Financial Forecasting: An Empirical Study On Box - Jenkins Methodology With Reference To The Indian Stock Market
Financial Forecasting: An Empirical Study On Box - Jenkins Methodology With Reference To The Indian Stock Market
Introduction
Statisticians George Box and Gwilym Jenkins developed a practical
approach to build ARIMA model, which best fit to a given time series
and also satisfy the parsimony principle. Their concept has
fundamental importance on the area of time series analysis and
forecasting. The Box-Jenkins methodology does not assume any
particular pattern in the historical data of the series to be forecasted.
Rather, it uses a three step iterative approach of model identification,
parameter estimation and diagnostic checking to determine the best
parsimonious model from a general class of ARIMA models This
three-step process is repeated several times until a satisfactory model is
finally selected. Then this model can be used for forecasting future
values of the time series.
The approach proposed by Box and Jenkins came to be known as the
Box-Jenkins methodology to ARIMA models, where the letter `I',
between AR and MA, stood for the `Integrated' and reflected the need
for differencing to make the series stationary. ARIMA models and the
Box-Jenkins methodology became highly popular in the 1970s among
academics, in particular when it was shown through empirical studies
www.pbr.co.in 115
Pacific Business Review International
(Cooper, 1972; Nelson, 1972; Elliot, 1973;Narasimham et Average (ARMA) models. The popularity of the ARIMA
al., 1974; McWhorter, 1975; for a survey see Armstrong, model is mainly due to its flexibility to represent several
1978) that they could outperform the large and complex varieties of time series with simplicity as well as the
econometric models, popular at that time, in a variety of associated Box-Jenkins methodology for optimal model
situations building process. But the severe limitation of these models is
the pre-assumed linear form of the associated time series
One of the most popular and frequently used stochastic time
which becomes inadequate in many practical situations. To
series models is the Autoregressive Integrated Moving
overcome this drawback, various non-linear stochastic
Average (ARIMA) model. The basic assumption made to
models have been proposed in literature however from
implement this model is that the considered time series is
implementation point of view these are not so straight-
linear and follows a particular known statistical distribution,
forward and simple as the ARIMA models.
such as the normal distribution. ARIMA model has
subclasses of other models, such as the Autoregressive The Box-Jenkins forecast method is schematically shown in
(AR), Moving Average (MA) and Autoregressive Moving Figure 1:
Figure 1: The Box-Jenkins methodology for optimal model selection
Literature Review
Stock market is basically non-linear in nature, prediction of forecasting ability in short-term period prediction. The
stock market plays important role in stock business. Data ARIMA model creates small forecasting errors in longer
mining, GBM, neural network and ARIMA can be experiment time period. In this paper, the author investigates
effectively used to uncover non-linearity of stock market. whether the length of the interval will influence the
The vital idea to successful stock market prediction is forecasting ability of the models or not.
achieving best result, also minimize inaccurate forecast
Mayank kumar B Patel and Sunil R Yalamalle(2014) aims at
stock price. For the past few years forecasting of stock return
using of Artificial Neural Network techniques to predict the
has been important field of research.
stock price of companies listed under LIX15 index of
Ayodele Ariyo Adebiyi, Aderemi Oluyinka Adewumi and National Stock Exchange (NSE). The results from the model
Charles Korede Ayo(2010) examines the forecasting will be used for comparison with the real data to ascertain the
performance of ARIMA and artificial neural networks accuracy of the model.
model with published stock data obtained from New York
Preethi and santhi(2012) surveys recent literature in the area
Stock Exchange. The empirical results obtained reveal the
of Neural Network, Data Mining, Hidden Markov Model
superiority of neural networks model over ARIMA model.
and Neuro-Fuzzy system used to predict the stock market
The findings further resolve and clarify contradictory
fluctuation. Neural Networks and Neuro-Fuzzy systems are
opinions reported in literature over the superiority of neural
identified to be the leading machine learning techniques in
networks and ARIMA model and vice versa.
stock market index prediction area. NN and Markov Model
Hsien-Lun Wong,Yi-Hsien Tu and Chi-chen Wang(2010) can be used exclusively in the finance markets and
used ARIMA model and vector ARMA model with fuzzy forecasting of stock price. In this paper, they propose a
time series method for forecasting. Fuzzy time series forecasting method to provide better an accuracy rather
method especially heuristic model performs better traditional method.
116 www.pbr.co.in
Volume 10 Issue 2, August 2017
Rene D. Estember and Michael John R. Marana (2016) are then examined for their ability to provide an effective
examined the potential of the Geometric Brownian Motion forecast of future values. A cross-validation technique was
(GBM) method as an accurate and effective forecasting used to improve the generalization ability of several models.
method compared to the Artificial Neural Network (ANN) The trading strategies guided by classification models
method. The number of days the volatility and drift are generate higher risk-adjusted profits than the buy-and-hold
moved were also determined and this was used to perform strategy as well as guided by the level-estimation based
the forecast of stock prices of holding companies registered forecast of the neural network and regression models. The
with the Philippine Stock Exchange and also compared to author decides to deploy the forecast the stock dividends,
the ANN method. it also indicates that the GBM method is transaction costs and individual-tax brackets to replicate the
more effective than the ANN method in forecasting stock realistic investment practices.
prices of these sample holding companies.
Abdulsalam sulaiman olaniyi, adewole, kayoed s, Jimoh
Nitin Merh, Vinod P. Saxena and Kamal Raj Pardasani R.G(2010) examined the moving average [MA] method to
(2010) an attempt is made to develop hybrid models of three uncover the patterns, relationship and to extract values of
layer feed forward back propagation ANN and ARIMA for variables from the database to predict the future values of
forecasting the future index value and trend of Indian stock other variables through the use of time series data. The
market viz. SENSEX, BSE IT, BSE Oil & Gas, BSE 100 and advantage of the MA method is a device for reducing
S& P CNX Nifty. Simulation results of hybrid models are fluctuations and obtaining trends with a fair degree of
compared with results of ANN based models and ARIMA accuracy. This techniques proven numeric forecasting
based models. method using regression analysis with the input of financial
information obtained from the daily activity equities
R.K. and Pawar D.D(2010) used to predicated the stock rate
published by Nigerian stock exchange.
because it is a challenging and daunting task to find out
which is more effective and accurate method so that a buy or Kuang Yu Huang, Chuen-Jiuan Jane(2009) used the
sell signal can be generated for given stocks. Predicting moving average autoregressive exogenous (ARX)
stock index with traditional time series analysis has proven prediction model is combined with grey system theory and
to be difficult an artificial neural network may be suitable for rough set theory to create an automatic stock market
the task. Neural network has the ability to extract useful forecasting and portfolio selection mechanism. Financial
information from large set of data. In this paper the author data were collected automatically every quarter and are
also presented a literature review on application of artificial input to an ARX prediction model for forecast the future
neural network in stock market Index prediction. trends. Clustered using a K means clustering algorithm and
then supplied to a RS classification module which selects
Jing Tao Yao and chew Lim tan(2011) explained artificial
appropriate investment stocks by a decision-making rules.
neural networks for classification, prediction and
The advantages are combining different forecasting
recognition. Neural network training is an art. Trading based
techniques to improve the efficiency and accuracy of
on neural network outputs, or trading strategy is also an art.
automatic prediction. Efficacies of the fusion models are
Authors discuss a seven-step neural network prediction
evaluated by comparing the forecasting accuracy of the
model building approach in this article. Pre and post data
ARX model with GM (1, 1) model. The hybrid model
processing/analysis skills, data sampling, training criteria
provides a highly accurate forecasting performance.
and model recommendation have been showed.
Assaleh et al. (2011) utilize two prediction models for
M. suresh Babu, N. Geethanjali and B. Sathyanarayana
forecasting securities' prices of two leading stocks in Dubai
(2011) examined the data mining techniques are able to
Financial Market (DFM). These stocks are: Emaar
uncover the hidden pattern, predict future trends and
Properties (EMAAR) and Dubai Islamic Bank
behaviors in financial market. Pattern matching techniques
(DIB).EMAAR is the leading real estate developer in the
is found to be descriptive in time series analysis. In this
Middle East and DIB is the world's first fully fledged Islamic
paper, author applied ant algorithm to accommodate a
bank. These stocks were chosen because they have sufficient
flexible and dynamic pattern-matching task in time series
historical data, actively traded, and each represents a
analysis. Apart from segment size the ant to sub-time-series
different sector in the UAE economy. The study uses daily
size affects the system performance. In this paper, the ratio
closing prices over the period from April 2000 to March
was set to 1 and also the ratio reduced to obtain a better
2006 (a total of 2176 data points).They use Artificial Neural
result.
Networks (ANN) and Polynomial Classifiers (PC) as
David Enke and Suraphan Thawornwong (2005) explained modeling techniques to predict stock prices from historical
machine learning for data mining to evaluate the predictive price data. This was the first time to apply PC to be used in
relationship of numerous financial and economic variables. stock prices
Neural network model used for estimation and classification
www.pbr.co.in 117
Pacific Business Review International
118 www.pbr.co.in
Volume 10 Issue 2, August 2017
Table 3: ARIMA (2,1,2) estimation output with close prices of Nifty 500 Index
parameters Coefficient Std. Error z p-value
const 1.59801 1.41445 1.130 0.2586
AR(1) 1.19122 0.0925221 12.87 <0.0001
AR(2) −0.759814 0.101019 −7.521 <0.0001
MR(1) −1.10509 0.104358 −10.59 <0.0001
MR(2) 0.692736 0.112597 6.152 <0.0001
Mean dependent var 1.606826 S.D. dependent var 62.37728
Mean of innovations 0.007381 S.D. of innovations 61.81107
Log-likelihood −11002.92 Akaike criterion (AIC) 22017.85
Schwarz criterion 22051.41 Hannan-Quinn 22030.18
Table 4:Statistical results of different ARIMA parameters for S&P 500 index and NIFTY 500 index
ARIMA S&P BSE 500 index NIFTY 500 index
AIC Mean of innovation AIC Mean of innovation
(1,0,0) 51215.16 1.0743 22064.82 0.6064
(1,0,1) 51158.89 0.8467 22042.97 0.4724
(2,0,0) 51158.90 0.8256 22043.00 0.4725
(0,0,1) 78903.08 ‐0.5784 31129.33 ‐0.0697
(0,0,2) 73512.27 ‐0.8330 29034.40 ‐0.1800
(1,1,0) 51134.33 0.0003 22023.12 ‐0.0022
(0,1,0) 51189.67 ‐0.0000 22044.99 0.0000
(0,1,1) 51134.24 0.0003 22023.62 ‐0.0012
(1,1,2) 51138.08 0.0007 22026.61 ‐0.0022
(2,1,0) 51136.19 ‐0.0006 22025.11 ‐0.0023
(2,1,2) 51137.41 0.0004 22017.85 0.0073
(1,1,1) 51136.15 0.0004 22025.11 ‐0.0021
www.pbr.co.in 119
Pacific Business Review International
The optimal model order is chosen by the number of model white noise or not, the Ljung-Box Q* statistic is compared
parameters, which minimizes either Akaike Information with the chi-square distribution with (h-m) degrees of
Criterion (AIC) or Bayesian Information Criterion (BIC).In freedom. Here h is the number of lags and m is the number of
the process of diagnosis it is verified whether the residual or parameters. Table 5 clearly demonstrates that the Q* statistic
error generated if white noise or not. The autocorrelations has same distribution as chi-square with (h-m) degrees of
specifying Q*-value, P-value (significance) at different freedom. Plots and the autocorrelations generated indicated
degrees of freedom having a lag of maximum 200 are shown that the model fits well.
in the table (Table 5). To determine whether the time series is
Table 5. Ljung‐Box test for the Verification of White Noise
INDEX (h-m) degree of freedom LB-statistics Q* p-value Is Series White Noise
Forecasting stage:
In forecasting stage, the best model selected can be where is the difference between the actual value
expressed as follows: and the forecast value of the series
Table 6.sample of empirical results of ARIMA(0,1,1) of S&P BSE 500 Index and ARIMA(2,1,2) of NIFTY 500
S&P BSE 500 index NIFTY 500 index
Sample period Actual value Predicted value Forecast error Actual value Predicted value Forecast error
2‐Jan‐17 11072.57 11049.1 0.00212 7002.5 6990.79 0.001672
3‐Jan‐17 11119.59 11051.3 0.006141 7028.75 6991.76 0.005263
4‐Jan‐17 11119.08 11053.6 0.005889 7030.75 6987.77 0.006113
5‐Jan‐17 11236.35 11055.8 0.016068 7106.9 6983.17 0.01741
6‐Jan‐17 11199.51 11058.1 0.012626 7083.1 6981.65 0.014323
9‐Jan‐17 11203.16 11060.3 0.012752 7085.15 6984.23 0.014244
10‐Jan‐17 11274.73 11062.6 0.018815 7132.2 6989.37 0.020026
11‐Jan‐17 11402.48 11064.8 0.029615 7214.15 6994.44 0.030455
12-Jan-17 11429.21 11067.1 0.031683 7231.85 6997.49 0.032407
13-Jan-17 11423.49 11069.3 0.031005 7228.3 6998.17 0.031837
16-Jan-17 11456.66 11071.6 0.03361 7247.95 6997.58 0.034544
17-Jan-17 11448.45 11073.8 0.032725 7243.65 6997.26 0.034015
18-Jan-17 11490.75 11076.1 0.036086 7269 6998.24 0.037249
19-Jan-17 11518.8 11078.3 0.038242 7287.75 7000.57 0.039406
20-Jan-17 11388.54 11080.6 0.027039 7202.75 7003.49 0.027664
23-Jan-17 11449.08 11082.8 0.031992 7242.55 7006.12 0.032645
24-Jan-17 11564.65 11085.1 0.041467 7315.25 7007.94 0.04201
25-Jan-17 11716.17 11087.3 0.053675 7417.7 7009.02 0.055095
27-Jan-17 11783.09 11089.6 0.058855 7455.15 7009.83 0.059733
30-Jan-17 11773.74 11091.8 0.05792 7452.15 7010.88 0.059214
31-Jan-17 11659.94 11094.1 0.048529 7379.3 7012.43 0.049716
1-Feb-17 11873.72 11096.3 0.065474 7516.05 7014.39 0.066745
2-Feb-17 11924.46 11098.6 0.069258 7547.1 7016.45 0.070312
3-Feb-17 11959.4 11100.8 0.071793 7569.9 7018.32 0.072865
6-Feb-17 12051.26 11103.1 0.078677 7628.1 7019.90 0.079732
7-Feb-17 12014.51 11105.3 0.075676 7603.2 7021.27 0.076538
8-Feb-17 12032.03 11107.6 0.076831 7614.75 7022.60 0.077764
9-Feb-17 12049.23 11109.8 0.077966 7623.6 7024.07 0.078641
10-Feb-17 12053.91 11112.1 0.078133 7631.25 7025.70 0.079351
13-Feb-17 12048.6 11114.3 0.077544 7626.95 7027.45 0.078603
This study experimented with different parameters of indicated in Table 6.ARIMA (0,1,1) is considered the best
autoregressive (p) and moving average (q) in order to for S&P BSE 500 index and ARIMA(2,1,2) is considered
determine the best model that will give best forecast as the best NIFTY 500 index is as shown in Table 2&3; hence
120 www.pbr.co.in
Volume 10 Issue 2, August 2017
it was selected as the best models based on the criteria listed The Box –jenkins methodology used ARIMA model to
in the previous section. The actual stock price and predicted calculate the errors between the actual close price and the
values are presented in Table 6, while Figure 4&5 gives the predicted close price generated by all the model. In this
graph of predicted price against actual stock price to see the study mean square percent error (MSPE) have been used for
performance of the ARIMA model selected. However, the each method of forecasting and each stock index. The table 7
forecast error is slightly low and impressive as the predicted shows results to find the best model from three models for
values are close to the actual values and move in the forecasting of stock prices in Indian stock market is
direction of the forecast values in many instances as shown ARIMA. Figure 4 & 5 displays the comparison between the
in Figure 2, which depicts the correlation of the level of actual close of S&P BSE 500 Index and NIFTY 500 index
accuracy. The forecast error is determined by forecasted close price generated by the ARIMA model
discussed above during Jan 2, 2017 to Feb 13, 2017.
Figure 2: Historical stock prices trend of S&P BSE 500 index for Feb.1st 1999 to Dec.30 2016
Figure 3: Historical stock prices trend of NIFTY 500 index for Sep.17th 2007 to Dec.30 2016
www.pbr.co.in 121
Pacific Business Review International
Figure 4: Graph of actual stock price versus predicted values for S&P BSE 500 index using ARIMA.
Figure 5 : Graph of actual stock price versus predicted values for NIFTY 500 index using ARIMA
Conclusion model which get better error more than currently model .
The results showed that the ARIMA model get the best
In the current research an attempt was made to study
MAPE (Mean Absolute Percentage Error) which is the
whether ARIMA model achieves better results than moving
measurement of this study.
average and Holt & winters exponential methods. The box-
jenkins methodology applied ARIMA model to forecast the References
stock prices for S&P BSE 500 Index and NIFTY 500 Index
G.E.P. Box, G. Jenkins, “Time Series Analysis, Forecasting
The analysis of the performance of the Indian stock market
and Control”, Holden-Day, San Francisco, CA, 1970.
with respect to time presents us a suitable time series
ARIMA model (0,1,1) and (2,1,2) which helps us in Dase R.K. and Pawar D.D., “Application of Artificial
predicting the approximate values of the future. The Neural Network for stock market predictions: A review
developed stock price predictive model with the ARIMA of literature” International Journal of Machine
indeed, the actual and predicted values of the developed Intelligence, ISSN: 0975– 2927, Volume 2, Issue 2,
stock price predictive model are slightly close. 2010, pp-14-17.
ARIMA(0,1,1) and ARIMA(2,1,2) as the best model based
JingTao YAO and Chew Lim TAN, “Guidelines for
on the fact that it satisfies all the conditions for the Goodness
Financial Prediction with Artificial neural networks
of Fit. In this study three models were used to apply into two
“.2011
stock market indices in order to find suitable forecasting
122 www.pbr.co.in
Volume 10 Issue 2, August 2017
David Enke and Suraphan Thawornwong, “The use of data 10.Whiteside II PE, J. (2008). "A practical application of
mining and neural networks for forecasting stock Monte Carlo simulation in forecasting". .AACE
market returns, 2005. International Transactions. 4: 1-12.
Abdulsalam sulaiman olaniyi, adewole, kayoed, Jimoh R.G, 11. Merh N., Saxena Vinod P., Raj Pardasani K.(2010). “A
“Stock Trend Prediction using Regression Analysis – A Comparison between Hybrid Approaches of ANN and
Data Mining Approach”, AJSS journal, ISSN 2222- ARIMA for Indian Stock Trend Forecasting”. Business
9833, 2010. Intelligence Journal,july,23-43.
M. Suresh babu, N.Geethanjali and B. Sathyanarayana, “ 12.Rene D. Estember, Michael John R. Maraña. (2016).
Forecasting of Indian Stock Market Index Using Data Forecasting of Stock Prices Using Brownian Motion –
Mining & Artificial Neural Nework”, International Monte Carlo Simulation. IEOM Society International,
journal of advance engineering & application, 2011. March 8-10,704-713.
7. Ching-Hsue cheng, Tai-Liang Chen, Liang-Ying Wei, “ A 13. Dima Waleed Hanna Alrabadi , Nada Ibrahim Abu
hybrid model based on rough set theory and genetic Aljarayesh. (2015). Forecasting Stock Market Returns
algorithms for stock price forecasting”, 2010, pp. 1610- Via Monte Carlo Simulation: The Case of Amman
1629. Stock Exchange. Jordan Journal of Business
Administration, Volume 11, No. 3,745-756.
8. Hsien-Lun Wong, Yi-Hsien Tu and Chi-Chen Wang,
“Application of fuzzy time series models for 14. Mayankkumar B Patel, Sunil R Yalamalle.(2014). Stock
forecasting the amount of Taiwan export”, Experts Price Prediction Using Artificial Neural Network
Systems with Applications, 2010, pp. 1456-1470. .International Journal of Innovative Research in
Science,Engineering and Technology, Vol. 3, Issue 6,
9. Assaleh, K. El-Baz, H. and Al-Salkhadi, S.
13755- 13762.
(2011)."Predicting stock prices using polynomial
classifiers: The case of Dubai financial market". Journal 15. Ayodele Ariyo Adebiyi, Aderemi Oluyinka Adewumi
of Intelligent Learning Systems and Applications, 3: and Charles Korede Ayo(2014). Comparison of ARIMA
82-89. and Artificial Neural Networks Models for Stock Price
Prediction. Journal of Applied Mathematics, vol. 2014,
1-7.
www.pbr.co.in 123