Predicting accounting fraud: Evidence from Japan (Accepted by The Japanese Accounting Review)


Release Date

September, 2016


This study develops a prediction model for identifying accounting fraud by analyzing the accounting information for Japanese firms. In particular, we (1) explore the characteristics of accounting fraud firms by analyzing financial information obtained from annual reports (yukashoken-houkokusho in Japanese) and (2) develop a model for predicting accounting fraud based on the characteristics of Japanese fraud firms. To identify the characteristic of fraud firms, we focus on 39 variables for the eight factors of “accruals quality,” “performance,” “nonfinancial measures,” “off-balance-sheet activities,” “market-related incentives,” “conservatism,” “real-activities manipulation,” and “Japanese-specific factors.” Through our univariate analysis and model building process, we find that “accrual quality,” “market-related incentives,” “real-activities manipulation,” “conservatism” and “Japanese-specific factors” are generally useful for detecting accounting fraud. We also conduct several analyses that test the predictive ability of our models, including (1) the detection rates of fraud firms, (2) Type I and Type II error rates, (3) marginal effect analysis on independent variables, and (4) robustness tests on time periods and industry clustering. We find that our models have generally higher predictive power in detecting accounting fraud. We expect that our models can be used widely in various accounting and finance practices.