Accounting for Estimation Risk When Pricing Under Adverse Selection

• Activity Number: 141
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Risk Analysis
• Author(s): Richard Neuberg* and Lauren Hannah
• Companies: and Columbia University
• Keywords: Pricing ; Prediction ; Adverse Selection ; Credit Scoring ; Kernelized Logistic Regression

Abstract

Financial product prices are set using mathematical models based on a set of inputs. We examine the challenges involved in estimating model parameters when the counterparty engages in adverse selection: minimizing overall estimation risk, assessing counterparty-specific estimation risk, and determining the necessary premium for irreducible estimation risk. Pricing regime credit scoring is chosen as an example. Using estimated probabilities of default in a plug-in interest rate estimator can cause systematic mispricing under adverse selection due to bias and variance. We suggest reducing these errors by (i) using an economic model which determines the premium for irreducible estimation risk through bootstrap or asymptotic distributional estimates, and (ii) introducing the use of kernelized logistic regression, a more accurate alternative to commonly used default probability estimators such as logistic regression in large samples, which also allows to better estimate loan-specific irreducible estimation risk. These methods are empirically examined on a panel data set from a German credit bureau, where we study dynamic dependencies such as prior rating migrations and defaults.