Exploiting locally stationary lead-lag relationships between variates improves multivariate time series forecasting accuracy.
