# A tibble: 10 × 16
country year growth gini secedu popgr govexp invra…¹ infrate gdppcap credit
<chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 Algeria 1965 NA NA NA NA NA NA NA NA NA
2 Algeria 1970 NA NA NA NA NA NA NA NA NA
3 Algeria 1975 NA NA NA NA NA NA NA NA NA
4 Algeria 1980 NA NA NA NA NA NA NA NA NA
5 Algeria 1985 NA 38.7 NA NA NA NA NA 4060 40
6 Algeria 1990 -2 35.3 65 2.6 31.2 25 21 4358 14
7 Algeria 1995 1 35.3 69 2.2 30 25 12 4813 8
8 Algeria 2000 NA NA NA NA NA NA NA NA NA
9 Argent… 1965 NA NA NA NA NA NA NA NA NA
10 Argent… 1970 NA NA NA NA NA NA NA NA NA
# … with 5 more variables: m2 <dbl>, povertyratio <dbl>, d_ssa <dbl>,
# d_la <dbl>, d_tran <dbl>, and abbreviated variable name ¹invratioInequality and Economic Growth
Jaromír Baxa & Eva Hromádková
Outline
Effects of unequal income distribution
Measuring economic inequality
Stylized facts on developments of inequality
Empirical analysis: Effects of growth on inequality
What do we know about inequality and growth
Conclusions and policy implications
Does inequality matter?
Inequality and wealth affect daily behavior: from consumer choice over housing to access to education and health care.
To what extent does it matter for individuals and for growth?
Optimistic view:
Inequality creates incentives (for better education, for searching for better job…)
Does inequality matter?
Pessimistic views:
Inequality is bad for growth due to lower accumulation of human capital: Branko Milanovic (Finance and Development, 2011, Sept.) argues that broadly accessible education is a necessary for increasing human capital but it is difficult to achieve it unless a society has a relatively even income distribution.
Inequality causes macro instability if spendings of the poors financed through debt creation.
Inequality might destroy social cohesion of the community, increases crime and decreases public safety. Robert Putnam (2000): inequality and trust are mutually enforcing.
Relation between income inequality and liberty – first and foremost, empowerment for self-actualization (so called positive liberty).
Measuring Economic Inequality
Income vs. wealth inequality
Gini coefficient and the Lorenz curve.
Lorenz curve: proportion of the total income of the population (y axis) that is cumulatively earned by the bottom x% of the population.
Usually approximated from discrete intervals.
Gini = A/(A+B) where A is the area above Lorenz curve, B below; interval (0;1) or in %.
See http://en.wikipedia.org/wiki/Gini_coefficient
Measuring Economic Inequality
Derivation of the Lorenz curve and Gini coefficient for global income in 2011 CMG Lee. Data from Table 1 on http://www-cdn.oxfam.org/s3fs-public/file_attachments/tb-economy-one-percent-methodology-180116-en.pdf by Oxfam.
Gini coefficient: Income inequality
Source: http://en.wikipedia.org/wiki/Gini_coefficient; created by Wikkiwonkk
Gini coefficient: Wealth inequality
Source: http://en.wikipedia.org/wiki/Gini_coefficient.
Evolution of inequality: Stylized facts
Rise in inequality from the onset of the industrial revolution in 1750. Inequality has been rising until 1920s. First in the U.K., then inequality within countries rising globally (although some convergence across countries appeared as well).
Decreasing inequality in the global north (1930 - 1980). Progressive taxation, rising role of the state and fear from communism lead to new consensus that has brought reversal in inequality trends. Wages kept similar growth rates as productivity and income gains splitted almost equaly across population.
Globalization as well as other policy changes in the north lead to increased inequality in the rich countries while decreasing in the emerging countries. Middle class in Western countries is one of the only groups whose income decreased. On the other hand income of super rich kept rising. In 2007, inequality in the U.S. was back at the level of 1920’s.
- Currently, inequality is at its highest levels since 1950’s. The increase in inequality is reported also in traditionally egalitarian countries like Germany or Denmark (OECD report Divided we Stand: Why Inequality Keeps Rising).
Notes: That fiscal policy became less redistributive is confirmed by Bastagli et al. (Income inequality and fiscal policy, IMF, 2012)
The Elephant Chart
Evolution of real income between 1988 and 2008
Source: National Inteligence Council Main Report 2017 (p.13), originally Lakner and Milanovic, 2013, WB.
Wages did not necessarily grow with productivity (R. Reich)
https://www.nytimes.com/2011/09/04/opinion/sunday/jobs-will-follow-a-strengthening-of-the-middle-class.html
Empirical exercise
- We will estimate the relationship between economic inequality and economic development.
- Hypothesis of the Kuznets’ Curve: As the economy develops, market forces first increase inequality and then decrease economic inequality => inverted U-shaped curve.
More precisely:
In poor countries, economic growth increases the income disparity between rich and poor people, in wealthier ones the growth narrows the difference.
Why? Workers migrate from agriculture to industrial sectors, high capital concentration at the initial stage of development; after higher impact of human capital, more people switch to higher paying sector and inequality decreases.
Estimation
- is there any correlation between Gini and GDP?
\[ Gini_i = \alpha + \beta_1 log (GDP_i) \] 2. Quadratic model (corresponds to expected inverted U-shaped relation)
\[ Gini_i = \alpha + \beta_1 log (GDP_i) + \beta_2 (log (GDP_i))^2 \] - To confirm the existence of the Kuznets curve, both coefficients \(\beta\) need to be significant.
- We will use panel data for estimation that allow to infer variation across and within countries.
Panel data methods
Pooled data: OLS model disregarding panel data structure. Might suffer from heteroscedasticity.
Fixed-effects model: Adds dummy variable to each cross-sectional unit. Effectively demeans over cross-sections. Assumes that slope coefficients are the same across units => Focus on development over time, cross sectional dimension minimized.
\[ y_{i,t} = \delta_i + X_{i,t} \beta + \nu_{i,t} \]
FE model is less efficient than OLS. F-test to confirm differences across the fixed effects.
Between-effects model: Mean response model where the means are calculated for each of the units. To use this model, one should have many units. => cross sectional dimension only
\[ \overline {y_t} = \delta_i + \overline {X_{i,t}} \beta + u_{i} \]
Panel data methods (Cont.)
- Random-effects model: Residuals for given subgroups distributed randomly, not true values, uncorrelated with all other variables in the model. Occurs in cases where individuals are drawn randomly from a large population, such as household studies.
\[ y_{i,t} = \alpha + X_{i,t} \beta + \mu_{i} + \nu_{i,t} \]
- People generally want to use the random-effects model because they wish to estimate the variables that are constant within unit and do not evolve over time. Weighted average of the fixed and between estimates.
Quick reference: https://www.aptech.com/blog/introduction-to-the-fundamentals-of-panel-data/
Data
Dataset by Garbis Iradian accompanying his paper Inequality, Poverty, and Growth: Cross-Country Evidence (IMF WP/05/28, 2005)
Gini coefficient, GDP, data on education, trade openness, regional dummies, savings on physical and human capital, government expenditures, population growth, inflation rates, credit and M2.
83 countries, observations from 1960 to 2000 in five years increments.
Unbalanced panel.
The original paper can be downloaded at https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Inequality-Poverty-and-Growth-Cross-Country-Evidence-17956
Data
rm(list = ls(all = TRUE))
library(readxl)
dataset_iradian2005 <- read_excel("C:/Users/Baxa/Documents/Vyuka/Macro/Empirics5Inequality/dataset_iradian2005.xls", sheet = "gretlext")
head(dataset_iradian2005,10)Data transformations and packages
#Data transformations
dataset_iradian2005$log_gdppcap = log(dataset_iradian2005$gdppcap)
dataset_iradian2005$log_gdppcap_sq <- dataset_iradian2005$log_gdppcap^2
dataset_iradian2005$log_gini = log(dataset_iradian2005$gini)
dataset_iradian2005$n = log(dataset_iradian2005$popgr/100)
dataset_iradian2005$sk = log(dataset_iradian2005$invratio/100)
dataset_iradian2005$sh = log(dataset_iradian2005$secedu/100)
dataset_iradian2005$gexp = log(dataset_iradian2005$govexp/100)
#for a bit unconventional experiment
dataset_iradian2005$log_gdppcap_3 <- dataset_iradian2005$log_gdppcap*dataset_iradian2005$log_gdppcap_sqPanel structure
- Defining the panel structure by means of index variables
data_inequality <- pdata.frame(dataset_iradian2005, index = c("country", "year"))
head(data_inequality,10) country year growth gini secedu popgr govexp invratio infrate
Algeria-1965 Algeria 1965 NA NA NA NA NA NA NA
Algeria-1970 Algeria 1970 NA NA NA NA NA NA NA
Algeria-1975 Algeria 1975 NA NA NA NA NA NA NA
Algeria-1980 Algeria 1980 NA NA NA NA NA NA NA
Algeria-1985 Algeria 1985 NA 38.7 NA NA NA NA NA
Algeria-1990 Algeria 1990 -2 35.3 65 2.6 31.2 25 21
Algeria-1995 Algeria 1995 1 35.3 69 2.2 30.0 25 12
Algeria-2000 Algeria 2000 NA NA NA NA NA NA NA
Argentina-1965 Argentina 1965 NA NA NA NA NA NA NA
Argentina-1970 Argentina 1970 NA NA NA NA NA NA NA
gdppcap credit m2 povertyratio d_ssa d_la d_tran log_gdppcap
Algeria-1965 NA NA NA NA 0 0 0 NA
Algeria-1970 NA NA NA NA 0 0 0 NA
Algeria-1975 NA NA NA NA 0 0 0 NA
Algeria-1980 NA NA NA NA 0 0 0 NA
Algeria-1985 4060 40 70 13.9 0 0 0 8.308938
Algeria-1990 4358 14 48 15.1 0 0 0 8.379769
Algeria-1995 4813 8 42 12.2 0 0 0 8.479076
Algeria-2000 NA NA NA NA 0 0 0 NA
Argentina-1965 NA NA NA NA 0 1 0 NA
Argentina-1970 NA NA NA NA 0 1 0 NA
log_gdppcap_sq log_gini n sk sh gexp
Algeria-1965 NA NA NA NA NA NA
Algeria-1970 NA NA NA NA NA NA
Algeria-1975 NA NA NA NA NA NA
Algeria-1980 NA NA NA NA NA NA
Algeria-1985 69.03845 3.655840 NA NA NA NA
Algeria-1990 70.22052 3.563883 -3.649659 -1.386294 -0.4307829 -1.164752
Algeria-1995 71.89473 3.563883 -3.816713 -1.386294 -0.3710637 -1.203973
Algeria-2000 NA NA NA NA NA NA
Argentina-1965 NA NA NA NA NA NA
Argentina-1970 NA NA NA NA NA NA
log_gdppcap_3
Algeria-1965 NA
Algeria-1970 NA
Algeria-1975 NA
Algeria-1980 NA
Algeria-1985 573.6363
Algeria-1990 588.4317
Algeria-1995 609.6008
Algeria-2000 NA
Argentina-1965 NA
Argentina-1970 NA- Checking dataset structure
Balanced Panel: n = 83, T = 8, N = 664Note that there are plenty of missing values in the panel => panel is unbalanced. Nevertheless, most panel data methods can handle unbalanced datasets correctly.
Estimation I: Pooled model
m1_pooled <- lm(data_inequality$log_gini ~ data_inequality$log_gdppcap + data_inequality$log_gdppcap_sq)
summary(m1_pooled)
Call:
lm(formula = data_inequality$log_gini ~ data_inequality$log_gdppcap +
data_inequality$log_gdppcap_sq)
Residuals:
Min 1Q Median 3Q Max
-0.60504 -0.16000 -0.01268 0.17494 0.47210
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.11966 0.69063 -0.173 0.863
data_inequality$log_gdppcap 1.00268 0.17069 5.874 1.01e-08 ***
data_inequality$log_gdppcap_sq -0.06537 0.01044 -6.261 1.15e-09 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.2289 on 341 degrees of freedom
(320 observations deleted due to missingness)
Multiple R-squared: 0.1626, Adjusted R-squared: 0.1577
F-statistic: 33.1 on 2 and 341 DF, p-value: 7.301e-14Estimation I: Pooled model
m1_pooled <- lm(data_inequality$log_gini ~ data_inequality$log_gdppcap + data_inequality$log_gdppcap_sq)
summary(m1_pooled)
Call:
lm(formula = data_inequality$log_gini ~ data_inequality$log_gdppcap +
data_inequality$log_gdppcap_sq)
Residuals:
Min 1Q Median 3Q Max
-0.60504 -0.16000 -0.01268 0.17494 0.47210
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.11966 0.69063 -0.173 0.863
data_inequality$log_gdppcap 1.00268 0.17069 5.874 1.01e-08 ***
data_inequality$log_gdppcap_sq -0.06537 0.01044 -6.261 1.15e-09 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.2289 on 341 degrees of freedom
(320 observations deleted due to missingness)
Multiple R-squared: 0.1626, Adjusted R-squared: 0.1577
F-statistic: 33.1 on 2 and 341 DF, p-value: 7.301e-14Alternatively, function plm from the plm package can be used:
m1_pooled_plm <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq, model = "pooling", data = data_inequality)
#plm accepts second powers etc within formula, but operator I() is necessary:
#inequality_pooled_plm <- plm(log_gini ~ log_gdppcap + I(log_gdppcap^2), model = "pooling", data = data_inequality)
# summary(m1_pooled_plm) #The results are exactly the same.The plm function can estimate fixed effects, random effects, between model etc., also with instrumental variables (use help for reference if needed).
Estimation II: Fixed effects
m2_fe <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq, model = "within", data = data_inequality)
summary(m2_fe)Oneway (individual) effect Within Model
Call:
plm(formula = log_gini ~ log_gdppcap + log_gdppcap_sq, data = data_inequality,
model = "within")
Unbalanced Panel: n = 83, T = 3-8, N = 344
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-0.3691338 -0.0466083 -0.0021988 0.0502955 0.2068199
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
log_gdppcap 0.1541724 0.1031818 1.4942 0.1363
log_gdppcap_sq -0.0097881 0.0062500 -1.5661 0.1185
Total Sum of Squares: 2.2628
Residual Sum of Squares: 2.2389
R-Squared: 0.010532
Adj. R-Squared: -0.31038
F-statistic: 1.37846 on 2 and 259 DF, p-value: 0.25381Estimation II: Fixed effects
m2_fe <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq, model = "within", data = data_inequality)
summary(m2_fe)Oneway (individual) effect Within Model
Call:
plm(formula = log_gini ~ log_gdppcap + log_gdppcap_sq, data = data_inequality,
model = "within")
Unbalanced Panel: n = 83, T = 3-8, N = 344
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-0.3691338 -0.0466083 -0.0021988 0.0502955 0.2068199
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
log_gdppcap 0.1541724 0.1031818 1.4942 0.1363
log_gdppcap_sq -0.0097881 0.0062500 -1.5661 0.1185
Total Sum of Squares: 2.2628
Residual Sum of Squares: 2.2389
R-Squared: 0.010532
Adj. R-Squared: -0.31038
F-statistic: 1.37846 on 2 and 259 DF, p-value: 0.25381Extracting fixed effects possible via command fixef .
F-test to infer whether the individual fixed effects are significantly different to each other:
Estimation II: Fixed effects (Cont.)
Model with time-fixed effects along with individual effects:
m3_fe_te <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq, model = "within", effect = "twoways", data = data_inequality)
#notice the option effect = "twoways" within the 'plm' command.
summary(m3_fe_te)Twoways effects Within Model
Call:
plm(formula = log_gini ~ log_gdppcap + log_gdppcap_sq, data = data_inequality,
effect = "twoways", model = "within")
Unbalanced Panel: n = 83, T = 3-8, N = 344
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-0.2832801 -0.0466421 -0.0016122 0.0445694 0.2104374
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
log_gdppcap 0.1615711 0.0971816 1.6626 0.09764 .
log_gdppcap_sq -0.0150714 0.0060567 -2.4884 0.01348 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Total Sum of Squares: 1.9803
Residual Sum of Squares: 1.8615
R-Squared: 0.059946
Adj. R-Squared: -0.27952
F-statistic: 8.03491 on 2 and 252 DF, p-value: 0.00041422
F test for twoways effects
data: log_gini ~ log_gdppcap + log_gdppcap_sq
F = 7.2978, df1 = 7, df2 = 252, p-value = 5.662e-08
alternative hypothesis: significant effectsEstimation II: Fixed effects (Cont.)
Adding more regressors:
data_inequality$n[!is.finite(data_inequality$n)] <- NA
#log created some -infinite values in variable n - for countries with no population growth between two periods
m4_fe_ext <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq + n + sk + sh + gexp, model = "within", data = data_inequality)
summary(m4_fe_ext)Oneway (individual) effect Within Model
Call:
plm(formula = log_gini ~ log_gdppcap + log_gdppcap_sq + n + sk +
sh + gexp, data = data_inequality, model = "within")
Unbalanced Panel: n = 69, T = 2-7, N = 230
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-0.1531375 -0.0329887 -0.0025931 0.0302148 0.1531375
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
log_gdppcap -0.0096052 0.1569586 -0.0612 0.9513
log_gdppcap_sq 0.0014105 0.0088218 0.1599 0.8732
n 0.0169966 0.0308162 0.5515 0.5821
sk 0.0216901 0.0481041 0.4509 0.6527
sh -0.0235601 0.0531237 -0.4435 0.6580
gexp -0.2404196 0.0501081 -4.7980 3.736e-06 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Total Sum of Squares: 0.75662
Residual Sum of Squares: 0.64097
R-Squared: 0.15284
Adj. R-Squared: -0.2516
F-statistic: 4.66087 on 6 and 155 DF, p-value: 0.00021925m4_pooled_ext <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq + n + sk + sh + gexp, model = "pooling", data = data_inequality)
#summary(m4_pooled_ext)
pFtest(m4_fe_ext, m4_pooled_ext)
F test for individual effects
data: log_gini ~ log_gdppcap + log_gdppcap_sq + n + sk + sh + gexp
F = 22.663, df1 = 68, df2 = 155, p-value < 2.2e-16
alternative hypothesis: significant effectsThese results are robust upon inclusion of time fixed-effects.
Estimation III: Between model
m6_bm <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq, model = "between", data = data_inequality)
summary(m6_bm)Oneway (individual) effect Between Model
Call:
plm(formula = log_gini ~ log_gdppcap + log_gdppcap_sq, data = data_inequality,
model = "between")
Unbalanced Panel: n = 83, T = 3-8, N = 344
Observations used in estimation: 83
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-0.383089 -0.138780 -0.030557 0.141447 0.417405
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
(Intercept) -1.709190 1.708096 -1.0006 0.3200178
log_gdppcap 1.427721 0.425172 3.3580 0.0012041 **
log_gdppcap_sq -0.093266 0.026154 -3.5660 0.0006154 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Total Sum of Squares: 4.6302
Residual Sum of Squares: 3.5477
R-Squared: 0.23379
Adj. R-Squared: 0.21464
F-statistic: 12.2052 on 2 and 80 DF, p-value: 2.3651e-05m6_bm_ext <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq + n + sk + sh + gexp, model = "between", data = data_inequality)
summary(m6_bm_ext)Oneway (individual) effect Between Model
Call:
plm(formula = log_gini ~ log_gdppcap + log_gdppcap_sq + n + sk +
sh + gexp, data = data_inequality, model = "between")
Unbalanced Panel: n = 69, T = 2-7, N = 230
Observations used in estimation: 69
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-0.303019 -0.091419 -0.013120 0.136599 0.353395
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
(Intercept) -0.541354 1.799862 -0.3008 0.76459
log_gdppcap 0.986219 0.430678 2.2899 0.02544 *
log_gdppcap_sq -0.054893 0.026717 -2.0546 0.04414 *
n 0.092021 0.039663 2.3201 0.02364 *
sk 0.027097 0.088112 0.3075 0.75947
sh -0.133976 0.065813 -2.0357 0.04606 *
gexp -0.169507 0.081900 -2.0697 0.04266 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Total Sum of Squares: 3.3944
Residual Sum of Squares: 1.6901
R-Squared: 0.5021
Adj. R-Squared: 0.45392
F-statistic: 10.4207 on 6 and 62 DF, p-value: 5.7863e-08Discussion
The results not robust across estimation method; some models confirm the non-linearity effect, others do not.
Fixed effects: almost nothing significant.
Between effects: intuitive results.
Possible explanations: - The Kuznets curve evident on cross-country perspective, but within countries this hypothesis does not hold.
Omitted variable bias - Inclusion of another variable might help.
Unbalanced dataset – too many missing observation causing that the dimension of time-evolution becomes imprecisely estimated.
Elephant-chart style dynamics?
Changes in inequality suggested by the „Elephant graph” implies rising inequality with even higher income.
m7_fe_y3 <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq + log_gdppcap_3, model = "within", data = data_inequality)
summary(m7_fe_y3)Oneway (individual) effect Within Model
Call:
plm(formula = log_gini ~ log_gdppcap + log_gdppcap_sq + log_gdppcap_3,
data = data_inequality, model = "within")
Unbalanced Panel: n = 83, T = 3-8, N = 344
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-0.3658830 -0.0444620 -0.0029675 0.0468553 0.2166873
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
log_gdppcap 3.2887884 0.8502588 3.8680 0.0001390 ***
log_gdppcap_sq -0.4021540 0.1058550 -3.7991 0.0001812 ***
log_gdppcap_3 0.0161160 0.0043407 3.7128 0.0002512 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Total Sum of Squares: 2.2628
Residual Sum of Squares: 2.1254
R-Squared: 0.060718
Adj. R-Squared: -0.24874
F-statistic: 5.55931 on 3 and 258 DF, p-value: 0.0010356m8_fe_y3_ext <- plm(log_gini ~ log_gdppcap + log_gdppcap_sq + log_gdppcap_3 + n + sk + sh + gexp, model = "within", data = data_inequality)
summary(m8_fe_y3_ext)Oneway (individual) effect Within Model
Call:
plm(formula = log_gini ~ log_gdppcap + log_gdppcap_sq + log_gdppcap_3 +
n + sk + sh + gexp, data = data_inequality, model = "within")
Unbalanced Panel: n = 69, T = 2-7, N = 230
Residuals:
Min. 1st Qu. Median 3rd Qu. Max.
-0.1551202 -0.0291448 -0.0018054 0.0275818 0.1551202
Coefficients:
Estimate Std. Error t-value Pr(>|t|)
log_gdppcap 4.1377817 1.0749059 3.8494 0.0001731 ***
log_gdppcap_sq -0.4987835 0.1286436 -3.8773 0.0001561 ***
log_gdppcap_3 0.0198670 0.0050985 3.8966 0.0001451 ***
n -0.0154631 0.0306500 -0.5045 0.6146259
sk 0.0173795 0.0460569 0.3773 0.7064345
sh -0.0528133 0.0513994 -1.0275 0.3057917
gexp -0.2198310 0.0482519 -4.5559 1.055e-05 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Total Sum of Squares: 0.75662
Residual Sum of Squares: 0.58345
R-Squared: 0.22887
Adj. R-Squared: -0.14667
F-statistic: 6.5297 on 7 and 154 DF, p-value: 9.4179e-07These results are robust upon inclusion of time fixed-effects as well.
Other results in the literature
Robert Barro, Inequality and Growth in a Panel of Countries, Journal of Economic Growth (2000).
p-values in parentheses
Kuznets’ curve confirmed.
What do we know I: Inequality and Poverty
Is the growth good for the poors? To what extent?
David Dollar and Aart Kraay (2002, Journal of Economic Growth) estimated the effect of economic growth on income of the lowest quintile and they found that their income rise proportionally to the overall GDP. So growth should be good for the poors.
Over time, the extreme poverty decreased substantially.
What do we know I: Inequality and Poverty
Source: National Inteligence Council Main Report 2017 (p.4)
What do we know II: Inequality and Growth
Is inequality good for growth?
Sarah Voitchovsky (JEconGrowth, 2005) suggests that it depends where the inequality is concentrated. While the inequality among the rich (measured as income of 90th percentile over 75th percentile) is good for growth, inequality among the poors isn’t (50th percentile over 10th percentile)
Berg, Ostry and Zettelmeyer (EBRD WP, 2011) ask „What makes growth sustained”. Using panel of 140 countries they find that lower inequality, democratic institutions, openness (etc.) are important determinants of growth sustainability.
Human capital accumulation is lower when inequality is high (Milanovic, Finance and Development, 2011)
Bagchi and Švejnar (JComparativeEconomics, 2015) showed that inequality has negative effects on growth, but once controlled for political corruption and rent-seeking, the negative effect of inequality on growth disappears. Hence, they conclude that only politically connected wealth inequality and economic growth have a negative relationship. (The impact of political corruption is evaluated using the Forbes magazine’ss list of billionaires.)
What do we know III: Is there an optimal level of inequality?
Source: Cornia – Court: Inequality, Growth and Poverty in the Era of Liberalization and Globalization, UNI-WIDER, 2001
What do we know IV: Inequality and the Great Recession
Mechanism how inequality contributes to macro instability: if inequality large, S>>I causes excess liquidity => risks of financial instability if access to credits is eased (Hyman Minsky). Why S>>I? The demand for luxury goods does not compensate the fall in demand for consumer goods.
Raghuram Rajan (Fault Lines, 2010) argues that inequality has been one of the key driving mechanisms behind the Great Recession of 2008/2009. He points to the fact that the inequality increased sharply since 1980’s because of higher demand for highly skilled workers while the routine jobs were partly automated and partly outsourced. The policy response that permitted rising consumption: easing access to credit => housing boom => illusion of wealth.
Then, tax-cuts and deregulation on financial markets contributed to rise of the income share of the top 1%: in the US rose from 8% of annual GDP to more than 20% of GDP => lack of demand, concentration of income leading to a savings glut simultaneously.
Savings from abroad financed booms in the EU periphery as well.
The instability argument pointed out also by other authors, i.e. James K. Galbraith: Inequality and Instability. A study of the World Economy Just Before the Crisis. 2012.
What do we know IV: Inequality and the Great Recession
Widely accepted that policy responses to the Great Recession increased inequality.
Ball, Leigh and Loungani (2011, IMF FD): “The research shows the pain is not borne equally. Fiscal consolidation reduces the slice of the pie going to wage-earners. For every 1 percent of GDP of fiscal consolidation, inflation-adjusted wage income typically shrinks by 0.9 percent, while inflation-adjusted profit and rents fall by only 0.3 percent. Also, while the decline in wage income persists over time, the decline in profits and rents is short-lived.” This implies that fiscal consolidations are connected with a risks of jobless recovery in many countries with negative effects on employment, human capital and risks of long-term growth.
Agnello and Sousa (2012, How does fiscal consolidation impact on income inequality); Woo et al. (2013, Distributional consequences of fiscal consolidation, IMF) confirm the finding that inequality is rising after periods of fiscal consolidations as well.
The impact of monetary policy still somewhat inconclusive. Why some effect to be expected: different sources of incomes and portfolios across income distribution react differently to policies. Increasing asset prices and profits contribute to rise of incomes and wealth of the rich.
http://voxeu.org/article/monetary-policy-and-inequality-us;
https://www.clevelandfed.org/en/newsroom-and-events/publications/economic-commentary/2017-economic-commentaries/ec-201701-monetary-policy-and-inequality.aspx
Policy implications
Should policy makers take care about economic inequality?
There is some evidence they shall do so, considering potentially positive effects for growth and redistributive nature of policies adopted in the aftermath of the Great Recession.
Inequality in the developed countries fostered by globalization as well.
Nevertheless, potential costs of inequality have been neglected by policy makers for long, and the effect on actual policy making materialized relatively recently.
Currently, many international institutions now share the view that inequality is a threat for future long-term growth and provide policy recommendations to mitigate inequality (for example, the OECD recommends reform of tax system favouring redistribution of income (both wage and capital income) and high quality of public services to assure equal access to high quality education and health system.
See http://www.oecd.org/social/name,59278,en.htm).
Key Points
Although most of the models in current macroeconomics is based on the assumption of heterogenous agents, economic inequality is an important phenomenon.
Inequality is measured by Gini coefficient (among other indicators such as ratio of incomes of upper and lower quintile etc.).
If nonlinear relationship shall be tested, squares or cubes of regressors can do the job (nonlinear terms significant => relationship confirmed).
The impact of growth on inequality supports the hypothesis of the Kuznets curve across countries, but the evidence from the within regression is weak at best.
Mechanisms how inequality might affect growth and contribute to the Great Recession + some of the stylized facts.
Simon Kuznets (1901-1985)
1971 The Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel “for his empirically founded interpretation of economic growth which has led to new and deepened insight into the economic and social structure and process of development”
His central thesis: underdeveloped countries of today possess characteristics different from those that industrialized countries faced before they developed => growth is not automatic.
http://en.wikipedia.org/wiki/Simon_Kuznets
http://cepa.newschool.edu/het/profiles/kuznets.htm