Outlier Detection In Bayesian CAPM Model

We extend a recent methodology proposed by Quintana and Iglesias (2003) based on product partition models. We measure the performance of an asset in a CAPM framework. We assume that the excess return is normally distributed and we impose a partition structure on the specific risks related to different time points. This allows us to remain in a normal setting consistent with a Mean- Variance analysis, even in presence of outlying observations. We apply an optimization algorithm to identify the partition that best resemble, in terms of a quadratic score function, the Bayesian estimates of the parameters of interest. In this way we separate standard observations from the atypical ones. The methodology is illustrated with reference to the IPSA data.