Materials Today: Proceedings xxx (xxxx) xxx

Contents lists available at ScienceDirect

Materials Today: Proceedings

journal homepage: www.elsevier.com/locate/matpr

Stock market analysis and prediction using time series analysis

Christy Jackson J. a,⇑, Prassanna J. a, Abdul Quadir Md. a, Sivakumar V. b
a
Vellore Institute of Technology, Chennai, Tamil Nadu 600127, India
b
Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology, Chennai 600062, India

a r t i c l e i n f o a b s t r a c t

Article history: Over the years the stock market has been considered a very risky investment by people around the globe.
Received 3 November 2020 This project aims to understand the historical data of the stock market and derive analysis from it to
Accepted 12 November 2020 reduce the gap of knowledge between the market behavior and the investor. A stock data comprises of
Available online xxxx
a lot of statistical terms which are difficult to understand by a normal person who wants to step into
stock market investments, this project aims at reducing the gap of knowledge. This study aims to tell
Keywords: the market scenario of the future by supporting it with statistical answers. Stock market volatility,
ARIMA
Daily returns, cumulative returns, Correlations between different stocks, Sharpe Ratio of the stocks,
Monte-Carlo
Fbprophet
CAGR value, Simple Moving Average are some important statistical terms to understand the risk of the
Cumulative return investment in the stocks. For the prediction of the future behavior of stocks work on ARIMA models,
Volatility Monte Carlo Method and Forecasting using Facebook’s prophet library have been used here.
CAGR Ó 2020 Elsevier Ltd. All rights reserved.
Sharpe ratio Selection and peer-review under responsibility of the scientific committee of the Emerging Trends in
Materials Science, Technology and Engineering.

1. Introduction were also not so easily available to the people. But Data Scientists,
statisticians, and tried to reduce the gap bit by bit over the years.
Time Series is a series of continuous data point’s index in order Now a day’s mutual fund applications use AI models, use certain
of date and time. Data connected through a continuous series of statistic benchmarks to make it easy for a newcomer to understand
time data. Analysis of time series data is done to extract meaning- it to some extent.
ful statistics and other characteristics of data. After statistical cal- Moreover, there has been a lot of studies done by people around
culations of time series data and after data analysis one can the globe to predict the future prices of the stocks but the stock
understand the behavior of the stock and deduce the amount of market does not solely depend on the historical data. It is also
risk involved in it before making any investments. affected by the sentiments of the people, which depend on some
Time series forecasting is a step further in making a valuable future events and so one cannot predict future events with 100%
step towards the understanding of future behavior. It refers to accuracy.
the use of models to predict future values based on previously The objective of this study is to understand and predict the
observed values. Models used in this study are Auto-Regressive stock behavior through statistical calculations and visualizations
Integrated Moving Average (ARIMA) model, Augmented Dickey- of historical data analysis. These objectives are:
Fuller Test tells about Stationarity of Time Series Data, Monte Carlo
Model is used to tell possible future predictions of stock for some a) Analysis of change in prices of the stock over time.
time, prophet library by Facebook is very robust in processing b) Comparative analysis of the daily and cumulative return of
the time series data and giving future predictions based on a daily the stocks.
trend of data, a weekly trend of data and yearly trend of data. c) Analysis using the Simple Moving Average of various stocks.
Stock market data over the years was considered to be very d) To find the correlation between different stocks’ closing
unpredictable and investment in the stock market was not very prices and daily returns.
difficult for newcomers. Moreover, so much data and analytics e) To find the Sharpe Ratio of the stocks and to learn how it can
be a helpful parameter while making investments.
f) To find the compounded annual growth rate of the stocks
⇑ Corresponding author.
over the last 10 years.
E-mail address: [email protected] (J. Christy Jackson).

https://doi.org/10.1016/j.matpr.2020.11.364
2214-7853/Ó 2020 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of the scientific committee of the Emerging Trends in Materials Science, Technology and Engineering.

Please cite this article as: J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al., Stock market analysis and prediction using time series analysis, Materials
Today: Proceedings, https://doi.org/10.1016/j.matpr.2020.11.364
J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al. Materials Today: Proceedings xxx (xxxx) xxx

g) To predict future stock behavior and future prices using High: Depicts the Highest value gained by the stocks on a par-
algorithms. ticular day.
Low: Depicts the Lowest value gained by the stocks in a partic-
ular day.
2. Related work
Adj. Close: The adjusted closing price is calculated after analy-
ses of the stock’s dividends, stock splits, and the new stock offer-
The motivation behind this topic of study was the gap of knowl-
ings which determine a new value of the stock know as adjusted
edge most people have before they start investing in the stock mar-
price.
ket. Every year smart investors make a good chunk of money out of
Volume: Volume, or trading volume, is the amount of a security
the stock market. Self-made billionaire Warren Buffet is one such
that was traded during a given time.
example who earned most of his wealth through smart investing.
Prediction of future stock behavior is another motivation. This
gap of investment knowledge needs to be reduced for a common 4. Statistical parameters
man.
The author Banarjee D. [1], has done his study on forecasting of 4.1. Daily stock return
National Stock Exchange data using the ARIMA model and has
done a comparative study on different values of p, d, and q on This parameter tells us how much the stocks gained or lost per
ARIMA and did a validation check of the forecasted stock price with day per share. It is calculated by subtracting the previous day’s
the actual stock price. In this study, ARIMA (1,0,1) gave the best fit closing price from todays’ closing price.
compared to other models. The authors Viswam, N., & Reddy, G. S.
[2], did a study on historical stock market data predicted future 4.2. Stock volatility
prices using the ARIMA model and they used the MACD model to
better analysis of the data. Volatility is a measure of the dispersion of returns for a given
The authors Angadi, M. C., & Kulkarni, A. P. [3], used the ARIMA stock or the market index. Mostly, the higher the volatility, the
model for prediction of stock prices, for p, d, q values they used riskier is the stock. It is often measured as either the standard devi-
auto ARIMA to get the best fit for the model. Obtained results ation or variance between returns from that same stock or the
reveal that the ARIMA model has a strong tendency to make short market index.
time predictions. The authors Devi, B. U., Sundar, D., & Alli, P. [4], In the stock markets, volatility is often associated with big
used NIFTY MIDCAP 50 as the index and selected the top four MID- swings in the price in either direction of the trend. For ex., when
CAP companies. They used ARMA and ARIMA models to predict the market gains and loses value more than one percent over a
future stock prices and used AIC and BIC criteria to get the best specific time, this is known as volatility of a market.
fit for the model. The authors Varghese, A., Tarhen, H., Shaikh, A., pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
DailyVolatilityFormula ¼ Variance ð1Þ
Banik, P., & Ramadasi, A. [5], used ARMA, Moving average, an
ANN model to predict the future prices of the stock and used max- pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
imum likelihood estimator and Yule-Walker estimation to check AnnualVolatilityFormula ¼ 252  Variance ð2Þ
the validation of their models.
The authors Sharma, A., Modak, S., & Sridhar, E. [6], using the 4.3. Cumulative stock return
LSTM model from RNN to predict the random nature of stock prices
in the future. The MSE value of the model came out to be signifi- This parameter tells us about how much the stocks gained or
cant and improved. lost per share over time, independent of the time. Cumulative
The authors Selvin, S., Vinayakumar, R., Gopalakrishnan, E. A., Return is equal to:
Menon, V. K., & Soman, K. P. [7], used NSE listed companies as
the stock price data. They used a sliding window approach on ðCurrentPriceÞ  ðOriginalPriceÞ
ð3Þ
non-linear model RNN, LSTM, and CNN and linear model ARIMA. OriginalPrice
The non-linear model outperformed the linear model during error
calculation. The authors Mondal, P., Shit, L., & Goswami, S. [8], did a 4.4. Compounded Annual Growth Rate (CAGR)
study on the effectiveness of the ARIMA model in forecasting secu-
rity values. They used Indian Stock market data from NSE for the CAGR is the rate of return by which tell us that this rate would
analysis. AIC has been used for selecting the best ARIMA model. be required for a company to grow from its starting value to its
The author’s Wang, J., & Wang, J. [9], used principle component ending value, it is assumed that the profits gained were again
analysis and Stochastic time effective neural networks for forecast- invested at the end of each business year of an investment firm.
ing the future and used MAE, MAPE, MSE, and RMSE to calculate  1=t
the performance of the model. V final
CAGR ¼ 1 ð4Þ
V Begin
3. Dataset description Where:
CAGR = compound annual growth rate
Data used for this project is day-wise historical time series data Vbegin = beginning value
of stock of the past 10 years in numerical form. Vfinal = final value
Dataset size – 10 Business years T = time in years
Data Source used – Yahoo finance
Dataset imported from yahoo finance consists of 7 columns 4.5. Correlation of stocks
consisting of Open, High, Low, Close, Adj. Close, Volume, and Date
as the index. Correlation(r) is a statistical measure that tells us the amount to
Date (Index of the Dataset): Dates of all Business Days in a year. which two variables move about each other. In finance, the corre-
Open: Depicts the Opening price of the Security. lation can measure the movement of a stock price with the stock
Close: Depicts the Closing price of the Security. market’s benchmark index.
2
J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al. Materials Today: Proceedings xxx (xxxx) xxx

P  
Rp = return of portfolio
ðX  X ÞðY  Y Þ
r ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffirffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð5Þ
P  2   2
Rf = risk free return rate
XX Y Y rp = standard deviation of the portfolio excess return

Where: 4.7. Simple moving average

r = the correlation coefficient

X = the average of the observations of variable X The simple moving average (SMA) is a simple method for tech-

Y = the average of the observations of variable Y nical analysis that smooths out the price data of the stock market
by updating the average price of stocks constantly. This average is
calculated over a specific period, like 10 days, 30 min, 20 weeks, or
4.6. Sharpe ratio
any period the investor chooses. Moving average techniques are
very popular and can be calculated for any time frame, suiting both
Sharpe ratio is a measure of risk-adjusted returns of a financial
long-term investments and short-term investments.
portfolio. Sharpe ratio is a measure of excess return one gets over
the risk-free rate, relative to its standard deviation [10]. A1 þ A2 þ    þ An
SMA ¼ ð7Þ
Rp  Rf n
SharpeRatio ¼ ð6Þ Where:
rp
A_n = The price of a stock during a period n
Where: n = the number of total periods

Fig. 1. Plots of stock prices overtime.

Fig. 2. Combined plot of stock for price comparisons.

3
J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al. Materials Today: Proceedings xxx (xxxx) xxx

Fig. 3. Comparative analysis of volume traded with a rolling mean of 30.

Fig. 4. Daily returns when plotted.

Fig. 5. Kernel density plot to compare volatility.

4
J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al. Materials Today: Proceedings xxx (xxxx) xxx

Table 1
Daily returns of stocks.

Date Reliance HDFC TCS INFY

21-05-2010 0.0044 0.013183 0.01548 0.007306
24-05-2010 0.027221 0.001095 0.00153 0.007223
25–05-2010 0.036181 0.0114 0.025153 0.025765
26–05-2010 0.022726 0.013306 0.054964 0.08604
27–05-2010 0.014087 0.037669 0.005217 0.00851
... ... ... ... ...
13-05-2020 0.021217 0.02895 0.000077 0.009452
14-05-2020 0.040429 0.036598 0.024261 0.051862
15-05-2020 0.016331 0.00621 0.004968 0.008889
18-05-2020 0.012779 0.057986 0.027841 0.017783
19-05-2020 0.022107 0.007171 0.001568 0.007079
2461 rows  4 columns

Fig. 7. Correlation of daily returns using heat map.

Fig. 6. Box plot to compare the volatility of stocks.

5. Analysis

Prices of stocks can be seen increasing with an upward trend in

plots for the 4 companies we selected for the selected period.
We can observe two Tech companies TCS and INFOSYS have
close similarities in the plots (Fig. 1) that there must some correla-
tion between these stocks because of their behavior whereas both
of them have a big difference in the Stock prices which can be
observed on the y-axis (Fig. 2. Fig. 3).

Fig. 8. Correlation of closing price using heat map.

Table 2
Cumulative return of the stocks over the years.

Date Reliance HDFC TCS INFY

20-05-2010 1.000000 1.000000 1.000000 1.000000
21-05-2010 0.9956 0.986817 0.98452 0.992694
24-05-2010 1.022701 0.985737 0.983014 0.999865
25-05-2010 0.985699 0.9745 0.958288 0.974103
26-05-2010 1.0081 0.987466 1.010959 1.057914
... ... ... ... ...
13-05-2020 3.485026 6.451662 7.129287 5.389721
14-05-2020 3.34413 6.215545 6.956325 5.110199
15-05-2020 3.398742 6.176945 6.921769 5.064777
18-05-2020 3.355308 5.818771 7.114477 5.154845
19-05-2020 3.281134 5.777042 7.12563 5.191338
2462 rows  4 columns

5
J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al. Materials Today: Proceedings xxx (xxxx) xxx

Fig. 9. Simple moving average for 10, 50, and 200 days.

5.1. Calculating daily returns

Daily returns can be considered as the value gained or lost by

stock to the previous day. We can see a graph of daily returns with
a mean close to zero (Fig. 4. Fig. 5. Table 1).
From here we analyzed and tell that Reliance stocks are most
volatile, and HDFC Bank Stocks are least volatile.
Through the box plot, we can observe that INFOSYS has a wider
graph due to some outliers, neglecting those above analysis is com-
pletely supported (Fig. 6).

5.2. Calculating cumulative return of the stocks

In the last 10 years, TCS gave us the maximum return, the

amount invested in TCS stocks has surged up to 7 times in the last
10 years. While Reliance gave us minimum return where prices
surge up to 3.3 times approx. With the least volatility among the
stocks HDFC has Stock seems to very promising with its returns
(Table 2).

5.3. Calculation of compounded annual growth rate of stocks and the

market index

Fig. 10. SMA of reliance.

In the past 10 years, TCS attained an excellent CAGR value and
Reliance have the least Annual growth rate. All four stocks have
6
J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al. Materials Today: Proceedings xxx (xxxx) xxx

Table 3
Annual Sharpe Ratio of last 9 years from 2011 to 2019.

Date Reliance HDFC TCS INFY

31-12-2011 1.290702 0.392527 0.124830 0.590615
31-12-2012 0.576007 2.691502 0.098416 0.444074
31-12-2013 0.068441 0.272582 2.776967 1.834415
31-12-2014 0.207587 1.953733 0.590855 1.130394
31-12-2015 0.353913 0.452002 0.383376 0.647585
31-12-2016 0.084834 0.356861 0.238275 0.484872
31-12-2017 2.624157 3.835127 0.588217 0.148690
31-12-2018 0.63751 0.527094 1.489296 1.08344
31-12-2019 1.063034 0.629196 0.364209 0.258321

Fig. 13. PACF plot of reliance stock.

Fig. 11. Annual sharpe ratio bar plot of last 9 years from 2011 to 2019.
tended to perform better than what the market has received over
the last 10 years.
Table 4
Results of Dickey-Fuller test.
5.3.1. Finding correlation
Parameters Values It can be observed, market index NIFTY has a good correlation
Test Statistics 2.036163 with the stocks. Moreover, both the tech companies seem to have
p - value 0.581605 a correlation which tends to slightly weak (Fig. 7. Fig. 8).
No. of lags used 21
The market index is closely related to all four companies. More-
Number of Observations used 2440
Critical Value (1%) 3.962485 over, we can observe the highest correlation between tech compa-
Critical Value (5%) 3.412291 nies (Fig. 9).
Critical Value (10%) 3.12811 Moving averages tells us that this is an average or minimum
possible return that we are likely to get shortly. Moving averages
change slowly in comparison to the actual price.
From the SMA plot of Reliance stock over 10, 50, and 200 days
we can analyze that 10 MA removes the noise from the plot while
50 days MA smoothens it over some time and help understand
market behavior over the next 50 days and the 200 days MA gives
us the trend.
As per the mean reversion strategy, it tells us that whenever the
actual price plot line crosses the MA line it states that the market is
going to rebound again towards the mean. So, in the above plot, we
can observe that in March-April 2020 when the prices fell below
the MA200 they are bound to arise in the future towards the mean.
SMA has been very helpful over the years for investing (Fig. 10).

5.4. Calculating sharpe ratio

Sharpe ratio is also known as Risk-Adjusted Return is the mea-

sure of the excess return of a portfolio of stocks over the risk-free
return. It tells us whether the risk taken to earn the return over the
risk-free possible return, from Investment security is worth taking
Fig. 12. ACF plot of reliance stock. a risk or not. It helps us in choosing the right stocks with minimum
7
J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al. Materials Today: Proceedings xxx (xxxx) xxx

Fig. 14. Seasonal decomposition of reliance stock into trend, seasonal, and residual components.

Fig. 15. ARIMA forecasting plot of future values with 95% confidence.

Fig. 16. Monte Carlo simulation of reliance of 1 business year.

risk good amount of return. Sharpe Ratio is a term that is calculated
annually, and it changes over time (Table 3).
What Sharpe Ratio is considered Good? Table 5
ASR < 1 – Bad (Not Worth taking the risk) Results of the ARIMA model.
1 < ASR < 2 – Acceptable
Parameters Values
2 < ASR < 3 – Good
MSE 0.013511207
ASR > 3 – Excellent
MAE 0.085270415
Fig. 11. RMSE 0.116237716
MAPE 0.011900728
6. Price prediction

6.1. Augmented dickey-fuller test which type of test is being used on the data, which is usually sta-
tionarity or trend-stationarity. For a large and complex set of time
In the AD Fuller test, time-series data is tested for the null series models AD Fuller test is used.
hypothesis. The null hypothesis states that a unit root is found in The ADF statistic measure which is used for the test is a nega-
time-series data. An alternate hypothesis depends on whether tive number and the more the number is negative, the hypothesis
8
J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al. Materials Today: Proceedings xxx (xxxx) xxx

Fig. 17. Prophet forecast of reliance of next one year.

will be rejected more strongly which means that there is a unit

root with some level of confidence. Results of the AD fuller test
are given in Table 4. p-value > 0.05 implies data is non stationary.

6.2. Finding ACF & PACF

The plot of ACF is a bar graph of coefficients of correlation which

is plotted between a time series and lags with itself. The PACF plot
is a plot between the series and the lags of itself which gives partial
correlation coefficients. ACF and PACF plots are used to identify the
number of AR and MA terms (Fig. 12. Fig. 13).

6.3. Seasonal decomposition of time series

The decomposition of Time series data into Seasonal, Trend, and

Residual component is known as the seasonal decomposition
(Fig. 14) [11].

6.4. ARIMA model

6.4.1. Auto-Arima function

Auto Arima function does the job of finding the best possible
values of p, d, q for our model. The model with the lowest AIC value
is selected and gives us the required values of p, d, q for our ARIMA
model. The results gave p = 0, d = 1, and q = 1 as the best fit for the
model.
ARIMA stands for Auto-Regressive Integrated Moving Average.
Here, the ‘‘Auto-Regressive term” means the lags in a stationary
series of a forecasting equation, ‘‘moving average term” means
the lags in the forecast errors, and ‘‘integrated” means the time ser-
ies data is made stationary by using a differencing method on the
series. Special cases of an ARIMA model are autoregressive models,
random-walk, exponential smoothing models, and random-trend
models (Fig. 15. Fig. 16. Table 5).

6.5. Monte carlo simulation

There are situations when a problem comprises of random vari-

ables which cannot be predicted easily then the probability of dif-
ferent possible outcomes are modeled, this is known as Monte
Carlo simulation. This method is useful in understanding the Fig. 18. Decomposition into a daily trend, weekly trend, and yearly trend.

9
J. Christy Jackson, J. Prassanna, Md. Abdul Quadir et al. Materials Today: Proceedings xxx (xxxx) xxx

impact of risk and uncertainty of prediction and forecasting models Declaration of Competing Interest
during analysis. Monte Carlo model can also be used to solve var-
ious problems such as in fields of engineering, supply chain, The authors declare that they have no known competing finan-
science, and finance. This model is also referred to as multiple cial interests or personal relationships that could have appeared
probability simulation models. to influence the work reported in this paper.

6.6. Fbprophet for forecasting References

Fbprophet is a very handy library made by Facebook for Time [1] D. Banerjee. (2014, January). Forecasting of Indian stock market using the
time-series ARIMA model. In 2014 2nd International Conference on Business
Series data. Here, The Blue line in the center implies the future and Information Management (ICBIM) (pp. 131-135). IEEE.
trend of average stock prices while the actual value can vary within [2] N. Viswam, G.S. Reddy. (2018). Stock market prediction using time series
the blue band. The blue band helps in tracking in what range can be analysis.
[3] M.C. Angadi, A.P. Kulkarni, Time series data analysis for stock market
the future stock prices can lie. Black scattered points are the actual prediction using data mining techniques with R, Int. J. Adv. Res. Comput. Sci.
stock data over the period. Using prophet data can be properly dis- 6 (6) (2015).
tributed into an overall trend, weekly trend, and yearly trend. Here [4] B.U. Devi, D. Sundar, P. Alli, An effective time series analysis for stock trend
prediction using the ARIMA model for nifty midcap-50, Int. J. Data Min. Knowl.
we can observe a yearly trend that has some seasonality (Fig. 17.
Manage. Process 3 (1) (2013) 65.
Fig. 18). [5] A. Varghese, H. Tarhen, A. Shaikh, P. Banik, A. Ramadasi, Stock market
prediction using time series, Int. J. Recent Innov. Trends Comput. Commun. 4
(5) (2016) 427–430.
7. Conclusion [6] A. Sharma, S. Modak, E. Sridhar. (2019). Data Visualization and Stock Market
and Prediction.
In this study, we could analyze and understand the risk [7] S. Selvin, R. Vinayakumar, E.A. Gopalakrishnan, V.K. Menon, K.P. Soman, in:
Stock price prediction using LSTM, RNN, and CNN-sliding window model, IEEE,
involved in the stock price and are well accounted for by statistical 2017, pp. 1643–1647.
terms such as CAGR, Sharpe Ratio volatility, and the cumulative [8] P. Mondal, L. Shit, S. Goswami, Study of effectiveness of time series modeling
return. This study has proved to be successful in comparison anal- (ARIMA) in forecasting stock prices, Int. J. Comput. Sci. Eng. Appl. 4 (2) (2014)
13–29.
ysis between stocks and get stock with lesser risk and good return.
[9] J. Wang, J. Wang, Forecasting stock market indexes using principle component
In this study, we worked with a linear model for the forecasting analysis and stochastic time effective neural networks, Neurocomputing 156
of the future price. But in there no evidence that stock data was lin- (2015) 68–78.
[10] How to Use the Sharpe Ratio to Analyze Portfolio Risk and Return. (2020).
ear, this motivates us to work on forecasting using the non-linear
Retrieved 22 May 2020, from https://www.investopedia.com/terms/s/
data shortly. Maybe with some statistical measurements, we can sharperatio.asp.
develop a model that can help in choosing the right stock. [11] J. Brownlee. (2020). What Is Time Series Forecasting? Retrieved 22 May 2020,
from https://machinelearningmastery.com/time-series-forecasting/.

CRediT authorship contribution statement

Christy Jackson J.: Conceptualization, Data Curation, Writing

Original Draft. Prassanna J.: Formal Analysis. Abdul Quadir Md.:
Methodology. Sivakumar V.: Project administration.

10