企業財務危機預警模型：整合式Logit模型

Abstract

本文針對台灣企業，以提升危機預測能力為目的，使用引入五類變數的二元Logit模型建立財務危機預警模型。我們整合傳統上由徵信人員同時考量眾多資訊的專家系統的精神，以及國內學者研究發現在台灣市場預測效果較佳的Logit模型，彙集過去研究發現具顯著影響危機預測能力的各種變數，包括：會計變數、Merton的KMV模型中代表市場變數的違約距離(DD)、公司治理變數、會計師變數，以及總體經濟變數，設計出一個同時符合傳統專家系統徵信考慮眾多因素的概念，又符合計量方法及客觀性的整合模型。我們檢驗未來一年企業發生財務危機的預測準確性，並觀察陸續引入各類變數時預測能力的變化，並再次驗證過去研究認為具影響預警能力的變數的顯著性。 實證結果發現，同時包含五種變數的整合模型，在in-sample預測總準確率為90.9%，與會計模型的90.0%相比略為提升；在out-sample的預測總準確率為95.6%，與會計模型93.4%相比小幅提升，尤其在危機公司預測的準確率由65.9%大幅提升至84.4%。結論發現，整合模型的五類變數在危機預測皆有顯著影響，其中會計變數是其中解釋能力最強的變數，市場變數次之，其餘變數的解釋能力則相對較弱。在預測能力方面，僅使用會計變數的預測準確率即可達到90%水準，顯示會計變數仍是最重要的變數，但在加入其他變數後仍可使模型解釋力更強並增加預測力，其中又以加入市場變數與加入總體經濟變數時的影響較大。

This paper aims to build a more predictable financial early warning system for Taiwan companies with Binary Logit model. We integrate traditional Expert system and Logit model, using five categories of variables including accounting variables, market variables, corporate governance variables, accountant variables, and macroeconomics variables, to build an integrating Logit model which conforms to expert system’s spirit that uses lots of information and does not lack in objectivity and theoretical background. We examine the one year later predict accuracy of corporate financial distress, and the changes in predict ability when continually adding each variable. We also examine the prior studies variables’ significance of affecting the predict ability.

We find that after considering the five categories of variables, the expert system integrating model’s in-sample predict accuracy is 90.9%, slightly higher than accounting model’s 90.0%; and out-sample predict accuracy is 95.6%, also higher than accounting model’s 93.4%, especially the predict accuracy in financial distress companies growing from 65.9% to 84.4%. We conclude that all the five categories of variables have significant effect to financial distress prediction, and accounting variables have the best explanatory power, market variables are the second best one, but the other variables’ explanatory power are much weaker. In predict ability, we can get 90% predict accuracy when only using accounting variables, showing that accounting variables are still the most important factor. We also can improve the predict accuracy and explanatory power when adding other variables, and the market variables and macroeconomic variables especially have the higher contribution.