Stress testing incorporating causal dependency [24]

Go to: Summary | Previous | Next   
Bullet points include: Given the joint distributions we might be able to identify the simplest (most plausible?) closely approximating Bayesian nets E.g. by seeking to minimise the relative entropy for all nets of a given maximum level of complexity (however defined), see Kemp (2010) section 3.8.5 Perhaps then developing stress tests by stressing the Bayesian net in some way Rebonato focuses on the reverse, using the Bayesian net to estimate the joint distribution. Needs us to: Identify relevant variables and specify causal relationships Provide relevant marginal and conditional probability tables Latter potentially derived via orders of magnitude and the bik described earlier

Contents | Prev | Next | ERM Lecture Series

Desktop view | Switch to Mobile