Page 8 - Azerbaijan State University of Economics
P. 8
THE JOURNAL OF ECONOMIC SCIENCES: THEORY AND PRACTICE, V.80, # 2, 2023, pp. 4-13
FORECASTING OF ENGEL CURVE COMPONENTS
In this section, for the forecasting purposes household income and expenses per capita
data are taken from State Statistical Committee of Azerbaijan (table 1). The data given
in table 1 match the middle class of population in the country (stratified based on
income per capita) and possibly represent the averaged Engel curve throughout the
country.
Table 1: Household income and expenses per capita
№ Years Income Expenses Expenses on food Expenses on food (%)
1 2008 108.9 114.6 65.2 0.57
2 2009 125.0 129.6 68.6 0.53
3 2010 144.2 147.4 71.1 0.48
4 2011 166.0 173.0 82.4 0.48
5 2012 190.9 202.0 87.3 0.43
6 2013 214.7 221.4 91.8 0.41
7 2014 230.0 234.9 95.6 0.41
8 2015 240.5 245.6 99.4 0.40
9 2016 257.8 264.7 107.1 0.40
10 2017 268.4 278.2 117.9 0.42
11 2018 276.0 286.0 119.7 0.42
12 2019 292.6 298.4 123.8 0.41
13 2020 291.4 297.8 129.2 0.43
14 2021 300.6 308.6 134.7 0.44
First, the stationarity of the time series is checked according to the identification
phase. Based on the correlogram and graphical analysis and ADF (Augmented Dickey
Fuller) test, it is defined that whether the time series is non-stationary or has an
increasing trend. In the given case the time series data are non-stationary, so based on
correlogram outputs in Eviews program package the ARIMA models were
constructed up to sixth differences.
Consequently the residuals are checked for White Noise using Ljung-Box Q Statistics;
AR roots should lie inside a unit circle showing covariance stationarity and MA roots
should lie inside a unit circle showing invertibility of ARMA process. Various models
constructed with separate values of p, d, and q parameters are selected by the Akakike,
Hannan-Quinn, and Schwarz statistics, whereby a smaller value identifies a better
parameterized model.We tested the model estimation parameters with Q-statistics. Where
the obtained probability values are greater than 0.05, the residuals are considered white
noise. Based on computed covariances, stationarity of the processes is ensured. Below in
tables 1-3 the ARIMA models are given, built on Box-Jenkins methodology.
8