Abstract We propose an alternative measure of system-wide connectedness to the popular generalized spillover index, based on generalized forecast error variance decompositions, of Diebold and Yilmaz (2012, 2014). Our measure relies on joint conditional forecasts to decompose variance, as opposed to the popular method's reliance on single-variable conditioning sets, and is a more precise measure of aggregate spillovers. We show in an application to US industry sector stock returns that the difference between the two measures can be substantial.