Testing fraud-detection tests: validation of the Z-score, the M-score and the Sloan ratio, and development and validation of a new model

Abstract

Truthfulness in accounting is fundamental to ensure effective decision-making and avoid negative spillovers on society. This thesis validates, with respect to calibration and discrimination, the Z-score, the M-score and the S-ratio as accounting fraud detection models, using the Hosmer – Lemeshow goodness-of-fit test and a receiver-operation-curve respectively, and finds that these are unable to properly discriminate between companies manipulating their accounts and companies not doing so. I have developed and validated a New Model that discriminates better than existing ones using changes in Accounts Payable, Current Assets and Leverage. These tackle the different red-flags associated with accounting fraud: working capital changes, deteriorating margins and sudden changes in leverage.