頁籤選單縮合
題名 | 具學習性之模糊專家系統在財務危機預測上之應用=The Application of Machine Learning Fuzzy Expert System in Financial Crisis Prediction |
---|---|
作者 | 林金賢; 陳育成; 劉沂佩; 鄭育書; | 書刊名 | 管理學報 |
卷期 | 21:3 2004.06[民93.06] |
頁次 | 頁291-309 |
分類號 | 494.7 |
關鍵詞 | 財務危機; 專家系統; 類神經網路; 模糊邏輯; Financial crisis; Expert system; Neural network; Fuzzy logic; |
語文 | 中文(Chinese) |
中文摘要 | 預測財務危機所使用的方法中,傳統的統計方法說明了自變數與應變數間的線性關係,卻法無法解釋變數間可能存在的非線性關係;專家系統提供了變數間較為詳細的規則,然而其中知識庫的擷取確是令人頭痛的問題;類神經網路雖然可以捕捉自變數與應變數間複雜的對應(mapping)關係,卻無法解釋變數間的因果關係。本研究嘗試結合人工智慧中,模糊邏輯對變數間的解釋能力,以及類神經網路的學習能力,建構一有效的財務危機預警系統。實證結果顯示,除了預測準確度較傳統統計方法為佳外,具學習性之模糊專家系統對即將發生財務危機之公司確能提供更早與漸強的警訊,其所獲取的知識庫也較傳統統計工具能提供更細膩的變數關係。在學術上,此方法對類似的預測問題提供了一可能的解決途徑。 |
英文摘要 | The causes leading to financial crisis can be due to the over-expectation about the market, or due to the inability to lower the production cost, or even due to the wrong decisions made by the enterprise. However, no matter what the real causes are, it should be possible to trace back to see the possibility of a financial crisis before it did happen. If it can be done through an early warning system to find out the probability of a financial crisis before it happened, it would be of great help for investors, managers, and the decision makers in the bank loan to take some appropriate actions. Therefore, a reliable and effective early warning system for financial crisis is a necessity from the practical point of view, and this problem has been a main issue in academia for a long time. The first analysis tool adopted in the literature about the financial crisis prediction was uni-variate discriminant analysis proposed by Beaver (1966), followed by the multi-variate discriminant analysis (MDA) and linear regression models (including the linear probability model, logit, and probit). Although the multi-variate discriminant analysis and linear regression models have become the most commonly used tools in the industry and academia, the statistical assumptions, such as linear relationship among the variables, explanatory variables following normal distributions, and independence among the variables have constrained their applications. Besides, the possible nonlinear relationships among the variables are not considered appropriately by these models. On the other hand, the development of the artificial intelligence has provided a new alternative to solve this problem. An expert system utilizes the IF-THEN rules in the knowledge base to filter the data. Since the rules in the knowledge base specify the relationships among the variables clearly, it is of great benefit for managers to know how the financial crisis happened, and what kind of actions can be done to improve or to avoid the situation. However, the problem is how to obtain the correct knowledge base. Besides, both the importance of each rule in the knowledge base and the importance of each parameter in each rule are all hard to decide. Fuzzy logic was initially proposed by Zadeh (1965). It is a tool that can be used to quantify the fuzzy phenomenon which cannot be expressed by the traditional binary logic. This tool is trying to express the relationship among the variables by using the fuzzy measurement and through the inference process. The basic structure of a fuzzy logic is similar to that of an expert system. Both of these two techniques use the ”IF-THEN” rules to express the relationships among the variables. The main difference between them is that fuzzy logic uses the linguistic variables instead of the numerical variables to do the inference. The consequence is that the obtained inference results can be much closer to the real life under some scenarios. However, the problem associated with fuzzy logic, similar to expert system, is that it is hard to obtain the correct knowledge base, either. Neural Network emulates human being's neuron operation to do the classification and prediction. It had been popular during 1950 through 1960. However, the application of it has been quiet for a while for its incapability of dealing with the nonlinear separation. Not until the back propagation neural network was proposed by Werbos (1974) to solve the nonlinear separation problem, the enthusiasm for neural network was brought up again (McClelland and Rumelhart, 1986). Back propagation is a breakthrough for the learning process in neural network, especially effective for the problem where the mapping among the input and output are complicated. Neural network is trying to find out the relationships among the variables through the learning process based on the historical data and trying to predict the output through the input variables based on the obtained relationships. Basically the expert system, fuzzy logic, and neural network are all a great help for a manager to make decisions. Expert system can embed the experience into the system, fuzzy logic can describe the problem by use of the close to human being's reasoning process and tolerate the inexactness and uncertainty associated with the data set, and neural network has the learning ability from the historical data. However the difficulty to obtain the correct knowledge base for both the expert system and fuzzy logic, and the complicated mapping function between the input and output variables which cannot be explained easily for neural network have constrained the applications of these three methods in management field. Therefore an expert system with learning ability and with the ability to tolerate with the uncertainty and inexactness in the data set will make itself an effective tool to solve problem in management field. The purpose of this research is to propose this hybrid, neuro fuzzy, to construct the financial crisis early warning system. This hybrid combines the explanatory functionality of an expert system, the tolerance of inexactness and uncertainty associated with the data set of a fuzzy logic, and the learning ability of a neural network. In addition to increasing the predicting accuracy and providing early warning signals, this proposed model is aimed to provide a knowledge base which will specify the relationships among the variables more specific, providing the managers more concrete suggestions to prevent the financial crisis. The empirical results show that in addition to providing more accurate prediction results and much lower misclassification cost, the proposed model, neuro fuzzy, can also provide earlier warning signals for financial crisis companies than the other techniques. The strength of the signals increase as the time comes closer to the crisis. Besides, neuro fuzzy can also provide more detailed relationships among the variables than the traditional statistical techniques. In addition to providing a new alternative to improve the accuracy of the financial crisis prediction, this paper is trying to present an alternative to explain the cause and effect relationship among the variables. The relationship among the variables described in the knowledge base is much more complicated than that contained in the traditional statistical techniques, especially that the knowledge base has presented the nonlinear relationships among the variables through the if-then rules. This paper has been a new try in applying neuro fuzzy to solve the financial crisis prediction problems. Although there have been a lot of applications of neural network to solve the managerial problems, the application of neuro fuzzy in management field is still in paucity. We believe that due to the uncertainty and inexactness characteristics associated with the data set in the management field, neuro fuzzy technique can be a much more appropriate tool to be applied in the management field than any other techniques. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。