以階層式自組織映射圖神經網路於動態金融環境之異常財務行為發現

Abstract

近年來台灣有許多體質不健全或營運不善的公司爆發財務危機事件，而最終蒙受損失為社會投資大眾並影響整體經濟環境。然而回顧以往學者的研究，皆以靜態的財務報表為分析的依據，或僅以總體經濟變數做單一構面的考量，本研究基於上述因素期望能以全盤式的考量在動態環境下企業之異常。而由於近年來有許多學者採用類神經網路之非監督式學習－自組織映射圖神經網路於財務危機上之應用均有不錯之成果，其特點即在於運用競爭式學習及自我組織映射的觀念，學習輸入資料的內在聚類規則，以達到資料類聚之目的。故本研究採用階層式的自組織映射圖神經網路，建構企業之財務趨勢檢定。

本研究之定義其異常財務行為泛指：於經濟新報資料庫危機事件中具有1.重整2.接管3.下市4.全額交割5.掏空挪用，以上五項因素其中一項。資料來源為2000年1月至2007年12月(96個月資料)，共724家公司。以總體經濟指標及財務指標資料，並透過一階變量處理及正規化，運用階層式自組織映射圖神經網路，建構企業異常財務行為趨勢，研究結果顯示以總體經濟指標及財務指標於企業異常財務行為趨勢，確實較優於只採用財務指標來得顯著，且採用階層式自組織映射圖神經網路更可以有效辨識企業之異常狀態。

That many domestic corporations have suffered financial crises recently victimizes shareholders of those corporations and negatively affects the overall economy. Prior researches on those financial crises are basically one-dimensional while basing on static balance sheets or variables of macroeconomics to support their conclusions. This research based on the above factors expectation can by omnibearing consideration under the dynamic environment the corporations anomaly.A number of scholars apply an unsupervised learning：Self-Organizing Map (SOM) of a neural network to their analyses of those financial crises taking advantage of characteristics of the competitive learning and the Self-Organizing Map , which help obtain the rules for internal clusters of the input data before categorizing them. As such, the study adopts a hierarchical SOM to detect the trend of corporation finances.

Anomaly financial behavior in the study borrows the definitions in Taiwan Economic Journal, which include (1) reorganization, (2) takeover, (3) delising, (4) full delivery, and (5) embezzlement. Over a 96-month period beginning from January 2000 to December 2007, seven hundred and twenty four(724) domestic companies have ever been subject to either one of the aforementioned anomalies.The study relies on the macroeconomics and finance indicators,”one-step variable” processes and normalizes them, and utilizes the Hierarchical SOM to establish the trend of the financial anomaly of corporations. The present research further shows the combination of macroeconomics indicators and finance indicators is superior to the use of finance indicators only in predicting anomaly corporation finance behaviors and the utilizing of hierarchical SOM helps identify anomaly of corporate operations more effectively.