Paper by Nikola Tarashev
Why should risk management systems account for parameter uncertainty? In order to answer this question, this paper lets an investor in a credit portfolio face non-diversifiable estimation-driven uncertainty about two parameters: probability of default and asset-return correlation. Bayesian inference reveals that – for realistic assumptions about the portfolio’s credit quality and the data underlying parameter estimates – this uncertainty substantially increases the tail risk perceived by the investor. Since incorporating parameter uncertainty in a measure of tail risk is computationally demanding, the paper also derives and analyzes a closed-form approximation to such a measure.