Statistics Seminar（2015-14）

Topic:Parsimonious, Accurate and Con fident Credit Rating

Speaker:Lijian Yang, Soochow University

Time:Wednesday, 14thOctober, 14:00-15:00

Location:K01 of Guanghua Hotel

Abstract:Berg (2007, Appl. Stoch. Models Bus. Ind.), Ryser & Denzler (2009, Financ. Mark. Portf. Manag.) had proposed to use the versatile generalized additive models (GAMs) for credting rating, by computing default/insolvency probability. A type of Bayesian information Criterion (BIC) is formulated for GAM that is asymptotically consistent for parsimonious variable selection without loss of prediction accuracy. We further propose simultaneous confidence corridors (SCCs) through the oracally efficient spline-backfitted kernel smoothing techniques Liu et al (2013 JASA), which characterize the global features of each additive components, and can be used for testing various hypotheses about their overall trends and entire shapes. Simulations and a real example provide illustration of the parsimony, accuracy and confidence features of our prcedures.