Wavelet analysis of bitcoin prices in 2015
So I finally got my quant platform up and sort of running, and so I’ve decided to work this by doing a quick wavelet analysis of bitcoin prices.
Please note this is coming off the top of my head, and the next post may be one in which I talk about how I messed all of the numbers.
To summarize briefly a wavelet is sort of a mix between a pulse and a wave. When I look at bitcoin prices, I see “waves” and a wavelet analysis let’s me see were the waves are and how strong those waves are. The reason this is useful for me is that if I know that the price of bitcoin is fluctuating +/- USD 2 over with a frequency of about 2 hours, that tells me where I should put my buy and sell orders.
Someone has done this before
http://arxiv.org/abs/1406.0268
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2547281
But he was using daily data, whereas I’m interested in minute by minute changes
So I finally get everything together and looking at the data at bitfinex, and this is what it looks like
The y axis is the period in (2**y) minutes. So 0, shows how strong the 1 minute bitcoin fluctuations are. 6 shows how strong the (2**6) = 64 minute fluctuations are.
The cool thing about this diagram is that it seems to say that there are no fluctuations that are longer than 8 hours. If you look at the volumes, you see that there is a lot of structure in the volume, and you have bursts of large volume. You can also see the pattern of changes at the one week level.
Something that is pretty interesting is that see that there are periods in which volume bubbles up. When volume increases so does short term volatility, but the increase in short run price volatility doesn’t make much different in the long term trend in prices.
One thing that these graphs seem to suggest is that bitcoin is in fact a lot less volatile than people assume. If you look at the bitcoin prices you’ll see that in fact, there is not that much volatility. The bitcoin prices have been moving in one direction and that is down, and it’s been a pretty steady trend.
The other thing that I’ve been investigating is to find evidence either for or against what I call the Hirner hypothesis
The Hirner hypothesis is that as bitcoin is used more often, that the price of bitcoin will do *down* rather than up. So what I’m doing right now is to do some more investigation to see if this makes sense or not. The basic problem is a philosophical one. Bitcoin prices have been going down. Bitcoin transactions have been going up. You can plot the two, but it would be like plotting the price of Bitcoin versus the distance of Voyager two from the sun. Perfect correlation that means nothing.
To argue that there is causation, you can do two things. First you can look for “bumps and wiggles.” If you have a bump and then you see the same bump in another data series, that might mean something (or not) see the cool web site Spurtious Correlations http://www.tylervigen.com/.
The second thing you can look for is magic numbers. If you divide the distance of Voyager 2 from the sun by the price of bitcoin, you get a totally random number that doesn’t mean anything. However, if you divide the number of bitcoin transactions with the price of bitcoin and you get “pi” or “2”, there might be something there.
The other thing I’m trying to figure out is how “reflexive” bitcoin prices are. Stocks are extremely reflexive. If you make everyone think that a stock is a good buy, everyone will buy it and this creates a feedback cycle that turns perception into reality. Politics and venture capital tends to be extremely reflexive, and VC’s live in a world where perception creates reality, and that impacts how people perceive bitcoin.
My strong suspicion is that bitcoin prices are not very reflexive. What that means is that I can convince everyone that bitcoin prices will go up, but that doesn’t cause them to go up. You can test for the amount of reflexivity finding some sort of “hype” index (google page views will do nicely) and see if there is a correlation with prices. One piece of evidence that prices aren’t reflexive is that you have periods of very high volume in 2015 which push up the price of bitcoin for a day, and then everything goes back to the long run trend.
The other thing about this analysis is that it tells you when and if the rules change. I have a strong suspicion that if you run this against 2014 data, you’ll get very different results. The Arxiv paper did this with daily data, but I can try rerunning this with minute data. Also running these sorts of things will tell you if the rules change.
bitquant 