企業信用風險一直以來都是學術界與業界不斷被討論的話題,中小企業數量占社會總企業數量的90%以上,在社會發展中非常重要。但由於中小企業存在各種缺陷,導致銀行不敢貸款給中小企業,以至于優秀的中小企業會因為無法貸款而倒閉。本研究認為,建立一套適合中小企業信用風險的評估模型,一方面可以有效區別中小企業的優劣,從而協助好的企業獲得經營上的融資挹注,使優秀的中小企業不會因為無法得到貸款而破產,有利於優秀中小企業的健康發展。另一方面,可以提高銀行的收入與風險承受能力,增加其優秀客戶數。 基於上述認知,本文對台灣中小型企業的信用違約風險進行了實證研究,採用XGBoost模型分析影響台灣中小企業之重要因素,建立中小企業財務風險的預警模型。本文有以下兩點發現:第一:本文總供挑選22個財務與非財務變數對中小企業的財務風險進行分析預測,發現在前十大變數排行中,僅有2個為非財務變數,其他均為財務變數,因此台灣中小型企業違約風險預測的關鍵預測因數主要依然是財務變數,但非財務變數(公司治理變數/審計品質變數)的作用一樣不可忽視。第二: XGBoost在保留了決策樹優點的同時,減少了決策樹過度擬合的問題。該方法在信用風險評估準確度上明顯優於決策樹與隨機漫步,且隨著訓練樣本的增加,該模型的優勢愈發明顯。 ;Based on the above cognition, this paper conducts an empirical study on the credit default risk of Taiwan′s OTC-listed SMEs; establishes and contrasts early warning models using XGBoost, Decision Tree and Random Forest. Two main findings unfold. Firstly, classification via XGBoost appears to be significantly superior to Decision Tree and Random Forest model in terms of prediction accuracy and the advantage using XGBoost becomes obvious when training sample increases. Given the inherited features from the Decision Tree, XGBoost further circumvents the common problem of overfitting and is thus worthy of attention for applications in credit assessment for SMEs. Secondly, we identify key predictors for the credit risk prediction of Taiwan′s SMEs. Among a total of 22 employed financial and non-financial variables, although 2 out of the top 10 important variables are found to be governance-related non-financial variables, financial variables remain crucial for SMEs’ credit worthiness.