Stats-Tec inc. Blockchain

It is well known in crypto currency markets that when Bitcoin moves up or down the rest of the market follows, mostly. Principal component analysis (PCA) can help determine how much of the market movement is correlated. Some groups of coins such as Ethereum, NEO, and at times DASH and Monero have moved differently from Bitcoin. These different groups can be identified (to some degree) in the eigen vectors of the PCA. For each eigen vector, the amplitude coefficients show its dominance in the market at any given time. This analysis uses the covariance matrix of the normalized dataset (as plotted above). Please wait for the data to load, execute the analysis and view the results below.

Results

Since crypo markets change dramatically, and only a short 3 month data window is followed here, comments are periodically moved from here the technical blog at statstec.ca. In PCA analysis, a centred version of the normalized data set can be rebuilt by multiplying the eigenvectors (EG listed in Appendix) by their amplitude scores (the curves 'a' above). Note that a negative EG1 coefficient multiplied by a negative amplitude score indicates price increase. In the plot above, the zero value of the amplitude curves represents a mean (sort of). From the results, typically over 90% of the information can be reconstructed with 4 eigenvectors. Also, EG1 represents typically about 80% of variations in the market. The coefficients of EG1 are all negative for every coin, except for Tether. It represents correlated market movements and perhaps indirectly money in the market.EG2 and E3 typically represents around 10% and 5% of variations in market, respectively. Here coins like NEM, and Binance, or Ethereum, NEO, and Stellar, are negatively correlated to Bitcoin and other coins. Amplitude curves 'a2' and 'a3' represent a certain proportion of money that moves toward or away from these coins to the others. (similar analysis applies to EG4).

Follow us on Twitter for uptodate commentary on the market using this analysis! @TecStats

Cluster analysis seeks to identify groups in the dataset: data points that are close to each other and far from other groups. The top 30 crypto-currencies do not form well defined different groups. Rather they form one big group with fuzzy edges and outliers. To try to form meaningful groups we have done 2 things: (1) the 2-day average price change and average volume are used as the clustering variable; and,(2) PCA analysis is preformed on this data set to reduce the number or variables for clustering. Here, a hybrid k-means clustering with seeds found from isolating and preclustering outliers (just a fun hack because because there is a lack of good Javascript clustering libraries) is used to make 4 clusters based on the first 6 eigenvectors. For interesting results, the 2-day average price change is presented with the 2-day average volume In the 4 cluster results below, positive value are daily positive price change (ie slope of price curve). Please try a few iterations to find interesting clusters. Pumping coins appear in small outlier cluster of 2 to 4 coins and show an up-tick at the end of the time series. The cluster assignments can be found below in the appendix. Go back to the first chart to plot individual cluster members.