Long Term and Short Term Investment Strategy for Predicting the Performance of BSE using MLP Model

Transcription

1 Indian Journal of Science and Technology, Vol 8(22), IPL0250, September 2015 ISSN (Print) : ISSN (Online) : Long Term and Short Term Investment Strategy for Predicting the Performance of BSE using MLP Model Sachin Kamley 1*, Shailesh Jaloree 2 and R. S. Thakur 3 1 Department of Computer Applications, S.A.T.I., Vidisha , Madhya Pradesh, India; skamley@gmail.com 2 Department of Applied Math s and Computer Science, S.A.T.I., Vidisha , Madhya Pradesh, India; shailesh_jaloree@rediffmail.com 3 Department of Computer Applications, M.A.N.I.T., Bhopal , Madhya Pradesh, India; ramthakur2000@yahoo.com Abstract Background/Objectives: Now a day s stock market has great influence in the daily life of people. The biggest problem of stock market is that it holds large uncertainties which are directly related to their long term and short term investment. The key objective of researchers in this area is to strengthen economy as well as profit maximization on individual share. Methods/Statistical Analysis: This study, adopted a most popular back propagation training algorithm for predicting the performance of Bombay Stock Exchange of India based on long term and short term basis. In this study, 3500 samples of nine different companies are considered are divided 60% for training, 20% for validation and remaining 20% for testing task. In the initial phase the data underwent five phases which includes variable selection, data preprocessing, training dataset, prediction and evaluation. However the research evaluated through MATLAB tool which consists of training performance, error rate, and output prices. After getting the output from various evaluations made on stock dataset. Findings: It is shown by experimental results the method has given more accurate prediction results. Besides, it takes less training time and epochs for prediction task. It is also found that long term investment also has more successive returns over short term investment. Applications/Improvements: In the present study back propagation method is employed for long term and short term prediction. The different experimental results are carried out on stock data sets. It is also observed by study for long span of time small scale and medium scale companies like HCL and Ambuja Cement have better performance compared to large scale companies like TCS. In contrast, short span of time large scale companies like TCS also have better performance. Keywords: BSE, Long Term Investment, Multi Layer Perceptron, Prediction, Short Term Investment 1. Introduction Stock price prediction is an important concern in these days and various soft computing techniques are conferred for this within the past. The emergences of Data Mining (DM) and Artificial Intelligence (AI) techniques have brought a new trend in the market. During this study Artificial Neural Network (ANN) technique with combination of Data Mining tool are applied on Bombay Stock Exchange (BSE) of India which is the most dominating stock exchange over the world 1. Typically ANN has the potential to map the input data values into output data values. Multilayer Perceptron (MLP) is the most generally used methodology of neural network with back propagation training algorithm. However, the MLP model has expertization to solve problem in both classification and prediction field and has been widely employed in various areas for the last four decades with greater prediction results 2. The proposed back propagation model is tested on completely nine different companies of BSE (i.e. large, *Author for correspondence

2 Long Term and Short Term Investment Strategy for Predicting the Performance of BSE using MLP Model medium and small scale organization). The network is trained for several times. However, best validation performance is chosen at epoch no. 12 which clearly indicates that proposed training model adequately mimics the trends and patterns of the market. The performance of companies evaluated based on long term and short term basis. For the short term stock performance day wise and monthly wise training results are considered. Similarly for long term performances half yearly, yearly and five yearly training results are considered. Moreover, for higher results some necessary fundamental variables capital gain and dividend paid (last five years intrinsic information of company) are considered 3,4. The rest of the paper is organized as follows: Section 2 focuses on data pre-processing features on stock data. Section 3 focuses on the proposed methodology and its usability in stock market. Section 4 focus on experimental results conducted by research study and at last Section 5 discusses the conclusion and future scopes of the stock market. 2. Data Preprocessing Stock market data is often noisy, incomplete and inconsistent and is likely to contain so many uncertainties. Data preprocessing is an important step which is used to resolve such issues and prepares the data for prediction task 5. The data employed in this study contains the range from Jan to Oct The data consists five important variables like open price, high price, low price, close price and volume 6. The companies are classified based on market capitalization. Companies have market capitalization value larger than 180 Cr. comes under the category of large scale, companies which have market capitalization range between 115 Cr. to 180 Cr. comes under medium scale category and companies whose market capitalization value less than 115 Cr. comes under the category of small scale category 1,7. The Table 1 shows this classification. The stock prices are too much long and not feasible for computation so next task is to apply data normalization step on stock data. The following formula is used to scale the price values within the range of [0, 1]. NP MINSP MAXSP MINSP (1) Where NP, MINSP, and MAXSP denotes New Price, Minimum Stock Price, and Maximum Stock Price. Stock market data contain so much variability and various types of risks are associated with data. So before training there are some quantitative measures performed on stock data which represent the market risk present in the data 8. The Table 2 shows this measure. Table 2 can be understood in following manner. Table 1. Classification of companies based on Market Capitalization Large scale company TCS ONGC Medium scale company HCL Tech. Tata Motors Small scale company Asian Paints Jindal Steel HDFC Ltd. NTPC Ambuja Cement Table 2. Quantative measure on Stock data Company Name Standard Deviation Mean Relative Standard Deviation Coefficient of Variation Standard Error Mean TCS ONGC HDFC Ltd HCL Tech Tata Motors NTPC Asian- Paints Jindal Steel Ambuja- Cement Skew Kurt 2 Vol 8 (22) September Indian Journal of Science and Technology

3 Sachin Kamley, Shailesh Jaloree and R. S. Thakur Where, Standard Deviation (SD): Standard Deviation is used to measure the market risk. The higher values of standard deviation represent the higher risk present in market. The Table 2. Shows that Asian Paints share has low market risk value and NTPC share has higher market risk. Mean: shows the mean of stock prices. The change in values due to market variability. Relative Standard Deviation (RSD): It is also another important measure which tells about the market risk more closely than SD. The RSD value shows that Ambuja Cement has higher risk value and ONGC has lower risk value. Coefficient of Variation (COV): It also tells more closely about the market risk. Standard Error Mean (SEM): SEM is another measure which is calculated around the mean. Skew and Kurt: Both the measures are used to find the data symmetry around the mean. The Table 2 shows that TCS has better value of Skew and Kurt i.e. TCS data is normally placed around the mean and NTPC shows about abnormal displacement of data. 3. Proposed Methodology In this study, Multilayer Feed Forward Neural Network (MFFNN) approach is adopted which is very effective method due to huge amount of training samples. A Multilayer Feed-Forward Neural Network consists of an input layer, one or more hidden layers, and an output layer 9. Back propagation algorithm is used for training in Multilayer Feed-Forward Neural Network which iteratively learns a data set of training tuples comparing the network s prediction for each tuple with the actual known target value. If performances of network are not satisfactory then weights and biases are modified so as to minimize the Mean Square Error (MSE) 10. However, the propagation is done in the backwards direction (i.e. output layer to hidden layer) so it is names as back propagation. The method is well suited for classification and prediction task. The algorithm 1 shows our proposed algorithm for training Algorithm 1: Back propagation Training Perform data normalization step on data (i.e. change the input and output values in the range of [0, 1]. Now set the number of neurons in the hidden layer in the range of [1, 21]. Initialize all the weights and biases for input neurons, hidden neurons and output neurons in the range of [-1, 1]. If stopping condition is false then do step 5 to 12. For each training samples do steps 6 to 11. Get input from input layer and transform it to corresponding layer i.e. hidden layer. Now hidden layer sums it weighted input signals. Apply activation function and propagated signal to the output unit. For each output unit sum its weighted input signal (i.e. signal from previous layer). Apply activation function to calculate the output signal. If each output unit gets a target pattern corresponding to input unit then declare training successful otherwise calculate the error at output unit (i.e. difference between actual output and predicted output). Error is back propagated to update the weights and biases so that network s error could be minimized. Stop training if errors are minimized. Figure 1 shows the flowchart of the proposed algorithm Experimental Results In this study 9 companies of BSE index are selected for training and testing purposes 6. However 60% data for selected for training, 20% for validation and remaining 20% for testing purpose for individual company. The input layer consists of 4 important variables, 1 hidden layer consists of 11 neurons, and 1 output layer consists of 1 output variable. The network is trained for four times for nine different companies. Therefore it is trained for total 36 times. The network is trained for four different prices like open price, close price, low price, and high price. Trainlm (Levenberg-Marquardt Algorithm) is used as training function, Training is used as learning function and Mean Square Error (MSE) is used as performance function. The network is trained for 36 times for different companies. The Figure 2 shows only 1 state of training. After network training performance measure is an important task. The Figure 3 shows the performance measure i.e. Mean Square Error (MSE) against the number of epochs. It is clearly shown by Figure 3 network got the best validation performance at epoch no. of 12 which is The next task is to select the predicted prices for getting the investment return on particular share based on Vol 8 (22) September Indian Journal of Science and Technology 3

4 Long Term and Short Term Investment Strategy for Predicting the Performance of BSE using MLP Model short term basis i.e. daily, monthly and long term basis half yearly, yearly, and five yearly bases 12,13. The future performance of individual company from current position is also evaluated. The testing sample of 365 days (i.e. 15- APR. 13 to 1-OCT. 14) are considered for evaluating the results on daily monthly, half yearly, and yearly basis 6. For evaluating the performance based on longest period (i.e. five yearly) more data samples are also considered. The following formulas are used to calculate the capital gain and rate of return (%) on individual share 8, 14. Capital Gain = Current Price of Stock - Last Price of Stock (2) Rate of Return at the End of Day/Month/Year Capital Gain = * 100 Last Priceof Stock (3) The Table 3 Shows the day wise Rate of Return for each company. The Figure 4 shows day wise performance of Indian companies. Table 3. Day wise Rate of return for individual company Company Name Day wise Rate of return (%) TCS 1.08 ONGC 1.48 HDFC LTD HCL Tech Tata Motors 1.62 NTPC 1.3 Jindal Steel 1.95 Asian Paints 1.33 Ambuja Cement 1.55 Table 4. Monthly wise Rate of return for individual company Company Name Monthly wise Rate of return (%) TCS 6.37 ONGC 1.51 HDFC LTD HCL Tech Tata Motors 6.22 NTPC 1.47 Jindal Steel -2.3 Asian Paints Ambuja Cement 4.09 Figure 1. Flow chart of Back propagation method. Figure 4 show that Tata Motors and Jindal Steel have more return on investment than other companies. A graph also shows that medium scale company (Tata Motors) and small scale company (Jindal Steel) have high return as compared to large scale (TCS, ONGC, HDFC Ltd.). The Table 4 shows the monthly wise Rate of Return for each company. The Figure 5 shows the monthly performance of Indian companies. Figure 5. shows that on monthly basis TCS, Tata Motors and Ambuja Cements has high rate of returns on investment which belongs to large scale, medium scale, and small scale. It is also clear by graph that Jindal Steel and Asian Paints have negative performance which belongs to small scale companies. The Table 5 shows half yearly performance of Indian companies. The Figure 6. shows half yearly performance of Indian companies. 4 Vol 8 (22) September Indian Journal of Science and Technology

5 Sachin Kamley, Shailesh Jaloree and R. S. Thakur Figure 2. Training state of network. Figure 4. Line graph showing the day wise trend of BSE Figure 3. Performance Graph of Network. Figure 6 shows that on half yearly basis TCS and Tata Motors have better performance so investing money in these shares for longest period of time would be beneficial. The HDFC, HCL Tech. and Ambuja Cement have investment return is positive but very low. So it may be 50% of risk on investing in these shares. The NTPC, Jindal Steel, and Asian Paints shares showing the future loss on share i.e. investing on these shares may be harmful. The Table 6 shows yearly performance of Indian companies. The Figure 7 shows yearly performance of Indian companies. Figure 5. Line graph showing the monthly trend of BSE Vol 8 (22) September Indian Journal of Science and Technology 5

6 Long Term and Short Term Investment Strategy for Predicting the Performance of BSE using MLP Model Figure 6. Line graph showing the half yearly trend of BSE Figure 7. Line graph showing the yearly trends of BSE Figure 7 shows that large scale companies TCS have strong return on investment (i.e. above 85%) than HDFC and ONGC. The medium scale company Tata Motors has also strong return as compared to HCL and NTPC. The small scale companies Jindal Steel and Asian Paints showing loss on future investment, but Ambuja Cement has performed well than other two. It is also observed by graph medium scale company Tata motors share competing with large scale company TCS. The Table 7 shows five yearly performances of Indian companies. The Figure 8 shows five yearly performances of Indian companies. Figure 8 shows that large scale company TCS has highest investment return than other corresponding company. In medium scale HCL Tech. has also a highest return but lower than TCS. In small scale only Ambuja Cement has better investment return than other corresponding companies. So investing money for TCS, HDFC Ltd. and Asian Paints would be beneficial but ONGC, Tata Motors, NTPC, Jindal Steel, and Asian paints share showing future loss. For getting the more accurate analysis yearly prices ( ) of stock are considered and capital gains on individual stocks are calculated 7,8,15. The Table 8 shows the average capital gain on stock for last five years. Stock prices fluctuate over time so fundamental variable also plays important role for evaluating the Figure 8. Line graph showing the five years trends of BSE stock market performance. Dividend yield is one of the important variables for knowing the future worth of the company 17,19. So we considered last 5 years dividends history from Bombay Stock Exchange (BSE) of India site 6. It is also used to calculate the expected return and future performances of company after two years from current position. The following formulas are used for calculation of expected return and future performance of company 8. 6 Vol 8 (22) September Indian Journal of Science and Technology

7 Sachin Kamley, Shailesh Jaloree and R. S. Thakur Table 5. Half yearly Rate of return for individual company Company name Half yearly Rate of return (%) TCS ONGC HDFC LTD HCL Tech Tata Motors NTPC Jindal Steel Asian Paints Ambuja Cement 4.6 Table 6. Yearly Rate of return for Indian companies Company Name Yearly Rate of return (%) TCS 89 ONGC HDFC LTD HCL Tech Tata Motors 89.5 NTPC 0 Jindal Steel Asian Paints Ambuja Cement Table 8. Company name ( ) Divided Paid+ Capital Gain Expected Return( r ) = *100 Buy Price of Stock (4) Future Stock Value after Two Years = Buy Price of Stock + (Buy Price* Expected Return) + (Result* Expected Return) (5) Where, Average capital gain on individual stock Result = Average capital gain for last five years TCS 414 ONGC -152 HDFC Ltd HCL Tech Tata Motors NTPC Jindal Steel -103 Asian Paints -203 Ambuja Cement 23.4 Buy Price* Expected Return The Table 9 shows yearly expected return and future performance of company. Table 7. Five yearly rates of return for Indian companies Company Name 5 Years Rate of return (%) TCS ONGC HDFC LTD HCL Tech Tata Motors NTPC Jindal Steel Asian Paints Ambuja Cement Figure 9. Index. Bar Graph Showing the Future Trends of BSE Vol 8 (22) September Indian Journal of Science and Technology 7

8 Long Term and Short Term Investment Strategy for Predicting the Performance of BSE using MLP Model Table 9. Future worth of BSE index Company name Average dividend for last five years Expected Return (R) Future price of company TCS ONGC HDFC Ltd HCL- Tech Tata- Motors NTPC Jindal- Steel Asian- Paints Ambuja- Cement The Figure 9 showing the future value of Indian companies (i.e. after 2 years). Figure 9 shows that TCS company has highest future value among all companies. It would be very beneficial to invest money in TCS share. Remaining companies also showing positive sign but very low as compare to TCS. So this research study will be helpful for stock users to select the best value of stock on right time. 5. Conclusion and Future Scopes In this study, nine different companies data are considered and back propagation method is employed for long term and short term prediction. The different experimental results are carried out on stock data sets. It is also observed by study for longspan of time small scale and medium scale companies like HCL and Ambuja Cement have also better performance as compared to large scale companies like TCS. In contrast short span of time large scale companies like TCS have also better performance. However, this study will be helpful to addresses various types of questions for stock users like what will be capital on individual share? How much return on individual share? and what will be future worth of company. Due to economic globalization stock users are increasing day by day so this study will be helpful to bring new trends in the market and removing the traditional misconception like large scale companies always perform better. In future U.S. and China economy will be considered for comparison with Indian economy and back propagation method with different integration model will be applied. 6. References 1. BSE History and Indices. [cited 2014 Oct 10] Available from: 2. Nilsson NJ. Artificial Intelligence: A new synthesis. USA: Morgan Kaufmann Publishers, Inc; Kaur S. Top ten technology predictions for 2014 and beyond. IETE Technical Review. 2013; 30(6): Brown DP, Jennings RH. On technical analysis. The Review of Financial Studies. 1989; Han J, Kamber M. Data mining: Concepts and techniques, 2nd ed. San Francisco, CA: Morgan Kaufmann; Online Stock Market Dataset [Internet] [cited 2014 Aug 5] Available from: 7. Das AP. Security Analysis and Portfolio Management, 2nd ed. New Delhi, India: I.K. International Publication; Gupta SK, Sharma RK. Financial management theory and practice. 6th ed. New Delhi, India: Kalyani Publishers; Sivanandam SN, Sumathi S, Deepa SN. Introduction to neural network using matlab th ed. New Delhi, India: Tata McGraw Hill Publishing Company Limited; Minsky ML, Papert S. Perceptrons: An Introduction To Computational Geometry. 3rd ed. Cambridge, Mass: MIT Press; Li F, Liu C. Application Study of BP Neural Network on Stock Market Prediction. 9th International Conference on Hybrid Intelligent Systems (IEEE); p Saeedmanesh M, Izadi T, Ahvar E, editors. HDM: A Hybrid Data Mining Technique for Stock Exchange Prediction. Proceedings of the International Multiconference of Engineers and Computer Scientists (IMECS). Hong Kong, China; Khan ZH, Tasnim SA, Md. Hussain A. Price Prediction of Share Market using Artificial Neural Network (ANN). International Journal of Computer Applications (IJCA). 2011; 22(2): Casu B, Fabbri D, Wilson JOS. Emerging issues in financial institutions and markets. The European Journal of Finance. 2014; 20(10): Yao J, Tan CJ, Poh HL. Neural networks for technical analysis: A study on KLCI. International Journal of Theoretical and Applied Finance. 1999; 2(2): Karim MR, Ahmed CF, Jeong B-S, Choi H-J. An efficient distributed programming model for mining useful patterns in big datasets. IETE Technical Review. 2013; 30(1): Vol 8 (22) September Indian Journal of Science and Technology

9 Sachin Kamley, Shailesh Jaloree and R. S. Thakur 17. Chen S-M. Forecasting enrolments based on fuzzy time series. Fuzzy Sets and System. 1996; 81(3): Hong T, Han I. Integrated approach of cognitive maps and neural networks using qualitative information on the World Wide Web: KBN Miner. Expert Systems. 2004; 21(5): Tung WL, Quek C, Cheng P, EWS G. A novel neural-fuzzy based early warning system for predicting bank failures. Neural Networks. 2004; 17(4): Vol 8 (22) September Indian Journal of Science and Technology 9