They seem to be computing the correlation of price levels (which are completely meaningless) rather than the correlation of price difference or returns.
Any two time series which broadly moved up over the same time period will look correlated if you compute the correlation of levels. But they could in fact be completely uncorrelated (and therefore have zero predictive power) when you compute the correlation on returns, which is what you should be doing.
Neat visualization, but as the website itself notes at the bottom (in fine print): "Correlation doesn't necessarily mean causation. Be careful when using correlations to draw causal inferences for trading or research."
Exactly. gotta read the fine print :) We think this could get more interesting with time though. It was really cool when we got our first couple days of data and were able to see the basic principles of economics at work by looking at the anti-correlation of supply with price!
When you lose money for two weeks straight, it's nice to be able to figure out if it's a statistical blip, or if there's a fundamental reason that the trade stopped working. If all you have is a statistical correlation, you can't do that, and you have to guess. If you have a causal relationship in mind, you can examine the market to see if there are reasons to believe that the causal relationship might have broken down.
There is a correlation to bitcoin repo on github. Is that the cumulative number of repo on github with "bitcoin" tag? or is it the activity on github.com/bitcoin?
Personally, I would think that correlation to the number of bitcoin repo on github does not make sense because that is the cumulative number. However, if you can correlate to average recent activities on all bitcoin repo on github, that could be very interesting. :)
Very interesting. Let's not forget that butter production in Bangladesh and U.S. cheese production, and sheep population in Bangladesh correlates hightly (99%) with the movements of the S&P 500.
This is great, but will only be of use when much more liquidity is bought to bitcoin. Right now, the Mt Gox and Newsweek debacles are the only things artificially pushing btc prices below their upwards paths. Why - is something that remains to be determined. Perhaps someone BIG needs to buy into bitcoin before the next rally ?
Not sure I understand the point... ? The reason for correlating against price is in hope of finding a leading indicator to movement so that you can better understand the relationship between bitcoin price and some other metric. This would in theory help you ultimatly make better returns by better predicting when to buy or sell...
Any two time series which move broadly in the same direction will be 'correlated' if you compute the correlation on price levels, whether they have any predictive power or not.
Correlation on price levels is meaningless, only correlation on price differences or returns is meaningful.
Sometimes dinosaurs are awesome, though. See the video of the older commentator at Bloomberg interviewing the daffy twit at Newsweek who "outed" Dorian Nakamoto:
"Ponzi scheme" has a very specific definition (early investors paid by late investors rather than from profit). While Bitcoin may certainly be in a bubble, a sharp rise in price does not qualify it as a ponzi any more than tulip bulbs or AAPL shares.
I think what he means is that early adopters are disproportionally rewarded if additional people use the currency, since mining gets exponentially harder.
The intrinsic value of USD is near zero. Useful in a post-apocalyptic as fire kindling, and not much else. Its perceived value is powered through the USA's massive military, political, economic, and financial power. Perhaps Bitcoin has a perceived value as a mechanism of subversion of such a hegemony.
Or maybe it's just a vehicle of speculation. Either way, no modern currency has intrinsic value.
Any two time series which broadly moved up over the same time period will look correlated if you compute the correlation of levels. But they could in fact be completely uncorrelated (and therefore have zero predictive power) when you compute the correlation on returns, which is what you should be doing.