FORECASTING BANGLADESHS INFLATION

USING THE BOX-JENKINS (ARIMA)

METHODOLOGY
PRESENTED BY
JACKSON SHAUMEN BISWAS
ID 20031
DEPARTMENT OF FINANCE
FACULTY OF BUSINESS STUDIES
UNIVERSITY OF DHAKA

JANUARY 2015

01/22/15

INTRODUCTION

01/22/15

INTRODUCTION
3

Control over inflation is the prime objective of regulators as inflation increases

uncertainty both in Consumers and Producers mind

Currently the financial regulatory authorities in Bangladesh are facing the twin
challenge of maintaining price stability while accommodating higher growth in
the economy.

During the past three decades, dramatic changes in the inflationary

environment have stimulated a wealth of studies on examining and evaluating
different methodologies in forecasting inflation.

One of the several approaches of inflation forecasting is using univariate time

series auto regressive integrated moving average (ARIMA) models.

This analysis assists in forecasting future values of inflation and to estimate

the impact of events or policy changes. Results from analysis can give
valuable information when formulating future policies.

01/22/15

OBJECTIVE OF THE STUDY

01/22/15

OBJECTIVE OF THE STUDY

5

To fulfill the partial requirement of the Evening MBA Program at the

Department of Finance, University of Dhaka

It serves as the basis for forecasting short term inflation of Bangladesh

using Autoregressive Integrated Moving Average (ARIMA) time series
models

01/22/15

DATA AND RESEARCH METHODOLOGY

01/22/15

DATA AND RESEARCH METHODOLOGY

7

Collected monthly Consumer Price Index data from January 2001 to

December 2012 (Data Source : Statistical Pocket Book and Bangladesh
Bank)

Specification and estimation of various possible types of ARIMA

models
1.Examination of the Data Series
(Trend Analysis, Autocorrelation and Partial Autocorrelation Analysis,
Seasonal Analysis)
2.Testing for Stationarity
(Augmented Dickey Fuller Test)
3.ARIMA Model Identification
(AR and MA order Identification, Identifying the order of
differentiation)
4.Model Comparison
( Akaike Information Criterion, Bayesian Information Criterion,
Variance, Standard Deviation, Mean Absolute Error and Mean Absolute
Percentage Error)
01/22/15

EXAMINATION OF THE DATA SERIES

01/22/15

TIME SERIES PLOTTING AND TREND ANALYSIS

9
Consumer Price Index

Trend Analysis Plot for CPI

Growth Curve Model
Yt = 117.425 (1.00597^t)

300
300

Variable
Actual
Fits

250

Accuracy Measures
MAPE
2.1460
MAD
3.9943
MSD
24.9082

CPI

CPI

250

200

200

150
150

100
1

14

28

42

56

70

84

98

112

126

140

Index

14

28

42

56

70

84

98

112

126

140

Index

First Difference Consumer Price Index

Time Series Plot and Trend Analysis Plot of

Monthly Consumer Price Index Data of
Bangladesh from January 2001- December
2012 shows trend like behavior

D_CPI

Visual depiction of Time series plot of first

order difference of Monthly Consumer Price
Index Data shows a non trend data set

-2
1

14

28

42

56

70

84

98

112

126

140

Index

01/22/15

AUTOCORRELATION AND PARTIAL AUTOCORRELATION ANALYSIS

Autocorrelation Function for CPI

Partial Autocorrelation Function for CPI

(with 5% significance limits for the autocorrelations)

(with 5% significance limits for the partial autocorrelations)

1.0

1.0

0.8

0.8

0.6

0.6

Partial Autocorrelation

Autocorrelation

10

0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8

0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8

-1.0

-1.0
1

10

15

20

25

30

35

10

15

Lag

Autocorrelation Function for D_CPI

25

30

35

Partial Autocorrelation Function for D_CPI

(with 5% significance limits for the autocorrelations)

(with 5% significance limits for the partial autocorrelations)

1.0

1.0

0.8

0.8

0.6

0.6

Partial Autocorrelation

Autocorrelation

20

Lag

0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8

0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8

-1.0

-1.0

10

15

20

Lag

25

30

35

10

15

20

25

30

35

Lag

If time series is stationary then its autocorrelation graph decays quite rapidly from its initial
value of unity at zero lag. Is case of no stationary data set autocorrelation dies out gradually
over time.
Autocorrelation of First order difference data set to Consumer Price Index drops quickly to zero
but has significant seasonal pattern
01/22/15

SEASONAL ANALYSIS
11

Seasonal Analysis for CPI

Multiplicative Model
Seasonal Indices

Detrended Data by Season

1.02

1.2

1.01

1.1

1.00

1.0

0.99
1

10

11

12

0.9

Percent Variation by Season

10

11

12

10

11

12

Residuals by Season

12

20
10

0
4
-10
0

10

11

12

Visual depiction of the graphical representation of seasonal analysis shows Strong Seasonality
in monthly data series of Consumer Price Index of Bangladesh

01/22/15

12

UNIT ROOT TEST (AUGMENTED DICKEY FULLER TEST)

01/22/15

UNIT ROOT TEST

13

ADF Test of Original Data Set of CPI

Exogenous: Constant
Lag Length: 2 (Automatic - based on SIC, maxlag=4)
Augmented Dickey-Fuller test statistic
Test critical values:

1% level
5% level
10% level

t-Statistic
2.185312
-3.477144
-2.881978
-2.577747

Prob.*
0.9999

t-Statistic
-7.209908
-3.476805
-2.881830
-2.577668

Prob.*
0.0000

ADF Test of First Difference Data Set of CPI

Exogenous: Constant
Lag Length: 0 (Automatic - based on SIC, maxlag=4)
Augmented Dickey-Fuller test statistic
Test critical values:

1% level
5% level
10% level

Here Null Hypothesis H0 = Data set has a unit root and Alternative Hypothesis H1 =
Data set does not follow unit root. If P-Value from ADF>0.05 then H0 accepted or H0
Rejected.
Original CPI data series has a unit root (P-Value of ADF which is 0.9999 > 0.05) but the
Data set of First Order Difference of CPI does not follow unit root (P-Value of ADF which
is 0.0000<0.05)
CPI data of Bangladesh has integrated at differentiation order 1.
01/22/15

14

IDENTIFICATION OF ARIMA MODEL

01/22/15

IDENTIFICATION OF ORDER OF DIFFERENTIATION IN ARIMA

15

Rule of Thumb

Rule 1: If the series has positive autocorrelations out to a high number of

lags, then it probably needs a higher order of differencing.

Rule 2: If the lag-1 autocorrelation is zero or negative, or the autocorrelations

are all small and pattern less, then the series does not need a higher order
of differencing. If the lag-1 autocorrelation is -0.5 or more negative, the series
may be over differenced.

Rule 3: If the series has a strong and consistent seasonal pattern, then you
should use an order of seasonal differencing--but never use more than one
order of seasonal differencing or more than 2 orders of total differencing
(seasonal and non seasonal).

Rule 4: The optimal order of differencing is often the order of differencing at

which standard deviation is lowest.

01/22/15

IDENTIFICATION OF ORDER OF DIFFERENTIATION IN ARIMA

16

Use a Non Seasonal First Order and a Seasonal First Order Differentiation
Autocorrelation of Residuals of a non seasonal ARIMA (0,1,0) Model
Autocorrelation
0.4593

Strong Seasonal
Pattern

Autocorrelation of Residuals of a seasonal ARIMA (0,1,0) (0,1,0)12 Model

Autocorrelation
0.0935

Autocorrelation of Residuals of ARIMA (0,1,0) shows strong seasonal pattern.

Lag 1 Autocorrelation of Residuals in ARIMA (0,1,0) model is very high. So a first order seasonal
differentiation should be introduced.
Lag 1 Autocorrelation of Residuals in ARIMA (0,1,0) (0,1,0)12 model is 0.0935 and the
Autocorrelations do not show visible pattern where Standard Deviation is 1.46 (Lowest)
01/22/15

IDENTIFICATION OF ORDER OF AR AND MA

17

Rule of Thumb

Rule 5: If the PACF of the differenced series displays a sharp cutoff and/or the
lag-1autocorrelation is positive--i.e., if the series appears slightly "under
differenced"--then consider adding an AR term to the model. The lag at which
the PACF cuts off is the indicated number of AR terms.

Rule 6: If the ACF of the differenced series displays a sharp cutoff and/or the
lag-1 autocorrelation is negative--i.e., if the series appears slightly "over
differenced"--then consider adding an MA term to the model. The lag at which
the ACF cuts off is the indicated number of MA terms.

Rule 7: If the autocorrelation at the seasonal period is positive, consider

adding an SAR term to the model. If the partial autocorrelation at the seasonal
period is negative, consider adding an SMA term to the model. Try to avoid
mixing SAR and SMA terms in the same model, and avoid using more than
one of either kind.

01/22/15

IDENTIFICATION OF ORDER OF AR AND MA

18

Use Non Seasonal AR, MR term and Seasonal MR term in the model

Autocorrelation of Residuals of a seasonal ARIMA (0,1,0) (0,1,0)12 Model

Sharp Cutof

Negative Autocorrelation in
Seasonal Lag

Partial Autocorrelation of Residuals of a seasonal ARIMA (0,1,0) (0,1,0)12 Model

Partial Autocorrelation is
Positive

Negative Partial Autocorrelation in

Seasonal Lag

Non Seasonal AR Term and MA Term should be incorporated in the model as Lag 1 PACF is
positive and ACF shows sharp cutoff.
Seasonal MA Term should be incorporated in the model as Seasonal Lag of PACF is negative.

01/22/15

IDENTIFICATION OF ORDER OF AR AND MA

19

Use Non Seasonal 2AR, 1MR term and Seasonal 1MR term in the model

Autocorrelation of Residuals of a seasonal ARIMA (2,1,1) (0,1,1)12 Model

No Sharp Cutof

Nearly Zero Autocorrelation in

Seasonal Lag

Partial Autocorrelation of Residuals of a seasonal ARIMA (2,1,1) (0,1,1)12 Model

Partial Autocorrelation is
Negative

Nearly Zero Partial Autocorrelation in

Seasonal Lag

Akaike Information Criterion is lowest (452.42), estimated white noise (Standard Deviation) is 1.30
for ARIMA (2,1,1) (0,1,1)12 Model and Adj R2 is 99.92%

01/22/15

20

COMPARISON OF MODELS

01/22/15

MODEL COMPARISON
21

ARIMA (2,1,1) (0,1,1)12 can be used to predict the short term inflation of Bangladesh
ARIMA Model
Seasonal ARIMA
(2,1,1) (0,1,1)12
Seasonal ARIMA
(2,1,1) (1,1,1)12
Seasonal ARIMA
(2,1,1) (2,1,1)12
Seasonal ARIMA
(0,1,1) (0,1,1)12
Seasonal ARIMA
(0,1,0) (0,1,1)12
Seasonal ARIMA
(0,1,0) (0,1,0)12
Non Seasonal ARIMA
(0,1,0)

AIC

SBC

SSE

Variance

S.D

MAPE

MAE

R
Square

452.42

466.80

213.86

1.70

1.30

0.45

0.89

99.924%

453.42

470.67

206.21

1.64

1.28

0.45

0.90

99.922%

453.71

473.83

188.71

1.52

1.23

0.45

0.90

99.923%

455.51

464.13

232.39

1.82

1.35

0.46

0.92

99.915%

457.01

462.76

239.82

1.86

1.36

0.48

0.96

99.912%

472.75

475.63

278.90

2.14

1.46

0.50

0.99

99.898%

560.29

563.25

415.33

2.92

1.71

0.66

1.26

99.869%

01/22/15

22

BEST FITTED MODEL

01/22/15

BEST FITTED MODEL

23

ARIMA (2,1,1) (0,1,1)12 model has high predictive power that means actual values
and predicted vales has high level of close match
Type
AR 1, 1
AR 1,2
MA 1,1
MA 2, 12
Intercept

Coef
1.098136
-0.223778
1.00
0.464201
0.139906

t Ratio
12.55
-2.52
41.33
4.44
10.02

Prob > [t]

0.01%
1.31%
0.01%
0.01%
0.01%

01/22/15

24

FINDINGS

01/22/15

FINDINGS
25

This research work is an attempt to select the accurate model among various
ARIMA estimated models.

A framework for ARIMA Modeling which includes Examination of data series,

testing of stationarity, ARIMA model identification and model comparison has
established.

Monthly Consumer Price Index data of Bangladesh has strong seasonal

pattern.

Accurate ARIMA model for forecasting Bangladeshs short term inflation has
identified which is ARIMA (2,1,1)(0,1,1)12

01/22/15

26

QUESTION AND ANSWER SESSION

01/22/15

27

THANKS YOU ALL

01/22/15