Empirical appendix of Public Expenditure Distribution, Voting, and Growth

Transcription

1 Empirical appendix of Public Expenditure Distribution, Voting, and Growth Lorenzo Burlon August 11, 2014 In this note we report the empirical exercises we conducted to motivate the theoretical insights of the paper. All Stata codes used in the exercises are available online at research/research_02.html. 1 Data We use Eurostat data for government expenditure by function, value added by branch, nominal and real GDP, public debt, population, and level of education. We merge these data with OECD data on the same variables in order to increase the pool of countries and years. We use also the World Input-Output Database (WIOD) for the national Input-Output tables in order to have a proxy for the complementarity across sectors. The final database consists of an unbalanced panel of 39 countries for 17 years, from 1995 to We cannot extend further the pool of countries because only those included in the analysis possess a classification of public expenditure by function (COFOG). 1.1 Government expenditure by function and value added by branch We get the data on public expenditure distributions from Eurostat and OECD. By merging the two databases, we cover 30 countries from 1995 to We use the annual government finance statistics, and in particular the general total government expenditure by function (COFOG) (gov a exp). We consider only total expenditure (TE), with no distinction between national accounts indicators such as intermediate and final consumption expenditures, compensation of employees, subsidies, property incomes, capital and other transfers, gross capital formation, or acquisitions/disposals of non-financial non-produced assets. Moreover, we consider only general expenditure (S13), with no detail on the part of expenditure that is due to the central, state, or local government, or social security funds. We use a 1

2 disaggregation of total expenditure by 10 functions (GF01-GF10). 1 The data are expressed in millions of ECU up to 1998 and in millions of euro from 1999 onwards. We get the data on value added distributions from Eurostat and OECD. We cover 30 countries from 1995 to We employ the annual national accounts by 10 branches, with aggregates at current prices (nama nace10 c). 2 We consider gross value added at basic prices. The unit is Millions of euro from 1999 and millions of ECU up to We compute the Gini coefficients for both the public expenditure and the value added distributions. The formula for the Gini coefficient is G t = 2ΣJ j=1 jf t(j) JΣ J j=1 f t(j) J + 1 J, where J is the total number of sectors, j is the rank of each sector in a given year, and f t (j) is the share of sector j in year t. Figure 1 reports the Gini cofficients for all the countries in our final sample over time. There seems to be a comovement between public expenditure concentration and value added concentration, with the exception of Finland and Malta. The correlation between public expenditure and value added is 0.13 on average. The correlation between value added and previous period s public expenditure is Moreover, both distributions are persistent over time and are not subject to sudden and drastic changes. Notable exceptions are, e.g., Iceland, Ireland, and Latvia during the crisis years. 1.2 Complementarity across sectors To our knowledge, there is no study in the literature that estimates the complementarity across sectors consistently for several countries. 4 Hence, we construct a uniform proxy for the level of complementarity/substitutability across sectors for different countries using the World Input-Output Database (WIOD). This database reports the Input-Output (I-O) tables of 35 industries in 39 countries from 1995 to If a sector delivers a large enough share of its production as intermediate goods for the production in another sector, then we consider that these two sectors are complementary in the production of the consumption good. Thus, we use as a proxy for the pairwise complementarity across sectors the 1 We conduct the same exercises also with a disaggregation by 69 functions (GF0101-GF1009) and with the detail of the Economic affairs function (GF04) into 9 sub-functions (GF0401-GF0409). There are no significant differences in the result, although the classification by 69 functions induced more volatile measures of concentration for several countries. 2 We can use a finer disaggregation into 64 branches. Results are similar as in the case of a finer public expenditure disaggregation, while the measures of concentration (the Gini coefficients) seem more volatile. For the OECD countries, we use the classification in 16 branches, which is the closest to the NACE classification used in Eurostat. 3 These average correlations refer to the 31 countries with at least 10 observations. 4 One recent exception is Atalay [2013], although its estimates for the elasticity of substitution across sectors apply only to one year (2002) and only to a pool of 7 countries. 5 An alternative would be the OECD Input-Output data, which covers a similar span of countries. 2

3 Australia Austria Belgium Bulgaria Canada Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Iceland Ireland Israel Italy Japan Korea Latvia Lithuania Luxembourg Malta Mexico Netherlands New Zealand Norway Poland Portugal Romania Slovak Republic Slovenia Spain Sweden Switzerland Turkey United Kingdom United States Year Value Added Gini Public Expenditure Gini Figure 1: The Gini coefficient across countries and over time. off-diagonal entries of the I-O tables. For each country in each year, we compute the ratio between offdiagonal and total entries of the I-O table. A high ratio indicates that on average the sectors in a certain country are complementary in the production of the final consumption goods. A low ratio indicates that sectors do not need the output of other sectors to realize their own production and are therefore substitutable in the production of the final consumption good. 6 See Jones [2013] for a recent analysis of the relation between the I-O structure of an economy and economic growth, and Acemoglu et al. [2012] for a study on the dependence of aggregate volatility on the I-O structure. Figure 2 reports the complementarity measure over time for each country. This measure seems to be almost constant for some countries (e.g., Italy) and evolving for others (e.g., Austria), although the changes over time are not dramatic (except for Luxemburg). 7 6 Among other limitations, this approach cannot account for final consumption goods complementarity on the demand side. Nevertheless, it provides a comparable and uniform measure of intermediate goods complementarity in production for 40 countries and across 15 years. 7 We compute the measure of complementarity using only domestic flows and using both domestic flows and imports from abroad, a detail that the WIOD database provides. We do not have the detail of the industry destination of exports, so 3

4 Australia Austria Belgium Brasil Bulgaria Canada China Complementarity measure from I-O tables Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary India Indonesia Ireland Italy Japan Korea Latvia Lithuania Luxembourg Malta Mexico Netherlands Poland Portugal Romania Russia Slovak Republic Slovenia Spain Sweden Taiwan Turkey United States Year Figure 2: The complementarity measure across countries and over time. 1.3 Complementarity and the evolution of the public expenditure distribution We merge the data from Eurostat and OECD on public expenditure and value added distribution with the data from WIOD on the complementarity across sectors for the different countries. We are left with a sample of 33 countries for 15 years, from 1995 to If the government proposes only growthenhancing reforms of the public expenditure distribution, our model predicts that countries whose sectors are more complementary tend to decrease the concentration of their public expenditure over time, while countries whose sectors are more substitutable tend to increase the concentration over time. Figure 3 shows the relation between the initial level of complementarity for each country and the change in the Gini coefficient of the public expenditure distribution. A higher initial complementarity predicts a lower we ignore them. The comparison between the two measures tells us that there is neither significant differences between the average over time for each country nor in the evolution of the measure over time for each country. Malta and Luxembourg are two exceptions. Hence, we can use one of the two measures. Given that we do not have data on the destination of exports, for consistency we use the version with data only on domestic inter-industry flows of intermediate goods. 4

5 Predicted Change in Gini Coefficient Complementarity in the first year Figure 3: The relation between the complementarity across sectors and the change in the Gini coefficient of public expenditure. The line is the linear fit of the data, the bands represent the 95% confidence interval. change in the Gini coefficient, which becomes even negative for a high enough complementarity. In other words, a higher initial complementarity predicts a diversification of public expenditure. A lower initial complementarity instead predicts a higher change in the Gini coefficient, that is, a concentration of public expenditure. Since we do not control for any country- or year-specific characteristics and the variables are arguably endogenous to each other, these correlations are not statistically significant and we cannot claim any causal relation. We move therefore to a regression analysis that controls for a series of observables. 2 Estimation We test the implications of the model by means of reduced-form regressions. First, we check whether there is a comovement between the public expenditure distribution and the value added distribution. Second, we test whether a change in the concentration of the public expenditure leads to higher or lower 5

6 GDP growth depending on whether the sectors are more substitutable or complementary. 2.1 The comovement between public expenditure and value added distributions Our model predicts that the public expenditure distribution drives the future distribution of value added, and that the current distribution of value added, which proxies the distribution of vested interests in the economy, influences the distribution of public expenditure. Hence, we test two relations. First, the Gini coefficient of the value added distribution of country i at time t is a function of the Gini coefficient of the public expenditure distribution in the previous year, that is, EQ. 1: Value Added Gini it = β Expenditure Gini it 1 + δ X it + γ t + γ i + ε it, where X it includes controls for a series of country/year characteristics such as nominal GDP level, population, general government consolidate gross debt as a percentage of GDP, percentage of active population (15-64 years), and percentage of active population with tertiary education, while γ t and γ i are time- and country-fixed effects. The element ε it represents the country-year errors. Second, the public expenditure Gini of country i at time t is a function of the value added Gini of the same period, that is, EQ. 2: Expenditure Gini it = β Value Added Gini it + δ X it + γ t + γ i + ε it. Table 1 reports the summary statistics in the regression sample of dependent and independent variables. The nominal GDP and the population are supposed to capture size effects at the country-year level. Real GDP growth and public debt keep track of the economic and public finance situation of the country, while real GDP per capita measures the development stage. The share of the population at working age proxies the demographic balance between working and non-working age population. Table 2 and Table 3 report the results of these regressions. Our regression sample consists of an unbalanced panel of 29 countries for 16 years, from 1997 to 2011, with a total of 422 observations. Model 1 in both regressions is simply the correlation between public expenditure and value added Ginis. The correlation is positive, thus confirming the graphical intuition of Figure 1 for which there is a comovement between the two variables. Model 2 introduces country-year characteristics. In order to fully control for country- and time-specific effects that may not be captured by our controls, we also introduce year and country dummies. We test for the hypothesis that either of these sets of dummies has zero effect, and the F -test in both cases rejects the null. The coefficient of the main regressors remains positive and statistically significant. The public expenditure distribution drives the future evolution of the value added distribution and the value added distribution influences the current public expenditure distribution, even after controlling for size, demographics, development level, economic and public finance situation, and any country- or time-fixed effects. 6

7 Variable Mean Std. Dev. Min. Max. N Value added Gini Expenditure Gini Nominal GDP Public debt Real GDP growth Population Real GDP per capita Share population Table 1: Summary statistics of the regression sample. The Gini coefficients are by construction between 0 and 1, where 0 represents a fully diversified distribution and 1 fully concentrated distribution. Nominal GDP is expressed in trillions of Euros. The general government consolidated gross debt is in percentage of GDP. Real GDP has reference year 2000, with 2000 exchange rates. Population is expressed in millions. The real GDP per capita is in millions of Euros per capita (e.g., the mean is Euros per capita). The share of the population between years old keeps track of the population at working age. 2.2 The effect of the public expenditure distribution on economic growth We test whether a change in the concentration of the public expenditure distribution decreases or increases growth depending on the level of complementarity across sectors. Our model predicts that if sectors are complementary, an increase in the Gini of the public expenditure distribution leads to an increase in future output. If instead sectors are substitutable, an increase in the expenditure Gini leads to a decrease in future output. Our regression model is Real GDP it = β c Expenditure Gini it 1 + γ t + γ i + ε it, where the time dummies γ t, the country dummies γ i, and error term ε it are defined as above. The index c on β c refers to the fact that we run two separate regressions for two distinct sets of observations. We partition the observations (the time-country indexes) depending on the corresponding level of complementarity. The complementarity is measured by our proxy from the I-O tables as the share of intersectoral flows over total flows. If the complementarity index is high, we assign the observation to the complementary-sectors subsample. If the complementarity index is low, we assign the observation to the substitutable-sectors subsample. Figure 4 reports the histogram of the complementarity index. By construction, the complementarity index is 0 for the minimum complementarity across sectors -perfect substitutability- and 1 for the maximum complementarity. Based on the histogram and on different specifications, we define an observation as belonging to the complementary-sectors case if the complementarity 7

8 Dependent variable: Value added Gini Model 1 Model 2 Lagged expenditure Gini 0.118*** 0.080** NGDP level Public debt 0.050*** Real GDP growth 0.060* Population Real GDP per capita 1.671*** Share population Country Dummies (29) No Yes Year Dummies (16) No Yes Adj.R-squared Observations * p < 0.10, ** p < 0.05, *** p < Table 2: Concentration of the value added distribution as a function of past public expenditure concentration. We report the corresponding standard error under each estimated coefficient. The regression includes a constant. index is greater than or equal to Our model predicts that the β coefficient is negative for the complementary-sectors case and positive for the substitutable-sectors case. An increase of the public expenditure concentration leads to lower growth if sectors are complementary and to higher growth if the sectors are substitutable. Table 4 reports the results of this regression. The first column of Table 4 reports the complementary-sectors case, while the second column reports the substitutable-sectors case. The complementary-sectors subsample consists of an unbalanced panel of 14 contries and 15 years (from 1995 to 2009), for a total of 100 observations. The substitutable-sectors 8 This value corresponds to slightly less than the 75-th percentile. 8

9 Dependent variable: Public expenditure Gini Model 1 Model 2 Value added Gini 0.180*** 0.139** NGDP level ** Public debt 0.038*** Real GDP growth *** Population 0.003* Real GDP per capita Share population ** Country Dummies (29) No Yes Year Dummies (16) No Yes Adj.R-squared Observations * p < 0.10, ** p < 0.05, *** p < Table 3: Concentration of the public expenditure distribution as a function of current value added concentration. We report the corresponding standard error under each estimated coefficient. The regression includes a constant. subsample consists of an unbalanced panel of 26 countries and 15 years, for a total of 258 observations. 9 We control in both cases for time- and country-speficic fixed effects, and we reject the hypothesis that these dummies are not significant. The coefficient of the lagged expenditure Gini is negative for the complementary-sectors case and positive for the substitutable-sectors case, as expected. This means that an increase in the concentration of the public expenditure leads to a higher (lower) future output if sectors are substitutable (complementary). 9 The WIOD I-O tables arrives until

10 Density Complementarity Level Figure 4: Histogram of the complementarity index. References Daron Acemoglu, Vasco M. Carvalho, Asuman Ozdaglar, and Alireza Tahbaz-Salehi. The network origins of aggregate fluctuations. Econometrica, 80(5): , Enghin Atalay. How important are sectoral shocks? Manuscript, University of Chicago, October Charles I. Jones. Misallocation, Input-Output Economics, and Economic Growth, volume II of Advances in Economics and Econometrics - Tenth World Congress. Cambridge University Press,

11 Dependent variable: Real GDP Complementary sectors Substitutable sectors Lagged expenditure Gini ** 1.713* Constant 0.228*** Country Dummies Yes (14) Yes (26) Year Dummies Yes (15) Yes (15) Adj.R-squared Observations * p < 0.10, ** p < 0.05, *** p < Table 4: Real GDP as a function of the concentration of the public expenditure in the previous period. We report the corresponding standard errors under each estimated coefficient. 11