透過您的圖書館登入
IP:3.22.71.188
  • 學位論文

台灣航運股股價與總體經濟的關聯性-類神經網路模型之應用

Application of Artificial Neural Network Forecasting Model of Macroeconomic Variables on Taiwan Transportation Stock Returns

指導教授 : 古永嘉
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


本研究以倒傳遞類神經網路應用於總體經濟指標與台灣航運類股股價報酬率的關聯性分析,研究期間為2000年至2010年9月之季資料,包含18家上市櫃航運股票報酬率及銷售淨額成長率和另外16種總體經濟指標,共2,236筆資料。 先以逐步迴歸法篩選最具有解釋能力的變數組合,再以類神經網路及基因演算法做檢測,採用外部效度的預測命中率作為績效衡量的方法。另外,用移動窗格法分析訓練期間逐次向後移動,檢視在期間長度條件固定不變時,變數資料的增減是否影響模型預測效果。最後,依最佳預測結果建構2種投資策略,藉以觀察報酬率是否如預期增加。 實證結果顯示: 一、 所選取的總體經濟指數在落後1-4期的情況下與航運股的股價報酬率大多具有顯著的影響。 二、 訓練基期在最佳狀況時,航運類股整體的平均預測命中率最高達60.80%。 三、 如以個別公司來分析,有四家公司之預測命中率超越整體平均命中率,達到72.22%。 四、 根據最佳預測模型所展開的投資策略模擬,以策略一的作多方式投資報酬率最好。在長達18個測試期中,個股最佳的平均年化報酬率高達155%,而且絕大多數的個股報酬率均優於同時期的大盤表現。

並列摘要


The study is using macroeconomic variables to forecast Taiwan transportation stocks returns through the application of back-propagation network approach. The period of study span is from the beginning of 2000 to September of 2010 with quarterly information. In total of 2,236 data, the model contains stock returns and net sales growth ratios of 18 listed transportation companies as well as other 16 microeconomics indices. Start with stepwise regression to select the most related variables combination. Following with artificial neural computing and genetic algorithm for testing and calculation, the predicted hit ratio (PHIT) of external validity is the tool for performance measurement. In addition, moving windows method is adopted to examine the forecast result when the training periods are tested and then moving forward to next period for testing. Finally, two investment strategies are built up based on the best sequence of prediction and see whether the approach can lead to increment of yield rate. The empirical study shows: 1. The result reveals a significant effect on the transportation stock returns by lagging one to four periods from most of the selected macroeconomic variables. 2. Under best scenario of training period, the average PHIT among all of the transportation stocks is up to 60.80% 3. If go down to company level, some of the companies’ PHIT are better than average result. There are 4 companies’ PHIT reach to 72.22%. 4. According to the best model result, we can develop a simulation through the investment strategies. The outcome shows the long position of strategy 1 defeats the other. In the testing period of 18 quarters, the best company shows an outstanding performance in return of 155%. In addition, most of the individual stock returns apparently beat the result of TAIEX in the same period.

參考文獻


1. 古永嘉、黃鐘億,「以改良式倒傳遞類神經網路運用成分股與技術指標預測台灣50指數報酬率之研究」,中華管理學報,第九卷第一期,PP. 37-56,2008。
8. 陳適宜,「基因類神經網路在臺股指數期貨的預測與蝶式交易策略研究」,國立台北大學企業管理學系碩士論文,2009。
12. 謝文馨,「總體經濟變數與股價指數之關聯性研究-以台灣為例」,國立成功大學企業管理學系碩博士班碩士論文,2007。
18. Kamijo and Tanigawa, "Stock Price Pattern Recognition-a Recurrent Neural Network Approach ", IEEE New York, 1990.
19. Kohonen, T., "An Introduction to Neural Computing", Neural Networks, 1(1), 1988, PP. 3-16.

被引用紀錄


吳俊德(2015)。航運類股價指數與總體經濟變數關係之研究〔碩士論文,國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU.2015.10040
林伯修(2013)。總體經濟領先指標對電信類股價長短期之影響效果–時間數列轉換函數模型之應用〔碩士論文,國立臺北大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0023-0307201313104000
左家文(2014)。以序列採礦方法探討景氣指標與進出口值的關聯〔碩士論文,國立中央大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0031-0412201512005955

延伸閱讀