應用倒傳遞與自組織映射圖神經網路於企業體質檢定模型之研究

Abstract

受到全球經濟環境的波動難以預測的影響下，台灣的企業在經營上已不像過去中小企業經營那樣容易，一旦遇到財務危機或整體經濟走勢不佳的利空因素而面臨倒閉時，其可能造成的龐大社會成本付出的連帶效應不容小覷。因此，若能夠在企業發生財務危機前，有效的從總體經濟指標和過去財務資料中可能引發危機的種種跡象中透過方法評估出經營狀況，提早客觀且有效地發現營運發生困難或體質不佳的企業，能幫助企業在財務爆發危機前找出因應對策，將可能造成的傷害降至最低。 自Beaver在1996年提出「以財務比率預測經營失敗」的研究後，財務危機預警的觀念與相關研究開始受到重視。因此，本研究提出以倒傳遞類神經網路為架構，搭配總體經濟指標與18項企業財務報表中的重要指標，嘗試從財務趨勢行為中，找出知識規則，推估企業下個月之財務表現。本研究運用倒傳遞類神經網路來做企業體質的總和評判，使類神經網路的輸出更具有可靠性，建立一個預測月內財務指數趨勢的企業體質預測模型，讓企業在面臨潛在的財務危機時，依據本模型所發出的警訊做出相對應的策略，以達到企業財務危機預警之目的。 有鑑於一般民眾對財務報表的解讀與認知多不如專業人士，本研究設計一套運用自組織映射圖網路來建立企業體質特徵模型圖，透過視覺化圖形，幫助一般民眾在投資決策時可以容易上手與解讀公司財務狀況。再搭配結合總體經濟因素與倒傳遞類神經網路的預測模型，相信能夠更精準且提早判斷出公司可能有危機的發生。

Under the impact of unpredictable global economic fluctuation, corporates management in Taiwan is not as easy as it was used to be. Once enterprises are encountering financial crisis or overall economic downward, they may bankrupt and lead to huge social cost and other domino effects. Therefore, if there are ways to effectively evaluate the possible financial crisis of the company through microeconomic index and past reports in advance may minimize the damage of the financial outbreak. After Beaver’s research in 1996, the idea of financial distress pre-warning and related research has started to be valued. Therefore, this research proposes a back-propagation neural network based model with microeconomic index and 18 important indexes of corporate financial reports, trying to find a trend in the finance behavior, and to inspect next month’s financial performance of the company. The results show that the rate of accuracy of the enterprise’s financial state evaluation is above 95%. Due to the fact that ordinary people’s knowledge of financial reports is not as good as the professionals, this research designs a corporate constitution character model by applying Self-Organizing Map. It is convinced that by using visualized graphs, people can understand the financial situation of a certain company in an easier way. In addition, coupled with microeconomic index and back-propagation neural networks, the prediction of when crisis may occur to a company can be more accurate and ahead of time.