برآورد اقلام تعهدی اختیاری با استفاده از الگوریتم ژنتیک گروه بندی شده

A number of different models have been suggested for detecting earnings management but the linear regression-based model presented by Jones (1991) is the most frequently used. The underlying assumption with the Jones model is that earnings are managed through accounting accruals. Typically, the companies for which earnings management is studied are grouped based on their industries. It is thus assumed that the accrual generating process for companies within a specific industry is similar. However, some studies have recently shown that this assumption does not necessarily hold. An alternative approach which returns a grouping which is, if not optimal, at least very close to optimal is the use of genetic algorithms. The purpose of this study is to assess the performance of the cross-sectional Jones accrual model when the data set firms are grouped using a grouping genetic algorithm. The results provide strong evidence that the grouping genetic algorithm method outperforms the various alternative grouping methods.

The occurrence of earnings management has been a widely studied subject for the past 30 years. One of the major challenges when examining possible earnings management is that the magnitude of it is difficult to assess. A number of different models have been suggested for detecting earnings management but the linear regression-based model presented by Jones (1991) is the most frequently used. The underlying assumption with the Jones model is that earnings are managed through accounting accruals. Typically, the companies for which earnings management is studied are grouped based on their industries. It is thus assumed that the accrual generating process for companies within a specific industry is similar. Recently, however, some studies have shown that this assumption does not necessarily hold. Dopuch, Mashruwala, Seethamraju, and Zach (2012), for example, showed that a violation of the homogenous accrual generating process within an industry causes measurement errors. In another study Ecker, Francis, Olsson, and Schipper (2011) showed that the performance of the Jones model is improved when lagged total assets are used as a grouping variable instead of the industry membership. Even though alternative methods have been used for grouping companies when using the Jones-model, none of them has clearly outperformed the grouping based on industry membership. An exhaustive search for the best possible grouping is in most cases impossible considering the large number of possible combinations even with moderate size data sets. An alternative approach that returns a grouping which is, if not optimal, at least very close to optimal is the use of genetic algorithms. Genetic algorithms have proven efficient in solving difficult problems such as the travelling salesman and the equal piles problems. Genetic algorithms have been used in a number of accounting applications. Back, Laitinen, and Sere (1996) used a genetic algorithm to determine the optimal predictors for a neural network-based bankruptcy prediction model. A similar study was carried out by Shin and Lee (2002) when they used a genetic algorithm to generate bankruptcy prediction rules. Hoogs, Kiehl, Lacomb, and Senturk (2007) presented a genetic algorithm approach for detecting financial statement fraud. Their model successfully classified 63% of the companies that had been accused by the Securities and Exchange Commission (SEC) for improperly recognizing revenue. The purpose of this study is to assess the performance of the cross-sectional Jones accrual model when the data set firms are grouped using a grouping genetic algorithm. The performance of the grouping genetic algorithm approach is compared with the performance of a number of other grouping techniques. The remainder of this study is organized as follows. The basic operating principle of the linear regression-based accrual models is covered in Section 2. In Section 3 an overview of both classic and grouping genetic algorithms is given. The research design is presented in Section 4 and the results from the empirical study are presented in Section 5. Section 6 concludes the study

The purpose of this study was to assess the performance of the cross-sectional Jones accrual model when the data set firms are grouped using a grouping genetic algorithm. The results provide strong evidence that the grouping genetic algorithm method outperforms the various alternative grouping methods. This is especially clear when evaluating the earnings management detection power at different levels of simulated earnings management. The results also show that alternative grouping methods, such as the frequently used two-digit SIC grouping, perform only marginally better compared with a random grouping approach. This study could be extended by including various modifications of the Jones-model as some studies have shown that there is a difference in performance between different versions of the Jones-model. Furthermore, a wider range of settings for the genetic algorithm could be assessed. Especially, lifting the constraint of a fixed number of groups could improve the performance of the genetic grouping algorithm approach.