A Multinomial Model of Trade Credit Risk using Data on Individual Trade-Credit Transactions

Using 2.4m individual trade-credit transactions we developed credit rating models based on financial ratios and on debtor company-level and transaction-level payment history. The aim is to allow creditors to classify potential transactions into several categories according to the expected number of days a debtor will be late to settle the invoice past its due date. We find that the most accurate (80% overall) model is the naive, no-change model. It predicts the number of days past the due date based on last preceding transactions with the debtor company. This model is also the cheapest from a creditor's point of view. The errors in classifications are asymmetric. The costly Type I errors is also lowest for this model. Financial ratios are not useful to expose very near-term credit risk of individual transactions, as the Type I errors for these models amount to 99%. The model with financial ratios produces the least Type II errors (0.57%)