econometrics ppt group 6
DESCRIPTION
TIME SERIES FORCASTINGTRANSCRIPT
GROUP 6ANURAAG MONDALAMAN NOGIA NITIN KUMARNITIN NATH SINGH
FORECASTING INDIA’S FOOD PRICE INDEX USING ARIMA MODEL
Objective of the Project
To model the monthly Commodity Food Price Index from the
year 2000 onwards using a time series analysis .
To forecast India's food price index for the next month(April)
using ARIMA model
Data Used
The data used has been obtained from www.indexmundi.com
The Commodity Food Price Index of 15 years(Monthly) has
been taken in the project.
The period taken into consideration is March 2000 to March
2015.
Commodity Food Price Index
“Everyone eats. As a result, everyone is affected to some
degree by food price changes.”
The Commodity Food Price Index is a measure of the monthly
change in international prices of a basket of food commodities.
It consists of the average of five commodity group price indices
(representing 55 quotations), weighted with the average export
shares of each of the groups for 2002-2004.
Commodity Food Price Index includes Cereal, Vegetable Oils,
Meat, Seafood, Sugar, Bananas, and Oranges Price Indices.
Data Description
Graph 1
Following is a plot of data for each month from March
2000 onwards. We can see the wide fluctuations in Food Price
Index in the year 2008 owing to the financial downturn that
was witnessed by the world economy which had an impact on
all the sectors.
3/1/
2000
8/1/
2000
1/1/
2001
6/1/
2001
11/1
/200
1
4/1/
2002
9/1/
2002
2/1/
2003
7/1/
2003
12/1
/200
3
5/1/
2004
10/1
/200
4
3/1/
2005
8/1/
2005
1/1/
2006
6/1/
2006
11/1
/200
6
4/1/
2007
9/1/
2007
2/1/
2008
7/1/
2008
12/1
/200
8
5/1/
2009
10/1
/200
9
3/1/
2010
8/1/
2010
1/1/
2011
6/1/
2011
11/1
/201
1
4/1/
2012
9/1/
2012
2/1/
2013
7/1/
2013
12/1
/201
3
5/1/
2014
10/1
/201
4
3/1/
2015
0
50
100
150
200
250
Price
Price
Graph 2
The next plot depicts the yearly mean price index from the
year 2000 onwards. It can be seen that the mean food price
Index has been rising from the year 2000 onwards.
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 20150
20
40
60
80
100
120
140
160
180
200
Mean Price Index
Mean Price Index
Modelling the series
The Box-Jenkins Methodology for univariate time-series has been used.
To obtain the results EViews8 was used. The three primary phases in building a Box-Jenkins Time
Series Model
Phase I-Identification Data Preparation by checking for and making series
stationary Model Selection by using ACF and PACF
Phase II-Estimation and Testing Estimation of model deriving its MLE parameter estimates. Diagnosticas check ACF/PACF of residuals
Phase III-Forecasting Use models to forecast.
Phase I-Identification
Checking for Stationarity
For checking stationarity we used the Augmented
Dickey-Fuller test (ADF) on the price series.
Using the results of the test we found that the
series is non stationary
Results on following slide:
Null Hypothesis: PRICE has a unit rootExogenous: NoneLag Length: 1 (Automatic - based on SIC, maxlag=13)
t-Statistic Prob.*
Augmented Dickey-Fuller test statistic 0.075322 0.7055Test critical values: 1% level -2.577945
5% level -1.94261410% level -1.615522
*MacKinnon (1996) one-sided p-values.
Augmented Dickey-Fuller Test EquationDependent Variable: D(PRICE)Method: Least SquaresDate: 04/22/15 Time: 00:45Sample (adjusted): 3 181Included observations: 179 after adjustments
Variable Coefficient Std. Error t-Statistic Prob.
PRICE(-1) 0.000171 0.002270 0.075322 0.9400D(PRICE(-1)) 0.466881 0.066794 6.989908 0.0000
R-squared 0.213725 Mean dependent var 0.333352Adjusted R-squared 0.209283 S.D. dependent var 4.562015S.E. of regression 4.056647 Akaike info criterion 5.649701Sum squared resid 2912.780 Schwarz criterion 5.685314Log likelihood -503.6482 Hannan-Quinn criter. 5.664142Durbin-Watson stat 1.979674
UNIT ROOT TEST FOR PRICE
In order to make the series stationary we differenced the original price series and generated a new series as price 1 with first difference
Next, we tested this new series for stationarity again by using Augmented Dickey-Fuller test (ADF)
By observing the results we found that the new series in stationary.
Results on following slide
Null Hypothesis: PRICE1 has a unit rootExogenous: NoneLag Length: 0 (Automatic - based on SIC, maxlag=13)
t-Statistic Prob.*
Augmented Dickey-Fuller test statistic -8.027300 0.0000Test critical values: 1% level -2.577945
5% level -1.94261410% level -1.615522
*MacKinnon (1996) one-sided p-values.
Augmented Dickey-Fuller Test EquationDependent Variable: D(PRICE1)Method: Least SquaresDate: 04/22/15 Time: 00:40Sample (adjusted): 3 181Included observations: 179 after adjustments
Variable Coefficient Std. Error t-Statistic Prob.
PRICE1(-1) -0.532686 0.066359 -8.027300 0.0000
R-squared 0.265751 Mean dependent var -0.034581Adjusted R-squared 0.265751 S.D. dependent var 4.720946S.E. of regression 4.045300 Akaike info criterion 5.638560Sum squared resid 2912.873 Schwarz criterion 5.656366Log likelihood -503.6511 Hannan-Quinn criter. 5.645780Durbin-Watson stat 1.980119
UNIT ROOT TEST FOR PRICE 1
60
80
100
120
140
160
180
200
25 50 75 100 125 150 175
PRICE
-30
-20
-10
0
10
20
25 50 75 100 125 150 175
PRICE1
Phase II-Estimation and Testing
Model estimation
For initial values of AR and MA to model a mean generating function, we use Correlogram ( ACF & PACF).
Further We try different combinations of AR and MA to make a suitable ARIMA model.
Date: 04/22/15 Time: 01:06Sample: 1 181Included observations: 180
Autocorrelation Partial Correlation AC PAC Q-Stat Prob
1 0.462 0.462 39.115 0.0002 0.194 -0.02... 46.061 0.0003 0.023 -0.07... 46.157 0.0004 0.064 0.109 46.908 0.0005 -0.02... -0.11... 47.065 0.0006 -0.12... -0.11... 50.020 0.0007 -0.21... -0.11... 58.883 0.0008 -0.26... -0.14... 72.063 0.0009 -0.16... 0.037 77.115 0.000
1... -0.03... 0.064 77.396 0.0001... 0.055 0.060 77.972 0.0001... -0.02... -0.09... 78.059 0.0001... -0.10... -0.13... 80.127 0.0001... -0.04... 0.020 80.458 0.0001... 0.069 0.044 81.401 0.0001... 0.031 -0.07... 81.590 0.0001... -0.00... 0.025 81.591 0.0001... -0.01... 0.017 81.648 0.0001... 0.063 0.076 82.450 0.0002... 0.061 -0.02... 83.213 0.0002... 0.028 -0.07... 83.378 0.0002... -0.04... -0.07... 83.857 0.0002... -0.01... 0.069 83.897 0.0002... -0.05... -0.06... 84.564 0.0002... -0.04... -0.01... 84.984 0.0002... -0.11... -0.11... 87.866 0.0002... -0.02... 0.119 88.018 0.0002... -0.11... -0.14... 90.986 0.0002... -0.12... -0.11... 94.593 0.0003... -0.12... -0.07... 98.224 0.0003... 0.004 0.102 98.227 0.0003... 0.072 0.088 99.372 0.0003... 0.090 0.026 101.20 0.0003... 0.153 0.051 106.43 0.0003... 0.161 0.041 112.32 0.0003... 0.155 0.008 117.80 0.000
Use ACF and PACF to identify appropriate models.
(2,1,2) ARIMA modelDependent Variable: PRICE1Method: Least SquaresDate: 04/22/15 Time: 00:58Sample (adjusted): 4 181Included observations: 178 after adjustmentsConvergence achieved after 38 iterationsMA Backcast: 2 3
Variable Coefficient Std. Error t-Statistic Prob.
C 0.323378 0.519277 0.622748 0.5343AR(1) -0.356228 0.265981 -1.339301 0.1822AR(2) 0.078652 0.199671 0.393908 0.6941MA(1) 0.849953 0.257539 3.300290 0.0012MA(2) 0.337116 0.144118 2.339173 0.0205
R-squared 0.233238 Mean dependent var 0.332135Adjusted R-squared 0.215509 S.D. dependent var 4.574855S.E. of regression 4.052016 Akaike info criterion 5.663994Sum squared resid 2840.459 Schwarz criterion 5.753370Log likelihood -499.0955 Hannan-Quinn criter. 5.700238F-statistic 13.15603 Durbin-Watson stat 2.002627Prob(F-statistic) 0.000000
Inverted AR Roots .15 -.51Inverted MA Roots -.42+.40i -.42-.40i
AR (1) and AR(2) are not significant.
(2,1,4) ARIMA modelDependent Variable: PRICE1Method: Least SquaresDate: 04/22/15 Time: 09:37Sample (adjusted): 4 181Included observations: 178 after adjustmentsConvergence achieved after 32 iterationsMA Backcast: 0 3
Variable Coefficient Std. Error t-Statistic Prob.
C 0.283847 0.617801 0.459446 0.6465AR(1) 1.158921 0.068222 16.98751 0.0000AR(2) -0.759547 0.063225 -12.01335 0.0000MA(1) -0.711078 0.091795 -7.746343 0.0000MA(2) 0.494012 0.095217 5.188289 0.0000MA(3) 0.163527 0.091151 1.794015 0.0746MA(4) 0.321004 0.084025 3.820340 0.0002
R-squared 0.295191 Mean dependent var 0.332135Adjusted R-squared 0.270460 S.D. dependent var 4.574855S.E. of regression 3.907525 Akaike info criterion 5.602217Sum squared resid 2610.956 Schwarz criterion 5.727343Log likelihood -491.5973 Hannan-Quinn criter. 5.652959F-statistic 11.93646 Durbin-Watson stat 2.027108Prob(F-statistic) 0.000000
Inverted AR Roots .58+.65i .58-.65iInverted MA Roots .66+.73i .66-.73i -.31+.48i -.31-.48i
MA(3) is not significant.
Dependent Variable: PRICE1Method: Least SquaresDate: 04/22/15 Time: 09:35Sample (adjusted): 5 181Included observations: 177 after adjustmentsConvergence achieved after 37 iterationsMA Backcast: 1 4
Variable Coefficient Std. Error t-Statistic Prob.
C 0.298671 0.588436 0.507567 0.6124AR(1) 0.907052 0.243977 3.717785 0.0003AR(2) -0.414339 0.325189 -1.274150 0.2044AR(3) -0.230172 0.217124 -1.060093 0.2906MA(1) -0.472554 0.228338 -2.069534 0.0400MA(2) 0.255119 0.234405 1.088369 0.2780MA(3) 0.283872 0.143706 1.975374 0.0499MA(4) 0.405752 0.100965 4.018761 0.0001
R-squared 0.297927 Mean dependent var 0.340960Adjusted R-squared 0.268847 S.D. dependent var 4.586313S.E. of regression 3.921640 Akaike info criterion 5.615041Sum squared resid 2599.095 Schwarz criterion 5.758596Log likelihood -488.9312 Hannan-Quinn criter. 5.673262F-statistic 10.24512 Durbin-Watson stat 2.000051Prob(F-statistic) 0.000000
Inverted AR Roots .60+.64i .60-.64i -.30Inverted MA Roots .66-.73i .66+.73i -.43-.48i -.43+.48i
(3,1,4) ARIMA model
AR(2), AR(3) & MA(2) is not significant.
(3,1,2) ARIMA model
Dependent Variable: PRICE1Method: Least SquaresDate: 04/22/15 Time: 00:56Sample (adjusted): 5 181Included observations: 177 after adjustmentsConvergence achieved after 19 iterationsMA Backcast: 3 4
Variable Coefficient Std. Error t-Statistic Prob.
C 0.333699 0.565676 0.589912 0.5560AR(1) -0.892963 0.081777 -10.91950 0.0000AR(2) -0.287029 0.102906 -2.789239 0.0059AR(3) 0.375963 0.077938 4.823885 0.0000MA(1) 1.396576 0.037482 37.26039 0.0000MA(2) 0.964087 0.036342 26.52815 0.0000
R-squared 0.246262 Mean dependent var 0.340960Adjusted R-squared 0.224223 S.D. dependent var 4.586313S.E. of regression 4.039541 Akaike info criterion 5.663450Sum squared resid 2790.360 Schwarz criterion 5.771116Log likelihood -495.2153 Hannan-Quinn criter. 5.707115F-statistic 11.17389 Durbin-Watson stat 2.003172Prob(F-statistic) 0.000000
Inverted AR Roots .43 -.66+.65i -.66-.65iInverted MA Roots -.70-.69i -.70+.69i
All the parameters are significant.
Diagnostic checking
Using (3,1,2) ARIMA model we find that all the parameters are significant, hence this is model can be used for forecasting.
We proceed to the next step i.e. diagnostic testing of the model.
Correlogram of residualsDate: 04/22/15 Time: 15:19Sample: 1 181Included observations: 161
Autocorrelation Partial Correlation AC PAC Q-Stat Prob
1 -0.49... -0.49... 40.527 0.0002 0.016 -0.30... 40.570 0.0003 -0.00... -0.21... 40.575 0.0004 -0.01... -0.17... 40.596 0.0005 -0.00... -0.15... 40.596 0.0006 -0.00... -0.13... 40.597 0.0007 0.021 -0.08... 40.669 0.0008 -0.03... -0.10... 40.836 0.0009 -0.00... -0.11... 40.837 0.000
1... 0.006 -0.11... 40.843 0.0001... 0.017 -0.08... 40.894 0.0001... 0.012 -0.03... 40.921 0.0001... -0.05... -0.10... 41.492 0.0001... -0.03... -0.21... 41.683 0.0001... 0.121 -0.07... 44.306 0.0001... -0.09... -0.14... 45.823 0.0001... 0.051 -0.11... 46.301 0.0001... -0.07... -0.23... 47.332 0.0001... 0.096 -0.15... 49.045 0.0002... 0.030 -0.02... 49.210 0.0002... -0.01... 0.064 49.240 0.0002... -0.09... -0.06... 50.976 0.0002... 0.052 -0.05... 51.491 0.0012... -0.04... -0.11... 51.836 0.0012... 0.132 0.111 55.175 0.0002... -0.19... -0.12... 62.600 0.0002... 0.169 0.002 68.160 0.0002... -0.08... 0.010 69.440 0.0002... 0.034 0.114 69.669 0.0003... -0.09... -0.08... 71.417 0.0003... 0.055 -0.11... 72.020 0.0003... 0.039 -0.05... 72.330 0.0003... -0.06... -0.02... 73.256 0.0003... 0.052 -0.04... 73.818 0.0003... 0.011 -0.03... 73.846 0.0003... 0.021 0.027 73.942 0.000
Unit root test for stationarity of residuals
Null Hypothesis: D(RESID) has a unit rootExogenous: NoneLag Length: 10 (Automatic - based on SIC, maxlag=13)
t-Statistic Prob.*
Augmented Dickey-Fuller test statistic -8.636451 0.0000Test critical values: 1% level -2.579315
5% level -1.94280510% level -1.615400
*MacKinnon (1996) one-sided p-values.
Augmented Dickey-Fuller Test EquationDependent Variable: D(RESID,2)Method: Least SquaresDate: 04/22/15 Time: 15:17Sample (adjusted): 20 181Included observations: 162 after adjustments
Variable Coefficient Std. Error t-Statistic Prob.
D(RESID(-1)) -8.352108 0.967076 -8.636451 0.0000D(RESID(-1),2) 6.394433 0.922708 6.930072 0.0000D(RESID(-2),2) 5.494756 0.853548 6.437550 0.0000D(RESID(-3),2) 4.688108 0.770407 6.085234 0.0000D(RESID(-4),2) 3.969352 0.681566 5.823872 0.0000D(RESID(-5),2) 3.240392 0.586127 5.528480 0.0000D(RESID(-6),2) 2.571765 0.483985 5.313730 0.0000D(RESID(-7),2) 1.932902 0.380934 5.074112 0.0000D(RESID(-8),2) 1.246360 0.277041 4.498833 0.0000D(RESID(-9),2) 0.639663 0.174266 3.670605 0.0003D(RESID(-10),2) 0.181622 0.080966 2.243192 0.0263
R-squared 0.832361 Mean dependent var 0.022037Adjusted R-squared 0.821259 S.D. dependent var 11.12582S.E. of regression 4.703752 Akaike info criterion 6.000084Sum squared resid 3340.918 Schwarz criterion 6.209736Log likelihood -475.0068 Hannan-Quinn criter. 6.085206Durbin-Watson stat 1.999566
Diagnostic Checking
From the unit root test for residuals we find that residuals are stationary
Adjusted R2 is around 82%
Phase III-Forecasting
-4.6
-5.64
-3.76
-4.2200
6-
0.67803
Change in the prices for the forecasted month -1.43478
Forecast price for april 2015= 142.52
Change in the price for the forecasted month =Coefficient + AR(i)*price(-1) + MA(i)*Resid(-1)
Using above model, the FPI values for the month of April has been predicted.
Variable Coefficient Std. Error t-Statistic Prob.
C 0.333699 0.565676 0.589912 0.556
AR(1) -0.89296 0.081777 -10.9195 0
AR(2) -0.28703 0.102906 -2.78924 0.0059
AR(3) 0.375963 0.077938 4.823885 0
MA(1) 1.396576 0.037482 37.26039 0
MA(2) 0.964087 0.036342 26.52815 0
Conclusion
The Commodity Food price index continues on following the downward trend as per the model.