資料載入中.....
|
請使用永久網址來引用或連結此文件:
http://ir.lib.ncu.edu.tw/handle/987654321/96309
|
| 題名: | 財務困境預測：綜合收入區分、比率拆分、增強模型結構與可解釋性;Financial Distress Prediction: Comprehensive Income Discrimination, Ratio Splitting, and Enhanced Model Structures with Interpretability |
| 作者: | 米右法;Rahmi, Asyrofa |
| 貢獻者: | 資訊工程學系 |
| 關鍵詞: | 財務困境預測;綜合收入;長期和短期屬性;串行組合方法;模型解釋;distress prediction;comprehensive income;long-term and short-term attributes;serial combination approach;model interpretation |
| 日期: | 2025-01-23 |
| 上傳時間: | 2025-04-09 17:39:22 (UTC+8) |
| 出版者: | 國立中央大學 |
| 摘要: | 本研究探討各種財務特徵對破產和財務困境的預測能力、一種新型財務困境模型及其可解釋性。首先，本研究將綜合收益及其組成部分，包括權責發生制會計下的應計項目和現金流量，納入破產預測模型。 鑑於綜合收益比傳統淨利更全面地反映公司的微觀和宏觀經濟風險，本研究將檢驗其與財務比率 (FR) 結合預測破產的鑑別力。其次，先前財務困境預測 (FDP) 的研究主要依賴財務比率，而未明確區分這些比率中的長期 (LT) 和短期 (ST) 屬性。 本研究引入一種新型模型，明確區分嵌入財務比率中的長期和短期會計屬性，以解決此限制。第三，先前使用堆疊技術的財務困境研究表明，不同的屬性發揮不同的作用。 這促使我們將屬性應用於不同的基礎模型並依序使用，因為序列組合方法可提高預測性能。 因此，我們提出了一種新型模型，即優化結構和分類器的修正序列組合 (MOSCS)，利用長期和短期財務比率、反映財務操縱的 Beneish 分數以及提供股價資訊的市場指標來改善財務困境預測。預測財務困境時對綜合收益及其組成部分的利用有限，可歸因於其預測能力的證據不足以及在不同地理環境中的應用有限。第四，我們調查這些模型的預測性能並提供對其可解釋性的見解。
利用美國破產和台灣財務困境數據，我們的發現證明了幾個關鍵貢獻。 首先，納入綜合損益及其組成部分顯著提高了破產預測準確性1.5%，並降低了錯誤分類成本——與僅使用傳統FR的模型相比，正確識別了額外的七家公司。 其次，區分FR中的LT和ST屬性顯著增強了財務困境預測，將準確性提高了1.5%，並將第二類錯誤降低了2.5%。 第三，所提出的MOSCS模型在預測財務困境方面優於先前的堆疊模型，達到74.56%的準確度，而先前模型為73.91%，且錯誤分類成本更低（92.10 vs. 94.44）。 此外，納入市場指標進一步提高了MOSCS的預測性能，達到75.02%的準確度，錯誤分類成本為90.42。
最後，本研究為財務利益相關者提供了寶貴的見解。 我們的分析表明，負OCI組成部分在綜合損益框架內對預測破產起著至關重要的作用。 對於財務困境預測，該模型證明了關注LT預測的“灰色區域”的重要性，並強調了謹慎優化模型結構以最大限度地減少財務困境公司錯誤分類的必要性。 這些發現有助於更深入地了解財務困境和破產預測，為財務決策提供寶貴的工具。
;This study investigates the predictive power of various financial features for bankruptcy and distress, a novel model for distress, and their interpretability. Firstly, this study incorporates comprehensive income and its components, including accruals and cash flows, to predict bankruptcy. Recognizing that comprehensive income more comprehensively captures a firm′s micro- and macro-economic risks than traditional net income, this study examines its discriminatory power in conjunction with financial ratios (FR) for predicting bankruptcy. Secondly, prior research in financial distress prediction (FDP) has primarily relied on FR without explicitly differentiating between long-term (LT) and short-term (ST) attributes within these ratios. This study introduces a novel model that explicitly distinguishes between LT and ST accounting attributes embedded within FR to address this limitation. Thirdly, a prior distress study employing stacking techniques suggests that distinct attributes play differential roles. This motivates us to apply the feature attributes on different base models and employ them sequentially, as the serial combination approach offers to improve the predictive performance. Therefore, we propose a novel model, Modified Serial Combination with Optimized Structure and Classifiers (MOSCS), leveraging LT and ST FR, Beneish scores reflecting financial manipulation, and market-based indicators offering stock price information to improve distress prediction. The limited utilization of comprehensive income and its components in predicting financial distress can be attributed to a perceived lack of evidence regarding its predictive power and limited applicability across diverse geographic contexts. Fourth, we investigate the predictive performance of these models and provide insights into their interpretability.
This study demonstrates several key contributions using US bankruptcy and Taiwanese distress data. Firstly, incorporating comprehensive income and its components can significantly improve bankruptcy prediction accuracy by 1.5% and reduce misclassification costs ─ correctly identifying an additional seven companies compared to a traditional FR model. Secondly, distinguishing between LT and ST attributes within FR significantly enhances distress prediction, improving accuracy by 1.5% and reducing Type II errors by 2.5%. Thirdly, the proposed MOSCS model outperforms a previous stacking model in predicting distress, achieving an accuracy of 74.56% compared to 73.91% and lower misclassification costs (92.10 vs. 94.44). Furthermore, including market-based features further improves the predictive performance of MOSCS, reaching an accuracy of 75.02% with a misclassification cost of 90.42.
Finally, this study provides valuable insights for financial stakeholders. Our analysis reveals that negative (harmful) OCI components are crucial in predicting bankruptcy within the comprehensive income framework. For distress prediction, the model demonstrates the importance of focusing on the ‘gray area’’ of LT predictions and highlights the need for careful model structure optimization to minimize the misclassification of distress companies. These findings contribute to a deeper understanding of financial distress and bankruptcy prediction, offering valuable tools for financial decision-making. |
| 顯示於類別: | [資訊工程研究所] 博碩士論文
|
文件中的檔案:
| 檔案 |
描述 |
大小 | 格式 | 瀏覽次數 |
|---|
| index.html | | 0Kb | HTML | 25 | 檢視/開啟 |
|
在NCUIR中所有的資料項目都受到原著作權保護.
|
