[6, 7], SVM, [7].,, SVM,, SVM,,,,,,, 0.01%,,,,.,,,, SVM,,,, [9], ( ) Soft Confidence-Weighted Learning (SCW)[8],, SCW[8], ( ),,, ( ),,,, , [

Transcription

1 SCW Predicting stock fluctuations using Two-level Mapping and SCW 1 Muhtar Fukuda Faculty of Environmental and Information Studies, Nagoya Sangyo University Abstract: Due to high uncertainty in the stock market, it is difficult to predict the future fluctuations of stock prices even if we use the state-of-the-art techniques of machine learning, such as Deep Learning. However, in some cases with choosing an appropriate machine learning algorithm, feature values and outputs for the prediction, we can have desirable predicted results, especially on short-term stock fluctuations about some market indices. Some initial reliable results have been achieved in our related work, by using Soft Confidence-Weighted (SCW) Leaning, which is one of online learning. In this paper, we propose a predicting method using two-level mapping and SCW. We will focus on feature transformations using the two-level mapping. The first one is based on the mathematical concept of the Singular Value Decomposition (SVD), to get strong convergence and higher accuracy. The second one is to make the predicted Fluctuation Strength (FS) more precisely, in which we use pre-learned outputs and do relearning. 1,,. ICT,,, SNS,,,,,,,, ) fukuda@nagoya-su.ac.jp (DL)[1] Deep Belief Network(DBN)[2] [3] [4],, DL,,,, (feature), Support Vector Machine (SVM)[5], SVM 89

2 [6, 7], SVM, [7].,, SVM,, SVM,,,,,,, 0.01%,,,,.,,,, SVM,,,, [9], ( ) Soft Confidence-Weighted Learning (SCW)[8],, SCW[8], ( ),,, ( ),,,, , [9], 6 1),, [10], ETF,,,,.,,,,., 2),,,, ( ),,,,,,,, SNS ( ), 1,,, 3) ( 1 (c)) ( 1 (b)) ( 1 (a)) 3600 ( 15 ),,. 4),,,, ( 1 (d)) ( 1 (e)) 90

3 5) ( 1 (a)) (feature), ( 1 (f)), 1: 6),,, ( (2)),, SCW, x t R d w R d, ŷ t = sgn(x t w), ŷ t {+1, 1} (1) x t, d, w,, ( ), FS (Fluctuation Strength) FS = x t w (2) w, FS, sgn(fs),,,, [9], DL,, 3, Web,,,,,., [9], ( 1 (f)) ( 1 (a)), 0, ( 1 (d)), 0 (+1), ( 1),,, 3.1, ETF,,,,.,, 91

4 ,., S, T, s S T s, s N s (T T s + N s ), T 0, N 0 Step0 s 0 S,, N s0 =0 Step1 S 1 = {s S T s T 0, N s N 0 }, s S 1, T = 3600, T 0 =0.99T = 3564, N 0 =1,, S Step2 S 1 P 1 %, S 2 s S 1 T, p t,v t,t=1, 2,...,T s AvgLiq(s) ( K ( )) 1 R 1 K+1 AvgLiq(s) = p t v t R k=0 t=1, K 0, 2 K+1 T, R = [ T 2 k ]. Step3 S 2 P 2 %, S 3 s S 2 T p t, t =1, 2,...,T,T +1 s AvgP C(s) ( K ( 1 R )) AvgP C(s) = (p t p t+1 ) 1 R p t+1 k=0 t=1, K 0, 2 K+1 T, R = [ T 2 k ]. K+1 Step4 S 3 {s0 } s 0 s P 3 %, s 0 S(s 0 ) Step s 0, s 0, S(s 0 ) = N, P 1 = P 2 = P 3, P 1 % P 2 % P 3 % S 1 = N P 1,P 2,P 3 N = 320, P 1 = P 2 = P 3 66., 320, s 0 s S(s 0 ) T + L p 0 t,p s t,t=1, 2,...,T,T+1,...,T+L, s 0 s AvgBeta(s 0 s) ( K AvgBeta(s 0 s) = α k β k β k α k = { p0 1 a 0 1 a 0, 1 β k = { ps 1 a s 1 a s, 1 a 0 t = 1 L L i=1 p 0 t+i, k=0 p 0 2 a 0 2 a 0,..., 2 p s 2 a s 2 a s,..., 2 a s t = 1 L ) 1 K+1 p 0 R a0 R a 0 } R p s R as R a s } R L i=1 p s t+i, K 0, 2 K+1 T, R = [ T 2 k ], L, a 0 t as t, t =1, 2,...,T L K(= 7), S(s 0 ), S(s 0 ) 30%, S 1 S (s 0 ) S(s 0 ), S(s 0 ),,. 3.2,,,, RSI (Relative Strength Index) [6], n [11] 92

5 , ( 1 (f)) ( ) t 0 p 0, L =3 2 stride + slack, p t, t =1, 2,..., L t 0 Step1 c t = (pt p0) p 0 100, t =1, 2,...,L Step2 t 1,t 2,t 3 {1, 2,...,L}, t, t {t 1,t 2,t 3 }, t t stride {c t1,c t2,c t3 },, stride Step3 {c t1,c t2,c t3 }, (l 1,l 2,l 3 ), 3 (h 1,h 2,h 3 ). stride =2, slack =2(L = 14), ( 1 (d), (e)) stride slack, s 0 S(s 0 )={s 1,s 2,...,s N },N= S(s 0 ), s S(s 0 ) t αt s =(lt1,l s t2,l s t3,h s s t1,h s t2,h s t3), x 0 t =(αt s1,αt s2,...,α s N t ), t =1, 2,...,T x 0 t, t = 1, 2,...,T X 0,, x 0 t,, 3.3, 2,,,, s 0,, ( 1 (a)) ( 1 (d)) PC(t) (Percentage change), p t t, M, p t 1,p t 2,,p t M, t =1, 2,...,T PC(t) = (a t p t ) 100, a t = 1 p t M, M p (t i) i=1 r 1,r 2,..., r [ T 3 ],..., r [ 2T 3 ],..., r T LC = r [ T 3 ] (Lower Criterion), UC = r [ 2T 3 ] (Upper Criterion) UC LC, { yt U +1 if PC(t) UC = 1 otherwise { yt D +1 if PC(t) LC = 1 otherwise t =1, 2,..., T, Y U =(y U 1,y U 2,...,y U T ), Y D =(y D 1,y D 2,...,y D T ) X = X 0 ( 3.2 ), {X, Y U } {X, Y D },, Y U Y D X Y = Y U ( Y D ),, [9], X 0 = UΣV (3), X = X 0 V, X Y SCW,, x 0 t, x t = x 0 t V, x t,. V SCW, x t R d w R d 93

6 , ŷ t = sgn(x t w), ŷ t {+1, 1} x t x t w 0,, w,,,, 1., 0, 3.3,,,, (2) FS, [9].,,, ( ),,,, SCW w =(w 1,w 2,,w n ), d =(d 1,d 2,,d T ), X 0 Vw = d (4), d f =(f 1,f 2,,f T ), f, d, (2) FS, (5) M X 0 VMw = d f f (5), W = w w, W = U 1 Σ 1 V 1, (3), M =Σ 1 U d f f wv 1 Σ 1 1 U 1 (6), M, x 0 t, t = 1 1, 2,...,T V M, x t 5 x t = x 0 t VM, t=1, 2,...,T (7) 2,, [10], ETF,, 100% 6%,,,.,,, ETF 3% 3.1 ETF, 1,2,, 5.1,, 1, 2, JASADQ, 94

7 3800, 3.1 Step1 1100, Step 320,, 370 Yahoo! [12] ,,,,,, %, ( ), (Yes) ( (Yes) ) ( A ), ( ) ( B ), B A 100 (%),,, ( 1 (d, e)) 3, (6 ), 3600, 1 1, 3%,, ( ),?? Yes Yes (%) : 2, Yahoo! 5.3, ( ) ( ), ( ) ( (2) ),, [9].,,,,,, 2 5.2?? : 2, 6 SCW[8],,, ( ) ( ), ( ) 95

8 1 3 ( ), ( ) 2,,, SCW [8],,, 0, ( ) 0,,,,,,,,,, [9],,, 30,,,, 320,, 2, [1] Hinton, G., Osindero, S. and Teh, Y. W. : A fast learning algorithm for deep belief nets, Neural Computation, Vol. 18, No. 7, pp (2006) [2] Bengio, Y., Lamblin, P., Popovici, D., and Larochelle, H.: Greedy Layer-Wise Training of Deep Networks, Advances in Neural Information Processing Systems 19 (NIPS 2006), accessed February 1, 2017, (2007) [3] Chao, J., Shen, F. and Zhao, J.: Forecasting exchange rate with deep belief networks, Proceedings of IEEE International Joint Conference on Neural Networks (IJCNN), pp (2011) [4] Yeh, S., Wang, C. and Tsai, M.: Corporate Default Prediction via Deep Learning, In The 34th International Symposium on Forecasting (ISF 14), (2014) [5] Boser, B. E., Guyon, I. M., and Vapnik, V. N.: A training algorithm for optimal margin classifiers, Proceedings of the fifth annual workshop on Computational learning theory (COLT 92), pp (1992) [6], :,, KBSE , pp (2011) [7] Shen, S., Jiang, H., and Zhang, T.: Stock market forecasting using machine learning algorithms, CS229 (Machine Learning) at Stanford University, accessed February 1, 2017, Zhang-StockMarketForecastingusingMachine LearningAlgorithms.pdf (2012) [8] Wang, J., Zhao, P., and Hoi, S. C. H.: Exact Soft Confidence-Weighted Learning, Proceedings of the 29th International Conference on Machine Learning (ICML 2012),pp (2012) [9] : SCW,, SIG-FIN , (2016) [10], [11], : Deep Belief Network,, SIG-FIN , (2014) [12] Yahoo!, 96