بررسي سودمندي روش هاي انتخاب متغير ريليف و همبستگي در پيش‌بيني و شناسايي متغيرهاي پيش بين بهينه تاثيرگذار بر سياست جسورانه مالياتي

Investigating the usefulness of Relief and correlation variable selection methods in predicting and identifying optimal predictor variables affecting aggressive tax policy

شواهد تجربي نشان مي‌دهد شركت‌ها در تلاش‌اند ماليات بر درآمد خود را كاهش داده يا به تعويق بيندازند. در اين پژوهش سياست جسورانه مالياتي به‌عنوان ابزار كاهش ماليات بررسي مي‌شود. پژوهش حاضر درصدد فراهم آوردن شواهد تجربي در خصوص مقايسه عملكرد روش‌هاي خطي مانند روش‌هاي مبتني بر همبستگي و ريليف با ساير مدل‌هاي قبل از كاهش متغيرها در پيش‌بيني معيارهاي سياست جسورانه مالياتي شركت‌هاي پذيرفته شده در بورس اوراق بهادار تهران است. هم‌چنين سودمندي روش‌هاي مبتني بر همبستگي و ريليف براي انتخاب متغيرهاي پيش‌بين بهينه، بررسي و مقايسه شد. به اين منظور دادههاي مربوط به 108 شركت براي دوره‌ زماني بين سال‌هاي 1389 تا 1397 با استفاده از روش داده‌هاي تابلويي، بررسي و آزمون شد. نتايج پژوهش حاكي از آن است كه بين برآورد كليه متغيرهاي پژوهش و بهره‌گيري از روش‌هاي‌كاهش متغير تفاوت معناداري وجود دارد. همچنين نتايج پژوهش نشان مي‌دهد كه سودمندي روش ريليف در مدل‌سازي عوامل مؤثر بر سياست جسورانه مالياتي بيشتر از روش‌ همبستگي است. علاوه بر اين نتايج نشان دادند كه عدم اطمينان متغيرهاي كلان اقتصادي تأثير مثبت و معنادار و عوامل كيفيت حسابرسي، رقابت بازار محصول، و ويژگي‌هاي مديران عامل تأثير منفي و معناداري بر سياست جسورانه مالياتي دارند.

The main issue in this research is the study of factors affecting aggressive tax policy and also modeling the factors affecting aggressive tax policy. The present study seeks to provide empirical evidence to compare the performance of methods such as correlation-based methods and relief with other models before reducing the variables in predicting aggressive tax policy criteria of companies listed on the Tehran Stock Exchange. In this regard, the usefulness of correlation and relief methods for selecting optimal predictor variables was also investigated and compared. 2.Hypotheses The purpose of this study is to investigate the performance of relief and correlation-based linear methods in predicting the optimal predictor variables affecting aggressive tax policy. According to this aim and the theoretical foundations, the research hypotheses were formulated as follows: Hypothesis 1: There is a significant difference between the predictive performance of aggressive tax policy using optimally selected predictor variables and the use of all predictor variables. Hypothesis 2: There is a significant difference between the usefulness of variable selection methods in predicting aggressive corporate tax policy. 3- Method For this purpose, data related to 108 companies listed on the Tehran Stock Exchange for the period between 1389 to 1397 were used. The fixed effects pattern regression method was evaluated and tested using the panel data method. The sample includes 108 companies that were selected by elimination method (screening). The data were first sorted and classified in Excel software and then analyzed using EVIEWS software. Also, relief and correlation-based methods were used using VEKA software to select the optimal independent variables affecting taxation 4- Results The results of testing the first hypothesis showed that the error of the first and second types and the coefficient of determination for all primary variables and the correlation-based method and the relief method are equal to (0.080, 0.060 and 0.597), respectively (0.090, 0.064 and 0.511) and (0.089, 0.064 and 0.522) which indicate a significant difference between the methods of reducing the variable and all the main variables; As a result, the first hypothesis is confirmed. Also, the error of the first and second types and the coefficient of determination for variable reduction methods including: correlation-based method, and relief method are equal to (0.090, 0.064 and 0.511) and (0.089, 0.064, respectively). an‎d 0.522) and shows that the usefulness of Relief method> is based on correlation method and therefore it can be said that among the methods of reducing the variable Relief method is better useful and thus the second hypothesis is confirmed. 5-Discussion and Conclusion The purpose of this study is to model the factors affecting tax planning in companies listed on the Tehran Stock Exchange. In this study, the measure of aggressive tax policy has been used to measure tax planning. Also in this research, two methods of variable reduction including (correlation-based method and Relief method) have been used to reduce the research variables and select the optimal variables. The results indicate that there is a significant difference between estimating all research variables and using variable reduction methods. The results also show that the usefulness of Relief method in modeling the factors affecting aggressive tax policy is more than the correlation method. In addition, the results indicate that among the factors affecting various tax criteria, the uncertainty of macroeconomic variables have a positive and significant effect on aggressive tax policy; In addition, the results showed that the uncertainty of macroeconomic variables has a positive and significant effect and the factors of audit quality, product market competition, and the characteristics of CEOs have a significant and negative effect on aggressive tax policy