**Status:** Maintenance. cryvo will soon work again as usual.

Is your coin before a transition?

alpha level r^{2}

.05

.1

.2

Complex dynamical systems – like cryptocurrency markets – have tipping points, where critical phase transitions can occur. To identify these transitions before they take place, can be of great value.

Conditional heteroskedasticity (r^{2})

An autoregressive model with p-order selection via best non-seasonal ARIMA model search based on AICc optimization is fitted to the time-series data. The metric phase transition indicator here is conditional heteroskedasticity (r^{2}), a measure of increasing variability conditional on variability at previous time steps.

Autocorrelation and variance

The data is first-differenced prior to analysis. The two trends are estimated by nonparametric Kendall tau correlations (τ).

Because of its nonparametric nature, the nonparametric drift-diffusion-jump (DDJ) model models a wide variety of unknown underlying time-series data generating processes and estimates drift, diffusion and jump metrics, that might indicate incoming phase transitions.

Conditional heteroskedasticity (r^{2})

Everything above the orange line in the bottom plot indicates potential phase transitions.

You can see how the phase transition indicator performed in the past for your coin. If it could reliably detect phase transitions in its value, you could look on the right side of the bottom plot, if there are any recent indicators for an incoming phase transition.

Autocorrelation and variance

The blue lines plot the following metrics:

**Autocorrelation:**An increase in autocorrelation indicates increasing similarity between consecutive observations, which might indicate an incoming phase transition.**Variance:**Can increase prior to a phase transition.

The blue lines plot the following metrics:

**Diffusion:**Small shocks at each point in time**Jumps:**Large intermittent shocks**Total variance:**Combines diffusion and jumps**Conditional variance:**Ascends to infinity at a critical tipping point caused by a phase transition

BTC

1w

15m

672

**Status:** Maintenance. cryvo will soon work again as usual.

How connected are your coins?

gamma hyperparameter

.01

.25

.5

Given that cryptocurrencies are highly intercorrelated, conditional independence association pattern – as visually represented by partial correlation networks – might be particularly insightful.

A regularized estimation of a partial correlation network is computed using the extended Bayesian information criterium (EBIC) based on first-differenced time-series data.

Blue lines indicate coins that are positively connected after controlling for all relationships between the selected coins.

Orange lines indicate coins that are negatively connected after controlling for all relationships between the selected coins.

The placement of nodes is not important. What matters, are the connections (edges) between the nodes.

Selected coins:

BTC ETH XMR XRP ZEC

1w

15m

3360

**Status:** Maintenance. cryvo will soon work again as usual.

How forecastable are your coins?

Pro mode Learn more Interpret

alpha level

.05

.1

.2

Forecastable component analysis is a dimension reduction method for multivariate time-series data, that takes temporal dependence explicitly into account and actively searches for the most forecastable subspace.

The forecastability measure (Ω) is based on the entropy of the spectral density of the time-series data. Higher entropy implies less forecastability, lower entropy implies more forecastability.

The forecastability measure (Ω) can be used to select coins for forecasting via neural networks.

The blue coins indicate significant forecastability according to the forecastability measure (Ω).

Selected coins:

BTC ETH XMR XRP ZEC

1w

15m

3360

**Status:** Maintenance. cryvo will soon work again as usual.

What is the future of your coin?

Networks to train

3

5

10

Forecast combination

mean

median

mode

Forecast interval

10

20

50

The forecastability measure (Ω) is based on the entropy of the spectral density of the time-series data. Higher entropy implies less forecastability, lower entropy implies more forecastability.

Multiple multilayer perceptron (MLP) neural networks with selection of hidden nodes via cross-validation are trained and their forecasts are combined.

More advanced neural network models might be available in the future.

The percent value quantifies the forecastability with a value range from 0% = 'not forecastable' to 100% = 'perfectly forecastable'. The visual percentage representation bar is always gray, i.e. not color-coded.

Interpret the results with caution. Only if patterns in the time-series data of the past allow for inferences about the future and the model accurately detects these patterns without overfitting, the results might be of some predictive value.

BTC

1w

15m

672

**Please note:** Computation time can vary significantly between coins.