Advances in Complex Systems

We modeled the stocks of the financial system as a set of many interacting like spins derived from binary daily returns. From the empirical observation of these returns, we used a Boltzmann machine to infer a distribution of states equivalent to a maximum entropy distribution. This model describes the interaction couplings between each stock pair in the system, which can be considered a complete network with $N$$(N-$1)/2 couplings. We then engage in a coupling removal process to find a critical graph that can describe the observed states of the system with the minimum number of edges. We interpret the critical graph as the backbone of the system, and it allows us to evaluate the importance of markets in their relation to others in the system. We also found that the structure of this critical graph is highly variable over time and appears to be dependent on the level of entropy of the system.

It is investigated how a possible racial bias in police officer’s searching behavior might be detected from the temporal variations in search rate and hit rate data. It is shown that this appears possible, conditions on data quality are derived, and a few tests on several aspects of racial bias are proposed.

In this study, we address an issue of stability of global production network by constructing a scheme combining classical Leontieff input–output analysis and modeling of contagion in financial networks. We propose a model of disruptive cascades in global economy with nonlinear effects similar to those in the financial contagion model proposed by Elliott, Golub and Jackson (2014). We apply the model to the analysis of stability of the global input–output sectoral network using the data from the World Input-Output Database (WIOD). We show that this sectoral network appears to be very stable with respect to sectoral shocks. However, by introducing synthetic substructure of economic sectors in the WIOD data we show that this stability is to a significant extent an artifact of working with aggregated data. Namely, we show that the impact of economic shocks may be much more pronounced if an underlying disaggregated network is sufficiently sparse.