Learning and Cross-Country Correlations in a Multi-Country DSGE Model

International spillovers in estimated multi-country DSGE models with trade are usually limited. The correlation of nominal and real variables across countries is small unless correlation of exogenous shocks is imposed. In this paper, I show that introducing adaptive learning (AL) with time-varying coefficients as in Slobodyan and Wouters (2012b and 2012a) increases the international correlation. I use an estimated large-scale model as in de Walque et al. (2017), which has reasonable forecasting performance under rational expectations (RE). The model features the euro area, the US, and an exogenous rest of the world, with endogenous exchange rate determination. I show that the increase in international correlation stems from the varying coefficients and the use of simple forecasting models. The increase in the correlation of international variables goes through two channels: larger shock spillovers through the exchange rate, and correlated adjustment of agents' forecasting model coefficients.