1 Prediction method for stock market Student : Dah-Sheng Lee Professor: Hahn-Ming Lee Date:30 January 2004
2 Outline Introduction Some Prediction method Neural Network(2000)[2] GA based Fuzzy Neural Network (2001)[3] Conclusions Reference
3 Introduction Input variable : stock price 、 turnover 、 technological index 、 financial index 、 market index etc… Output variable : stock price 、 buy-sell decision 、 trend 、 portfolio etc… Other factory : 漲跌幅限制、恐怖攻擊、天然災害、人為炒作或 交易成本對報酬率的影響等等
4 Some Prediction method Statistics : –Auto Regressive Moving Average(ARMA) Machine Learning: –Neural Network[1] –Fuzzy and Gray[3] –Support Vector Machine[4]
5 Neural Network(2000)[2] The Neural Network has 1 hidden layer, 8 input unit and 2 output unit ( ) Tokyo Stock Exchange Price Indexes (TOPIX) from 11/1995 to 10/1997 Input unit : 1.Changes of TOPIX 2.PBR 3.Changes of the turnover by foreign traders 4.Changes of the currency rate (Yen - Dollar) 5.Changes of the turnover Tokyo Stock Market 6.etc.
6 Neural Network(2000) cont. Output unit: highest and lowest value of TOPIX till 4 weeks in the future
7 Neural Network(2000) cont. Learning Algorithm
8 Neural Network(2000) cont. Simulation Result Utilize the above chromosome for the dealings of TOPIX during the period from November 1996 to October In this period, TOPIX changed from 1556 yen to 1360 yen. This means that we would have lost about 13% of our investments if we had not executed dealings. However, we have succeeded in keeping our investments by following the dealing rule obtained by Gas. (The total amount of money has been changed as follows: billion yen -> billion yen.)
9 GA based Fuzzy Neural Network (2001)[3] The System consist of –factors identification (technical indexes) –qualitative model (GA fuzzy neural network) –decision integration (artificial neural network) Index of Taiwan Stock market –Training samples are from 1/1/1994 to 12/31/1995 –Testing samples are from 1/1/1996 to 4/30/1997
10 GA based Fuzzy Neural Network (2001)(cont.)
11 GA based Fuzzy Neural Network (2001) ---factors identification This part collect 42 kinds of technical indexes and non-quantitative information The 42 kinds of technical indexes are
12 GA based Fuzzy Neural Network (2001) ---factors identification (cont…) The non-quantitative information include related economics journals, government technical reports and newspaper from 1991 to 1997 The experienced experts eliminated the unnecessary events and then divided the useful events into six dimensions (political,financial,economic,message,technical, and international) The questionnaire for each event has the following format: –IF event A occurs, THEN it’s effect on the stock market is from to.
13 GA based Fuzzy Neural Network (2001) ---qualitative model The fuzzy method is employed to capture the stock experts’ knowledge GA used in this model with parameters below –Fitness function Where N denotes the number of the population and value is set to be 50 Ti represents the i-th desired output Yi represents the i-th actual output format of Chromosome is 8-digit value on the basis of 2
14 GA based Fuzzy Neural Network (2001) ---qualitative model (cont…) The “Dimensional GFNN” combines all events of specific dimension occurred and Integrated by using an “Integrated GFNN” The GA parameters in “Dimensional GFNN” is –Generations : 1000 –Crossover rate: 0.2 –Crossover type: two-point crossover –Mutation rate: 0.8
15 GA based Fuzzy Neural Network (2001) ---qualitative model (cont…) The GA parameters in “Integration Dimensional GFNN” is –Generations : 1000 –Crossover rate: 0.2 –Crossover type: two-point crossover –Mutation rate: 0.8
16 GA based Fuzzy Neural Network (2001) ---decision integration (cont…) Both the quantitative and qualitative factors are inputs of ANN, and should normalized in [0,1] The ANN including “time effect” input node In this system,two different out-puts, O1 and O2, are verified
17 Conclusions Is there any new features we can figure for the existing system performance improvement? –For example, in terms of the stock index in Taiwan, the politics is a very effective factor. Can the methods above be applied to forecast the price or trend of new financial merchandise? –Plenty of new financial merchandises are elaborated under the fast growing financial engineering, We can try to forecast the trend and price of them.
18 Reference [1] “An intelligent forecasting system of stock price using neural networks” Baba, N.; Kozaki, M.; Neural Networks, IJCNN., International Joint Conference on, Volume: 1, 7-11 June 1992 Page(s): vol.1 [2] “Utilization of neural networks and GAs for constructing reliable decision support systems to deal stocks” Baba, N.; Inoue, N.; Asakawa, H.; Neural Networks, IJCNN 2000, Proceedings of the IEEE-INNS-ENNS International Joint Conference on, Volume: 5, July 2000 Page(s): vol.5 [3] ”An intelligent stock trading decision support system through integration of genetic algorithm based fuzzy neural network and artificial neural network” Kuo, R.J.; Chen, C.H.; Hwang, Y.C. Fuzzy Sets and Systems Volume: 118, Issue: 1, February 16, 2001, Page(s): [4] “Stock selection using support vector machines” Fan, A. ; Palaniswami, M., IEEE Neural Networks, Proceedings. IJCNN '01. International Joint Conference on page(s): vol.3
19 Reference [5] “A simple neural network for ARMA(p; q) time series” H. Brian Hwarng ; H. T. Ang, Omega, 2001, page(s): 319–333 vol. 29