This paper assesses the economic value of modeling conditional correlations for mean–variance portfolio optimization. Using sector returns in three major markets we show that the predictability of models describing empirical regularities in correlations such as time-variation, asymmetry and structural breaks leads to significant performance gains over the static covariance strategy. Investors would be willing to pay a fee of up to 983 basis points to switch from the static to the dynamic correlation portfolio and about 100 basis points more for capturing asymmetries and shifts in correlations. The gains are robust to the crisis, transaction costs and are most pronounced for monthly rebalancing.

We develop a stress-testing network model calibrated to the largest banks and investment funds in over-the-counter (OTC) derivatives markets. We examine the impact of the mandatory collateralisation of bilateral OTC derivatives on liquidity, counterparty, and systemic risks, as well as the impact of market frictions on participants’ ability to withstand liquidity shocks. The collateralisation of bilateral trades reduces counterparty and systemic risks but increases the prominence of liquidity-driven defaults and the potential for the central counterparty to transmit losses. Frictions such as fire sales, delayed payments, and no partial payments by defaulted counterparties greatly increase liquidity risk and systemic losses.