An investigation in BIST on f-score model and pentagon theory for the detection of fraud risk

Big accounting scandals around the world have increased the interest in detecting fraud risk of companies which are traded in the stock market. Various fraud detection methods have been created in the literature by investigating the causes of these scandals and examining their transactions. It is clear that the detection of fraudulent transactions of companies are important in many aspects. Fraudulent financial reporting can cause huge losses for the parties involved. Regulations and laws enacted to prevent this might be insufficient. Companies can hide their fraudulent transactions with improper records and methods by finding gaps in the accounting system and taking advantage of them. As a result, detection of fraudulent transactions becomes difficult. The aim of this research is to give an approach to the detection of fraud risks of companies traded in Istanbul Stock Market and to examine the effect of the variables used in the research in detecting fraud risks on traded companies in Istanbul Stock Market. In this study, the effect of the factors in the fraud theory, which started as a fraud triangle and developed as a fraud pentagon, on fraudulent transactions were investigated. After that, F-score model was investigated for the prediction of fraudulent transactions which was developed by Dechow et al. In order to create models for the fraud detection, two groups were formed among the companies that are traded in Istanbul Stock Market and do not have financial activities. These chosen companies separated two groups as manipulators and non-manipulators. The annual report results of these companies between the years 2017-2021 were examined. SPSS statistical program was used for the analysis of the data in the study. With this program, logistic regression analysis was applied to examine the effects and relationships of pressure, opportunity, rationalization, arrogance and competence factors in the fraud pentagon theory in determining fraud risk probabilities. As a result of the analysis, external pressure, financial target and financial stability, which are examined as sub-factors of the pressure variable and other two fraud factors which are opportunity and rationalization was significant in terms of determining the probability of fraud risk, but no significant relationship was found with competence and arrogance fraud factors in the model. Logistic regression analysis was applied also to examine the effects and relationships of Dechow F-score financial variables on determining the probability of fraud risk. These financial variables are change in accounts receivable, change in intangible assets, change in cash sales, change in return on assets, changes in inventories and RSST accruals. As a result of this analysis, in the model only the change in account receivables and the change in intangible assets was significant in terms of determining the possibility of fraud risk. A new model has been established with the change in accounts receivable and the change in intangible assets and the variables that are significant in the fraud pentagon which are external pressure, financial target, financial stability, opportunity and rationalization. As a result of this analysis, the change in accounts receivable, external pressure, financial target, opportunity and rationalization were found to be significant within the model. However, intangible assets and financial stability were not found significant in the model. In the next stage, F-score values were calculated for the first model constructed with significant fraud pentagon factors and for the second model constructed with Dechow F-score variables and significant fraud pentagon factors, and the success of these values in distinguishing manipulator and non-manipulator companies was examined. Last step was to compare methods which are logistic regression analysis and artificial neural network in terms of correct classification of the companies grouped according to their fraud risks. Classification accuracy rates were analyzed and compared with artificial neural networks and logistic regression analysis separately for model 1 and model 2. In both models, it has been obtained that the artificial neural networks perform a higher rate of correct classification, even if it is a small percentage. The purpose of this research is to create models by using financial and non-financial data in determining the fraud risks of companies traded in Istanbul Stock Market. The calculation of F-score values for the determination of fraud risk with the models created is considered to be an important and original study in terms of statistically examining which variables can guide the detection of fraud risk in Istanbul Stock Market.