It usually appears in the small print: “The company will contact credit rating agencies in its consideration of your application.” For older products it is easy for lenders to compare your profile against historical data to rate your suitability. Mathematicians have now devised a credit scoring technique for newly launched products where little data is available for analysis.
“There are basically two classes of customer,” notes Professor David Hand from the Department of Mathematics at Imperial College, London. “There are the ‘bads’ – those that default in some way before the end of the loan term – and the ‘goods’ who keep up their payments. Once you have accumulated a body of data for a loan product, you can build some sort of model that uses application form data and credit bureau ratings to classify applicants as good or bad.”
But for new products no data are available. Worse still, you normally have to wait until the end of the loan term before you know that a customer really is ‘good’. Professor Hand has applied a branch of statistics called survival analysis to borrowers’ records that should help banks score applicants more accurately.
“Survival analysis takes into account customers who are likely to default before the end of the loan term,” Professor Hand explains, “not just the customers who have actually defaulted at by the time of the calculation. When we used our survival analysis for lookahead scoring far fewer accepted customers turned bad.”
Professor Hand hopes to further refine his methods. For instance, he wants to allow for the low probability of borrowers going ‘bad’ when reaching the end of the loan term. “Already we have shown that survival analysis, using just a small body of data, is a powerful technique to improve applicant credit scoring for new loan products. It should help lenders accept few ‘bad’ customers and develop more comprehensive predictive models faster.”