Nonexistence of Markovian time dynamics for graphical models of correlated default (Q415636)

The paper under review studies if it is possible to begin at time 0 with a configuration in which every spin is \(-1\) and then flip spins one at a time from \(-1\) to \(+1\) according to Markovian dynamics so that the configuration of spins at each time is described by an Ising model and at time \(T\) the configuration is distributed according to the prescribed Ising model. The authors show that there is a nonexistence result of Markovian dynamics for three graphical models of correlated defaults. Investors are exposed to credit risk due to the possibility that the borrower will not pay back all or part of the interest or principal as promised. The models of default are the structural and reduced-form ones. The difficulty of modeling default probabilities is compounded for complex financial instruments, in essence, bundling together a group of borrowers. It is then no longer sufficient to determine the default probability for a single firm, it becomes necessary to model the joint probabilities that various subsets of firms in the basket will default. The model studied in this paper was originally presented in a preprint by \textit{I. O. Fliliz} et al. [``Graphical models for correlated defaults'' (2008; \url{arXiv:0809.1393})]. A finite graph \(G\) with vertex set \(V\) and edge set \(E\), each vertex \(v \in V\) represents a firm, the edge is the interdependence between two firms. The default random variable \(I_v\) takes value 1 if the firm \(v\) defaults and value 0 otherwise for each vertex \(v\). The probability of a given pattern \(\epsilon = (\epsilon_v)_{v\in V}\) of defaults is \[ P \{ I_v = \epsilon_v, v\in V\} = \frac{1}{Z} \exp (H(\epsilon)), \] where \(H(\epsilon) = \sum_{u\in V}\alpha_u \epsilon_u + \sum_{[v,w]\in E}\beta_{[v,w]}\epsilon_v \epsilon_w\) is called a Hamiltonian, \(Z\) is a normalizer for the probability, \(\alpha_u\) is the individual propensity of the firm \(u\) to default, the parameter \(\beta_{[v,w]}\) for an edge \([v,w]\in E\) is positive if the joint default of firms \(v\) and \(w\) is favored, is negative if the joint default is discouraged. With substitution spins \(Y_v = 2I_v -1\) for each \(v\in V\), the model \(P\{Y_v=\pm 1, v\in V\}\) is the usual Ising model associated to the graph \((G=(V, E); \alpha, \beta)\). The model does not have dynamical description of default and correlated default. The authors study the existence of time-homogeneous Markov chain \(X_t\) on the graphical model. For a subset \(A\subset V\), let \(B=A\cup \{v\}\), \(v\) is not in \(A\). The Komogrov forward equations for the chain X with empty set initial can be expressed in (2.3) of the paper. Hence both parameters \(\alpha_u, \beta_{[v,w]}\) are determined by explicit ODE's in Section 2. For a complete graph \(G=K_N\) on \(N\geq 4\) vertices, \(\alpha_u\) and \(\beta_{[v,w]}\) are independent of vertices, the graphical model admits Markovian dynamics if and only if the firms default independently by Proposition 3.1. For a complete bipartite graph \(G=K_{M, N}\) with disjoint vertices \(V_0\) (\(|V_0|=M\geq 3\)) and \(V_1\) (\(|V_1|=N\geq 3\)), the graphical model admits Markovian dynamics if and only if the firms default independently by Proposition 4.1. Section 2 gives the model in general, and section 3 is devoted to the proof of Proposition 3.1 and section 4 to the proof of Proposition 4.1. Section 5 shows that the same result for another type of complete bipartite graphs. The model does not admit time-homogeneous Markovian dynamics. All three models in this paper cannot admit a time-homogeneous Markovian dynamics. The paper does not address if there is a time-inhomogeneous Markovian dynamics since the correlated default has the time-dependent joint probability for two firms default. The complete graph and complete bipartite graph are very special graphs. It would be interesting to see the explicit time-homogeneous Markovian dynamics on other graphs, say perfect graphs, symmetric graphs.