untitled

Size: px

Start display at page:

Download "untitled"

舒戈
7 years ago
Views:

1 理論 Graduate Institute of Information Management College of Business Chinese Culture University Master Thesis 類 Apply Integrated Classification Model to Taiwan Stock Trend Prediction Yu-Chun Liao Advisor: Chein-Shung Hwang,PhD 99 年 6 JUNE, 2010

Integrated Classification Model to Taiwan Stock Trend

2 路來路路來不論精論不論論論利理晴精見論更歷念勵論利歷識李識林勵不了省力更老女勵 i

3 論類數 60 ( ) 立理 98 年度 2 論論利率理? 金不? 累理理識更理. 了臨利聯立類 SVM 類 KNN 類類切利類率降不類利類更利略類略易數 279 易數類略易數 87 易 27.6 數不類類類 ii

4 Apply Integrated Classification Model to Taiwan Stock Trend Prediction Student: Yu-Chun Liao Advisor: Prof.Chein-Shung Hwang Chinese Culture University ABSTRACT In the era of low interest rate, how does one learn to invest and manage the wealth? How to retain the capital on hands and not let it depreciate through inflation? Effective investment suddenly becomes the subject that everyone cares about. Adage like The accumulation of wealth starts from investment and wealth management is well said under backdrop of high conscientiousness for wealth management nowadays. Since investment merchandises have considerably been diversified and stock market naturally becomes one of the most popular targets for investment monies. Nonetheless, how to hedge the risk so as to locate the best timing for targeting your investment, to acquire high investment returns and low risk exposures as the outcomes for wealth management, would be the absolute necessities for all investor studies. Therefore, investors not only select the stocks of their own interest as the stock investment targets, but also will jointly face the dilemma of when for the actual transactions to take place. Hence, all kinds of technical analysis methods for stock market are widely researched and applied as well. This research employs a variety of technical indicators in the sense of complementing each other. In addition, this research also intends to locate the relevancies between technical indicators and the market trends, and predict the timing for adequate investment between the relationships from a variety of these indicators. During research, Bayesian, SVM, KNN classifications were established accordingly. And these three classification algorithm modules were used to predict the validity of timing for the buy-transaction in the stock market. Lastly, the composite multiple classification module will be used to predict the time points for buy and sell transactions so as to facilitate the investors in increasing the investment returns in addition to lower the unnecessary risks. iii

How to retain the capital on hands and not let it depreciate through inflation? Effective investment suddenly becomes the subject that everyone cares about.

5 This research has been proven through lab experiments that, in the performance for integrated classification model used in the research, the average return ratio for integrated prediction for the market index is better than that from using the single classifier approach. Within the investment strategies selected by voting method, for those aggressive investment strategies that invest during identical large categories, the respective trading lasts 279 days with points gain of the market index during each transaction. And those conservative types of investment strategy lasting 87 days which would trade during identical small categories, they managed with 27.6 points gain of the market index. We can see that all of these can effectively facilitate the investors in making the right decisions so as to prevent from exposures to unnecessary investment risks. Keyword Technical analysis, stock, KNN classification, SVM classification, Bayesian classification iv

Within the investment strategies selected by voting method, for those aggressive investment strategies that invest during identical large categories, the respective trading lasts 279 days with 12.

6 錄錄錄錄論論類類料練論來論來參 v

7 錄料理料理料理廉料理料理類料數類類類類類類類略類略 vi

8 錄論例數滑異廉類量類略類 vii

9 論 Benjamin Graham(1973) 量了來累理累降力年來理識易金理理不金理利都理降力力累不理理不累降? 力理不來理樂累更不來不理來力金略不不 1

10 若來度利率理? 金不? 理識更理了臨行年利率不降利率裡金不? 了行理臨利易略行臨不數 2

11 異不都狀不都不狀理不理行理利不理年來若不數若異不切論利 MACD RSI OBV 量 KD 3

12 W%R 廉 K 量立利類歷料來立更利識切率降不利立類 (Bayes 類 ) 量 (SVM 類 ) K (KNN 類 ) 類利類切都率降不類歷料來立行聯例來歷料量年料來來 4

13 類 (Bayes 類 ) 量 (SVM 類 ) K (KNN 類 ) 類論類利類類不 MACD RSI OBV 量 KD W%R 廉 K 量易料 88 年年 4 30 易 5

14 論 (MACD RSI KD W%R 廉 OBV 量 K 量 ) 立 3 類 (Bayes 類 SVM 類 KNN 類 ) 利金行來論 1-1 說類行論論 6

15 論說流理流行略了流數參數說略行略論論論了理參 7

16 來理論列參金龍 (1998 ) - 來說來說力力力力不力度見金行行 8

17 例易不力力行不數離不若不不不落 9

18 金金金數行量數數數不金數金數値不列數不論金降例金復金益數量金數量金不數量 10

19 來理論都料來來 KD OBT 廉 MACD RSI 都數說 : 歷易易不歷便來 (Edward Magee, 1966) 理歷行不論歷數理理不切不 11

20 說利量料來來來來都料來來 KD MACD RSI 都數說歷易易不列 1. 歷 2. 易 3. 念 4. 理量不不利類來率率 12

21 7 (MACD RSI KD W%R 廉 OBV 量 K 量 ) 理 KD KD (Stochastic Line) George C Lane 1957 年 1986 年 KD Stochastic Line 念量 KD KD 裡 KD 數行 61% KD KD KD 立理 (1998) KD MACD RSI WMS% R 論 KD 廉良 13

22 ( )KD 說 RSV(Raw Stochastic Value) RSV ( C ( H { L ) *100 L }) n { n} = n} { n C n n L {n} n H {n} n C n L {n} 度 H {n} L {n} 聂 n 念力 RSV 力度 RSV 數 K * t = Kt 1 + a RSVt K t t 力 RSV D * t = Dt 1 + a Kt D t t K t 滑 K[0]=D[0]=50 a 都 1/3 例 14

23 數滑異數滑異 (Moving Average Convergence-Divergence MACD) (Gerald Appel) 勒 (W.Fredrick Hitschler) 1979 年易 (Stock Market Trading System) 理利 DIF MACD 蘭立,, 來, MACD 理兩數滑 Exponential Moving Average EMA 來兩離 DIF 利離離數滑來 MACD 不降論率了 MACD 精 MACD 落 MACD 利不離率數領量 15

24 ( )MACD 數滑異說 (1) 數 P t Pt H + Lt + 2 * C ) 4 ( t t = H t t L t t C t t (2) 數 En 2 + * ( Pt En (1 + n) t = Ent 1 t 1 ) En t P t n 數 (EMA) n (3) 離 DIF t = Ent Emt (4)MACD DIF h 滑 MACD 2 + * ( DIFt MACD (1 + h) t = MACDt 1 t 1 ) 16

25 數滑異 (Relative Strength Index,RSI) 1978 年. (Welles Wilder JR.) 易念 New Concepts in Technical Trading Systems 行數來理利力 RSI 度度例 RSI 0~100 RSI 降靈領數行 RSI 度度利 RSI RSI 來. 17

26 RSI, KD 來 ( )RSI 說 RSI (1 + RS) = RS 度 RS = AU AD n n AU n n 數 AU n Ui i= n = 1 n U i i 數 AD n n 數 AD n n Di i= n = 1 D i i 數 RSI RSI AUn 100 * ( AUn + ADn) = 18

27 廉廉 ( Williams Overbought/Oversold Index) 拉. 廉 (Larry Williams) 1973 年 " (How I Made A Million Dollars)" 廉狀數理來率廉廉度 19

28 ( ) 廉說 W % R ( Ct Ht) *100 ( Ht Lt) = C t t H t t L t t 廉量量 On Balance Volume OBV 1963 年蘭 Joseph E Granville 利 (New Key to Stock Marker Profit) 量 OBV 量來 OBV 量量數量來量力 20

29 量 OBV 量 OBV 量量理量不 OBV 量不 ( )OBV 量說 OBV * t = Volt Supt Supi ( Ci Li) ( Hi Li) = i=0 t,vol t t 量, Sup t t 21

30 六 K K (Candlestick Charts) 都 K 理錄拉留留錄來來 K K K 來說不易 22

31 - 不兩來若若略略率 (1)MACD( ) 理龍 (1994) (2000) Van Home & Parker (1967) 龍 ( 79 年 ) 林良年 1999 MACD 2-1 列 MACD 料料理龍年數 23

32 數類數料料龍年數林良數料數類數 (2) RSI( ) 理 Jesnen & Benington 隆 Levy Levy 林年龍年年龍年 RSI 2-2 列 RSI 料 24

33 料理料易不易隆年率料年年料林年率龍年數料 25

34 料龍年數年數數 (3) KD( ) 理龍 KD 2-3 列 KD 料 26

35 料理龍年年年年年年數類數 (4) W%R( 廉 ) 理廉 2-4 列 W%R 料廉料理 W%R 廉年年數 W%R 廉 27

36 W%R 廉年易來率略 W%R 廉 (5) 理李良林林 2-5 列料料理廉廉廉行行參數不兩來率廉易略 28

37 李良年年數料略林年年數劣林年年略論不裡不論參數來兩來易兩兩不量更更靈 29

38 來參更不精易略識 30

39 類 KNN 類 KNN 類 (K-Nearest Neighborhood) 練料 (training data) 便來了練料 (test data) 類料練料 K 練料 K 練料類 K 練料料離來料類 (Larkey and Croft,1996) KNN 立類料類理 2-7 例 A A 離 B C K 2 A C2 KNN 類練料量練料量若料理類量類度 31

40 類類類 (Naive Bayesian Classifier) 理論來類類類理 (Bayesian Theorem) 率類立來類練類練念念類料行類料類類類練練理來類料 (2-16) 理 P( X C) P( C) P( X ) P ( C X ) = 32

41 類 X 例 c 類 X 例 C 類率 =(C 類 X 例率 ) (C 類率 )/(x 例率 ) 類利率來料類料不類類類率類類不來料類量量 (Support Vector Machine, SVM) 1995 年 Vapnik SVM (structural risk minimization, SRM) 理年來類理度量類類率 2-8 量類量類量類 33

42 Vapnik 量 VC (trade-off) 路練量 (quadratic programming) 量了 34

43 利聯立類 SVM 類 KNN 類類切利類率降不料易 88 年年 4 30 易 11 年歷易料 3-1 歷易料易料料練 35

44 易料練易料類立類類立類 36

45 類類練類立易錄易 -7%~7% 16 ( 類 ) 類立類料料理理易類類類類類類 37

46 類理參數類類參數參數錄類類類料料理理略易類類類錄類略類 38

47 類料理錄類類類類參類類類類錄類類料料理理易類類類類 39

48 料易料易 88 年 1 99 年 4 易料料易率類利練料料兩 ( ) 練料數 2569 ( 易數 ) ( ) 料練料 ( ) 料數 279 ( 易數 ) ( ) 類料數 7% -7% 16 類料數

49 類料數類類練數數利 MACD RSI KD OBV 廉 K 量理 0~ 料

50 量 42

51 練利練料料料利 k-fold cross-validation 率類類率立類利 7 利類錄數利, 利類錄數數利兩利類兩數利數類 3-fold cross-validation 數 4-3 G1 G1 料 G2 G3 練料 G2 G2 料 G1 G3 練料 G3 G3 料 G1 G2 練料 ( ) Bayes 數 Bayes 類 4-3 macd 廉 OBV 數都利兩兩 macd wms macd 43

52 obv macd 廉 obv macd 數 4-3 BAYES 類量 ( ) SVM 數 SVM 類 4-4 數都不 7 類 44

53 量 ( ) KNN 數 KNN 類 4-5 macd 廉 kd 數都利兩兩 macd 廉 macd kd 廉 kd macd 廉 kd kd 廉數類量 45

54 利練料 ( 2569 料 ) 類類錄利類類料 ( 279 料 ) 類 (1) 類兩類 (2) 類 1~8 (3) 類 9~16 (4) 易數易數 (5) 利數易數利數易列類類類不列略略類類類略略 46

55 類 ( ) Bayes 類類易數利數類類易數利數 ( ) KNN 類類易數利數 47

56 略 ( 類 ) 略 Bayes 類 KNN 類類略 Bayes 類 SVM 類類略 KNN 類 SVM 類類略類類略類類 2 ( 3 ) (1) 略 Bayes KNN 兩類易數利數 (2) 略 Bayes SVM 兩類易數利數 48

57 (3) 略 KNN SVM 兩類易數利數 (5) 略 BAYES KNN SVM 類類易數利數 (6) 略略略易數利數 49

58 略 ( 類 ) 略 Bayes 類 KNN 類類略 Bayes 類 SVM 類類略 KNN 類 SVM 類類略類類略類類 2 ( 3 ) (1) 略 Bayes KNN 兩類易數利數 (2) 略 Bayes SVM 兩類易數利數 50

59 (3) 略 KNN SVM 兩類易數利數 (4) 略 Bayes KNN SVM 類類易數利數 (5) 略略略易數利數 51

60 理 4-1 略數略略略略類兩兩略略類略略兩兩略例數 279 易 87 易略利數利數利數略略略略略略 52

61 4-2 類數類數說類不略類來利數利數略略類 4-2 略利數 27.6 略利數易數 279 利金略易數 87 利數 2401 數 (1 50 ) 易金 2,401*50=120,050 易 ( 易 5 ) 87*5*50=21,750 利 98,300 53

62 略易數 279 利數 3593 數 (1 50 ) 易金 3,593*50=179,650 易 ( 易 5 ) 279*5*50=69,750 利 109,900 易數 279 略利略略利略都易力略 87 行易利略不率 54

63 論來論類利類更利略類略易數 279 易數類略易數 87 易 27.6 數不不不類類不利不不不不類不參利不類連度率類連度來來利更類不列更更類不不類類連度不來來參 55

64 參力理類神路立理論料例論李良理論金龍北金林良立論林利理論林論祥料理理論林利率論立理論 56

65 離率理論龍立金論立論神路類數類立論立金論隆益立理論數切理論論葉料立理論龍量立理論金類神路立論例理論 57

66 立北論類神路略理論料例立理論易略立金論立論林金易力例立金論 58

67 Al-Kofahi, Khalid, Tyrrell, Alex, Vachher, Arun, Travers, Tim and Jackson, Peter(2001), Combining multiple classifiers for text categorization,proceedings of the tenth international conference on Information and knowledge management, Atlanta, Georgia, USA. Demirekler, M., and Altin,cay, Hakan (2002), Plurality voting-based multiple classifier systems: statistically independent with respect to dependent classifier sets., Pattern Recognition, 35(11),pp Edwards, Robert D. and Magee, John (1966),Technical Analysis of Stock Trends,5thed.Boston Mass John Magee Inc,5-6. Gerald, A. and Frederick, W. H. (1990), Stock market trading systems.greenville,sc Traders Press. Graham, Benjamin (1973), Intelligent Investor: a book of practical counsel, 4th rev. ed, HARPER & ROW, Publishers, New York. Jensen, Michael C. and Benington, George A. (1970), Random walks and technical Theories Some Additional Evidences, Journal of finance,vol.25,pp Larkey, L., Croft, W. B. (1996), Combining Classifiers in Text Categorization, Proceedings of the 19th International Conference on Research and Development Information Retrieval (SIGIR96), Zurich, Switzerland, pp Levy, R.A. (1967), Random Walks Reality or Myth, Finance Analysts F Jouranl.Vo123,PP Tsochantaridis, I., Hofmann, T. and Joachims, T. and Altun, Y. (2004), Support Vector Learning for Interdependent and Structured Output Spaces, Proceedings of the twenty-first international conference on Machine learning,p

68 Tsochantaridis, I., Joachims, T. and Hofmann, T. and Altun, Y. (2005), Large Margin Methods for Structured and Interdependent Output Variables,Journal of Machine Learning Research (JMLR), pp Wilder, J. and Welles, Jr, (1978), New concepts in technical trading system.mcleansvill,nc Trend Research Ltd. Wyckoff, Peter (1970), Technical Analysis,Edited by Zard, F.F. and Kerekes, G.T. The Stock Market Handbook I Homewood Dow Jones-Irwin Inc. 60

untitled

untitled 逸老年 1 錄錄...I 錄...II... III... 1... 1... 1... 2... 4... 4... 4... 6... 7... 7... 7... 8... 9... 9...10 流... 11... 13...13...15...16...17... 21...21...21 度...27 論... 29...29 來...29 I 參料... 30 料...30