Finansal Piyasalarda Hisse Fiyatlarının Derin Öğrenme ve Yapay Sinir Ağı Yöntemleri ile Tahmin Edilmesi; S&P 500 Endeksi Örneği

Gelişen teknolojiler sayesinde günümüzde bilgisayarların gücü artmış ve bununla birlikte farklı amaçlara hizmet eden birçok algoritma geliştirilmiştir. Bu algoritmalar birçok alanda olduğu gibi finans alanında da sıkça kullanılmakta ve karar vermenin farklı boyutlarında destekleyici bir rol üstlenmektedir. Özellikle ortaya çıkması muhtemel durumların önceden öngörülmesinin hayati önem taşıdığı borsa işlemlerinde tahmin yöntemlerine sıkça başvurulmaktadır. Bu çalışmada Yahoo Finans üzerinden elde edilen S&P 500 endeksine ait veriler kullanılarak derin öğrenme ve sığ öğrenme yöntemleri yardımıyla geleceğe yönelik fiyat tahminleme çalışması gerçekleştirilmiştir. Bu bağlamda 12.08.2000 ile 13.8.2020 tarihleri arasındaki günlük fiyat verileri ilk 19 sene (veri setinin %95’i)eğitim, son 1 sene (ver setinin %5’i) test olacak şekilde ayrılarak uzun kısa süreli bellek (LSTM) ve çok katmanlı algılayıcılar (MLP) yöntemleri tahmin gerçekleştirilmiştir. Eğitim, test ve tüm veri kök ortalama karesel hatalarının LSTM ağı için sırasıyla 17.3, 65.3 ve 22 dolar, MLP ağı için sırası ile 16.1,61.2 ve 20.6 dolar bulunmuştur. Bu da kullanılan her iki yöntemde elde edilen eğitim ve test hatalarının birbirine yakın sonuçlar verdiğini ve bu yöntemlerin tahmin çalışmaları için uygun seçenekler olduğunu göstermektedir.

Deep Learning and Artificial Neural Network Estimation of Stock Prices in Financial Markets; S&P 500 Index Application

Thanks to the developing technologies, the power of computers has increased and many algorithms have been developed that serve different purposes. These algorithms are frequently used in finance as well as in many other fields and play a supportive role in different dimensions of decision-making. Forecasting methods are frequently used especially in stock exchange transactions where predicting possible situations in advance is vital. In this study, the future price estimation study was carried out with the help of deep learning and shallow learning methods using the data of the S&P 500 index obtained through Yahoo Finance. In this context, the daily price data between 12.08.2000 and 13.8.2020 is divided into the first 19 years (95% of the data set) training, the last 1 year (5% of the data set) as a test, and long short term memory (LSTM) and multilayer perceptron (MLP) methods have been estimated. Training, test, and all data root mean square errors were 17.3, 65.3 and 22 dollars for the LSTM network, 16.1,61.2 and 20.6 dollars for the MLP network, respectively. This shows that the training and test errors obtained in both methods used give similar results and these methods are suitable options for prediction studies.

___

  • [1] X. Li, Y. Li, X.Y. Liu, C.D. Wang, "Risk Management via Anomaly Circumvent: Mnemonic Deep Learning for Midterm Stock Prediction," arXiv, 2019
  • [2] S. Kim, M. Kang, "Financial series prediction using Attention LSTM," Arvix, 2019
  • [3] S.S. Namini, A.S Namin, “Forecasting economics and financial time series: ARIMA vs. LSTM,” ArXiv, vol. abs/1803.06386, pp. n.page, 2018
  • [4] M. Roondiwala, H. Patel, S. Varma, “Predicting Stock Prices Using LSTM,” International Journal of Science and Research (IJSR), vol. 6, pp. 1754-1756, 2017
  • [5] A. H. Manurung, W. Budiharto, H. Prabowo,” Algorithm and Modeling of Stock Prices Forecasting Based on Long Short-Term Memory (LSTM),” International Journal of Innovative Computing Information and Control (ICIC), vol. 12, no.12, pp. 1277-1283, 2018
  • [6] Q.Zhuge, L. Xu, G.Zhang,” LSTM neural network with emotional analysis for prediction of stock price,” Engineering Letters, vol. 25, pp. 167-175, 2017
  • [7] T. Fischer, C. Krauss, "Deep learning with long short-term memory networks for financial market predictions", European Journal of Operational Research, v.270(2), ss. 654-669, 2018
  • [8] K. Bayındır, "Financial Time Series Prediction With LSTM Recurrent Neural Networks," Yüksek lisans tezi, Bilgisayar mühendisliği bilimi, Fen Bilimleri Enstitüsü, Bahçeşehir Üniversitesi, İstanbul, 2017
  • [9] Deng L., An overview of deep-structured learning for information processing, presented at the APSIPA ASC 2011 Xi’an., Xi’an, China ,October 18-21,2011,98052
  • [10] A.Serwa,” Studying the Effect of Activation Function on Classification Accuracy Using Deep Artificial Neural Networks,” Journal of Remote Sensing & GIS,vol. 06, 2017
  • [11] Accessed date:18 September 2020.[Online].Available: https://missinglink.ai/guides/neural-network-concepts/perceptrons-and-multi-layer-perceptrons-the-artificial-neuron-at-the-core-of-deep-learning/.
  • [12] S. Hochreiter, J. Schmidhuber, “Long short-term memory,” Neural computation, vol. 9, no.8, pp. 1735-1780, 1997.
  • [13] Accessed date: 26 June 2020.[Online].Available: http://colah.github.io/posts/2015-08- Understanding-LSTMs/
  • [14] Yan, S., Understanding LSTM and its diagrams., https://medium.com/mlreview/understanding-lstm-and-its-diagrams-37e2f46f1714, (accessed Sept.1,2020)
  • [15] X. H. Le, H. V .Ho, G. Lee, S.Jung, "Application of long short-term memory (LSTM) neural network for flood forecasting", Water, vol.11, ss. 1387, 2019
  • [16] K. A. Althelaya, E.El-Alfy, S. Mohammed, "Evaluation of bidirectional lstm for short-and long-term stock market prediction," 9th international conference on information and communication systems (ICICS), ss.151-156, 2018
  • [17] J. Cao, Z. Li, J. Li, " Financial time series forecasting model based on CEEMDAN and LSTM," Physica A: Statistical Mechanics and its Applications, vol. 519, 2018
  • [18] D. K. Kılıç, Ö. Uğur, "Multiresolution analysis of S&P500 time series. Annals of Operations Research, Annals of Operations Research, ss.197-216, 2018