Can We Explain the Sign-Switching Behavior of Cross-Country Interest Rate Correlations?

Abstract

This paper considers the well established empirical fact that conditional correlations among cross-country interest rates switch signs. Switching implies an alternation of coupling and decoupling of global bond markets over time. This evidence is robust to alternative estimation schemes. Here we use a seminonparametric (SNP) model with a BEKK-GARCH variance function to estimate conditional second moments both to confirm these results and to provide auxiliary models for structural estimation of term structure models. Using an extensive historical analysis, we find that major driving forces behind the sign-switching behavior of conditional correlations between the Eurodollar rate and the Euroyen rate are synchronization and dis-synchronization of business cycles and coordination and discoordination of monetary policies triggered by international policies and financial market crashes. Especially, we find that the two interest rates are more likely to couple when both the U.S. and Japan slip into a recession while the likelihood of decoupling is highest when both economies are in expansion. We also explore whether proposed International Affine Term Structure Models (IATSMs) and International Quadratic Term Structure Models (IQTSMs) are able to reproduce the sign-switching behavior of conditional correlations among cross-country interest rates. We find that a small subset of the IATSMs can generate sign-switching behavior but only by forgoing their ability to describe other features such as the positivity of nominal interest rates, heteroskedasticity in volatility, and correlations among underlying state variables. In contrast, the IQTSMs are able to generate it without limiting their ability to describe other dynamic features. Using the MCMC-Efficient Method of Moments (EMM), we test the empirical performance of the models in reproducing the sign-switching behavior of conditional correlations. The result suggests that the IATSMs conclusively fail to capture it while the IQTSMs are relatively successful but fail to reproduce ephemeral ones.