Subjectivity in Conventional Tail Measures : An Exploratory Model with 'Risks & Biases’
Many commonly used models for measuring risks depend on some input parameters whose values are generally set by judgment. The approach is practical as judgment or intuition reduces complexity to a manageable level. Faster decisions can be taken in a business environment where timeliness is more important than precision. A critical input parameter, for example, in the case of tail measures would have been the risk tolerance level or the probability level. Paradoxically, scholars pre- and post-global financial crisis commonly possess opposite views on the magnitude of the risk tolerance level: less than the 99th percentile was the pre-crisis prescription, and more than or equal to 99th percentile was the prescription post-global financial crisis. The issue frequently brought up by critics in post-March 2020 political-economic discourse has been whether the conservative regulatory prescriptions were triggered by the debilitating impacts of the “Great Recession” of 2007-10. While it is difficult to confirm that the same had been an ultracrepidarian criticism, it is equally difficult to ascertain whether or not the low profitability of a business firm was due to its functioning in a low-risk environment post-March 2020. Additionally, higher administrative costs and sometimes escalated legal costs led to either extremely low profits or losses, creating a bad reputation, degradation in credit ratings, and adversely affected funding sources over time. Conversely, the present paper proposes a framework wherein both the risk tolerance level and the Value-at-Risk level are determined simultaneously by the model. The approach is helpful in reducing the bias in tail measures evolved due to the conjoint stream of policies, i) regulatory prescriptions and compliance directives, those that made more conservative unexpectedly after March 25, 2020, and ii) the selection of a risk model which was not having a good fit, specifically at higher quantiles
