Imminence of Event and Scoring Models in Assessing Portfolio Risk

• Activity Number: 24
• Type: Contributed
• Date/Time: Sunday, August 8, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Business and Economics Statistics Section
• Author(s): Timothy H. Lee*+
• Companies: CLConsulting
• Keywords: classification ; score ; performance window

Abstract

Statistical classification is a popular method for assessing likelihood of event (e.g., default in payment or response to solicitation) based on each individual's current profiles. A common way of using such an approach is to create scores, discriminants, or similar quantities as bases for classifying individuals into groups. It is sometimes expected that a good score would reflect the imminence of occurrence of event and can separate one group from the other with satisfactory accuracy as well. The following issues were investigated: (i) Do scores that are developed using same performance windows between profile time and occurrence of event outperform scores developed with varying distances? (ii) How can the imminence of event be incorporated with scores for portfolio decision-making? (iii) How to measure model performance incorporating both accuracy of separation and sensitivity to imminence of event. Finally, proposed a dual concordance rate as a way of measuring model performance and discussed the issues of using short-term time series data for classification, in practitioners' perspective. For the analysis, time series data is generated.