Financial Crisis Forecasting-Comparing Logit Model with Neural Network

[[abstract]]摘要 過去研究常使用公司財務報表，用以建立企業危機預警(corporate bankruptcy)模型；然而，許多發生危機公司的資料往往較不健全，在使用上和搜集上較容易受到限制，而且，對已發生危機公司所提供之財務報表，其可信度仍是另一個值得思考的問題。所以，單純以公司所提供之財務報資料來建立預警模型，似乎容易受到公司提供資訊的可靠性問題，而使得模型受到質疑。因此，本研究首次加入新的市場風險指標VaR(Value at Risk)，以增加利用VaR來衡量公司的市場風險程度。 過去過內多數研究者皆表示台灣股票市場為一有效率的市場，根據效率市場假說，若股票市場是有效率的，則公司的股價應已反應所有已公開資訊。基於此一理由，相信利用VaR較能合理且正確的對公司危機發生做一預測。本研究探討使用羅吉斯模型與類神經方法分別以財務變數所建構的模型、加入景氣因素及VaR值的預測能力比較，並了解加入景氣因素及VaR值是否能提高預測能力。 研究結果可歸納以下幾點： 1. 隨著時間增加，模型的預測能力就越差。 2. 危機發生前一年預測能力最高，顯著的變數為營業用現金流量比率、應收帳款週轉率及每股盈餘，顯見這三個財務比率為主要影響危機發生的主要變數。 3. 加入景氣因素並無法有效提高預測能力。 4. 加入由GARCH(1,1)所求算之VaR值，能夠有效提高危機發生前一年預測能力。 5. GARCH(1,1)較傳統變異數較能有效捕捉股票市場波動的情形。 6. 類神經的預測能力並沒有比羅吉斯模型要來的好。[[abstract]]ABSTRACT In the past, we used to set up a financial crisis forecasting model by using a company’s financial statement. However, it has limitation in using and collecting data because the data of many companies that had crises are usually incomplete. Furthermore, the credibility of the financial statements provided by those companies is another question for us to think. According to this, setting up a business crisis precaution with only the financial statements of a company tends to be query because of the credibility of the data. Therefore, this research adds a new market risk index VaR(Value at Risk), as the proxy of market risk. In the past, most researchers found that Taiwan’s stock market is efficient. Based on efficiency market hypothesis, if the stock market is efficient, the stock’s price of the company has reflected all public information. On account of this reason, I believe we can make more reasonable and more correct forecast of a company’s crisis by using VaR. In this report, we discuss comparing Logit Model with Neural Network’s forecast ability when add financial variable, business cycle and VaR. Whether the forecast ability will increase by adding business cycle and VaR is studied. We can generalize the conclusions to the following: 1.The more the time passes, the worse the model’s ability of forecast is. 2.The greatest ability of forecast is on the year before the crisis happened. The most important variables that influence the crisis are business cash flow ratio、account receivable turnover ratio and earning per share。 3.Adding the element of boom cannot rise the ability of forecast efficiently. 4.Adding GARCH(1,1) count VaR can efficiently rise the ability of forecast the year before the crisis happens. 5.GARCH(1,1) can catches the volatility of the market more effectively than traditional variables. 6.Neural Network’s ability of forecast isn’t better than Logit mod