Sovereign Credit Default Swaps and Macroeconomic Fundamentals

We provide evidence in favour of a significant non-linear, time-varying dependence between sovereign credit default swap (CDS) spreads and macroeconomic fundamentals for OECD countries. Macroeconomic conditions alone explain more than 80% of the out-of-sample variation in CDS spreads when non-linear random forest regressions are used. This is almost double the predictive performance of sparse and dense linear statistical learning methods. We test the coherence of the expected CDS spreads across different pure out-of-sample scenarios, e.g. training a random forest regression on EU countries and predicting CDS spreads for non-EU economies. The results suggest that non-linear machine learning methods enable “shadow” sovereign CDS pricing for countries and historical periods in which tradable sovereign CDS contracts are not available