the abscissa computed from the time rather than from the datum's index)?

For stock and commodity prices, it makes sense to omit the intervals when there is no trade, such as weekends and nights.  That's because the market actually disappears during those intervals -- so that Friday 4:00 pm coincides with Monday 9:30 am in "Wall Street time".  For Bitcoin prices, however, the global network effectively keeps the market running continuousy, even when one exchange is down or has no transactions. 

By the way, in a market that is continuously active, the "opening price" and "closing price" become rather arbitrary data points.  Thus the color of each candle should perhaps be defined by comparing its closing price with the previous closing, rather than with its opening price.  Or perhaps one should redefine the opening and closing prices by linear interpolation between adjacent transactions.  But presumably that would be too messy to implement, and too confusing to people who are used to the current scheme.

I should add that I find your charts wonderful, and that is why I am complaining so much. 

Perhaps someone will be interested in this comparison between the prices of MtGOX, Bitstamp, and BTC-China:



I should have used Huobi instead of BTC-China, since the latter appears to be nearly dead now.   However the website where I got the data (http://bitcoincharts.com/charts/) does not work with Huobi.  Is there an alternative source?

Hm, would you consider using UTC for display too?

Bitcoin price events and discussions are global in scope, not particularly related to the reader's locale.  It may be easier for everybody if everybody uses the same clock.

Oops, sorry, forget it.
I got confused because other charts that I have been loking at use UTC rather than local time, and I assumed your charts used it too. Stupid of me.

OK, would you then consider shifting the price column (for both bids and asks) down by half a line? That way it would be obvious what price range each volume entry refers to.

One disadvantage of the current setup is that the intervals shift with time.  Sometimes the 2h chart plot shows 12:00:00 to 13:59:59, sometimes it shows 13:00:00 to 14:59:59, depending on when one opens the chart.

This inconsistency makes it hard to discuss things like volume spikes, price drops, which market is the leader, etc.: I may see one large spike, the other guy reading my post may see two half-size ones.

So the suggestion is that the beginning of the intervals be always synchronized to a multiple of the time step, independently of the current time. That is, the intervals in the 1d chart should always start at 00:00, those on the 3h chart start at 00:00, 03:00, 06:00, etc., those on the 1w chart should start always on Sunday 00:00, and so on.

I have been assuming that the date and time shown on the summary line at upper left (or when selecting a candle with the crosshairs) is the start of the time interval in question.  Is it not?

Great job, thanks!

The prices in the ask depth table at the upper right seem to be off by 1 step.  For example when they are

  5000 475
  4950 320
  4900 112
  4850  24

4867

presumably they should instead be

  5050 475
  5000 320
  4950 112
  4900  24

4867

The bid depth prices seem correct.

Usually Huobi is nearly dead from 2:00 am to 6:00 am local time  (17:00 to 21:00 UTC).  But today volume increased to daytime leves at 4:15 am (19:15 UTC), and the price jumped up from ~4800 CNY to ~4900 CNY at the same time.  It then jumped again to ~5000 CNY at 5:30 am (20:30 UTC).

I deduce, dear Watson, that this rally was born in the "West" (perhaps at BTC-e) and was then carried to Huobi by arbitrage, or was generated at Huobi by "Western" traders.

EDIT: volume at Huobi was down to sleep metabolism again at 6:am (21:00 UTC). Wops, not really.

Very interesting.  Could this be the reason for the alleged difficulty of moving USD out of Mt Gox?

A government's attitude towards cryptocoins presumably depends on their perceived effect on the country's economy - or, rather, what it thinks the economy should be.

Cryptocoins have been used mostly for speculation so far, but are being promoted as a means to move money across national borders, bypassing bank fees and money export regulations.  That may explain why bitcoin became so popular in China, at a time when its price was stagnating on the West.

Most countries do not care for speculation and zero-sum games, even if they ruin some people's lives.  Financial gambling tends to concentrate wealth, and politicians would rather serve one billionaire than 1000 millionaires. Such schemes get banned only when banks or powerful people lose big money with them.

Most governments do care for international money transfers. Countries that are likely to be on the receiving end (such as the US, Singapore, and tax heavens) will probably love cryptocoins, while countries likely to be on the other end (like China, India, and Latin American countries) will want to restrict or ban them, as they do with other payment methods.

I suspect that China is planning to ban cryptocoins, but is doing it slowly so that big investors have the time to sell their coins to smaller fish.  I presume that they would rather deal with 10,000 workers who lost 10,000 USD each than with one tycoon who lost 100 million because of the ban.