Title: Estimation of risk measures
Authors: Thorsten Schmidt - University Freiburg (Germany) [presenting]
Abstract: While risk measures are a topic of paramount importance, the estimation of risk measures has long been neglected in the literature. Most of the practically used estimation procedures introduce a bias in the sense that the risk is underestimated. This is confirmed in backtesting procedures, where the performance shows potential of improvement. We present unbiased estimators which do not suffer from this deficiency. Moreover, we present a new estimation procedure based on deep neural networks which allows us to obtain unbiased estimators in a numerically efficient way when explicit formulae are not available.