Here the wallet signal leads the price signal, on day 120 in a cycle of approximately 4-6 days, where both signals are also strongly correlated. In the case of unruly distributions, the results are enhanced in combination with other methods (e.g. Granger causality \cite{Auinger_2015}, Bayesian structural time series models \cite{Brodersen_2015}, among others). The key realization is that although market behavior (network dynamics) is not the same as market conditions (market structure), the intelligent agent can use the same graphical metaphor to gain context, and predict the counterfactual response. But the risk is that correlation at training time does not ensure correlation at test time, the AI should therefore be aware of divergences.
Use of low complexity property testing methods by decentralized blockchain agents
Neighborhood Asymmetry
Once the AI is context aware (of the trust space in terms of entities, and of the relevant variables given various causality tests) information metrics derived from the data space itself are utilized to measure the actual information content of each sample. The neighborhood asymmetry method \cite{Vladislavleva2015} sums the vectors from the data record to the neighbors implicitly defined by the supplied data matrix and returns the length of this resulting neighborhood directionality normalized by the number of neighbors. In this way, this metric is primarily concerned with the symmetry of the neighbor distribution but also contains a contribution from the distance to each of the neighbors. Figure 6 shows the symmetry for the BTC off-chain economy modeled in terms of on-chain economy variables (fees) in the period of August 2nd, 2017 to January 24th, 2018; it tracks over-the-counter exchanges (OTC), wallet services, paper wallet generators, block explorers, among many others.