Rotational Invariant Importance Sampling
Time and place:
12:30 PM on Tuesday, May 12th, 2009; NAC 4113
Description:
Speaker: Ranja Reda (Vienna University of Technology)
A new importance sampling technique to increase the efficiency of rare event simulation. By using a multidimensional rotational invariant auxiliary density, this method can be applied to estimate risk measures for credit risk portfolios where the method of importance sampling via mean shifting is not applicable. This is especially the case in inhomogeneous portfolios, i.e. portfolios with multiple areas of high losses. Furthermore, rotational invariant importance sampling allows calculating contributions to risk measures of different customer clusters in credit porfolios quicker and more precisely.
Our new model, rotational invariant importance sampling has already been implemented in Bank Austria's system. Their main credit portfolio with about 2000 clients is already measured through our new simulation technique. As you can imagine there was lot of work to be done as far as the calibration of the model, finding the optimal parameters for the new importance sampling density, etc.
