Inferring the dependence structure of complex networks from the observation of the non-linear dynamics of its components is among the common, yet far from resolved challenges faced when studying real-world complex systems. While a range of methods using the ordinal patterns framework has been proposed to particularly tackle the problem of dependence inference in the presence of non-linearity, they come with important restrictions in the scope of their application. Hereby, we introduce the sign patterns as an extension of the ordinal patterns, arising from a more flexible symbolization which is able to encode longer sequences with lower number of symbols. After transforming time series into sequences of sign patterns, we derive improved estimates for statistical quantities by considering necessary constraints on the probabilities of occurrence of combinations of symbols in a symbolic process with prohibited transitions. We utilize these to design an asymptotic chi-squared test to evaluate dependence between two time series and then apply it to the construction of climate networks, illustrating that the developed method can capture both linear and non-linear dependences, while avoiding bias present in the naive application of the often used Pearson correlation coefficient or mutual information.

You do not currently have access to this content.