PARTICLE FILTERING FOR RISK MANAGEMENT AND REPLICATION OF A RISK APPETITE INDICATOR

The authors propose a new approach to estimating stochastic volatility parameters. Traditional methods maximize the conditional likelihood. The proposed model optimizes two criteria: the deviation of the observed residuals PDF from the theoretical PDF and the in-sample predictive power of the volatility estimate. The resulting model yields better results than GARCH and than Harvey et al.’s stochastic volatility model. Two more applications of this innovation are also examined. First, volatility fit residuals for two assets are combined to estimate dynamic correlation. The model aptly estimates dynamic correlation when it is significant – though with some lag. Second, these models are used to replicate the RBC Risk Appetite Indicator. Results show that even though the authors are missing 40% of the inputs to the risk indicator, the replication strategy adequately replicates the indicator. We expect these three models to be of significant use to the SIAS team.