混合式電腦輔助稽核技術於審計風險之應用

A hybrid computer assisted auditing techniques in auditing risk management

徐銘甫(Ming-Fu Hsu)

指導教授 ： 白炳豐 ； 王銘杰

國立暨南國際大學/管理學院/國際企業學系/博士(2012年)

摘要

財務報表審查不健全將損害全球經濟市場的運作、降低投資者的信心。專業的會計機構指出審計失敗是因為查核人員缺乏足夠的知識與欠缺有效之查核工具。此外，審計判斷的一致性取決於查核人員的專業知識與問題解決能力，在現今急遽變化的經濟環境下，採用一般的查核工具往往無法找出問題所在。審計失敗大多與高階管理階層相關，其可以越過傳統的內部控制與抵制審計委員會有效之運作，更重要的是其瞭解傳統標準查核工作之限制，因此，本研究發展出一個新的審計風險偵測模型，它整合了多元特徵擷取技術、機器學習與知識擷取技術。多元特徵擷取技術的概念源自於整體學習，其優點為克服單一技術的內部缺陷，並採用多準則技術有系統地選出最有資訊內涵的變數及其相關機制之組合。支援向量機是近年來相當受到注目的新影技術，其分類效果在許多領用上都有顯著的表現，但其缺點為缺乏可理解姓，本研究採用知識擷取方法來解決支援向量機黑盒子的缺點，並將其決策步驟轉換為易於瞭解的規則，進一步提升其應用的廣度。此外，本研究近一步探討資訊透明度對審計失敗的影響，此結果可以提供相關當局瞭解此指標的重要性。本研究之結果可提供內部與外部審計人員依照相關的規則妥善分配有限的審計資源，亦可以提供投資者充分的財務資訊。

關鍵字

財務報表 ； 多元特徵擷取技術 ； 機器學習 ； 知識擷取 ； 多準則決策 ； 透明度與資訊揭露

並列摘要

To protect the global economic market, fraudulent financial statements (FFS) detection is essential. Recently, FFS have begun to grow extremely, which has deteriorated the confidence of investors and shocked the financial systems. Professional literature indicated that failure in detecting FFS rested with auditor’s insufficient capability and lacked of effective assisted mechanism. Auditing judgment consistency has proven that it is subject to auditor’s work experience and the ability of problem solving, so that leads the auditing decisions encountered in today’s turbulent business environment to cover with a layer. In addition, most FFS is caused by top managers who have the authority to override the internal controls and deploys de facto power against audit committee. Such managers understand the limitation of an audit and the insufficient of standard auditing procedures in detecting FFS. There is an urgent need for another effective detecting mechanism. The study proposed a hybrid model to reduce these risks. The model integrates multiple feature selection combination which was grounded on ensemble learning, support vector machine (SVM) and knowledge extraction approaches. The advantage of multiple feature selection can eliminate the errors made by singular approach and determine appropriate features and mechanisms by multiple criteria decision making (MCDM) technique. The SVM has superior forecasting accuracy comes with a critical defects is lacking of interpretability. Thus, the knowledge extraction approaches were employed to tackle with the obscure nature of SVM and yield comprehensive rules as well as enhance its empirical application. The proposed model, which is supported by real example, can assist both internal and external auditors who must allocate limited auditing resource. The decision rules derived from the proposed model can be viewed as a roadmap to modify the personal capital structure. In addition, the investigation further examines the effectiveness of corporate transparency and information disclosure index on FFS. The governors can consider the potential implication and formulate future policy to sound the stability of financial market.

並列關鍵字

Financial statement ； Multiple feature selection ； Machine learning ； Knowledge generation ； Multiple criteria decision making ； Transparency and information disclosure

參考文獻

[1]. J. L. Abbot, Y. Park, S. Parker, The effects of audit committee activity and independence on corporate fraud. Managerial Finance, 26(11), 2000, 55–67.

[2]. K. J. Archer, R. V. Kimes, Empirical characterization of random forest variable importance measures. Computational Statistics & Data Analysis, 52(4), 2008, 2249-2260.

[3]. N. Barakat, J. Diederich, Learning-based rule-extraction from support vector machines: performance on benchmark data sets, in: Proceedings of the Conference on Neuro-Computing and Evolving Intelligence, Knowledge Engineering and Discovery Research Institute (KEDRI), Auckland, New Zealand, 2004.

[4]. N. Barakat, A.P. Bradley, Rule extraction from support vector machines: a sequential covering approach, IEEE Transactions on Knowledge and Data Engineering, 19, (2007), 729–741.

[5]. N. Barakat, A. P. Bradley, Rule extraction from support vector machines: A review. Neurocomputing, 74(1-3), 2010, 178-190.