使用機器學習演算法預測企業財務危機

Using machine learning algorithms to predict financial distress

周永昱(Yung-Yu Chou)

指導教授 ： 石百達

國立臺灣大學/管理學院/財務金融學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202101194

摘要

相較於過去文獻多追求財務危機預測之最佳模型或是最佳變數組合。本研究探討資料預處理對於建立財務危機預測模型之重要性。透過不同定義之產業別來進行樣本配對，產生配對樣本，以建立財務危機預測模型，比較財務危機預測模型的表現。
研究結果顯示，使用較佳的配對樣本進行訓練，可以提升模型的整體表現，且結果於不同的樣本配對比例、分類模型皆穩固。在五個衡量指標：準確率、召回率、精確率、F1-score、ROC AUC中，新產業所訓練出之模型絕大多數都異於且優於舊產業。此外，隨機森林在除精確率外，其餘的衡量指標皆為最佳的模型。

關鍵字

財務危機 ； 機器學習 ； 樣本配對 ； 資料預處理 ； 邏輯斯迴歸 ； 支持向量機 ； 隨機森林 ； K-近鄰演算法

並列摘要

This study which is different from previous literature that concentrated on seeking the best model or variables combinations explores the importance of data preprocessing for the establishment of financial distress forecasting models. Samples are paired through different industrial classification systems to generate matching samples to build different financial distress prediction models and compare the performance of models.
The research results show that using better-paired samples for training can improve the model’s overall performance, and the results are stable in different sample pairing ratios and classification models. Among the five measurement indicators: accuracy, recall, precision, F1-score, and ROC AUC, most of the models trained by the new industrial classification system are different from and better than those of the old one. In addition, except for the accuracy rate, the random forest is the best model for all other metrics.

並列關鍵字

Financial distress ； Machine learning ； Sample Matching ； Data Preprocessing ； Logistic regression ； Support vector machines ； Random forest ； K-Nearest Neighbor

參考文獻

1. Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. The Journal of Finance, 23(4), 589-609.

2. Beaver, W. H. (1966). Financial ratios as predictors of failure. Journal of Accounting Research, 71-111.

3. Boyacioglu, M. A., Kara, Y., Baykan, Ö. K. (2009). Predicting bank financial failures using neural networks, support vector machines and multivariate statistical methods: A comparative analysis in the sample of savings deposit insurance fund (SDIF) transferred banks in Turkey. Expert Systems with Applications, 36(2), 3355-3366.

4. Dong, W., Liao, S. S., Fang, B., Cheng, X., Chen, Z., Fan, W. (2014, June). The Detection of Fraudulent Financial Statements: an Integrated Language Model. In PACIS (p. 383).

5. Dong, W., Liao, S., Zhang, Z. (2018). Leveraging financial social media data for corporate fraud detection. Journal of Management Information Systems, 35(2), 461-487.