#CorrelateThis - The Neural Activity of Mice Indicates Cryptocurrency Price Fluctuations by@dankhomenko

#CorrelateThis - The Neural Activity of Mice Indicates Cryptocurrency Price Fluctuations

Neurobiologist Gvido Mayer found that 70% of neurons correlated with Bitcoin and Ether prices. The results were caused by “pointless correlations” - with the correlation between two signals that develop slowly over time, the chances of finding a significant correlation between them are much higher than those that do not. Only 4.9% were correlated to a random number vector, close to a false positive effect, which is within the expected level of error. The neurons in the brains of mice, of course, couldn’t encode it.
image
Dan Khomenko Hacker Noon profile picture

Dan Khomenko

CEO of NFTSTARS.APP and SpaceSwap.app An entrepreneur and founder of a successful IT engineering company.

The other day, I came across a weird but terribly interesting piece of research. It was conducted by Gvido Mayer, a neurobiologist who works at the International Brain Laboratory in the Shampalim Center in Lisbon, Portugal.

In his research, Gvido reported the discovery of neurons that showed a neural correlation to the price fluctuations of the main cryptocurrencies, at the time the research was conducted. The neurobiologist used the public data set of the Allen Institute’s neuropixel records to correlate the frequency of pulsation (impulses) of single neurons with the price of Bitcoin and Ether.

The dataset contained a burst of activity of 40.010 neurons recorded in 58 mice that were stationary and passively observing visual stimuli. 

The results of the research

Gvido found that out of the 40.010 single neurons recorded, 70% correlated with Bitcoin and Ether prices. 

35% of the neurons showed a correlation, which is well above the expected false positive factor of 5%. 

The results were caused by “pointless correlations” - with the correlation between two signals that develop slowly over time, the chances of finding a significant correlation between them are much higher than those that do not. 

Let's look at the graphs: 

image

The neural correlation of the price of cryptocurrency is illustrated above. 

(А) An example of the neurons from four different brain areas showed a strong correlation to the Bitcoin prices (the upper graph) and Ether (the lower graph) at the time of examination. The pulse rate and the price of the cryptocurrencies were fixed in 60-second intervals. 

image

(В) Distribution of Pearson correlation coefficients, characterizing a linear relationship between the two quantities, was centered on zero but revealed a large proportion of neurons positively and negatively correlated with the cryptocurrency prices. The random number vector was also used for comparison.

image

(С) About 70% of neurons showed a significant correlation to the price of Bitcoin and Ether, while only 4.9% of neurons were correlated to a vector of random numbers, close to a false positive effect.  

Thus, many neurons have shown a strong correlation between their pulsation rate and cryptocurrency prices (see graph A). At the same time, a random vector denoted a weak correlation to the pulsation rate (graph B). 

For an unusually high percentage of neurons, the correlation with the price of cryptocurrencies is significant: Bitcoin - 70.5%, Ether - 68,8%! At the same time, the correlation with a random vector was only present in 4.9% of neurons, which is within the expected level of error.

And yet... Why such a large proportion of neurons have a significant correlation to the price of cryptocurrency?  

The neurons in the brains of mice, of course, couldn’t encode it. And I think that mice (almost certainly) can’t read and interpret complex financial data😉.

The explanation is that both the frequency of neuron impulses and the price of cryptocurrencies slowly evolve with time. The temporary constant of these auto-correlations was simply similar. Since the two correlated signals have this statistical property, the chances of finding a strong correlation between them are much higher than with the usual level of false-positive results. 

In addition, a large number of neurons in the data set allowed the observation of strong dependencies only by chance; due to temporary auto-correlations in both signals, conventional methods, such as multiple comparisons, could not reduce the false positive factor to acceptable levels.

Let us not forget that the “pointless correlations” are a kind of trap, which is very difficult to avoid when studying such signals as neural activity.  

Most of the study warns that the potential mixing of meaningless correlations should be taken seriously. Otherwise, one might erroneously conclude that 70% of neurons in the mouse code correlate to cryptocurrency prices.

Dan Khomenko Hacker Noon profile picture
by Dan Khomenko @dankhomenko. CEO of NFTSTARS.APP and SpaceSwap.app An entrepreneur and founder of a successful IT engineering company.Read my stories

Tags

Join Hacker Noon

Create your free account to unlock your custom reading experience.