تأثير كاهش داده‌ها با استفاده از تحليل عاملي بر دقت مدل‌هاي پيش‌بيني ورشكستگي

Data reduction influence on the accuracy of prediction failure company models

هدف از اين پژوهش تعيين الگوهايي با استفاده از نسبت‌هاي مالي براي بالا بردن توان تصميم‌گيري استفاده‌كنندگان از صورت‌هاي مالي در پيش‌بيني ورشكستگي شركت‌ها است. در اين پژوهش از 55 نسبت مالي پركاربرد استفاده‌شده و با استفاده از تحليل عاملي اكتشافي به 12 عامل تبديل شده است. سپس با استفاده از مدل لوجيت و شبكه‌هاي عصبي صحت پيش‌بيني ورشكستگي با استفاده از 12 عامل به‌دست‌آمده، مورد بررسي قرار گرفته است. جامعه‌ي آماري شامل دو گروه، 40 شركت ورشكسته و 82 شركت غير ورشكسته، است. اطلاعات استفاده‌شده مربوط به دوره‌ي زماني 1393-1387 است و نتيجه‌هاي پژوهش حاكي از آن است كه 12 عامل به‌ دست‌ آمده با بهره‌گيري از هر دو مدل، داراي توان بالايي در پيش‌بيني ورشكستگي شركت‌ها است و نيز الگوي مبتني بر شبكه‌ي عصبي داراي بالاترين دقت است.

Introduction The main goal of providing financial information is to improve the ability of decision makers to make financial and economic decisions. Therefore, the accuracy of the information provided increases the possibility of the accuracy of decisions. Understand and predict the likelihood of bankruptcy and attempt to improve these models has been the most important financial concern of decision-makers over the past decade being the subject of extensive researches conducted by financial and accounting researchers. In our country this subject has always attracted the attention of researchers. Previous studies used different models and techniques to predict bankruptcy, but the constant technological changes and economic developments on one hand, and the complexity of the business environment in the country due to economic sanctions on the other hand have faced durability of existing companies with a serious threat and thus the necessity to provide appropriate models and use new techniques for accurate prediction of corporate bankruptcies have doubled in Iran. Accordingly, it is necessary to carry out more intense scientific research in the field. Thus, the current paper aims to explore new bankruptcy models that fit Iran economic environment using factor analysis techniques, neural network and logit. Research Hypothesis To predict the bankruptcy of companies accepted in Tehran Stock Exchange Market, we use analysis factor as a data reduction method. In this study we examined the accuracy of prediction failure models with these factors. The Following hypothesis can be made in this regard: 1) Exploratory factor utilizes the logit model's ability to predict bankruptcy. 2) Exploratory factor utilizes neural networks, have the ability to predict bankruptcy. 3) Neural networks have higher ability than logit model in recognition of bankruptcy companies. Methods This is an applied EX- post factor correlational study. This research consisted of firms listed in Tehran Stock Exchange during the period from 2008 to 2015. Accordingly, a list of companies has been prepared to study including 40 bankrupt ones that were subject to Article 141 of the commercial code and article 41 of stock exchange directions. It also should be noted that using the information related to six years before bankruptcy for each company required them to be studied during the periods from 2008 to 2015. Then we used Q.Tobin to select 82 companies that are not bankrupt. We used 55 financial ratio and analysis factor to reduction data (12 factor). Dependent variable: the probability of bankruptcy of a firm; zero for bankrupt companies and one for others. Independent variables:12 Factors Extraction of 55 financial ratios. Results In this study, 55 financial ratios were reduced to 12 factors by factor analysis, so that each factor contained factor loads of variables. Then, it was reviewed by using logit and neural network. In addition, Q.tobin’s criteria was used to unbankrupt companies & article 141 of commercial code and article 41 of stock exchange directions to select bankrupt companies. To analyze data, SPSS 23 & MATLAB softwares were used as results were determined by comparing. Exploratory factors have a high capability in predicting bankrupt companies. In this respect, both models enjoyed high accuracy. As a whole, neural network method has high accuracy in classified companies. Thus, in both models, the highest accuracy is dedicated to data about a year before the base year. Discussion and Conclusion The findings show that the use of factor analysis to convert various financial ratios to several homogeneous factors and the use of neural network techniques and logit for data analysis improve the ability of models to predict corporate bankruptcy. Therefore, to determine and estimate companies bankruptcy, it is suggested the investors, including potential investors and financial analyzers use this techniques, variables and used models in this study. Using the neural network model is more emphasized by researchers, of course. To review the ongoing concern of companies' activity auditors can also use above-mentioned models. With the help of results and presented model in current article, banks and other financial and credit institutions as well as creditors in general can examine the decision on granting the loan to the companies, especially companies with consecutive & accumulated losses. Moreover, they can make decisions that are more reasonable as well as prevent wasting their capitals.