Working Paper: NBER ID: w26814
Authors: Bomin Jiang; Roberto Rigobon; Munther A. Dahleh
Abstract: In this paper, we develop a methodology to estimate hidden linear networks when only an aggregate outcome is observed. The aggregate observable variable is a linear mixture of the different networks and it is assumed that each network corresponds to the transmission mechanism of different shocks. We implement the methodology to estimate financial networks among US financial institutions. Credit Default Swap rates are the observable variable and we show that more than one network is needed to understand the dynamic behavior exhibited in the data.
Keywords: financial network; identification through heteroskedasticity; mixture model; EM algorithm
JEL Codes: E0; E44; G1; G21
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Estimation of financial networks (G19) | Understanding systemic risk (F65) |
| Interconnectedness of banks (F65) | Pathways for shock propagation (E44) |
| Heteroskedasticity in data (C21) | Identification of financial networks (G29) |
| Larger shocks (E32) | Nonlinear responses in the network (C45) |
| Correlations in CDS spreads (C10) | Underlying risk transmission mechanisms (F65) |
| Propagation of shocks (E32) | Effect on different banks (G21) |