應用Logistic迴歸建構盈餘管理預測模型

Applying Logistic Regression to Construct Earnings Management Prediction Model

汪政憲(Cheng-Hsien Wang)

指導教授 ： 陳達新

共同指導教授 ： 林婷鈴

國立臺北大學/商學院/企業管理學系/碩士(2011年)

摘要

國內外的金融舞弊事件增加，使得「盈餘管理」的議題更受到重視。文獻顯示管理因素中影響「盈餘管理」的因子相當多，但卻未被直接用來預測企業的「盈餘管理」行為。本研究運用資料探勘技術中之邏輯斯迴歸，期望能建構一套有效預測盈餘管理模型，提供決策者判斷企業是否進行盈餘管理之參考。本研究以2004年到2008年台灣上市、上櫃公司為研究對象，以裁決性應計數作為公司是否進行「盈餘管理」之代理變數，並根據過去文獻整理出對盈餘管理具有顯著性影響的20個自變數。首先，利用接受者操作特徵曲線(receiver operating characteristic curve，簡稱ROC曲線)下的面積(area under curve，簡稱AUC)來判別模型的鑑別力，再利用Youden’s index及數學商高(畢氏)定理來選擇最佳的臨界值，並考量模型預測錯誤成本高低來評估模式績效。另外，為增強模式整體之正確預測率，經由刪除預測機率值較低之樣本進行特徵篩選。實證結果發現，經由邏輯斯逐步迴歸，篩選出績效門檻、前期裁決性應計數、獨立董事比、股價淨值比、董事會規模、研究發展費用率、極端盈餘表現、財務危機、營收成長率與經理人持股比例等10個變數，所建構出的盈餘管理預測模型，整體正確預測率可達82.48%。

關鍵字

盈餘管理；邏輯斯迴歸；ROC曲線；AUC；Youden’s index

並列摘要

Increasing in domestic and international financial fraud has brought more attention to the issues of earnings management. Many related studies have focused on identifying factors which significantly affect earnings management. However, these factors have not been used directly to forecast the level of earnings management of companies in advance. This study aims to develop an earnings management prediction model, which applies logistic regression of data mining techniques to provide suggestions for outside decision makers. The samples comprised of listed and OTC firms in Taiwan during the fiscal years from 2004 to 2008. Discretionary accruals are used as the proxy for the magnitude of earnings management. Twenty independent variables with significant impact on earnings management are sorted out from the literature. First of all, we used techniques based on the Receiver Operating Characteristic (ROC) curve to cope with different cut-off values. And then, we evaluated the discriminatory power by the Area Under the Curve (AUC). Youden's index and Pythagorean theorem are used to find an optimal cut-off value, and the measure of the misclassification cost is used to evaluate model performance. In addition, to enhance the prediction power of the model, we perform feature selection by removing the samples with lower predicted probability. Ten variables in the feature subset were chosen from empirical results using the stepwise method of logistic regression. The model built on the ten variables was able to achieve a prediction accuracy rate of 82.48%.

並列關鍵字

Earnings management；logistic regression；ROC curve；AUC；Youden’s index

參考文獻

Hsieh, W. T. and Wu, T. C. (2006). Determinants and market reaction of assets impairment in Taiwan. Taiwan Accounting Review, 6(1), 59-95.

林有志、邱炳雲、何韋霆(2009)，「盈餘門檻與盈餘管理行為之研究」，會計與財金研究，2卷2期：頁1-16。

Barton, J. (2001). Does the use of financial derivatives affect earnings management decisions?. The Accounting Review, 76(1), 1-26.

黃志仁、廖彩伶、陳于格(2009)，「現金增資之盈餘管理行為：裁決性應計項目與業外損益之整合性決策」，當代會計，10卷1期：頁63-98。

楊朝旭、吳幸蓁(2003)，「總經理薪酬績效敏感性、績效門檻與盈餘管理關聯性之研究」，會計評論，36期：頁55-87。