BOX-JENKINS

METHODOLOGY

Introductory Business Forecasting: a 1

practical approach
 ARIMA was first introduced by
George E. P. Box (University of Wisconsin, USA)
and
Gwilym M. Jenkins (University of Lancaster, UK) in
1976.

•The term ARIMA is in short stands for

the combination that comprises of
Autoregressive/Integrated/Moving Average models.

• ARIMA modelling is commonly applied to

time-series analysis, forecasting and control.

Introductory Business Forecasting: a 2

practical approach
 Stages in the ARIMA Model
Development
Stage Process
1 Model Identification
2 Model Estimation and
Validation
3 Model Application

Introductory Business Forecasting: a 3

practical approach
1 Model identification
 Common statistics used to identify the model
type is the autocorrelation (ACF) and the partial
autocorrelation coefficients (PACF).
 Although the class of model suited to a
particular data series can be determined by
simply plotting the ACF and PACF, however
this may not necessarily be the best and
ultimate procedure. This is because human
eyes and minds may in many cases can be
deceiving and inefficient. Therefore, in order to
select the best fitted model, one needs to run
several models and by applying certain
statistical test procedures one should then be
able to determine the best fitted model.
Introductory Business Forecasting: a 4
practical approach
2 Model estimation
and validation
 Estimating the models, performing the
necessary diagnostic testing procedures and
selecting the ‘best’ model for forecasting
purposes.
 More specifically, the process is to search for
the estimated parameter values that minimize
the differences between the actual and the
forecast values. This is done in a rather
iterative manner. At each step in the process a
new and improved model is obtained and these
iterations are repeated until a model with the
minimum random errors are obtained.

Introductory Business Forecasting: a 5

practical approach
3 Model application

 If all test criteria are met and that the

model’s fitness has been confirmed, it is
then ready to be used to generate the
forecasts values.
 develop a system to monitor the forecast
values produced.

Introductory Business Forecasting: a 6

practical approach
Assumption of Box-Jenkins
Methodology
 The application of the Box-Jenkins methodology
lies on the assumption that concerns the
characteristic of the initial data series.
Basically, it is assumed that the data series is
stationary. Where such assumption is not met,
then the necessary procedures are performed
in order to achieve stationarity in the series.
Incidentally, for most economics or business
data series, non-stationarity is the norm.

Introductory Business Forecasting: a 7

practical approach
The stationary series
 A series is said to be ‘stationary’ if it fluctuates
randomly around some fixed values, generally
either around the mean value of the series or it
could be some other constant values or even
zero value.
 More specifically, it can be said that, ‘a series is
stationary if it does not show growth or decline
over time’. In other words, the data series does
not indicate presence of trend component.

Introductory Business Forecasting: a 8

practical approach
The stationary series

 Example:

200

100

-100

-200

Introductory Business Forecasting: a 9

practical approach
The non-stationary series
 i. a series that is not constant around the mean
or level (due to trend or seasonal pattern) and
hence can be expressed as deterministic
function of time t.
 Example: A Non-Stationary Series in Mean

Introductory Business Forecasting: a 10

practical approach
The non-stationary series
 ii. a series which in addition to being time
dependent as in (i) also shows an increase in
the variability of the observations through
time.
Example: A Series Non-Stationary in
Variance

Introductory Business Forecasting: a 11

practical approach
Transformations (Non
stationary to stationary)
 A simple procedure used to remove the
non-stationarity in time series is to
perform the differencing.
 log transformation is commonly used to
stabilise the variance.

Introductory Business Forecasting: a 12

practical approach
Test for non-stationarity
(Unit Root test)
 The non-stationarity of a series can be
determined by,
> either through simple observation of
the plotted data or
> more accurately by using statistical
test procedures. In practise the test
based on the Augmented Dickey-Fuller
(ADF) procedure

Introductory Business Forecasting: a 13

practical approach
Differencing

 A non-stationary series can easily be

made stationary by the process of
differencing. This is analogous to the
process of removing the trend pattern
from the actual data. This is usually
referred to as detrending process, and
the resulting time series data is called
detrended series which is now
stationary.
Introductory Business Forecasting: a 14
practical approach
Differencing (d)

 Occasionally, taking the first difference

may not be sufficient to induce
stationarity in the series. This calls for
the second degree differencing.
 A series that requires first difference to
be stationary is said to be integrated of
order one or first order integrated series.

Introductory Business Forecasting: a 15

practical approach
The order of differencing
The first order differencing is
defined as the difference between
the current value, yt , and the
preceding value, yt  1 . This is
written as;
y t  y t  y t  1
where the symbol  denotes the
differencing process.
Introductory Business Forecasting: a 16
practical approach
Example: Values in the First and Second Difference
Time t yt yt  1 yt  yt  yt  1 2 yt  yt  yt  1
(1) (2) (3) (4) (5)

1 17 - - -
2 20 17 3 -
3 23 20 3 0
4 26 23 3 0
5 28 26 2 -1
6 29 28 1 -1
7 32 29 3 2
8 34 32 2 -1
9 39 34 5 3
10 42 39 3 -2
11 46 42 4 1
12 48 46 2 -2
13 51 48 3 1
14 58 51 7 4
15 67 58 9 2
16 71 67 4 -5
17 76 71 5 1
18 78 76 2 -3
19 81 78 3 1
20 85 81 4 1
21 83 85 -2 -6

Introductory Business Forecasting: a 17

practical approach
The Basic Models of the Box-
Jenkins Methodology
 There are four basic models involved:
1. The Autoregressive (AR) model,
2. The Moving Average (MA) model, and
3. Mixed Autoregressive and Moving
Average (ARMA) model.
4. Mixed Autoregressive, Integrated and
Moving Average (ARIMA) model.

Introductory Business Forecasting: a 18

practical approach
 1. The autoregressive (AR)
model
 In the AR model the current value of the variable is
defined as a function of its previous values plus an
error term.
 Mathematically, it is written as,
y t   1 y t  1   2 y t  2  .......... ......   p y t  p   t

 Above model is the general form of the order AR

process where p denotes the number of lagged
terms of the dependent variable.

Introductory Business Forecasting: a 19

practical approach
2. The moving average (MA)
model
 Previously the moving average is the
average of the actual observations,
whereas in this case it is a function of the
error terms.
yt    t  1 t  1   2 t  2  ..........   q t  q

Introductory Business Forecasting: a 20

practical approach
 3. The mixed autoregressive
moving average (ARMA)
model
 . Under the assumption of stationarity, the mixed
autoregressive and moving average model of the Box-
Jenkins methodology is known as ARMA model. In
other words the series is assumed stationary (no need
for differencing) and the ARMA model is written as;
y t   1 y t  1   2 y t  2  ......   p y t  p

  1 t  1   2  t  2  ......   q  t  q   t

Introductory Business Forecasting: a 21

practical approach
4. The mixed autoregressive
integrated moving average
(ARIMA) model
 When the stationarity assumption of the
variable is not met, then the ARIMA modelling
is formulated. In this formulation, it is
necessary that the data series needs, firstly, to
be differenced in order to achieve stationarity.
The model, thus, obtained is represented in
general term as ARIMA(p,d,q), where as stated
earlier the symbol ‘d’ denotes the number of
time the variable needs to be differenced in
order to achieve stationarity.

Introductory Business Forecasting: a 22

practical approach
 ARIMA with seasonal
component – ARIMA (p,d,q)
(P,D,Q)s
 For series with seasonal component presents, then
additional differencing is necessary to be performed in
order to eliminate the seasonality effect. This is called
‘seasonal differencing’ and is performed as follows:
Let z t be the seasonally differenced series such that z t  y t  y t  12 for
monthly data series and z t  y t  y t  4 for quarterly data series. However, if
even after performing the seasonal difference, the resulting series, z t remained
non-stationary, then further differencing is performed. This is called the first
order non-seasonal differencing, that is, wt  z t  z t  1 .

Introductory Business Forecasting: a 23

practical approach
 Steps in the identification of
the Box-Jenkin’s model
1. Construct a time plot for the original data series. Try to
identify any unusual observation. If exists, decide whether a
transformation is necessary. If necessary, transform to
achieve stationary in variance, for example do log
transformations. Plot also the ACF and PACF to confirm
the stationarity condition of the series.
2. If data series appears non-stationary, perform the first
difference. For non-seasonal data series the first difference
is sufficient. If seasonality exists, perform also the seasonal
difference. If the non-seasonality condition persists after
the seasonal difference, perform the non-seasonal
difference. Plot also the ACF and PACF to confirm the
stationarity condition of the series.
Introductory Business Forecasting: a 24
practical approach
 Steps in the identification of
the Box-Jenkin’s model
3. When stationarity condition has been achieved, examined the ACF and
PACF to see whether any discernible pattern of the data series exist.
The ACF and PACF may reveal the type of AR or MA model
appropriate (see Section 7.6.5 for details). Model with seasonal
component is suggested by larger autocorrelation/partial
autocorrelation at seasonal lags, example at lags 12, 24 for monthly
data and at lags 4, 8 etc. for quarterly data.
4. In the model identification stage, it may be slightly easier to select pure
AR or pure MA models. But when selecting a mixed ARIMA model,
the decision to decide on the values of p and q is much more difficult.
More so for model with seasonal component. Hence, it is worth
considering several possible models. To finally determine the best
fitted model, one needs to use several statistical measures such as
the MSE, AIC/BIC or the Box-Pierce (Ljung-Box) statistic.

Introductory Business Forecasting: a 25

practical approach
 General Guidelines for model
identification
i. If the autocorrelation functions (ACF) decay exponentially (dies down
slowly) and the partial autocorrelation coefficients function (PACF)
have spikes, the process can best be captured by an AR model. In
this case, the order of the AR model (or the value of p) equals the
number of significant spikes in the PACF.
ii. If the partial autocorrelation coefficients decay and the
autocorrelation coefficients have spikes, the process can best be
captured by MA model. The order of the MA model (or the value of
q) equals the number of significant spikes in the ACF.
iii. If both the autocorrelation and partial autocorrelation correlograms
are characterized by irregular patterns, the process can best be
captured by an ARMA model where the order of the model (or the
values of p and q) equals the number of significant spikes in the
PACF and ACF, respectively.

Introductory Business Forecasting: a 26

practical approach
EXAMPLE:
ACF (Original Data yt) – Dies down Slowly – not stationary
– need to do differencing
CPI

1.0 Coefficient
Upper
Confidence
Limit
Lower
Confidence
ACF

0.5 Limit

0.0

-0.5

-1.0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Lag Number

PACF (Original Data yt ) – have spike at lag 1 – this series can be made
stationary after performing first difference
CPI
Partial ACF

1.0 Coefficient
Upper
Confidence
Limit
Lower
Confidence
0.5 Limit

0.0

-0.5

-1.0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Lag Number

Introductory Business Forecasting: a 27

practical approach
 EXAMPLE:
ACF First difference – dies down quickly – stationary – MA(1) or MA(2)
CPI

1.0 Coefficient
Upper
Confidence
Limit
Lower
Confidence
0.5 Limit
ACF

0.0

-0.5

-1.0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Lag Number

PACF First Difference – Spike Lag 1 – AR (1)

CPI
Partial ACF

1.0 Coefficient
Upper
Confidence
Limit
Lower
Confidence
0.5 Limit

0.0

-0.5

-1.0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Lag Number

Possible ARIMA models – ARIMA (1,1,1) or ARIMA (1,1,2)

Introductory Business Forecasting: a 28

practical approach
Estimation and Validation of the
Box-Jenkins Model
 Several possible models – need to
choose the ‘best’ fitted model.
 Two common statistical measure used
when validating the ARIMA models are
the AIC (Akaike’s Information Criteria)
and The Box-Pierce Q statistic.

Introductory Business Forecasting: a 29

practical approach