A portifolio that contains many European, American, and more complex options demands extensive simulaitons to determine its price and risk measure (e.g., VaR, CVaR). We use regression to appoximate the response surface of those option pricing simulators, and thus effecient risk estimation is made possible.
Using this approach we can achieve 20x speedup over traditional PDE- or Monte Carlo-based method.