考慮景氣趨勢應用粒子群最佳化與支援向量機於財務危機預測

Using Particles Swarm Optimization and Support Vector Machines with the Trends of Business Cycle for Financial Crisis Prediction

陳佳郁(Chia-Yu Chen)

指導教授 ： 謝俊宏

共同指導教授 ： 陳牧言(Mu-Yen Chen)

國立臺中科技大學/資訊學群/資訊工程系碩士班/碩士(2011年)

https://doi.org/10.6826/NUTC.2011.00025

摘要

本論文以2000年至2010年3月的台灣電子產業為樣本，主要原因是該產業為台灣股市成交量最大的類股，也是台灣在國際間競爭力的來源，對於台灣股市而言具有舉足輕重的影響力，樣本集為33家正常公司和33家財務危機公司，共八季的528筆記錄，其中每筆紀錄包括了13個財務比率指標，並將數據進行正規化。本論文分成二大實驗:(1) 未考慮景氣循環之財務比率指標當變數導入粒子群最佳化演算法(Particle Swarm Optimization, PSO)-支援向量機(Support Vector Machines,SVM)、格點搜尋法(Grid search,Grid)-SVM和基因演算法(Genetic Algorithm, GA )-SVM模型並預測準確率且找出最佳的參數;(2) 考慮景氣循環跟財務比率指標導入 PSO-SVM、Grid-SVM和GA-SVM模型並預測準確率間找出最佳的參數。以上兩大實驗均利用不同季資料量和交叉驗證進行實驗，檢驗是否會造成預測準確率的影響。最後，實驗顯示單財務比率指標導入模型會依季資料量增加使準確率增加，加入景氣循環導入模型後，在於短期季資料量準確達到最高，PSO-SVM模型預測總平均準確率95.84%較Grid-SVM模型預測總平均準確率95.58%和GA-SVM模型預測總平均準確率95.67%高，而PSO-SVM模型在15次交叉驗證中達到最高。更加證明PSO-SVM模型在預測企業財務危機是一個很好的方法，可以當成投資人投資前的決策參考。

關鍵字

粒子群最佳化演算法 ； 支援向量機 ； 格點搜尋法 ； 基因演算法

並列摘要

In this thesis, we apply particle swarm optimization (PSO) into support vector machine (SVM) for financial bankruptcy prediction model construction. Besides, due to electronic industry has the most volumes in Taiwan stock market, and also is the source for international competitiveness of Taiwan. Therefore, we adopt Taiwan’s electronics industry as our experiment dataset from Jan. 2000 to March 2010. We collect 33 non-bankruptcy companies and 33 bankruptcy companies as our experiment samples. Besides, the dataset is composed a total of 528 records for eight quarters prior to the occurrence of financial distress, in which each record including 13 financial ratios indicators. The goals for this thesis are divided into two experiments: (1) Using financial ratio index (without considering business cycle) as variable for the Implementation of PSO-SVM, Grid (Grid Search)-SVM and GA (Generic Algorithm)-SVM model to forecast accuracy and to identify optimum parameter; (2) Considering business cycle and financial ratio index to implement PSO-SVM, Grid-SVM and GA-SVM model and forecast accuracy and identify optimum parameter. The above two major experiments used different quarterly data volume and cross-validation to conduct experiments on whether it will affect the accuracy of the forecast. Finally, the experiments indicate that accuracy increased as quarterly data volume increased for single financial ratio index model. After adding business cycle model, the forecast accuracy reaches the highest point on short-term quarterly data volume. The general average of forecast accuracy of PSO-SVM model is 95.84% which is higher than the general average accuracy rate of Grid-SVM model (95.58%) and GA-SVM model (95.67%), and PSO-SVM model reaches highest point on the 15th time cross-validation experiments. It proves that the PSO-SVM model for prediction of corporate financial crisis is a good solution and can also help investors to make the correct investment making.

並列關鍵字

Particle swarm optimization ； Support vector machine ； Grid Search ； Generic Algorithm

參考文獻

22. 許晉雄、鄒慶士(2009)，多準則投資組合最佳化分析：結合K-MOPSO 及TOPSIS，東吳經濟商學學報，第六十五期，pp.109-138。

32. 劉易昌(2004)，支援向量機於財務預測上之應用，靜宜大學資訊管理學系研究所，碩士論文。

34. Altman, E.I.,(1968),”Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy.” Journal of Finance, Vol.23(4),pp.589-609.

35. Andreou, E.,(2001),”Bully/victim problems and their association with coping behaviour in conflictual peer interactions among school-age children.” Educational Psychology Vol.21: pp.59–66.

36. Antonio C. Briza and Prospero C. Naval Jr.,(2008), “Design of stock trading system for historical market data using multiobjective particle swarm optimization of technical indicators,” Proceedings of the 2008 GECCO conference companion on Genetic and evolutionary computation, Atlanta, GA, USA, pp. 1871-1878.