Stochastic kriging for conditional value-at-risk and its sensitivities

Measuring risks in asset portfolios has been one of the central topics in the financial industry. Since the introduction of coherent risk measures, studies on risk measurement have flourished and measures beyond value-at-risk, such as expected shortfall, have been adopted by academics and practitioners. However, the complexity of financial products makes it very difficult and time consuming to perform the numerical tasks necessary to compute these risk measures. In this paper, we introduce a stochastic kriging metamodel-based method for efficient estimation of risks and their sensitivities. In particular, this method uses gradient estimators of assets in a portfolio and gives the best linear unbiased predictor of the risk sensitivities with minimum mean squared error. Numerical comparisons of the proposed method with two other stochastic kriging based approaches demonstrate the promising role that the proposed method can play in the estimation of financial risk.