Retail default prediction by using sequential minimal optimization technique

Research output: Contribution to journalArticlepeer-review


This paper employed sequential minimal optimization (SMO) to develop default prediction model in the US retail market. Principal components analysis is used for variable reduction purposes. Four standard credit scoring techniques—naïve Bayes, logistic regression, recursive partitioning and artificial neural network—are compared to SMO, using a sample of 195 healthy firms and 51 distressed firms over five time periods between 1994 and 2002. The five techniques perform well in predicting default particularly one year before financial distress. Furthermore, the prediction still remains sound even 5 years before default. No single methodology has the absolute best classification ability, as the model performance varies in terms of different time periods and variable groups. External influences have greater impacts on the naïve Bayes than other techniques. In terms of similarity with Moody's ranking, SMO excelled over other techniques in most of the time periods.
Original languageEnglish
Pages (from-to)651-666
JournalJournal of Forecasting
Issue number8
Publication statusPublished - Dec 2009


  • credit risk
  • default prediction
  • sequential minimal optimization
  • multivariate statistics
  • artificial intelligence


Dive into the research topics of 'Retail default prediction by using sequential minimal optimization technique'. Together they form a unique fingerprint.

Cite this