| 研究生: |
郭泓霆 Kuo, Hung-Ting |
|---|---|
| 論文名稱: |
以循環神經網路模型增進新台幣匯率的短期預測能力 Improving Prediction Performance of Short-term Exchange Rate of Taiwan by Using Recurrent Neural Network |
| 指導教授: | 徐士勛 |
| 口試委員: |
黃裕烈
徐之強 |
| 學位類別: |
碩士
Master |
| 系所名稱: |
社會科學學院 - 經濟學系 Department of Economics |
| 論文出版年: | 2018 |
| 畢業學年度: | 106 |
| 語文別: | 中文 |
| 論文頁數: | 34 |
| 中文關鍵詞: | 匯率預測 、樣本外預測 、神經網路模型 |
| DOI URL: | http://doi.org/10.6814/THE.NCCU.ECONO.010.2018.F06 |
| 相關次數: | 點閱:133 下載:26 |
| 分享至: |
| 查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
以往針對匯率的預測,傳統的計量方法會將資料本身的歷史資訊以線性估計方式建模。但隨著各國外匯交易往來頻繁,影響我國外匯價格的因素日趨複雜,線性估計模型的預測誤差亦不斷擴大,因此本文採用兩種不同的神經網路模型來預測我國外匯價格,參考 Bao et al.(2017) 所提出循環性類神經網路模型,藉由模型非線性估計的方法與自編碼器的降噪方法來達到更好的預測效果。
為進行有效的比較,本文比較傳統的Autoregressive Distributed Lag Model 計量模型與兩種神經網路模型架構類神經網路模型 (Fully Connected Neural Network) 與堆疊式自編碼器 (Stacked Autoencoder) 搭配循環神經網絡 (Recurrent Neural Network),提供有系統的變數選擇,資料預先處理,資料轉換,模型建構,參數調整優化與樣本外預測評估。評估的方法採均方誤差來衡量模型樣本內與樣本外預測的優劣,接著本文分別估計上述三種模型 1 日、7 日、30 日短中期的預測結果並將其與隨機漫步模型比較。
結果顯示神經網路模型於樣本外預測皆優於於隨機漫步模型。另外自編碼器搭配循環神經網路模型以其優異的訊息傳遞與資訊降噪能力,更是在 7 日與 30 日的預測結果上遠優於其他模型。
1 緒論 1
2 台幣兌美元匯率 實證模型建構 3
2.1 Autoregressivedistributedlagmodel ............ 4
2.2 類神經網路模型 ............. 6
2.2.1 Fully Connected Feedforward Network 模型 .... 7
2.2.2 模型訓練與參數估計................... 9
2.3 LongShort-TermMemoryNetworks............... 10
2.3.1 LSTM神經元....................... 12
2.3.2 LSTM模型........................ 14
2.4 StackedAutoencoders ................. 16
2.4.1 單層Autoencoder模型.................. 17
2.4.2 多層Autoencoder模型.................. 18
3 實證方法 19
3.1 資料說明 ............................. 19
3.2 研究方法 ............................. 20
4 實證結果 23
4.1 1日匯率預測結果........................ 24
4.2 7日匯率預測結果........................ 26
4.3 30日匯率預測結果 ...................... 27
5 結論 29
[1] Adya, M., and Collopy, F. (1998). “How effective are neural networks at forecasting and prediction ? A review and evaluation”. Journal of forecasting J. Forecast., 17, 481–495.
[2] Bao, W., Yue, J., and Rao, Y. (2017). “A deep learning framework for financial time series using stacked autoencoders and long-short term-memory”, PLOS ONE, 12(7).
[3] Bengio, Y., Simard, P., and Frasconi, P. (1994). “Learning long-term dependencies with gradient descent is difficult”, Neural Networks, 5(2),157-166
[4] Caire, P., Hatabian, G. and Muller, C. (1992). “Progress in forecast- ing by neural networks”. Neural Networks, 2, 540-545.
[5] Connor, J., Martin, R., and Atlas, L. (1994). “Recurrent neural net- works and robust time series prediction” . Neural Networks, 5(2), 240–254.
[6] Contribution, O. (1989). “On the approximate realization of contin- uous mappings by neural networks”. Neural Networks, 2, 183–192.
[7] Cybenko, G. (1989). “Approximation by superpositions of a sigmoidal function”. Mathematics of Control, Signals, and Systems, 2, 303–314.
[8] Enders, W. (2014). Applied econometric time series, 4th Edition. New York, United States : Wiley.
[9] Granger, C.W.J., and A.P. Andersen. (1978). An introduction to bi- linear time series models (Vandenhoeck and Ruprecht, GSttingen).
[10] Graves, A. (2012). Supervised sequence labeling with recurrent neural networks. Berlin, Germany : Springer.
[11] Hornik, K. (1991). “Approximation capabilities of muitilayer feedfor- ward networks”. Neural Networks, 4(2), 251–257.
[12] Hornik, K. (1993). “Some new results on neural network approxima- tion”. Neural Networks, 6(8), 1069-1072.
[13] Hornik, K. Stinchcombe, M., and White, H. (1989). “Multilayer feed- forward networks are universal approximators”. Neural Networks, 2(5), 359–366.
[14] Kim, T. Y., Oh, K. J., Kim, C., and Do, J. D. (2004). “Artificial neu- ral networks for non-stationary time series”. Neurocomputing, 61(1– 4), 439-447.
[15] Kingma, D. P., and Ba, J. L. (2015). “Adam A method for stochastic optimization”. ICLR, 1-15.
[16] Kuan, C. (2006). Artificial neural networks. IEAS Working Paper : academic research 06-A010, Institute of Economics, Academia Sinica, Taipei, Taiwan.
[17] Lipton, Z. C., Berkowitz, J., and Elkan, C. (2015). A crit- ical review of recurrent neural networks for sequence learning. arXiv.1506.00019[cs.LG]
[18] Refenes A. N. , M. Azema-Barac, L. Chen, and S. A. Karoussos. (1993). “Currency exchange rate prediction and neural network design Strategies”. Neural Comput Applic, 1(1), 46-58
[19] Tong, K., and Lim, K. S. (1980). “Threshold autoregression, limit cycles and cyclical data”. Royal Statistical Society, 42(3), 245-292.
[20] Vincent, P. (2010). Stacked denoising autoencoders : learning useful representations in a deep network with a local denoising criterion, Paper presented at the 27th International Conference on Machine Learning, 11, 3371–3408.
[21] Weigend, A.S., Huberman, B.A. and Rumelhart, D.E., (1992). Pre- dicting sunspots and exchange rates with connectionist networks. In:
M. Casdagli and S. Eubank (Editors), Nonlinear Modelling and Fore- casting, SFI Studies in the Sciences of Complexity, Proc. Vol. XII. Addison-Wesley, Redwood City, pp. 395-432.
[22] Zhang, G., Patuwo, E. B., Hu, M. Y. (1998). “Forecasting with arti- ficial neural networks : The state of the art”. International Journal of Forecasting, 14, 35–62.