應用一般迴歸神經網路法構建財務危機預警模式

Abstract

企業若發生財務危機往往會導致一般投資人損失慘重，為了讓投資人能夠在企業發生財務危機前，就能夠儘早有所察覺，以免投資不當，因此有必要構建一套財務危機預警模式。目前已有許多中外文獻運用各種方法，例如：二分類檢定法(dichotomons classification test)、區別分析(discriminate analysis)、Probit分析、Logit分析、倒傳遞類神經網路(back-propagation neural network，BPNN)、模糊理論（fuzzy theory）等，來構建財務危機預警模式。由於財務變數的一些特殊特性，使得研究的結果幾乎都傾向由BPNN所建立的財務危機預警模式有較高的正確預警率，但類神經的預測方法並不只有一種，其中一般迴歸神經網路法(general regression neural network；GRNN)是較晚期才發展的類神經預測方法，且已有學者證明GRNN於各個領域之應用均較BPN佳，但GRNN法至今尚未在中外文獻中見其用在構建財務危機預警模式上，因此本研究利用一般投資大眾皆可收集到的財務資料，利用 GRNN法構建一個財務危機預警模式，提供一般投資大眾使用，並利用實際財務資料，比較本研究方法和BPNN之財務危機預警模式之正確預警率，結果顯示本研究之GRNN財務危機預警模式較BPNN財務危機預警模式之正確預警率為高，證實本研究所提之GRNN方法確實為一有效之構建財務危機預警模式的方法。

Investors always encounter a loss when the financial distress of enterprises occurs. Hence, to prevent the investitive loss, establishing a financial distress warning model for investors is necessary. Many studies used various methods such as: dichotomons classification test, discriminate analysis, Probit analysis, Logit analysis, Back-propagation Neural Networks(BPNN), Fuzzy theory etc, were employed to establish the model. Because of some specific properties of financial ratio variables, all of above methods suggested to use the BPNN method to get high accuracy on financial distress warning model. However, a recently developed neural network, general regression neural networks (GRNN), has been proven to have a higher predictive power than BPNN. Therefore, this study utilizes the GRNN method to establish the financial distress warning model based on a real set of financial ratio data and compares the effectiveness of both methods. The results indicate that the GRNN model has better early warning accuracy than that of the BPNN model.