Neural Network Technique for Prediction of Time Series

(1)

ISSN (Print) : 2319 – 2526, Volume-2, Issue-1, 2013

106

Neural Network Technique for Prediction of Time Series

Tuhin Mukherjee¹ & Md Asif Jalal²

1Department of Business Administration, University of Kalyani, Nadia, West Bengal, India

2Department of Computer Science, Pondicherry University, India E-mail: [email protected]¹, [email protected]²

Abstract – As like as any other soft computing algorithm, Artificial Neural Network (ANN), is recently used by researchers to solve challenging research problems.

Popularly, it is used for all those cases which are data intensive. ANN can be used to develop the complex relations between variables from missing data. This paper, concentrates on the time series prediction problem with the use of ANN. Both feed forward and feedback networks are used in its empirical part. Clearly, it suggest the benefit of ANN (i.e. the use of back propagation learning algorithm) over other traditional techniques of time series prediction.

Keywords – Artificial Neural Network (ANN) , Feed Forward ANN, Feed Back ANN, Back Propagation

I. INTRODUCTION

The prediction of the price movement of the stocks is one of the issues on research of stock market.

Because the neural network can find a model proved by the training set, so the training set and the settings of the parameters become extremely important. In addition, because the stock market‟s dynamics are very quick and the model for this system may change in the short term, the more recent data should be given much weight to on the consideration of the market. On the other side, the old data should be lower estimated by the network, without loosing much of the general characteristics of the model of the domain.

While numerous scientific attempts have been made, no method has been discovered to accurately predict stock price movement. The difficulty lies in the complexity of modeling human behavior. This paper is an attempt to use ANN approach for financial forecasting (direction of price movement as well as future estimate).

The rest of this paper describes difficulties in financial forecasting and suitability of ANN modeling in this respect, review of literature, experiment of this study comprising of searching for optimal parameter selection of ANN approach, and a final conclusion indicating

major findings with new area of investigation as well as limitation of this research study.

II. DIFFICULTIES IN FINANCIAL FORECASTING AND SUITABILITY OF ANN MODELING

Stock market has long been considered a high return investment field. Due to the fact that stock markets are affected by many highly interrelated economical, political and even psychological factors that interact with each other in a very complex fashion, it is very difficult to forecast the movement in stock market.

Predicting is telling about the future which will incur certain error. To produce a meaningful prediction, the error incurred must be minimum. There are several ways used by investors to predict stock market returns such a technical analysis, fundamental analysis and mathematical models. However these techniques incapable of determining the exact forecast price. Due to these imperfection factor current studies using soft computing techniques (Soft Computing represents that area of Computing adapted from the physical sciences.) such as Granular Computing, Rough sets, Neural Networks, Fuzzy sets and Genetic Algorithms are highly used to improve the prediction accuracy and computational efficiency compared to earlier techniques.

With the advancement being made in computer and telecommunication technologies today, the world‟s major economies and financial markets and becoming more and more globalize. As this trend accelerates, financial markets are becoming more and more interrelated and fundamental factors will become increasingly critical to financial market analysis. In the global marketplace, the prevailing methods of technical analysis where a single market is modeled through historical simulation and back testing of its own past price (or volume) behavior is rapidly losing its competitive advantages. Institution and individual

(2)

107 traders both are increasingly applying new technologies to financial forecasting. Recent research shows that these nonlinear domains can be modeled more accurately with these technologies (like ANN) than with the linear statistical and single-market methods that have been the mainstay of technical analysis throughout the past decade.

Another advantage of ANN implementation is that the processing is distributed among many nodes. Even if some of the nodes fail to function properly, the effect on the overall performance of the system will not be significant. This assertion can be verified by turning off m randomly selected hidden layer nodes and observing the resulting effect on the system performance.

However due to their large number of inputs, network pruning is important to remove redundant input nodes and speed up training and recall. Essential features of a neural network are: The network topology, Computational functions, and Training algorithm.

Decisions on the target output with respect to concerned inputs will select these features along with their respective parameters like learning rate, number of hidden layers, and number of nodes in each layer etc.

Financial neural network must be trained to learn the data and generalize, while being prevented from overtraining and memorizing the data. Once trained, the network parameters (weights) will be kept fixed. The model is then used with the input data set for prediction.

A neural network can be designed to predict the direction, magnitude or just turning points in the stock price movement.

III. LITERATURE REVIEW

Research reviewed in this area generally attempts to predict the future data points of some time series using historical data sets. Possible time series include: Base time series data (e.g. closing prices) or time series derived from base data (e.g. indicators which are frequently used in Technical Analysis). There are many studies that attempted to predict future values of a series from the past values of that same series or using data from different series. The studies those are representative of the current research in the time series prediction include (Chan and Foo, 1995; Quah and Srinivasan, 2000; Yao and Poh, 1995; Hobbs and Bourbakis, 1995;Austin Looney et al. , 1997;

FalasCharitouet al. ,1994). These studies consider data from both fundamental and technical analysis. For example Falaset al. (1994) used ANNs to attempt to predict future earnings based on reported accounting variables. They found no significant benefit using ANNs and concluded that accounting variables chosen were not appropriate earning predictors. Quah and

Srinivasan (2000) used mainly accounting variables to predict excess returns (with limited success). Chan and Foo (1995) used ANNs to predict future time series values of stock prices, and used these „future‟ values to compute a variety of Technical Indicators. They concluded that the networks ability to predict, allows a trader to enter a trade a day or two before it is signaled by regular technical indicators, and this accounts for the substantially increased profit potential of the market participants.

IV. EXPERIMENT OF THIS STUDY

The data used in this research is the daily stock prices . In our previous works, we have used closing stock prices for similar research but to check consistency of the result, the present paper uses average daily prices as calculated by observing prices in each hour . The data set covers the period from 1st Jan 2008 to 31st Dec 2011 (4 calendar years). A randomly selected 100 stocks ( whichhave been participating in BSE 500 within the period of our research study) have been considered in the sample. Industry wise classification of our sample is given in table 1.0.

Main aim of this paper is to compare the predictive ability of GNN based financial forecasting models with that of ARCH/GARCH models in the context of Indian stock market.

Firstly, it is necessary to formulate null hypothesis which will be tested against its alternate hypothesis. Taking the null hypothesis that there is no difference in predictive ability of two models (i.e.

GNN and ARCH/GARCH), we can formulate hypothesis as:

H_O: µ(GNN) = µ(ARCH/GARCH) H_A: µ(GNN) ≠ µ(ARCH/GARCH)

As the sample size is large, so z-test will be suitable for difference in means, assuming the populations to be normal and shall work out the test statistic z as under (C.R.Kothari,2003):

Z = (µ (s₁ )- µ (s₂ ))/ ( σ(s₁)²/n₁+ σ(s2)²/n₂)^1/2 Since the population variances are not known, so we have used the sample variances, considering the sample variances as the estimates of the population variances.

Using the statistical software package SPSS (version 10.0),we get the computed z-values for different error metrics in following table 1.1.

As alternate hypothesis is two sided, so we shall apply a two-tailed test for determining the rejection regions. At 5% level of significance, it comes to as under (using normal curve area table) : R: |z| > 1.96.

(3)

108 V. RESULT OF THIS EXPERIMENT

Evidently it is clear from table 1.1 that, all the observed values of z w.r.t. different error terms of each company came consistent (i.e. either all rejects null hypothesis or all fail to reject), which is quite natural.

Moreover, it is observed from z statistic that 60% of the sample companies lie in the rejection region and hence reject the null hypothesis. So we can conclude that 60%

of our sample stocks imply that predictive abilities of two models (GNN and ARCH/GARCH) differ significantly. This summary is represented in following table 1.2. This finding is very closed and consistent with Austin (1997).

VI. ACKNOWLEDGEMENT

This paper is continuation of our previous research papers and really an advancement in interdisciplinary research field of computer science and financial management. All the authors contribute in this paper truly and sincerely. Valuable references are given along with this paper.

VII. REFERENCES

[1] Abdullah, M. H. L. b. and V. Ganapathy (2002). “Neural Network Ensemble for Financial Trend Prediction”. Tencon 2000: Proceedings:

Theme: Intelligent Systems and Technologies for the new millennium.

[2] Austin, M. , C. Looney et al. (1997). “Security Market Timing Using Neural Network Models”.

New Review of Applied Expert System, Vol. 3, pp. 3-14.

[3] Baba, N. and H. Handa (1995) “Utilization of Neural Network for Constructing a User Friendly Decision Support System to Deal Stock”. IEEE International Conference on Neural Networks.

Table- 1.0

Industry-wise Classification of Stocks in our sample

Industry No. of Companies

Housing Related 7

Automobiles and ancillary 1

Pharmaceutical 3

Refineries 6

Software 7

Aluminium 1

Diversified 4

Power 8

Industry No. of Companies

Transport Equipments 6

Transport Services 3

Capital Goods 7

Finance 12

Media & Publishing 6 Chemical & Petrochemical 3

Telecom 3

Healthcare 3

Textile 3

Metal & Mining 1

Agriculture 3

FMCG 5

Tourism 3

Others 5

Total 100

Source : Case study.Results computed.

Table 1.1

z-Values for GNN Vs ARCH/GARCH Sl.

No.

Company -

Sector AAE MSE Max

AE 1. *ABB LTD.-

Capital Goods -2.8022 -2.6522 -2.7209 2 *ACC LTD. –

Housing Related -2.9893 -2.7722 -2.8007 3

Ackruti City Limited- Housing Related

-1.6504 -1.6756 -1.0013

4

Adani

Enterprises Ltd.- Diversified

-1.0238 -1.5528 -1.0922 5 *Adani Power

Ltd. - Power -2.7077 -2.7388 -2.8144 6

ADITYA BIRLA NUVO LTD.-Diversified

-1.5500 -1.6451 -1.0964

7

*Ahluwalia Contracts(India) Ltd.- Housing Related

-2.7074 -2.7440 -2.8774

8

AIA Engineering Ltd.- Capital Goods

-1.0929 -1.6222 -1.0172 9 Allahabad Bank

– Finance -1.0463 -1.3878 -1.0945 10 Allcargo Global

Logistics Ltd.- -1.0187 -1.6798 -1.0005

(4)

109 Sl.

No.

Company -

AE Transport

Service 11

*Ambuja Cement Ltd.- Housing Related

-2.4545 -2.7810 -2.1353 12 *Andhra Bank –

Finance -2.8246 -2.5610 -2.8167 13 Aptech Ltd. –

Software -1.0834 -1.6888 -1.0973 14

*AurobindoPhar ma Ltd.- Pharmaceutical

-2.8049 -2.7330 -2.8122

15

Aventis Pharma Ltd.-

Pharmaceutical

-1.1176 -1.6258 -1.0965 16 *Axis Bank Ltd.-

Finance -2.8809 -2.7520 -2.8553 17

*Ashok Leyland Ltd. – Transport Equipments

-2.8073 -2.7795 -2.8173

18

Bajaj Auto Ltd. – Transport Equipments

-1.3054 -1.6745 -1.3576 19 Bajaj Finance

Ltd. – Finance -1.8064 -1.6271 -1.0775 20 *Bajaj Hindustan

Ltd.- Agriculture -2.7597 -2.7175 -2.6673 21

Ballarpur Industries Ltd. – Other

-1.1238 -1.6732 -1.8965 22 BalmerLawrie&

Co. Ltd.-Other -0.0037 -1.6851 -1.3975 23 *Bartronics India

Ltd.- Other -3.8024 -2.7276 -2.5163 24

*BASF India Ltd. - Chemical

& Petrochemical

-3.8553 -2.7764 -2.8673

25

*Berger Paints India Ltd. - Chemical &

Petrochemical

-3.8673 -2.5559 -2.8143

26 BF Utilities Ltd.

– Power -0.0784 -1.6756 -1.0934 27

Bharat Forge Ltd.- Transport Equipments

-1.0168 -1.6382 -1.0974 28 *BPCL –

Refineries -2.7966 -2.4982 -2.8118 29 BhartiAirtel Ltd.

- Telecom -1.0787 -1.6722 -1.0185 30 Bilcare Ltd. –

Helthcare -1.1009 -1.6397 -1.0970

Sl.

No.

Company -

AE 31 *BOC India Ltd.

– Refineries -2.6087 -2.6512 -2.2370 32

Bombay Dyeing

& Mfg. Co. Ltd.- Textile

-1.2065 -1.6458 -1.0975

33

*Bombay Rayon Fashions Ltd.- Textile

-2.6000 -2.7560 -2.8174

34

*Century Textiles &

Industries Ltd. – Diversified

-2.8556 -2.7890 -2.8153

35 CESC Ltd. –

Power -1.6711 -1.6978 -1.0275 36

Chambal Fertilisers&

Chemicals Ltd- Agriculture

-1.0089 -1.8658 -1.0945

37 *CIPLA –

Pharmaceutical -2.8012 -2.4730 -2.8553 38 *CMC Ltd. –

Software -2.8000 -2.7627 -2.4973 39

*Coal India Ltd.

– Metal &

Mining

-2.8083 -2.3120 -2.5373

40

Colgate- Pamolive(India) Ltd. – FMCG

-1.0069 -1.6145 -1.1375

41

Container Corporation of India Ltd.- Transport Service

-1.7134 -1.6087 -1.0875

42 *Cox & Kings

Ltd. – Tourism -2.6823 -2.7046 -2.6573 43 *CRISIL Lltd. –

Finance -2.5826 -2.7556 -2.8373 44

*D.B.Corp Ltd.- Media &

Publishing

-2.3462 -2.6520 -2.8993 45 Dabur India Ltd.

– FMCG -1.5639 -1.5708 -1.0085

46

*Deccan Chronicle Holdings Ltd.- Media &

Publishing

-2.2326 -2.8270 -2.1153

47 *Delta Corp Ltd.

– Other -2.8768 -2.7570 -2.8363 48 *Dish Tv India

Ltd. – Media & -2.8733 -2.7173 -2.2873

(5)

110 Sl.

No.

Company -

AE Publishing

49 *DLF – Housing

Related -2.7623 -2.7784 -2.8128 50

E.I.D. Parry (India) Ltd.- Agriculture

-1.0744 -1.6705 -1.0936 51 *EIH Ltd. –

Tourism -2.8773 -2.7503 -2.8665 52 *Emami Ltd. –

FMCG -2.6783 -2.6502 -2.8172 53 Engineers India

Ltd. – Other -1.4264 -1.6702 -1.0933 54

* Eicher Motors Ltd. – Transport Equipments

-2.8893 -2.7716 -2.8173

55

*Entertainment Network(India) Ltd. –Media &

Publishing

-2.8067 -2.7746 -2.8763

56

*Escorts Ltd.- Transport Equipments

-2.7623 -2.7982 -2.8143 57 * Essar Oil Ltd.-

Refineries -2.7323 -2.7584 -2.7553 58

Exide Industries Ltd. – Transport Equipments

-1.1034 -1.6758 -1.0025 59 * FDC Ltd. –

Helthcare -2.8027 -2.7797 -2.6523 60 *Federal Bank –

Finance -2.8068 -2.7900 -2.8175 61

Financial Technologies (India) Ltd.- Software

-1.0032 -1.5656 -1.0973

62 *Finolex Cables

Ltd. –Telecom -2.8635 -2.7934 -2.8345 63

Finolex Industries Ltd.- Chemical

&Petrochemical

-1.0789 -1.6733 -1.0966

64

*Firstsource Solutions Ltd.- Software

-2.7461 -2.7966 -2.5436 65 *Gail (India) Ltd.

– Refineries -2.8764 -2.6549 -2.7622 66 * Geodesic Ltd.

–Software -2.7654 -2.7679 -2.8128 67 *Gillette India

Ltd. – FMCG -2.8789 -2.9875 -2.8173 68 GMR

Infrastructure -1.0784 -1.6791 -1.0911

Sl.

No.

Company -

AE Ltd.-Power

69

Graphite India Ltd.-Capital Goods

-1.2678 -1.6760 -1.0957

70

*Grasim Industries Ltd.- Textile

-2.8889 -2.7780 -2.6724

71

*GTL Infrastructure Ltd.-Telecom

-2.8074 -2.7979 -2.8524

72

* Gujarat Gas Co.Ltd.- Refineries

-2.8789 -2.7658 -2.8100

73

GVK Power and Infrastructure Ltd.-Power

-1.2574 -1.6265 -1.0971

74

*Havells India Ltd.-Capital Goods

-2.7653 -2.7864 -2.8453

75

* HCL Technologies Ltd.-Software

-2.4563 -2.6780 -2.8165 76 HDFC Bank

Ltd.-Finance -1.5074 -1.6728 -1.0977 77 HEG Ltd.-

Capital Goods -1.0034 -1.6428 -1.1188 78

*Hero Honda Motors Ltd.- Automobiles

-2.5023 -2.5630 -2.8448

79

*Hindalco Industries Ltd.- Aluminium

-2.9823 -2.1220 -2.6338

80

* Hindustan Unilever Ltd.- Diversified

-2.8653 -2.9870 -2.8358 81 *HMT Ltd.-

Capital Goods -2.7613 -2.7337 -2.8100 82

* HT Media Ltd.-Media and Publishing

-2.7643 -2.1358 -2.5101 83 I.C.S.A(India)

Ltd.-Finance -1.0854 -1.6738 -1.0909 84

*IBN18 Broadcast Ltd.- Media and Publishing

-2.8073 -2.7789 -2.4100

85 *ICICI Bank

Ltd.-Finance -2.8923 -2.7865 -2.7573 86 ICRA Ltd.-

Finance -1.0037 -1.6711 -1.0575 87 *IDBI Bank

Ltd.-Finance -2.8021 -2.7165 -2.7551

(6)

111 Sl.

No.

Company -

AE 88

India Cements Ltd.-Housing Related

-1.0098 -1.6794 -1.0877

89

* Indiabulls Power Ltd.- Power

-2.8076 -2.7338 -2.8123 90 *Indian Hotels

Co.Ltd.-Tourism -2.8053 -2.5642 -2.8165 91

*Indian Oil Corporation Ltd.- Refineries

-2.8675 -2.7796 -2.8072

92

*INDOSOLAR Ltd.-Capital Goods

-2.8028 -2.7563 -2.8253

93

*Infosys Technologies Ltd.-Software

-2.8011 -2.7904 -2.8422 94 ING Vysya Bank

Ltd.-Finance -1.0071 -1.6700 -1.0581 95

*Ipca Laboratories Ltd.-Healthcare

-2.8001 -2.7002 -2.8001 96 *ITC Ltd.-

FMCG -2.8645 -2.7335 -2.6575 97 J.K.Cement Ltd.-

Housing related -1.7216 -1.7676 -1.8401 98

Jet Airways (India) Ltd.- Transport Services

-0.6517 -0.4920 -0.8140

99 JSW Energy

Ltd.-Power -0.8041 -0.7971 -0.2342 100 NTPC Ltd.-

Power -1.0999 -1.6748 -1.0222

*5% level-60 Companies

Source: Case study. Results computed.

Table 1.2

Summary of z-test (at 5% level) for predictive ability of GNN and ARCH/GARCH

Number of stocks rejecting

Null hypothesis 65 65%

Number of stocks fail to reject null

hypothesis 35 35%

Total 100 100%

Source: Case study. Result Computed.

