保留意見對預測企業財務危機的增額解釋能力

Abstract

近年來企業經營環境丕變，企業面臨財務危機的可能性益形增加，而 企業失敗影響的層面相當廣大，因此本研究擬建立一個可靠的模式得以讓 利害關係人有所依循，避免遭受損失。 以往研究者大多以財報資料與 股票市場資料建立模式，但現實台灣市場的效率性是令人質疑的，且股價 變動受到太多外來因素的影響，因此本研究採用財務報表資料。除了財報 資料，本研究亦認為會計師的查帳報告書的保留意見對預測企業財務危機 應具有解釋能力，因此本研究將納入會計師是否出具保留意見的類別變數 和單純僅以財報資料所建構的預測模式做比較，檢驗會計師的保留意見是 否真具有增額解釋的能力。而在研究方法上，由於真實資料並不符合常態 與線性的假設，若採用傳統的統計方法進行分析將會有偏誤產生，而類神 經網路是仿生物神經網路的資料處理系統其無須常態與線性的假設，較符 合真實情形，故本研究將採用類神經網路之倒傳遞網路與傳統的統計方法 來作比較，看何者的效果較佳。 研究解果顯示，無論運用傳統統計方 法(區別分析與Logistic分析)或倒傳遞網路，會計師的保留意見對預測財 務危機均具有增額解釋能力；再比較傳統統計方法及倒傳遞網路上，發現 無論是僅單純以財務比率建構或是財務比率納入保留意見變數、無論是在 測試樣本抑或訓練樣本上，倒傳遞網路的效果均較統計方法為佳。 The probabilities for business to encounter financial difficulties are increasing because of the tremendously changing environment in recent years. Therefore, this paper want to construct a reliable model for all stakeholders to avoid unnecessary losses. Past research build prediction models by financial ratios or stock prices, but it is a debate that Taiwan is an efficacy market, and the change in stock prices are affected by a lot of uncontrollable factors. So the paper only keep financial ratios as useful variables to construct model. ! Additional financial ratios, the paper believe that CPAs' qualified opinions have incremental explanatory power in prediction of business failure. The paper examines does qualified opinions have incremental explanatory power by comparing models which build only by financial ratios and financial ratios adding qualified opinions categorical variable. In methodology, the real data from financial statements usually do not qualify the assumptions of normal distribution and linear. There will have biases if we use traditional statistical analysis. However, if we use neural network to construct prediction model, we can ignore the above two assumptions, and are more likely the way we make judgement. The paper will compare models which construct by neural network and by traditional statistical methods(logistic and MDA). The results indicate that no matter model constructs by BPN or traditional statistical analysis, qualified opinions have incremental explanatory power. Compare with BPN and traditional statistical models, the paper finds that the prediction ability of BPN is better than of traditional statistical analysis, regarding model constructing by only financial ratios or financial ratios adding qualified opinions, no matter in training sample or test sample.