This paper proposes an empirical framework to assess the ad-hoc tail risk connectedness across financial markets. I discuss the role of systemic risk as a predictor of financial tail risks in the quan- tile regression framework introduced in Adrian et al. (2019). I independently estimate the expected shortfall, also called conditional value-at-risk (CVaR), for four asset classes. Surprisingly, systemic stress has little informational power over future tail outcomes in those asset classes, even if the overall picture is not homogeneous across markets. The CVaR measures constructed as the tails of predictive conditional distributions are then taken to a generalized vector autoregression (G-VAR) framework in which forecast-error variance decompositions are invariant to the variable ordering as in Diebold & Yilmaz (2012) to build a directional left-tail risk spillovers measure. I find strong evidence of high dependence between the tails of the distributions across financial markets with risks of contagion during episodes of high volatility.