Annals of the University of Petroşani, Economics, 10(4), 2010, 317-326


ABSTRACT: The hereby paper deals with an econometric analysis of the interconnection of consumption per capita and GDP within 27 European Union member states during a three years’ period. The main purpose of the paper is to show the manner Eviews program is able to handle such an approach, the stages to be followed as well as the carrying out of certain previsions. KEY WORDS: Gross Domestic Product (GDP); Method of Least Squares (MLS); Ordinary Least Squares (OLS); Eviews Program JEL CLASSIFICATION: C24, E20

1. INTRODUCTION The subject of the paper regards the analysis of interconnection between consumption per capita and GDP per capita, at a European level, perceived being the expression of the Keynesian consumption function theory, interpreted as a linear dependence between private consumption and disposable income. In order to draw out this econometric analysis, the authors have employed data sources belonging to 27 member states of the European Union, during a three years period; the analysis has as an endogenous variable the consumption per capita and as an exogenous variable the GDP per capita. The measuring unit used for the emerging data is Euro per capita, calculated according to current prices. Due to this reason, the data have been re-interpreted according to the index of the Consumption Prices in order to provide the similar prices of year 2000 and in order to ensure data comparability. The data to be processed are displayed by Table 1.

Assist.Prof., Ph.D. Student, University of Petrosani, Romania, Assist.Prof., Ph.D. Student, University of Petrosani, Romania,

Ct is the total consumption per capita.GDPt is The Gross Domestic Product per capita. Stoicuţa.. it is going to be solved owing to the Eviews 7.318 Stanciu. Table 1. .0 software package. THE MATHEMATICAL MODEL This econometric model employs the following notations for the variables that enter its composition: . 2. .P. N.E. The data to be processed Country Belgium Bulgaria Czech Republic Denmark Germany (including ex-GDR from 1991) Estonia Ireland Greece Spain France Italy Cyprus Latvia Lithuania Luxembourg Hungary Malta Netherlands Austria Poland Portugal Romania Slovenia Slovakia Finland Sweden United Kingdom GDP/capita 2001 13065 1044 3493 17331 13241 2702 16134 7156 8889 12756 11484 8105 2175 1980 27007 3184 5714 14568 13661 2917 6659 1266 6233 2436 13731 14579 14315 consumption/capita 2000 2001 2002 18400 9851 5203 1500 934 559 4400 2517 1519 23700 12623 6691 19500 3400 17200 11400 12000 18700 16400 11700 3000 3100 28000 3700 9100 19000 19200 4000 9900 1600 8200 3100 17800 22700 22900 10355 2013 10011 6408 6760 10048 8967 6579 1794 1726 15644 2416 4980 10600 10142 2396 5496 1076 4757 1882 9597 10973 12152 5352 1187 5658 3641 3705 5344 4807 3554 1025 952 8687 1560 2551 5702 5259 1263 2992 703 2660 1146 5083 5942 6509 2000 24600 1700 6000 32500 25100 4500 27600 12600 15700 23700 20900 14500 3600 3500 50200 5000 10800 26300 25900 4900 12000 1800 10800 4100 25500 30000 27200 2002 6937 618 2043 9098 6855 1573 9207 4005 4931 6741 6164 4292 1230 1106 14884 2061 3067 7710 7198 1478 3596 814 3557 1448 7148 7782 7590 Data Source: Eurostat The mathematical model of this application is displayed below. A.

and the link’s type is linear. the autocorrelation coefficients (AC) start at 1 or -1 and decrease very slowly. Descriptive statistics for GDP per capita In order to test the normal distribution of the two data series. we are going to descriptively analyze the data series used by this model. The analysis of the corelograms shows that the series are stationary. first. the .44 1123.843041 16 Series: PIB_L Sample 1 81 Observations 81 12 Mean Median Maximum Minimum Std. minimum and maximum values.97 30723. Descriptive statistics for consumption per capita Figure 2. we have displayed the histograms of the two series where one can notice the mean. Skewness Kurtosis 16917. Dev.55 14893. There is a direct link between the two variables (consumption / per capita and GDP / per capita). Accordingly.62 52640.552 0.93 12088. and c are the parameters to be determined after applying the method of the smallest squares As the mathematical model has been stated. median. The analysis of the corelo-grams shows that there is a direct link between the two variables (consumption and GDP). one can see that the distribution of consumption is platikurtotic and the distribution of GDP is leptokurtotic. kurtosis. As the value of the probability is smaller than 10%.700123 3. Q-static and its associated probability represent a statistical test which has as null hypothesis that fact that there is no autocorrelation up to lag k. we are going to define the following econometric model: Ct = b + a ⋅ GDPt + ε t (1) where a and b.178904 1.949709 Probability 0. they are exhibited by the graphs below: 14 12 10 8 6 4 2 0 5000 10000 15000 20000 25000 30000 Series: CONS_L Sample 1 81 Observations 81 Mean Median Maximum Minimum Std.243764 8 4 Jarque-Bera 4. On the other hand. and Jarque-Bera test. Within the histograms displayed above. standard deviation.084175 0 0 10000 20000 30000 40000 50000 Jarque-Bera 6. Dev.80 1301.817867 Probability 0.European Econometric Analysis of the Interconnection of Consumption … 319 Using the above variables. In case the probability associated with Q-static test is higher than the relevance level. and the type of the link is linear. one may assert with certainty that the null hypothesis is rejected. after processing the two data series.724 8057.155 12077. such facts are also shown by the graphical representation of the data cloud.87 0. such facts are also shown by the graphic representation of the data cloud. Skewness Kurtosis 12563. asymmetry coefficient.033076 Figure 1. information regarding the statistical indices is extracted.

Stoicuţa. the t-Statistic test. the more its value is closer to 1. the probabilities for the data series are higher than the relevance level in case of a large number of lags. the coefficients are considered statistically significant. we shall employ the method of the smallest squares and the following command in Eviews: ls cons_l c pib_l The values of the parameters a and b determined according to the method of the smallest squares employing Eviews program are displayed by Figure no. As R 2 in our case represents 0.320 Stanciu.074 + 0. Due to the fact that this coefficient represents 1. in our case. Eviews reports the standard error of the coefficient (Std. This coefficient displays values ranging between 0 and 1. therefore the above condition is met. As. the best specified its regression. one can state that .P. Error). Eviews also reports R 2 value (R-Squared) that shows the percentage of the total variance of the dependent variable which is due to the independent variable. Graphical representation of the data cloud for the two variables After the descriptive analysis of the data series. then the DW value is about 2.. we are going to really solve the econometric model given by relation (1). one can state that the data for these lags are auto-correlated.E. In case the errors are not correlated. Another given coefficient is the Durbin-Watson one which tests the serial correlation of the errors. null hypothesis is rejected and alternative hypothesis is accepted . N. The rest are uncorrelated. A.466252.955078. 60000 50000 40000 P IB _ L 30000 20000 10000 0 0 10000 20000 CONS_L 30000 40000 Figure 4. It shows that the regression line given by relation (1) can be written as follows: Ct = 1534.there is autocorrelation up to lag k. and the probability associated with it. Assuming that the relevance level represents 5% and. In order to determine the two parameters. 5. the probabilities of the t-Statistic test are below this level.651977 ⋅ GDPt (2) For each independent and constant variable. as known.

.α = 0. Eviews window and the determining of the parameters’ values of the mathematical model Variable C PIB_L R-squared Adjusted R-squared S. Table 2. indicates that hypothesis H0 (according to which the model is not correctly specified) is going to be rejected. we are going to check up the hypotheses of the model of linear regression.E. 3.81985 1679.0000 12563.954510 1718.3096 1. σ2).015908 t-Statistic 4.651977 0.0000 0. the model being a correct one.466252 Std.There is a significant positive linear dependence between GDP per capita and consumption per capita (regression curve is positive and significantly departs from zero) . which corresponds to the regression curve) . The above representations help us conclude the following: . 0.0000.0. Error 329. Errors’ mean is null In order to test this hypothesis we are going to draw out the histogram of errors.98318 Prob.549 2. THE CHECKING UP OF THE HYPOTHESES OF THE LINEAR REGRESSION MODEL The testing of the validity of hypotheses drawn out within the linear regression model is similar with the analysis of autocorrelation.The relation t2 = F is checked up between the statistic value of F and t.648847 40.000000 Mean dependent var S.93 8057. in other words. similar to Gauss’s bell. emphasizing errors’ average as well as the fact that errors follow a normal distribution N (0.76073 17.9902 0.955078 0. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic) In order to validate the model. of regression Sum squared resid Log likelihood Durbin-Watson stat Coefficient 1534.955078% of the variant of consumption series is explained by the estimated regression model.European Econometric Analysis of the Interconnection of Consumption … 321 there is a serial correlation of the errors or.33E+08 -717.1. 3. and normality.D.552 17.621 0.074 0. . hetero-skedasticity. errors are auto-correlated and the model should be improved.

A. one can notice the below results: Table 3.48151 Probability Probability 0.5 85.828250 37. of regression Sum squared resid Log likelihood Durbin-Watson stat Coefficient 3008193.740190 3246282.18 X 10 -12.P.5601 0. After applying White test. -514.746685 0.86025 32.000000 0. so that the errors’ mean is approximately zero.00000 Mean dependent var S.0000 0. Stoicuţa. 6368811. normality is checked up.019935 0.9587 60.000000 12 8 4 0 Figure 5.884395 5. 3.E. Dev.327 1707. 32.32253 0.22E+14 -1327. except for lag 27. Errors’ histogram Errors’ histogram shows that the mean represents 2. 16 Series: RESID Sample 1 81 Observations 81 Mean Median Maximum Minimum Std.840 1. Skewness Kurtosis Jarque-Bera Probability -6000 -4000 -2000 0 2000 4000 -2.322 20 Stanciu.000000 Test Equation: Dependent Variable: RESID^2 Method: Least Squares Date: 01/14/10 Time: 17:56 Sample: 1 81 Included observations: 81 Variable C PIB_L PIB_L^2 R-squared Adjusted R-squared S.654839 -6.2.950507 Std.E.0000 2880487.727 -6263.D.001891 t-Statistic 3. The series of errors is homo-skedastic The corel-gram of quadratic errors shows that errors are hetero-skedastic due to the fact that quadratic errors are auto-correlated.. The checking up of homo-skedasticity hypothesis F-statistic Obs*R-squared 114. 8.9587 0. Error 823071.030764 10. N. 0.18E-12 204.0005 0.54417 Prob.774 -0. dependent var Akaike info criterion Schwarz criterion F-statistic Prob(F-statistic) .8719 4092.55570 0. which means that errors are normally distributed.94894 114.

dL].0005 only in case of the free term.66 in the table). where dL=1. parameters are estimated using ordinary least squares (OLS): r 2 = 3008193-514. as well as F test (indicating that 0. After performing this test one may notice the following: the coefficient of autocorrelation is small (0.3.032). it is 0). this is the reason of performing the Durbin-Watson test (as the terms of its implementation are carried out).European Econometric Analysis of the Interconnection of Consumption … 323 White test is a statistic test which is based on the following regression model: rt2 = a + b ⋅ GDP _ Lt + c ⋅ GDP _ L2 t (3) After running the EViews program. so that we should accept that all coefficients are different from 0). dU=1.5601 ⋅ GDP _ L +0. DW belonging to the interval [0. Non-auto-correlation of errors 6000 4000 2000 R _EQ 1 0 -2000 -4000 -6000 -8000 -8000 -4000 R_EQ1(-1) 0 4000 Figure 6.247). and the model should be improved . probability p is small (0.019935 ⋅ GDP _ L2 (4) The test of T significance for each coefficient (p-value represents maximum 0. Accordingly. recommending the use of other methods in order to estimate the parameters.00 error occurs in case we reject the null hypothesis. 3. estimating ρ in the model et = ρet-1+vt. in case of the rest terms. Graphical representation of errors’ correlation The above graphic of the correlation between residues and delayed residues does not show whether the correlation is evident. Durbin-Watson statistics represents 1. auto-correlation hypothesis is not carried out.61. shows that homo-skedasticity is violated and hetero-skedasticity is present.51 (which shoes that order I auto-correlation is positive.

7445 0.113240 t-Statistic 2.E.056882 1666.P. dependent var Akaike info criterion Schwarz criterion Durbin-Watson stat As the implementation of DW test determines poor results.370438 -0. On the other hand. so that there is no error to reject the null hypothesis (of normal distribution p = 0). 3. A.183765 Std.444769 2.D. 0.838881 Mean dependent var S. RESID PIB_L RESID 1.626694 6. Error 322. The above five hypotheses determine us to conclude that the linear regression model should be improved so that it obeys the hypotheses of a simple linear regression model as the admittance of this model might determine errors as follows: .291912 -0.0146 0.6577 0..44E+08 -2. Checking up of the hypothesis of errors’ auto-correlation Variable R_EQ1(-1) R-squared Adjusted R-squared S.015658 0.5. of regression Sum squared resid Log likelihood Coefficient Std.086090 being higher than 0.006964 0.0320 -10.19E+08 -706. N.5568 -0.973303 Probability Probability 0.031245 0.498573 -1.247263 0.E.476 2.995 17.116831 0. skewness is close to 0 and kurtosis is close to 3.71701 1.8): F-statistic Obs*R-squared 3.37*10-8. Errors’ normality Errors’ normality is displayed by errors’ histogram which shows that we deal with a normal probability density (Gauss’s bell). 0.1236 3.118046 t-Statistic 0.7121 0.ε) = -2.4893 0. The data series for these features are not stochastic Due to the fact that cov(x.37E-08 2880487.183525 Prob.4. one can state that this hypothesis is carried out.324 Stanciu.34110 1715. Table 4.056882 0. the fact asserts that this hypothesis is checked up. we have performed the Breusch-Godfrey Serial Correlation LM test to confirm them (its returning R2 value of 0. as it is shown by the matrix of covariant between the error vector and the data vector that shows the GDP. Error 0.37E-08 PIB_L -2.030603 Test Equation: Dependent Variable: RESID Method: Least Squares Date: 01/16/10 Time: 18:40 Presample missing value lagged residuals set to zero. Stoicuţa.68723 17. Variable C PIB_L RESID(-1) RESID(-2) Coefficient 119.556722 Prob.

02.3473%. the Durbin-Watson statistic DW ∈ [d L .84%. shows no autocorrelation The hypothesis of the homo-skedasticity is not carried out.0. p statistics = 0.047 ⋅ ln(PIB_L)ij-1 .002 + 0. and the Durbin-Watson statistics ranging within the interval recommending that the hereby auto-correlation is not confirmed (DW Є [dL.European Econometric Analysis of the Interconnection of Consumption … 325 .20. . the probability of carrying out the null hypothesis representing 0. Another econometric model that shows even better results has the following - shape: ln(CONS_L)ij = a1 + a 2 ⋅ ln(PIB_L)ij + a 3 ⋅ ln(PIB_L)ij-1 + a 4 ⋅ ln(CONS_L)ij-1 +ε ij . due to its delayed values) has higher probability of carrying out the null hypothesis due to the free term and ln(PIB_L) .014 ⋅ ln(CONS_L)ij-1 . we suggest a model equivalent in the meaning of interpretation and in certain items of elasticity: ln(CONS _ L)i j = b + a ⋅ ln( PIB _ L)ij + ε ij (5) The estimating of the parameters owing to OLS determines the following relation: ln ( CONS _ L )i j = 0. According to the above facts.Inefficient indices of estimations.The auto-correlation hypothesis is carried out so that the probability of the null hypothesis representing 0. and the Durbin-Watson statistics ranging within the interval that recommends that the hereby auto-correlation is not confirmed (DW Є [dL. the probability of attaining the null hypothesis representing maximum 0. reasons: - (8) This econometric model is better than the previous one due to the following Errors follow a normal distribution: mean 0 and repartition σ2 .936 ⋅ ln( PIB _ L)ij reasons: - (6) This econometric model is better than the previous one due to the following The auto-correlation hypothesis is carried out. (7) The estimating of the parameters owing to OLS determines the following relation: ln(CONS_L)ij = 0. Brush-Gefrey test (more recommended under theses circumstances.04.337 + 0.Over-evaluation of the determination report.84%. 4-d U ]. 4-dU] The hypothesis of the autocorrelation is fulfilled. statistics ρ = -0. 4-dU] = 2. accordingly.938 ⋅ ln(PIB_L)ij + +0. the probability by the null hypothesis being 0.Erroneous results after implementing the t-Student test that show an increased significance of indices. .02.

having p = 0. . the White test. MAKING PREDICTIONS We are going to make predictions for the previously displayed . T.. Academy Publishing House. Stoica. http://epp. 5edition. Bourbonnais. Greene..492 Euro per capita).europa.16 in case of ln (PIB_L)..851484.047758 (which represents the equivalent of 1006.326 Stanciu. The analysis of the model shows that in case of certain states belonging to the EU the errors are smaller. 4.277 Euro per capita in 2003 Romania. O. A. and according to the regression model. (2008) Econometrie. Bilă. Bucharest [6]. one gets a consumption expressed in a natural logarithm of 6.M.55 in case of the free term and 0. A. with a real value of 7. Economics. CONCLUSIONS The analysis contained by paragraph 8 demonstrates that it is more useful to approach the dependency in terms of elasticity considering the diversity of employed data (EU countries with different levels of development). pp. Bucharest [2]. N. The predictions are going to be made for the years 2003 and 2004. Andrei. B. Economics Publishing House. M. Subsequently. Stoicuţa. in proportion of 4. Vogelvang.eurostat. N. Stoicuţa.H. (2009) Adjusting Economic of Romania’s GDP Using Econometric Model of the System: Budget Expenditure-GDP. in case of a GDP representing 1150. (2005) Econometrics-Theory and Applications with Eviews. 5.E. according to the regression line shown above. Prentice Hall [7]. and is negatively influenced by the evolution of the previous year consumption. MicroCAD [5]. (2009) Econometric Modeling of the Energy System of the European Union.7% as a result of the PIB_LOC changes of the previous year. Stoicuţa.8 % due to the change of the PIB_LOC of the current year. The above model shows that the change of consumption in case of a unit increase of the income is going to alter in proportion of 93. Annals of the University of Petrosani. (2000) Econometric Analysis. N. P.The homo-skedasticity hypothesis is attained.. (1980) Identificarea si estimarea parametrilor sistemelor.P.. Stoicuţa. REFERENCES: [1]. Prentice Hall Inc [3]. International Scientific Conference XXIII. 9(2). Tertisco. 291-300 [4]. W.