Comparing the Performance of Linear Models and Artificial Intelligence to Predict the Manipulation of Financial Statements: Empirical Evidence from Dechow Model and Bayes Networks

Manipulation of information and financial statements of companies leads to the occurrence of fraud and restatement of financial statements, which causes the loss of people's capital and ultimately leads to the bankruptcy of companies. These issues have raised concerns about the quality of information in financial statements and the decision-making of companies' financial information users. Due to the importance of this issue, predicting the occurrence of these manipulations and the factors affecting them have always been the focus of analysts, investors, managers and researchers. The purpose of this research is to compare the performance of linear models and artificial intelligence in the field of predicting manipulation of financial statements. Dechow's linear model and Bayes networks artificial intelligence model were considered to compare the performance for predicting manipulation of financial statements. This research is applied in terms of purpose, post-event in terms of data, and descriptive-correlation in terms of analysis. The population of the research was all the companies listed on the Tehran Stock Exchange in the period from 2017 to 2022, and the samples were selected using the systematic elimination method with restrictions. The criteria for selecting companies with manipulation of financial statements is the presence of an unqualified audit opinion with a conditional clause subject to distortion in financial data or the existence of tax disputes with the tax area according to the income tax reserve note and tax file and the condition clause of the audit report or the existence of significant annual adjustments financial updates were provided. Research data has been collected by means of library and document mining methods and has been analyzed with Eviews software. The results showed that the Dechow model with 68.88% correctness and Bayes networks with 90% correctness have the ability to predict the manipulation of financial statements among companies listed on the Tehran Stock Exchange. Also, according to the research results, the performance of artificial intelligence models such as Bayes networks is better than linear models such as the Dechow model.