1.1 Introduction – Information Technology Industry and NIFTY IT
1.2 Impact of IT on the Indian Economy
1.3 Performance of IT industry in the Past Decade
2 DESIGN OF THE STUDY
2.1 Review of Literature
2.2 Statement of the Problem
2.5 Scope of the Study
2.6 Research Methodology
2.7 Research Tools
2.8 Data Collection Method
2.9 Plan of Analysis
2.10 Limitation of the Study
3 PROFILE OF THE COMPANY
3.1 Information Technology Industry
3.1.2 Industry Structure of the IT Sector
3.1.3 Growth of IT Sector
3.1.4 Market Size of IT Industry
3.1.5 Challenges in IT industry
3.1.6 Government Regulations
3.1.7 Future Prospects of IT Industry
3.1.8 Future Trends in 2021
3.1.9 India’s IT Sector – Growing Opportunities for Investment
3.2 National Stock Exchange
3.3 NIFTY IT Index
3.4 Porter’s Five Force
3.5 Company Profile
4 DATA ANALYSIS
4.1 Summary Statistics
4.1.1 Normality Test
4.2 Time Series Plot
4.3 Dicky Fuller Test
4.5 Model 2: ARIMA
4.6 Hypothesis Testing
5 SUMMARY OF FINDINGS, CONCLUSION, SUGGESTION
5.1 Summary of Findings
LIST OF TABLES
Abbildung in dieser Leseprobe nicht enthalten
LIST OF FIGURES
Abbildung in dieser Leseprobe nicht enthalten
CERTIFICATE BY THE GUIDE
Certified that this dissertation titled “FORECASTING INDIA’S NIFTY IT INDEX” is based on an original study conducted by Dhanya M L under my guidance. She has attended the required guidance sessions held. This project report has not formed a basis for the award of any other Degree / Diploma of any University or Institution.
SIGNATURE OF THE GUIDE :
NAME OF THE GUIDE : Dr. Rajveer S Rawlin
DESIGNATION : Assistant Professor
QUALIFICATION : PhD, M.B.A, B. Tech
I hereby declare that the Project Report on “FORECASTING OF INDIA’S NIFTY IT INDEX” under the guidance of Dr. Rajveer S Rawlin submitted in partial fulfillment of the requirements for the degree of POST GRADUATE DIPLOMA IN MANAGEMENT is my original work and the same has not been submitted earlier for the award of any other Degree/Diploma/Fellowship.
Place: Bangalore Dhanya M L
Date: Reg. No. 191111
I extend my special gratitude to our Director (Academics, Research & Administration) DR. MANASA NAGABHUSHANAM and Head-Department of Finance Dr. Triveni P for inspiring me to take up this project and also for their able guidance and support in completing this Project.
I wish to acknowledge my sincere gratitude and indebtedness to my project guide Dr. Rajveer S Rawlin of RAMAIAH INSTITUTE OF MANAGEMENT,
Bangalore for his valuable guidance and constructive suggestions in the preparation of the project report.
Student Name: Dhanya M L
Reg. No: 191111
Before investing in the stock, an investor analyses a particular company's stock to predict the return. Predicting stock price trends in the stock market is one of the most challenging and highly demanding tasks. Prediction of these trends will be beneficial for investors, professional analysts and researchers, managers. In the short run, short-term prediction of the stock market faces lots of ups and downs. In the long run, the forecast of the stock market is more accurate and statistically significant. Stock prices are highly fluctuating due to being impacted by several factors such as company fundamentals, interest rates, Supply, Political stability, international events, foreign exchange rate, etc. Investors make use of various prediction models to determine future stock prices. Fundamental, sentimental and technical analysis are widely used strategies in predicting stock prices. Fundamental analysis is useful in understanding the current value or intrinsic value of a company. Technical analysis helps in identifying the entry and exit positions in the stock market. Using machine learning coupled with statistical techniques is the new trend in predicting share prices.
NIFTY IT index captures the performance of the Indian Information Technology (IT) companies. The NIFTY IT index consists of 10 companies listed on the National Stock Exchange (NSE). IT sector in India has been recording tremendous growth over the years, where it accounts for a growth rate of 7.5 percent per annum. These stocks have a weightage of around 16.3% in the NIFTY index.
Time series analysis is a statistical tool used in forecasting the price movement of an asset. For the study, the dataset was collected over a period to identify trends, seasonality, irregularity, and cyclic nature of a stock to predict future prices. The purpose of the study is to evaluate the effectiveness of the ARIMA model for forecasting stock prices. For checking whether stationarity present in the variables Dicky Fuller Test was used in the study. If the variables are not stationary based on the results, logarithms of variables are collected. According to the results, time series analysis can be useful for predicting stock prices, and that the model used is statistically significant overall.
CHAPTER- I INTRODUCTION
The twenty-first century is known as the Digital age or age of information technology. The twenty-first century is known as the Digital age or age of information technology. Globally, India is a knowledge-based economy because of its impressive IT (Information Technology) sector or industry. In today's society, the use of information and communication (ICT) has made the world a better place to stay, making our lives easy than ever before. Today we use virtual technology from whether we want a piece of information or make a call to those that live miles away from us.
In India, two significant stock exchanges are present. These two stock exchanges are the Bombay Stock Exchange (BSE) SENSEX and National Stock Exchange (NSE), NIFTY. NIFTY is a benchmark index for the equity market of India. Stock market indices also provide a barometer to measure the direction or trend of market behavior. NIFTY 50 comprises 14 sectors of the country's economy and offers possible opportunities to the investment manager in one portfolio. NIFTY IT is one of those sectors. NIFTY 50 delivered negative returns in only two years since the beginning – in 2011 (-24.62%) due to high inflation, interest rate hike, rupee depreciation, and slowing economic growth and in 2015 (-4.06%) due to low corporate earnings and the Bharatiya Janata Party's loss in Delhi and Bihar state elections.
The NIFTY IT Index reveals how IT companies are performing in India. The NIFTY IT index comprises ten IT companies listed on the National Stock Exchange (NSE). These companies are top-performing companies in the IT Sector. It is necessary to forecast the NIFTY IT index because it helps an investor to decide its investments.
The prediction of NIFTY IT or any index is necessary cause it indicates which stocks are having tremendous growth or declining in the market, whether they will provide returns on investment or minimize the risk on the investment. It is also useful in assessing various companies' performance. The NIFTY IT Index is computed using a float-adjusted and market-capitalized method.
The stock market offers companies a means of raising capital for future expansion and enables an investor to earn a return on the investment. Before investing in the equity market, an investor analyses a particular company to predict its performance to get a return. Amid the fluctuations in the stock market, predicting price trends is the most complex and challenging tasks. This task is carried out by equity analysts, investors, researchers, and managers for various purposes. Stock market performance witnesses’ bullish trend, bearish trend, stagnant growth in a day, weak, month, year, In the short run, short term prediction of the stock market faces lots of ups and downtrend. In the long run, the stock market forecast is a comprehensive analysis of the price of the company's shares, its influential factors, its trends, and the risk- return ratio. To earn profits by investing in the stock market an accurate prediction is significant. To minimize the risk on the investment, forecasting the stock market is one of the finest strategies. Therefore, research on stock market prediction has significant theoretical importance on investment decisions and large-scale application perspectives.
Bengaluru, Hyderabad, Chennai, Delhi & NCR, Chandigarh, Mumbai and Pune, Kolkata, Ahmedabad, Kochi, Thiruvananthapuram are the key IT centers in India.
1.2 IMPACT OF INFORMATION TECHNOLOGY (IT) ON THE INDIAN ECONOMY
IT sector incorporates two main components: IT offerings and Business Process Outsourcing (BPO) – consulting (IT-BPM). The IT-BPM industry had contributed to India's GDP of 7.7% in 2020 from the least 1.2% in 1998. Information technology (IT) services make up 51% of total service sector revenues (banking, financial services, and insurance).
IT services and products have become necessary to improve productivity for other sectors. The IT sector is likely to accelerate the ramp-up, efficiency, and development of the Indian economy. This sector has helped India's economic growth and, over the years, has made governance more effective and responsive.
IT sector has a broad perspective for accelerating the growth, performance, and development of the economy of India further. It has contributed to the development of the Indian economy and has made governance more effective and reactive over the years. The services provided by the government are much simpler and less costly because of the technological progress of the government.
1.3 PERFORMANCE OF IT INDUSTRY IN THE PAST DECADE
Global markets have experienced fluctuations over the past decade, from 2009 to 2020. The global economy was just coming out of the financial downturn are the leading industries, amongst all sectors that have led the rally during the last decade of NSE NIFTY. IT was a lot more erratic in the middle years before picking up.
During the boom of cloud technology and digital services, Indian IT companies have struggled. These factors led to the downturn of the IT industry at the start of the decade. In addition to this limitation on IT budgets also contributed to the low economic growth of India.
The industry generated 120 billion U.S. dollars in the 2013-14 financial year. Since then, there has been a steady rise in the country's end-user IT spending, which by the end of 2016 amounted to US$ 79 billion. During the fiscal year, over three million direct jobs generated by the IT- BPM industry across the country. The 2018 year saw a structural revival of the IT industry, including the up-gradation of digital services through functions such as engineering, R&D (Research and Development). During the 2018 financial year, the IT-BPM service sector had an export value of more than three times the export value of software products and engineering services.
Abbildung in dieser Leseprobe nicht enthalten
Figure 1: NIFTY Sector Indexes Performance from 2010 to 2020
Source: Singhal, Harsh. “NIFTY in the Last DECADE (2010–2019) - Historical Analysis
In the present scenario, when the whole world is struggling to deal with the COVID-19 pandemic. The COVID-19 pandemic is impacting economies around the world. The Indian IT sector continues to display optimistic signs and can resolve this unparalleled tragedy.
This pandemic has impacted the orientation of IT activities towards new ways of doing business. It has speeded up operations and adopted a holistic approach to digital transformation, cost optimization, and automation.
CHAPTER II DESIGN OF THE STUDY
2.1 REVIEW OF LITERATURE
1. Enke, D., Grauer, M, & Mehdiv, N. (2011): Stock Market Prediction with Multiple Regression, Fuzzy Type-2 Clustering, and Neural networks
Stock market forecasting analysis poses many obstacles and possibilities, with individual stock or index predictions focused on either forecasting the amount (value) of future market prices or the direction of market price movement. This paper introduces a three-stage method of stock market prediction. This three-stage model was "Multiple Regression Analysis, Clustering of Fuzzy Type 2 and a Neural Network of Fuzzy Type 2 " models for stock price prediction.
The objective of the study is to estimate stock price by using the proposed hybrid model. In this research paper, authors used in their models "Fuzzy type-2 Clustering or Fuzzy type-2 Inference Neural Networks based on Differential Evolution (DE)" which to forecast stock prices by any examined researchers.
The data collection will be secondary. The input variables for the 30 years, beginning in the fiscal year 1980-81 through 2009-10, were collected from the S&P index. Only those businesses for which data are available are considered a sample for the entire throughout the analysis. Incorrect t-statistics and p-values were excluded from the study. The sample size considered was 200 firms.
Multiple regression analysis is used in this research paper to decrease the dimensionality of the variable set. It has also been useful to understand which of the input variables are relevant and driving price changes in the stock market direction. A multiple regression model has been performed on 25 financial and economic variables to define the relationship between these variables and the S&P 500 market price. In the second stage, "Type-2 Fuzzy Clustering based on Differential Evolution" is used to create a prediction model. A "Fuzzy type-2 Neural Network" is used for the third stage to implement the rationale for future stock price prediction. The results of the network simulation show that the proposed model outperforms the suggested model. The method "Incremental algorithms based on evolution" was used to solve the issue of clustering optimization. For model improvisation, the fuzzy neuronal network inference method was provided with evolutionary differential optimization.
Independent Variable: 3-month T-bill (T-Bill3) rate, 3-month Certificate of Deposit (CDR3) rate, past S&P 500 (SP500) Index price level, past Money Supply (M1) level, recent Industrial Production (IP) reading, and most recent Producer Price Index (PPI).
Dependent Variable: Market Price of S&P 500 Index
In multiple regression, p-values below 0.005 indicate that independent variables have resulted in a significant predictive capacity. The multiple regression relationship showed that positive changes in the T-bill, past index price level of S&P 500, recent Industrial Production, Money Supply in the past (M1) resulted in a positive impact on the stock market forecast of the S&P 500 Index price for the following month. And similarly, positive changes in the Certificate of Deposit and recent Producer Price Index (PPI) have negative effects on the stock market level forecast for the Index price of the S&P for the next month.
In the prediction analysis, using the "type 2 Fuzzy Clustering method" gives the cluster centers a better location and ultimately results in a better model of fuzzy law. This analysis will capture more ambiguity while providing a more robust response to inaccurate results. In the study, the "Fuzzy type-2 approach compared to a Fuzzy type-1 approach". Under which simulations resulted in a lower prediction error when it forecasted the stock market price. The model's R- square value also showed that the model describes 0.994, indicating that future equity market prices are at 99.4%. Finally, the simulation reveals that the proposed model outperforms the traditional stock market forecasting models.
2. Hassan, M. R., & Nath, B. (2005): Stock market forecasting using Hidden Markov Model: A new approach
Stock price, stock market prediction have been good financial subjects that attracted the researchers' attention for many years. In the AI community, it has been one of the biggest challenges. The stock market is highly volatile and non-linear by its very nature. Stock Price fluctuations depend upon several factors like equity, rate of interest, shares, options, warrants, mergers, and ownership of large financial corporations, etc.
The purpose of this research is to forecast the following day's closing price for a specific share of a company in the stock market using "Hidden Markov Model. In this paper, the "Hidden Markov Model" is used to predict some of the stocks of interconnected airline markets. The researchers have developed the "Hidden Markov Model" for forecasting time series. As a result of its ability to model dynamic systems, the "hidden Markov model" is widely used for the recognition of model and problem classifications. In this article, the researchers examined trends in the historical data set. Then they inserted the appropriate neighboring prices to the dataset and predicted the next day's exchange.
Data collection was secondary. The secondary market was collected from Southwest Airlines for 1.5 years (approximately) from September 17, 2002, to December 16, 2004. The observations of the input data are continuous rather than discrete. The sample size is 4 Airline Firms (British Airlines, Delta Airlines, Southwest Airlines, and Ryanair Holdings Ltd.)
Independent Variable: Opening, closing, highest price, and the lowest prices of a stock
Dependent Variable: Next day's closing price of a stock
Past 1.5 years data of 4 different Airlines stock prices were utilized to train the Hidden Markov Model and Competence of the model tested using observation of the last 3 months. After this step has been carried out, researchers predicted the following day's closing price of the stock price using the "Hidden Markov Model". Similarly, using the Artificial Neural Network (ANN), the current closing price of the shares of companies were predicted.
In this study, researchers assumed that the following day's stock price would follow the same historical data pattern of a stock price.
In this study, the researchers found that the use of the "Hidden Markov Model (HMM)" to predict the price of shares will have a significant statistical basis. The models that were considered were statistically significant in allowing for strong management of the new datasets.
It also has computational competency to build and forecast similar trends in the dataset with efficiency and effectiveness. In the research results, it is shown that the next day's stock price behavior follows the historical data set pattern of a stock price. The results obtained using "HMM" are significantly good and it offers a scope for the prediction of stock markets.
3. Hiransha, M., Gopalakrishnan, E. A., Menon, V. K., & Soman, K. P. (2018): NSE Stock Market Prediction Using Deep-Learning Models
Researchers in different fields have used neural networks under data mining techniques, as it is one of the advanced data mining techniques. Researchers used four forms of deep learning architecture in this research paper, i.e. 'Multilayer Perception (MLP), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN)’ to predict a company's stock price based on previously available data.
They have considered the closing price of two different stock markets on a day-wise basis that is 'National Stock Exchange of India (NSE) and New York Stock Exchange (NYSE)'. Data collection was secondary. Secondary data is derived from the S&P Index share price and input variables for 30 years starting from the financial year 1980- 81 to 2009-10. Historical data on the shares considered in the NYSE were collected from January 3, 2011, to December 30, 2016, and were reported in U.S. dollars ($). Most traded equities from three different sectors are the auto, banking, and NSE IT sectors. The sample size was five firms (Maruti, Hindustan Computers Limited (HCL), and Axis Bank from the National Stock Exchange stock market and Bank of America (BAC), and Chesapeake Energy (CHK) from the NYSE stock market).
Independent Variable: Day-wise closing price of considered stocks
Dependent Variable: Prediction of stock prices of Maruti, HCL, Axis Bank, Bank of America (BAC), and Chesapeake Energy (CHK)
Artificial Neural Networks is regarded as one of the non-linear statistical tools. Normally, it consists of three types of layers: input layers, masked layers, and output layers. In this study, the researchers extracted only the daily closing price of the shares considered as investors decide to buy the shares according to the closing price of the stock market. The extracted data were then subjected to normalization to unify the data within the range of 0 and 1. This procedure was performed to aggregate all stock data. Similarly, the desynchronization process was performed while acquiring the original predicted values. They trained four Artificial Neural Networks (ANN) such as ''Multilayer Perception (MLP), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN)'' with the stock price of Tata Motors from NSE. The models obtained were used for predicting the
stock price of Maruti, HCL, and AXIS Bank from the National Stock Exchange stock market and also for predicting the stock price of Bank of America (BAC) and Chesapeake Energy (CHK) from the New York Stock Exchange (NYSE) stock market.
The results obtained were compared with those of the ARIMA model, where it can be concluded that the DL models exceed the ARIMA model. Neural Network Architecture may detect non-linear patterns within the data. The MLP network succeeded in capturing the model in the case of Maruti, while RNN and LSTM did not recognize the seasonal model, and the model was almost captured by CNN. In HCL, the MLP model has been successfully captured the network to a certain degree and failed to capture the rest. The trend was nearly successfully acknowledged by RNN where LSTM, CNN failed to capture the system transition. MLP, RNN, LSTM, and CNN" captured the pattern to some extent in the case of Axis Bank and failed to capture residue." In the case of Bank of America, the "MLP, RNN, and LSTM" network captured patterns at the end and failed to capture them at the beginning, while "CNN" captured the pattern in almost three other networks compared to. "MLP and RNN" did not recognize the pattern to some degree in the case of Chesapeake Energy, and "LSTM" caught the pattern only at the end, not at the beginning, although "CNN" performed better compared to the other three networks.
Research suggests that CNN has outperformed other models as well as the existing linear model and can easily identify existing patterns in both stock markets. From the results, it has been observed that the models which were considered in the study are capable of identifying the pattern existing in National Stock Exchange (NSE) and New York Stock Exchange (NYSE) stock markets.
4. Cakra, Y. E., & Distiawan Trisedya, B. (2016): Stock price prediction using linear regression based on sentiment analysis
Given that the prediction of stock prices depends on demand, this is seen as a difficult task. Efficient Market Hypothesis (EMH) stock prices, however, were also triggered by the new information. This paper was influenced by previous research conducted by Bollen, Mao, and Zeng (2010, in which they have concluded that there is a connection between sentimental analysis such as opinions of people on Twitter and stock prices.
Researchers forecasted stock market prices of the Indonesian stock market using simple sentimental analysis. In this search, they used five algorithms as a vector support machine (SVM), Naive Bayes, Decision Tree, Random Forest, and Neural Network (with single- layer perception). In this article, the Naive Bayes and Random Forest algorithm used to classify tweets to calculate sentiment regarding a company.
This study used two data kinds – the opening and closing prices of companies whose prices fluctuate every day and which are quoted in Indonesia and The data contained opinions (shared via Twitter) on selected products manufactured by the companies that were considered. In this study, the data collected on the views expressed in the Tweets were classified into three categories: positive, negative, and neutral. Data collection was secondary. Secondary data was collected for two weeks from April 14th, 2015 to April 30th, 2015. The sample size of the study was 13 firms in Indonesia.
The company's share price was forecasted based on the results of the sentiment analysis. The linear regression approach was used to establish the forecast model. This predictive model was used to predict price movements, percentage, and price. In this study, share price analysis and sentiment analysis were combined to create a predictive model.
Independent Variable: opinions shared via Twitter and open stock price, the close stock price of the companies for each day
Dependent Variable: Price fluctuation predictions, margin percentage prediction, and price prediction
The results of the sentiment analysis showed that of the five algorithms used in the study, Naïve Bayes (56.50) and Random Forest (60.39) were the most accurate. These two forecasting models were then used to classify all tweets into three categories. Among the classified three categories of tweets (positive, negative neutral), positive tweets decrease the 𝑅2 value. In price fluctuation prediction, Naïve Bayes (67.37) and Random Forest (66.34) produced the highest accuracy in predicting whether the upcoming price will go up or down. In margin percentage prediction, few data fitted with the models produced (𝑅2 = 0). In price prediction, lots of data fitted with the models produced (𝑅2 = 1). The experiment shows that with 0.9989 and 0.9983 coefficient of determination, the use of prior stock price and hybrid function as predictors offers the best prediction.
5. Hassan, M. R., Nath, B., & Kirley, M. (2007): A fusion model of HMM, ANN, and GA for stock market forecasting
Researchers forecast the behavior of the financial market in this research paper by introducing a fusion model by integrating the Hidden Markov Model (HMM) and Artificial Neural Networks (ANN) and Genetic Algorithms (GA) to predict the behavior of the financial market. This research discussed the weakness of the basic model (Hidden Markov Model (HMM) used in the previous study, (Hassan and Nath, 2005) does not effectively recognize the associated models. HMM suffers from low initialization and interpolation settings by localizing only a single sequence of observations.
Apple Computer Inc., International Business Machines Corporation (IBM), and Dell Inc. have used stock values by taking into account the four characteristics of these stocks: open, high, low, and near prices. Secondary data from the period from 10 February 2003 to 10 September 2004 was obtained.
In this analysis, the Artificial Neural Network (ANN) was used to transform the Hidden Markov Model sequences of input observation (daily market prices) (HMM). The initial parameters of the Hidden Markov Model were optimized using genetic algorithms. Later, qualified HMM is used in the historical data to classify and locate comparative trends.
In this analysis, two alternative techniques are explored for building one-day forecast value based on the HMM defined trends. The researchers interpolated the near-trend data values in the first approach to obtain the prediction value obtained by Hassan and Nath in the previous research (2005). In the second approach of the proposed model, they did not inter- palate the values as is done in the Hassan and Nath (2005).
Independent Variable: Open, high, low, and close price of selected stocks
Dependent Variable: Next day's closing price
The accuracy of the proposed merger model for Apple Computer Inc and Dell Inc is close to that of the ARIMA model. The proposed fusion model was slightly more accurate for International Business Machines Corporation (IBM) than for ARIMA. The forecasts obtained were compared with a standard forecast system. The result shows that the greatest accuracy was created by the predictive model. The proposed model is also better than the production of the base model (Hassan and Nath 2005), where only one HMM is used in a new approach to stock market forecasting.
Before the result is achieved, the performance of the fusion method is measured in terms of Average Absolute Percent Error (MAPE). The proposed model can be used without analyzing the dataset (i.e., seasonality test, regime analysis, or cycle analysis before adopting the model) before the forecasting stock market. By comparing the obtained result with a popular statistical forecasting tool, it has been observed that the forecasting ability of the fusion model is as good as that of the ARIMA model.
6. Yeh, C. Y., Huang, C. W., & Lee, S. J. (2011): A multiple-kernel support vector regression approach for stock market price forecasting
Stock market prediction is a difficult task as it depends on both economic and non-economic factors. In previous stock market research, researchers proposed a linear model based upon a statistical approach. These are often inadequate for forecasting the stock market and are stationary. To overcome these constraints, researchers proposed a regression model. This regression model integrates multi-kernel learning and Support Vector Regression (SVR) to deal with stock price forecasting. Support Vector Regression (SVR) is a non-linear regression method. Based on kernel it attempts to locate in high-dimensional space a regression hyperplane with small risk. This research incorporates minimum sequential optimization and gradient projection methodology. They also developed a two-stage multiple-kernel learning algorithm. This algorithm allows various parameters to be combined to improve overall system performance.
The gathered data was secondary. The datasets were considered from the Taiwan Capitalization Weighted Stock Index (TAIEX). It is a stock market index for companies traded on the Taiwan Stock Exchange. Secondary data was collected from October 2002 to December 2005 were collected.
The user does not need to specify the hyperparameter settings in advance in this algorithm. This feature will improve the performance of the system. A regression model that integrates multiple-kernel learning and Support Vector Regression (SVR) is proposed in this study to deal with stock price forecasting. SVR is required for Kernel function hyperparameters to be tuned manually. A two-phase multi-kernel learning algorithm was developed to optimally combine multi-kernel arrays for SVR.
The performance of the proposed method has been compared with that of other techniques, i.e., single kernel support vector regression (SKSVR), autoregressive integrated moving average (ARIMA) model, and TSK fuzzy neural network (FNN). In the contrast between Multiple- Kernel Support Vector Regression (MKSVR), Single Kernel Support Vector Regression (SKSVR), and Auto-Regressive Integrated Moving Average, Multiple-Kernel Support Vector Regression (MKSVR) performed better (ARIMA). It has been found from experimental studies that this approach performs better than other methods and provides good financial time series forecasting accuracy.
7. Gupta, A., & Dhingra, B. (2012): Stock market prediction using Hidden Markov Models
Profit prospects for arbitrage are one of the motivating factors for doing stock market research. Researchers provided the Maximum a Posteriori Hidden Markov Model method in this research paper for forecasting stock values given historical data for the next day. To model the stock data as a time series, the Continuous Hidden Markov Model (CHMM) was used.
TATA steel, Apple Inc., IBM Corporation, and Dell Inc. were considered four distinct stocks to test the strategy. A separate HMM is being trained for each stock. Secondary data from 12 August 2002 to 4 November 2009 were collected.
Independent Variable: Fractional change in stock value and the intra-day high, low, opening, and closing values of the stock
Dependent Variable: Overall possible stock values for the next day
They have assumed in the current approach that the model for one specific stock is independent of other stocks on the stock market and four secret underlying states that emit four observable observations, i.e., fractional high, fractional low, fractional transition. The probabilities of pollution are modeled as Gaussian Mixture Models (GMM) which can be applied in the future to forecast stock prices. The choice of attributes or feature selection is important in this approach. The knowledge is quantized in the current study to form observation vectors for a full day. The regular fractional change in the stock value and the fractional deviation of the stock's intra-day high and low values were used to train the continuous HMM and then to make a Maximum A Posteriori (MAP) decision for the next day on all possible stock values.
Fractional modifications have been used to model the changes in stock details that have remained consistent over the years. In this analysis, the fractional deviation of both high and low values is calculated as a good indicator since it provides a direction for the volatility of stock markets.
They tested the method on several stocks and compared the performance of the Hidden Markov Model to some of the current stock forecast models of the Hidden Markov Model-fuzzy model, ARIMA, and Artificial Neural Network (ANN). The precision of the proposed model is verified using the Mean Absolute Percentage Error (MAPE) metric. They also maximized the probability of a set of observations for all potential future values expected. For Apple Inc. and IBM Corporation, the Maximum a Posteriori-Hidden Markov Model (MAP-HMM) outperforms other models and performs comparably for Dell Inc. By using discrete Limit, a Posteriori-Hidden Markov Model (MAP-HMM) model to display the actual value along with the Tata Steel and Apple Inc. forecast value, it was found that this model outperformed Tata Steel and Apple Inc. in forecasting stock prices.
While it is assumed in this model that the stock of one company is independent of other stocks in the market, these stocks are highly correlated with each other, and stocks in other markets are also correlated to some degree.
8. Ouahilal, M., Mohajir, M. E., Chahhou, M., & El Mohajir, B. E. (2016): Optimizing stock market price prediction using a hybrid approach based on HP. filter and support vector regression
In this research paper, researchers proposed a novel hybrid approach incorporating Support Vector and Hodrick-Prescott filters. It is combined to optimize forecasting of Integrated Asset Management (IAM) by taking into account the date, open price, close price, high price, low price, and stock price value.
Routine data were collected from 2004 to 2016 as part of this analysis. The sample size is 2840. The objective of the study is to enhance the forecasting of stock prices by understanding the historical data of the IAM (Support Vector and Hodrick-Prescott filter). In this analysis, the closing price is the input of forecasting. It is the final price in a day for securities trades at a stock exchange.
Independent Variable: Date, open price, close price, low price, and Volume
Dependent Variable: Close price for different amounts of time in the future
In the stock market, the stocks fluctuate each day. The change in stock price will be difficult to understand given the fluctuation in the same. Noise filtration is critical in getting a better understanding of the trend. In this research, Hodrick-Prescott had used for this purpose. It was more responsive to long-term fluctuations. It uses the HP. filter to decompose the time series into several series with similar frequencies. In this analysis, enhancing and optimizing SVR model predictions is built with the help of the HP filter. It will parse and normalizes the data by filtering and eliminating all existing noise in financial time series to help generate better result accuracy. The error rate is calculated between the real and the predicted value. The Mean Average Percentage (MAPE) error in this study is used to calculate the error.
They have conducted several experiments using Maroc Telecom (IAM) financial time series to perform this approach. The Hodrick-Prescott filter is used in the study to filter the noise, standardize the data value of each attribute separately and decompose time series into multiple series with similar frequencies. In this analysis, to enhance and optimize SVR model predictions, and HP. the filter is used. It will parse and normalize data by filtering and eliminating all existing noise in financial time series. It helps to deliver the results more accurately.
The combination of the SVR model and the HP filter delivers better results in this study. It is because the MAPE error by this model is the lowest of all the proposed error rates. It is a powerful predictive tool for the stock price and financial time series as it provides better results. The stock IAM price and Hodrick-Prescott filter (HP filter) components trend is ultra-smooth and performed very well in predicting the future price direction of Integrated Asset Management (IAM). The cycle component's extreme values suggested a positive trend reversal.
This study demonstrated that applying the regression model to the financial sector is considered a sound approach in the context of stock market forecasting
9. Bini, B. S., & Mathew, T. (2016): Clustering and Regression Techniques for Stock Prediction
Prices on the stock market are volatile. It is a difficult task for sellers and buyers to foresee the future value of the market. In this study, researchers proposed an analysis system that allows individuals to identify more profitable businesses using the data mining approach (clustering and regression).
Clustering is descriptive and predictive analysis is predictive regression. The predictive method used in the study (regression) projects data values where the descriptive method (clustering) techniques recognize the relationship between them.
Data mining techniques designed in this study is to overcome uncertainties in the stock market. In this study, the clustering technique identified the top companies on the stock market. The main objective is to forecast the future stock price using the regression techniques for those identified companies.
From National Stock Exchange (NSE) the dataset is gathered (NSE). Companies such as Wipro, TCS, Rolta, Polaris, Persistent, NIIT TECH, Naukri, Mindtree, INFY, and HCL TECH collected this data. Data sets are collected six-month cycle.
The collected dataset undergoes various techniques of clustering. K-means algorithm, Agglomerative algorithm, EM algorithm, and density-based DBSCAN (Density-Based Spatial Clustering of Applications with Noise) were the algorithms used in this study. K-means algorithm is used in the partitioning technique and the agglomerative algorithm is used in the hierarchical technique.
The EM algorithm used in the model-based technique and the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) approach was used in the density-based method. The validation index has also been used to evaluate the efficiency of various clustering techniques, such as partitioning technique, hierarchical technique, model-based technique, and density-based technique. The C-Index, Jaccard Index, Rand Index, and Silhouette-Index is used in the validating index.
While analyzing the results, it can be analyzed that obtained by the indexes, the K-means and EM algorithm shows better performance than agglomerative and DBSCAN (Density-Based Spatial Clustering of Applications with Noise) since the C-index of K-means and EM was having a value near to zero whereas Jaccard Index, Rand, and Silhouette-index was having a value near to 1. From the prediction results, it has been observed that the future price of the prediction month was approximately equal to the actual stock prices of the predicted month.
For future work, researchers have suggested that forecasting the stock market even better an online stock prediction system to be developed by using Partitioning based or model-based technique along with multiple regression techniques.
10. Patel, J., Shah, S., Thakkar, P., & Kotecha, K. (2015): Predicting stock market index using a fusion of machine learning techniques
Researchers concentrated on predicting the potential value of the stock market index in this research paper. Although the stock market is fluctuating, technical analysis believed that almost all information is related to stocks will reflect in recent prices. For this reason, it is easy to forecast stock prices. It was observed from the previous study that current stock prediction methods only use one layer of prediction that takes statistical parameters as inputs and gives the final result. Researchers developed a two-stage prediction model to overcome this limitation, and it can also decrease the error in a phase-wise way.
This paper proposes a two-stage fusion method in the first stage that involves Help Vector Regression (SVR), Artificial Neural Network (ANN), Random Forest (RF). In the second stage of the analysis, SVR fusion prediction models are used, resulting in the fusion of SVR-ANN, SVR-RF, and SVR-SVR.
Two indices were selected from India's stock markets, CNX Nifty and S&P Bombay Stock Exchange (BSE) Sensex, for experimental evaluation. For experiments, historical data from 10 years considered for experiments from January 2003 to December 2012. For three types of the dataset, forecasts are made on 1-10, 15, and 30 days in advance.
Independent Variable: Close, high, low, and opening prices of CNX Nifty and S&P BSE Sensex.
Dependent Variable: Prediction results for 1-10, 15, and 30 days.
To assess the performance of these prediction models, Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), relative Root Mean Squared Error (MMSE) and Mean Squared Error (MSE) were used as evaluation measures.
In this research, based on average predictive performance, the researchers also compared the performance of single-stage models with two-stage models for CNX Nifty and S&P BSE Sensex. As the prediction made for a greater number of days in advance, the comparison between these two models is substantially evident. The predictive performance of these hybrid models is compared with the scenarios at a stage where ANN, RF, and SVR are used on them and also proposed two-stage fusion models. SVR-SVR models exhibit a moderate improvement over SVR.
In the first stage of the results, it was observed that tuning experiments for each Support Vector Regression, the transformation of input space through the RBF kernel performs better than the polynomial kernel. The SVR-ANN and SVR-RF models outperform the Artificial Neural Networks (ANN) and Random Forest (RF) models for all forecast tasks for both data sets. The SVR-ANN model outperforms all other models for stock market indices. SVR-SVR outperforms Support Vector Regression for all prediction tasks, except for prediction tasks up to 3-4 days in advance (SVR).
From the results, it has been observed that two-stage hybrid models perform significantly better than single-stage prediction models.
2.2 STATEMENT OF THE PROBLEM
India's IT industry has made great strides over the past decade and has supported the country's economic growth. The IT sector is on a growing trend. India's IT sector is expanding rapidly. Investing in stocks of the IT sector is very common among investors. Due to the COVID-19 pandemic, the IT sector's stocks were more volatile than usual. Increased potential investment returns generally go hand-in-hand with higher risk. The risk associated with such investments may be very high or low, but the element of risk depends upon the performance of that stock in the market, the value of the company, and the industry to which the company belongs. Therefore, an investor would benefit from a precise prediction of the stock price, and he should analyze the stock price trends and risk and return associated with the stocks. Given the IT sector has produced stable returns over time it would be useful to forecast the IT index from an investor’s perspective.
1. Forecasting prices of India’s NIFTY IT index
2. Comparing forecasted values with actual prices
Null Hypothesis (Ho) – There is no significant difference between the forecasted values and the actual values.
Alternative Hypothesis (H1) – There is a significant difference between the forecasted values and the actual values.
2.5 SCOPE OF THE STUDY
In this study, India’s Information Technology (IT) index prices are studied weekly for the past 10 years from 2011 to 2020. This is used to forecast prices for 2021.
2.6 RESEARCH METHODOLOGY
Historical data for the NIFTY IT index is obtained from 2011 to 2020. To perform forecasting ; Time ; Series Modelling is used. An appropriate ARIMA model is fit to forecast prices.
2.7 RESEARCH TOOLS
Time Series Analysis: Time series analysis helps in forecasting future prices by looking at past trends.
ARIMA Model: In Time Series forecasting, Autoregressive Integrated Moving Average
Modelling (ARIMA) is used for stock price prediction. The Autoregressive Integrated Moving Average Model is a form of regression analysis that forecasts a dependent variable from its past values or from its past values and other independent values and their past values.