The second negative term is a Chinese exchange, of the centralized type. That is, as demand for the exchange of CNY/digital asset increases, this may be exercising some downward pressure on price. The negative effect is slightly larger when people use the centralized exchange, rather than a Ripple gateway. This may suggest that some operatives turned to Ripple as a haven when the Bitcoin exchanges where hit, although prices are still susceptible to movements of XRP assets in and out of the economy via a gateway. But strong demand from the middle market supersedes the fears of those uncomfortable with all-time-highs, and this is why the usage of the Ripple wallet exploded on Dec 13–14th, in tandem with the spike in price volatility of XRP — alongside with attractive dynamics of the BTC/XRP pair. It is also important to note what the AI does not see: the lack of any oscillatory term in the formula (sin, cos) hints at the lack of strong regularities (periodicities) during the eight months of the analysis. The other valuable observation here is that in the absence of access to many of the Ripple ledger statistics that will normally allow identifying large holders, the usage of the wallet allows to single out the mid-market as a force driving the market (and this wallet service is predominantly accessed from the US).
 

Time series analysis and prediction

The application of genetic programming to the study of behavior and causation in cryptocurrency markets is not only an analytic artifact, but it is fundamentally aligned to the nature of the problem.   
Taleb and Douady \cite{Taleb_2013} explain that the natural selection of an evolutionary process is particularly antifragile since a more volatile environment increases the survival rate of robust species and eliminates those whose superiority over other species is highly dependent on environmental parameters. In the context of cryptocurrencies, we could use temporal correlations in blockchain traffic to gauge the response of a given object (e.g. fees) to the volatility of an external stressor that affects it, but another approach is to simply study the response of the market (as measured by a common risk metric, such as volatility) to the actual market behavior (with the consumption of services and information measured by HTTPs requests, including endogenous variables such as the activity at the customer service channels of the wallet and the exchange itself, mining pools, mining profitability feeds, and so on; in this toy example we use just a subset of those variables). The driving variables are modeled using symbolic model ensembles, as in Figure 1, which is based in daily time series for the period of November 1st, 2016 to May 9th, 2018.