Do liquidity variables improve out-of-sample prediction of sovereign spreads during crisis periods?

•Comprehensive study of out-of-sample yield spread predictability.•Mean difference test for analyzing the cross-sectional differences among our forecasting models.•Our proposed liquidity models are useful for forecasting yield spreads in crisis periods.

This paper addresses the out-of-sample prediction of European Monetary Union yield spread changes. We extend the Longstaff and Schwartz (1995) approach by using liquidity variables, namely funding liquidity as measured by European Central Bank's unconventional monetary policy as well as a commonly used market liquidity proxy. Our out-of-sample results highlight that the economic forecasting models outperform the autoregressive moving average benchmark during times of crisis, when liquidity-based models yield superior predictions. However, the economic models do not yield forecasting gains during the pre-crisis period. Hence, our results provide evidence for the usefulness of economic models in predicting sovereign spreads during crisis periods.