This paper applies novel tools from spatial econometrics to measure, quantify and predict sovereign CDS spreads. Network risk is modelled by making each sovereign's CDS spread a function of the CDS spreads of its “neighbors” in the financial network. The main findings of the paper are: (1) the network model improves forecasting accuracy by 15 % to 20%; (2) exogenous financial shocks propagate in the network of sovereigns and 40 % to 50% of the total effect is due to indirect (network) effects. These findings suggest an alternative explanation to the well-known credit spread puzzle. To rationalize the findings the paper develops a simple structural network model of sovereign credit risk with financial cross-holdings and multiple equilibria