Application of Artificial Neural Network Hybrid Models with Metaheuristic Algorithms (PSO, ICA) in Earnings Management Predicting

Metaheuristic approaches are inspired mainly by the order and rules of natural organisms. Today, these approaches have been widely used in various branches. According to the importance of forecasting, understanding the methods in earnings management predicting can provide useful information for the beneficiaries. The variety of obtained factors due to the results of linear patterns for measuring earnings management has caused investors to hesitate the reported earnings quality. Therefore, the porpose of this research is to provide a better templet for earnings management predicting. In the first step, using the pattern of neural networks, the linear model was optimized, then Particle Swarm Optimization and Imperialist Competitive Algorithm were used to optimize the pattern. The empirical overviews of 620 observations (year-company) accepted in the Tehran Stock Exchange during the years 2010 to 2015 indicate usefulaness and positive impact of combined methods on the performance of earnings management prediction, there is also a difference in meaning between the usefulness of linear and nonlinear methods. In other words, using predictive algorithms in predicting earnings management, the prediction accuracy increases with the elimination of inefficient variables. In addition, the findings of the research indicate a better and suitable performance of Imperialist Competitive Algorithm than other patterns in the efficiency of the management variables with accuracy (95/8%).