利用資料探勘技術辨別財務重編公司的特性

Abstract

本研究利用資料探勘技術探討重編企業的重要特質。研究樣本為2002年至2012年台灣上市公司。為了提高分群分析的準確性，共有三種分群算法（K均值，兩步法和自我組織圖（SOM）被應用來分析。在找到最佳分群模型後，將所有樣本放進最佳分群模型應用決策樹技術來分析結果。結果發現，K均值方法為最佳分群模型，並進一步將所有資料分成三組。利用決策樹分析技術發現，財務重編企業有以下特徵，包括（1）經會計師簽證之企業繼續經營之營運支出的能力被證明可能是有疑問的，（2）大多數主要來自非四大會計師事務所簽證，（3）簽証說明可能表現出較低的股東權益報酬率。當公司擁有上述三個特徵之一時，重編財務報表的可能性提高。

This study applies data mining technology to investigate the characteristics of companies restating financial reports. The research sample includes these companies listed in Taiwan market, chosen from 2002 to 2012. To improve the accuracy of the cluster analysis, this paper applies three kinds of clustering algorithms (K-means, Two-step, and Self organization map (SOM)) to conduct the analysis. The results show that the optimal clustering model is the K-Means method and the data is divided into three groups by applying the decision tree technique to locate the critical characteristics of the restating firms. The results find that the firms with financial restatement possess the following characteristics: (1) the capability of a company to continue the operational spending which is certified by the accountant may prove to be doubtful, (2) most aforementioned certifications are mainly issued from the non-big four CPA firms, and (3) the certified statement may show a lower return on equity. When a company possessed one of these three aforementioned characteristics, it is noted that the higher likelihood of a firm to restate the financial statements.