چکیده:
The main goal of this research is to accurately analyze the profit management using the neural networks and decision tree and comparing them with the linear models. For this purpose eleven variables effecting the earnings management as independent variables and discretionary accruals as a dependent variable have been used. In this research 55 companies from 2006 through 2009 were analyzed in a seasonal way. Regression Panel Method of linear model and Generalized Feed Forward network and CART were used through neural network and decision tree were used. The results of the research indicated that the neural network method and decision tree in the prediction of earnings management compared to the more precise linear methods and have a lower level of error. Meanwhile, earnings management with prior discretionary accruals and performance threshold and the firm performance, size, earnings persistence in both methods has the highest connection.
فایل کامل مقاله :
دریافت (523.4 kB)
هدف اصلی تحقیق حاضر بررسی دقت پیشبینی مدیریت سود با استفاده از شبکههای عصبی و درخت تصمیمگیری و مقایسه آن با مدلهای خطی است. برای این منظور از یازده متغیر تأثيرگذار بر مدیریت سود بهعنوان متغیرهای مستقل و اقلام تعهدی اختیاری بهعنوان متغیر وابسته استفاده شده است. در این تحقیق تعداد 55 شرکت از سال 1385 تا سال 1388 به صورت فصلی مورد بررسی قرار گرفت. از روش رگرسیون پنلی جهت مدل خطی و از شبکه عصبی پیشخور تعمیم یافته و درخت تصمیم کارت جهت بررسی از طریق شبکه عصبی و درخت تصمیم استفاده شد. نتايج حاصل از اين تحقيق نشان داد که روش شبکه عصبی و درخت تصمیم در پیشبینی مدیریت سود نسبت به روشهای خطی دقیق تر و دارای سطح خطای کمتری است. ضمناً مدیریت سود با متغیرهای اقلام تعهدی اختیاری و غیراختیاری دوره قبل و عملکرد شرکت ، اندازه، تداوم سود در هر دو روش دارای بیشترین ارتباط است.
چکیده (انگلیسی):The main goal of this research is to accurately analyze the profit management using the neural networks and decision tree and comparing them with the linear models. For this purpose eleven variables effecting the earnings management as independent variables and discretionary accruals as a dependent variable have been used. In this research 55 companies from 2006 through 2009 were analyzed in a seasonal way. Regression Panel Method of linear model and Generalized Feed Forward network and CART were used through neural network and decision tree were used. The results of the research indicated that the neural network method and decision tree in the prediction of earnings management compared to the more precise linear methods and have a lower level of error. Meanwhile, earnings management with prior discretionary accruals and performance threshold and the firm performance, size, earnings persistence in both methods has the highest connection.
فایل کامل مقاله :
دریافت (523.4 kB)
نکردی من برم شام 
