Dự báo giá chứng khoán sử dụng công nghệ mạng nơron.

_ K ^ HOi thto ICT.rda'06 Proceedings ofICT.rda'06 Hanoi May 20-21.2006 DU* BAO GIA CHLTNG KHOAN SU* DUNG C6NG NGHE MANG NORON Forecast Stock market using Neural network Nguyen Quang Hoan, Hoang Thj Lan Phovng Tom tdt Bai bao nghien cuu khd nang sit dung mgng noron du bao gid chung khoan, v&i thugt hoc Backpropagation cho vi$c dao tgo mgng Va trinh bay mot sd kit qud thir nghiem cua chucmg trinh mo phong Tiir khda: Mgng Noron nhdn tgo, du bdo chirng khodn Abstract This paper is a research on ability to apply Neural network in forecasting Stock market using Backpropagation for training Finally, paper give out some results of demo program of Forecasting Stock market Keywords: Neural network Forecast stock market GlOfI THIEU Bai toan dir bao gia ca thi truang chung khoan da dugrc nghien cuu va ap dung mpt so thj truemg nhu: NewYork, Tokyo, DownJonc.vdi dp chinh xac khoang 68%-* 90% [3,7,5] Vift nam da co mpt so noi nghien cuu ve d[f bao nhu: Uy ban Chung khoan Nha nuoc Tuy nhien, thj trudng chung khoan Vift nam mdi lap nen dy bao gia chung khoan se duoc quan tam nhifu hon Cac phuong phap thyrc hifn dy bao rat da d^ng, moi phuong phap thich hgrp voi tung bai toan cy the Mpt sd phuang phap dien hinh da dugc thyc hifn de dy bao nhu: chpn de "fit" so lieu ma ta co cic mo hinh dy bao khac Cac ham dy bio tieu bieu nhat la ham tuyen tinh, ham mil va ham Logistic tuang ung vdi cic mo hinh dy bao tuyen tinh, mo hinh dy bio ham md va mo hinh dy bao ham Logistic [10,3] a Mo hinh tuyen tinh La mo hinh don gian nhat vdi ham dupc dung de "fit" la ham tuyen tinh: y=mx+b Theo phuang phap thi hf so m, b se dugc tinh: "d-O' XZ^XZ^') (E.)(E(^1-(S^)(Z^) (1.1) (1.2) 1.1 Phuranig phap ngoai suy Phuang phap ngo^i suy la mpt nhii'ng phucmg phap don gian nhat diing de dy bdo, CO su dyng so lifu thong ke qua khur lam dau vao Cac so lifu qua khu se dupe "fit" theo mpt ham nao dd ho^ie su dyng mang Noron thong minh vdi mpt tryc x la tryc thoi gian, mpt tryc y la cac so lifu qua khur Cac gia trj tuong lai se dugrc dy bio bing each tinh gii trj ciia ham tai cic thoi diem tuong lai Tuy theo ham duac lya b Mo hinh hdm mH Mo hinh ham mu bien de tinh cic dai lugng tang truong nhu dan so Phuang trinh cua ham mu dugc hifu dien: y=b*m^x Lay hai ve ta se co: Ln(y)=x*ln(m)+ln(b) (1.3) Khi dimg phuong phip binh phuang toi thifu de "fit " ham tren vdi s6 lifu qua khu di Ky y^u HQi thto ICT.rda'06 Proceedings of ICT.rda'06 Hanoi dugc bien doi tuong umg, ta tim dugc cac hf so a=ln(m) va c=ln(b), m la cic hf so cho vifc "fit" ham tuyen tinh dS neu tren Khi ffi va sg dugc tinh: m—e vam=e (1.4) c Mo hinh mgng Noron La mo hinh co kha ning "hpc" tu cac du lifu qui khur, co thf cap nhat cac tham so Mo hinh m^ng Noron cd thf sir dyng ca ham tuyen tinh va phi tuyen cho dao tao mang Neu lya chgn dugc cac tham so tdi uu thi la mo hinh xap xi rat tot ducmg cong djch chuyen ciia doi tugng can dy bao Cy the ve nmng Noron s6 dugc trinh bay t^i phan khoin thi vifc tfnh toin hay tri thuc ciia thj trudng li khd chung ta cung chua hieu day dii va cac qui luat ciia chiing C chuyen gia chi thyc sy tot tri thurc cua chiing va khong thyc thdng cd loi va thong tin khong d Trong cic phuang phip tren cd uu diem la: dieu khien dQ I hon, tong hgp va thyc hifn "Cac cd dao t?o" Vi vay, mang noro cho moi trudng thi trudng chung k Bii bao thyc hifn nghien i nghifm mpt phuang phip dy bi mang Noron vdi thuat hgc Backprc iTNG DVNG MANG NORON < BAO GIA CHUivG KHOAN 1.2 Mo hinh kinh te lirgrng Mo hinh kinh te lugng la mo hinh vdi nhif u bien mo ta sy phy thugc cua cic d^i lugng can dy bio tren co sd cac thong so kinh tf xa hOi Mo hinh kinh te lugng dugc hifu dien bdi ham sau: Tap du* lifu gii ca chiirng khoii thap tren mang Internet tai vahoo.charts.com vdi hon 2000 n chung khoin tir cic nam 1996 den r DH lifu sau dugc chuin hoa se vio cho m?ing Noron thyc hifn "1 y=a,*x,+a2*X2+ +a„*x„+b (1.5) Sau dao t^o mang dugc sur dyng < Trong y li d^i lugng can dy bio; xj, X2 Gii chiirng khoin d thdi diem tiep the X3 x„ la cic thdng so kinh ti xi hgi lien quan Cic tham s6 aj.a2 a„ xic dinh sy 2.1 Xfiy dvng cau triic mang phy thugc cua d^i lugng dy bio vio cic thong Cau true m?ng BackPropagation s6kinhte-x3hgi[10] cho bai toan li m^ng ldp: mgt ldp > ldp an va mgt Idp Tuy nhien vifc j 1.3 D^ bio theo hf thong chuyen gia mgt clu true toi uu phy thugc vi Md hinh chuyen gia dya tren dinh gii ciia n^if m ciia nha phit trien, va dya tre chuyen gia llnh vyc dy bio Cic so lifu thir nghifm [1] dy bio cua cic chuyen gia dua dugc xem xet, dinh gii va tong hgp de dira ket qui dy bio cuoi cimg [3] Hf thong chuyen gia siir dyng cac lufit cua chuyen gia, va dua moi quan hf vdi cac mo hinh toin hgc Mo hinh chuyfn gia co thf phat trien ket hgp vdi mo hinh: Phan tich ket hgp {Conjoint Analysis), Ty moi {Bootstrapping) va Kinh te lugng {Econometric) [10] Uu diem ciia hf chuyen gia la siir dyng lu^t suy dien de dua kft qua Khi ip dyng hf thong chuyen gia cho thj trudng chumg 158 IBnhl: Mgng backpropagation idp HOi thto ICT.rda'06 Proceedings of ICT.rda'06 Hanoi May 20-21.2006 Hinh mo ta cau triic mang backpropagation ldp Cd n noron tai idp vio, p noron ldp an, m noron ldp ra, Vj, la A trpng so gitla ldp vao va ldp an, wjk la trgng so ^Wjk giOa ldp an va ldp ( ^ ^ , 0^^ va do: X la tip vector dau vao (x/, ,x„); / la t|p vector dich tai dau {t/ t„); Xj la gia trj tai noron dau vao thu i; Zjla gii trj cua noron diu thir k; tk la gii trj dich cua noron dau thii' k Dau vao ciia noron ldp an thiiry: n (2.1) zin /=o // la ham truyen t^i Idp an, dd gia trj ciia noron thury cua ldp an la: z/=fj{zinj) Dau vao cua Noron ldp thu VOI: _ ^ ^ a la hang sd hpc (2.5) - ~ " T ~ , a Hieu chinh trpng so cho lap Wjk - Ta co: "• ' ^ *=o (2.6) dit "*='*">'* dE _ dE dUk ta cd: (2.7) dE - Tinh : ta cd vdi gii tri M, va Uj 8u^ (/>y)thi: — - ^ , va tir(2.6) nen: p *:>"ãô* = Z ^7^ J'' (2.2) 7=0 f2 li ham truyen t^i ldp ra, gii trj ciia noron thir k cua ldp la: yk-f2{yink)- Cic him truyen sir dung bai toin la ham sigmoid cho ldp an va tuyen tinh cho Idp ra: ^'^1 + e-^ (2.3) f(x)=x (2.4) ^ 5M^ - Tinh 5W: ^ j k ^ j k 2.2 Thuat toan huan luyen lan truyen ngugc (Backpropagation) Ggi E li sai so binh phuang giira dau vi dich ciia mang dang huan luyf n Ta CO dg bien doi ciia trpng so Wjk dugc (2.8) n^ * ^ j k ^ j k (2.9) t, la gii trj dpc lap so vdi trpng so nen:—*-=0 dw^v (2.10) dE (mpi k=1.2 m) nen: tinh theo cong thuc p Nhu viy t^i mdi md hinh huan luyf n: ^Jk ^ik (2.11) nhung ta cd: y, = fiiy'^k) • Do dd: Ky y^u HQi thto ICT.rda'06 Proceedings of ICT.rda'06 Hanoi May Aw ^jk ^jk ^«* ^jk ' vdi =-^— ykH%,rn y/ = l,2,„/? ^^•^^^ (; - Tinh dgo hdm f^ix) p Ta lai cd: yin, =Y,^j^iy '^'"°"^ chuomg trinh/^ la ham tu '-' ' ' y=fii.x)=x (2.13) Nen: N dyin, ^^y^^' d ^ Vay: &Wjk ^ * flW-^ ^^^ ^'^' ' (2.14) • • Nhdn xet: Cic trpng so m?ng Noron dpc lap dw., riifj-=jork=knhau, nghia li: g^.^ [Oif j V yor k ^ k' a& ^ ^, ^^jk = ^ ^ = ^ * ^ y =«(^* ->^^ ^^ V^-l,2„m ^^^^^ y/=l,2„,/j C^/; n/r^f /ro/t^ so cAo ldp an _ ' " \ u ' Ta CO cong thuc: (2.15) ^ ^ Suy ra: Tir cdng thiire (2.6),(2.8),(2.12),(2.15) Ta ^°"- dE ^ = -d^Zy vdi: (2.17) t=i c\ii *=i c^j (2.16) Sk={tk-yk)f',{yink) (2-22) nhung y, = f2{yin,) nen: ^=_y« ^20^) ^_y.„ M ^ ) ^ = _ y j a^, tr* a.^ ^ t r ' d)in, ^ d^, tr* a.^ ^'-^^ theo cong thirc (2.14): ym, = 2^Zj^-^^, JL± = ^^ _± ma v,y va Wjk la bien dpc lap nen: (2.24) i