Feature Selection, Deep Neural Network and Trend Prediction

doi:10.1007/s12204-018-1938-5

Abstract

Abstract: The literature generally agrees that longer-horizon (over a month) predictions make more sense than short-horizon ones. However, it’s an especially challenging task due to the lack of data (in unit of long horizon) and economic data have a low S/N ratio. We hypothesize that the stock trend is largely dictated by driving factors which are filtered by psychological factors and work on behavioral factors: representative indicators from these three aspects would be adequate in trend prediction. We then extend the Stepwise Regression Analysis (SRA) algorithm to constrained SRA (cSRA) to carry out a further feature selection and lag optimization. During modeling stage, we introduce the Deep Neural Network (DNN) model in stock prediction under the suspicion that economic interactions are too complex for shallow networks to capture. Our experiments indeed show that deep structures generally perform better than shallow ones. Instead of comparing to a kitchen sink model, where over-fitting can easily happen with a shortage of data, we turn around and use a model ensemble approach which indirectly demonstrates our proposed method is adequate.

Key words: feature selection| trend prediction| constrained Stepwise Regression Analysis (cSRA)| Deep Neural Network (DNN)

摘要： The literature generally agrees that longer-horizon (over a month) predictions make more sense than short-horizon ones. However, it’s an especially challenging task due to the lack of data (in unit of long horizon) and economic data have a low S/N ratio. We hypothesize that the stock trend is largely dictated by driving factors which are filtered by psychological factors and work on behavioral factors: representative indicators from these three aspects would be adequate in trend prediction. We then extend the Stepwise Regression Analysis (SRA) algorithm to constrained SRA (cSRA) to carry out a further feature selection and lag optimization. During modeling stage, we introduce the Deep Neural Network (DNN) model in stock prediction under the suspicion that economic interactions are too complex for shallow networks to capture. Our experiments indeed show that deep structures generally perform better than shallow ones. Instead of comparing to a kitchen sink model, where over-fitting can easily happen with a shortage of data, we turn around and use a model ensemble approach which indirectly demonstrates our proposed method is adequate.

关键词: feature selection| trend prediction| constrained Stepwise Regression Analysis (cSRA)| Deep Neural Network (DNN)

CLC Number:

O 213

FANG Yan (方艳). Feature Selection, Deep Neural Network and Trend Prediction[J]. Journal of Shanghai Jiao Tong University (Science), 2018, 23(2): 297-307.

References

[1] KANDEL S, STAMBAUGH, R F. Modeling expectedstock returns for long and short horizons [OL]. (2000-10-19) [2016-01-15]. http://ideas.repec.org/plfth/pennfi/42-88.html. [2] CAMPBELL J Y, LO A W C, MACKINLAY A C.The econometrics of financial markets [M]. New Jersey:Princeton University Press, 1997. [3] CAMPBELL J Y, YOGO M. Efficient tests of stockreturn predictability [J]. Journal of Financial Economics,2006, 81(1): 27-60. [4] ANG A, BEKAERT G. Stock return predictability: Isit there? [J]. Review of Financial Studies, 2007, 20(3):651-707. [5] GARDNER E S. Exponential smoothing: The state ofthe art—Part II [J]. International Journal of Forecasting,2006, 22(4): 637-666. [6] FRANSES P H, DIJK D, LUCAS A. Short patchesof outliers, ARCH and volatility modelling [J]. AppliedFinancial Economics, 2004, 14(4): 221-231. [7] BALABAN E, BAYAR A, FAFF R W. Forecastingstock market volatility: Further international evidence[J]. The European Journal of Finance, 2006, 12(2):171-188. [8] ATSALAKIS G S, VALAVANIS K P. Surveying stockmarket forecasting techniques—Part II: Soft computingmethods [J]. Expert Systems with Applications,2009, 36(3): 5932-5941. [9] HINTON G E. Learning multiple layers of representation[J]. Trends in Cognitive Sciences, 2007, 11(10):428-434. [10] BENGIO Y. Learning deep architectures for AI [J].Foundations and Trends? in Machine Learning, 2009,2(1): 1-127. [11] BABA N, SUTO H. Utilization of artificial neural networksand the TD-learning method for constructing intelligentdecision support systems [J]. European Journalof Operational Research, 2000, 122(2): 501-508. [12] BURKHOLDER T J, LIEBER R L. Stepwise regressionis an alternative to splines for fitting noisy data[J]. Journal of Biomechanics, 1996, 29(2): 235-238. [13] CHATFIELD C. What is the ‘best’method of forecasting?[J]. Journal of Applied Statistics, 1988, 15(1):19-38. [14] ZHANG G P. Time series forecasting using a hybridARIMA and neural network model [J]. Neurocomputing,2003, 50: 159-175. [15] HADAVANDI E, SHAVANDI H, GHANBARI A. Integrationof genetic fuzzy systems and artificial neuralnetworks for stock price forecasting [J]. Knowledge-Based Systems, 2010, 23(8): 800-808. [16] WANG J J, WANG J Z, ZHANG Z G, et al. Stockindex forecasting based on a hybrid model [J]. Omega,2012, 40(6): 758-766. [17] LEE T H, YANG W. Granger-causality in quantilesbetween financial markets: Using copula approach [J].International Review of Financial Analysis, 2014, 33:70-78. [18] LYNCH M. The Investment Clock. Special report,2004. [19] LO A W, MAMAYSKY H, WANG J. Foundations oftechnical analysis: Computational algorithms, statisticalinference, and empirical implementation [J]. TheJournal of Finance, 2000, 55(4): 1705-1770. [20] ZORIN A, BORISOV A. Modelling Riga Stock ExchangeIndex using neural networks [C]// Proceedingsof the International Conference Traditions and Innovationsin Sustainable Development of Society. Rezekne,Latvia: Information Technologies, 2002: 312-320.