State-Dependent Contagion Risk : Using Micro Data from Swedish Banks

In this paper we exploit a unique data set on banks’ interbank exposures to estimate the risk of contagion between the four largest banks in Sweden representing roughly 80% of total bank assets. For each quarter from 1999, the banks have reported their largest counterparty exposures broken down on (i) unsecured loans (ii) securities (iii) net and gross derivative positions and (iv) repo positions. Hence, contrary to earlier work, we can construct the actual matrix of bilateral, unsecured exposures between banks to study potential contagion effects. We first use market data to model the interdependence of banks’ asset values, using the Gaussian copula. Following Merton’s (1973) approach, a default is defined as when asset values shrink below the value of outstanding debt. Default events are then simulated through Monte Carlo simulation. Given an event of at least one default, we use the data on interbank exposures to estimate the likelihood of successive defaults. We show that, in striking contrast to when the initial default occurs for idiosyncratic reasons, state-dependent contagion is very likely. Our exposure data reveals that exposures between banks are very asymmetric. This means that attempts to re-construct bilateral exposures from totals common in the literature could underestimate the risk of default contagion. Corresponding author; lars.frisell@riksbank.se.