Post on 06-Jul-2020
transcript
Inflation and the Rich After the Global
Financial Crisis
Branimir Jovanovic
National Bank of the Republic of Macedonia
jovanovicb@nbrm.mk
February 13, 2014
VERY PRELIMINARY
ABSTRACT
We investigate the relationship between in�ation and the change in the top 10 percent
share of income after the global �nancial crisis of 2007-2008 in 42 countries. Contrary to
the recent literature, we �nd that higher in�ation is associated with greater reduction in the
top 10 share. We dismiss omitted variables or reverse causality as potential explanations for
the association. The main reason why would in�ation reduce inequality is that low-income
individuals have been protected from in�ation by minimum wage increases, di¤erently from
high-income individuals. Higher social transfers, due to higher revenues brought by in�ation,
may have also played a role in some countries.
JEL: D31, E31, G01
Keywords: in�ation, inequality, income distribution, top10 share, �nancial crisis, Great
Recession
1
I. Introduction
One of the elements of the consensus in macroeconomics before the global �nancial cri-
sis of 2007-2008 was that in�ation should be low and stable (Goodfriend (2007), Woodford
(2009)). In accordance with this, many central banks adopted in�ation targets of around 2
percent. With the emergence of the �nancial crisis, some challenged this consensus. Blan-
chard, Dell�Ariccia and Mauro (2010) questioned whether the distortions from higher in�ation
are really so high, and if they are higher than the potential bene�ts of avoiding the zero inter-
est rate bound. Rogo¤ (2008) proposed to use in�ation as one of the instruments for reducing
public debts, while Aizenman and Marion (2011) argued that raising in�ation to 6% could
erode US public indebtedness by 20 percent within 4 years.
Recent research argue that optimal in�ation in standard new Keynesian models remains
below or around 2 percent even when one allows for zero lower bound on the interest rate,
due to the low frequency of its appearance (Coibion, Gorodnichenko and Wieland (2012),
Schmitt-Grohé and Uribe (2010)). The situation is far from clear, however, because adding
certain features to the models easily increases the optimal rate of in�ation. For example,
Di Bartolomeo, Acocella and Tirelli (2013) demonstrates that when in�ation serves as a form
of taxation for �nancing social transfers, the optimal in�ation rate increases.
Another strand of literature that has attracted a lot of attention recently is income in-
equality. Many have pointed at rising income inequality in the U.S. as the fundamental cause
of the global �nancial crisis of 2007-2008 (see Stiglitz (2009), Milanovic (2009), Wade (2009),
Fitoussi and Saraceno (2010) and Rajan (2010)). Although the studies di¤er slightly in the
elaborations, the main idea is that the shift of income towards the rich and the fall of the
middle class in the U.S. created political pressures that were accommodated by easier credit.
This created a credit boom and eventually led to a �nancial crisis in the U.S., which spread
to the rest of the world. Ranciere and Kumhof (2010) develop a dynamic stochastic general
equilibrium model in which increase in inequality leads to a �nancial crisis.
Coibion et al. (2012) have linked these two strands of literature, analyzing the contribution
of monetary policy e¤orts for lowering the in�ation to income inequality. They have found
2
that the disin�ationary policies in the U.S. in the early 1980s have lead to a persistently
higher inequality in the 1990s.
Coibion et al. (2012) are not the �rst to analyze the relationship between in�ation and
income inequality. The in�ation-inequality nexus has attracted researchers� attention for
some time, and older studies have usually found that in�ation is likely to reduce inequality.
Bach and Ando (1957) analyze the redistributional e¤ects of in�ation in the US in 1939-1952,
while Bach and Stephenson (1974) in 1946-1971, �nding that, in general, in�ation has shifted
income from business pro�ts to wages and salaries, and from lenders to borrowers, and,
hence, has a¤ected equality favorably. Nordhaus (1973) attempts to determine the e¤ects of
anticipated in�ation on lifetime income, using three theoretical model frameworks - Classical,
Neoclassical and Keynesian. He �nds that in�ation is an equalizing factor in all the three
systems, although he notes that the simulation technique used is rather complicated and
sensitive to the assumptions.
More recent studies, on the other hand, have established a positive correlation between
in�ation and inequality across countries during 1960s-1990s (Beetsma and van der Ploeg
(1996), Dolmas, Hu¤man and Wynne (2000), Bulir (2001), Albanesi (2007), Crowe (2006)).
However, the arguments presented therein suggest that it is high inequality that leads to
higher in�ation. As Beetsma and van der Ploeg (1996) argue, in democratic societies with
high inequality, the median voter would prefer pro-poor policies, i.e. policies that redistribute
income from the rich to the poor, such as unanticipated in�ation, which would imply that
higher inequality results in higher in�ation. Similar mechanism is proposed by Dolmas,
Hu¤man and Wynne (2000), who argue that with higher inequality, the median voter prefers
higher in�ation as a way of �nancing higher government expenditures. Crowe (2006) and
Albanesi (2007) also provide models in which higher inequality raises in�ation, although for
di¤erent reasons. In their models, greater income inequality leads to greater inequality in
political in�uence, with the rich having greater political power. If the rich perceive that
in�ation will favour them, because the in�ation tax may hurt the poor disproportionately
more, then they may push for in�ationary policies.
3
This study will try to contribute to these debates, by investigating the relationship be-
tween in�ation and the top 10 percent share of income, during the Great Recession, on a
sample of 42 countries. Results suggest that higher in�ation during the Great Recession has
been accompanied by a decline in the top 10 share of income. The relationship is not likely
to be due to omitted variables or reverse causality. The main explanation is that countries
that have experienced higher in�ation during this period have also seen an increase in their
minimum wages, which has protected low-income individuals�income, but has not a¤ected
the income of the top 10 percent. In addition, higher in�ation may have brought "windfall"
revenues for some governments, which have been used for higher social transfers.
The paper proceeds as follows. Section II presents the methodology and the data used.
Section III investigates the relationship between in�ation and the change in the top 10 share
of income, using cross-country data, and assesses whether it can be interpreted causally.
Section IV delves more deeply into the underlying mechanisms that explain the relationship,
using both cross-country and household-survey data. Section V concludes.
II. Methodology and data
To assess the relationship between the change in inequality and the in�ation, simple
regression will be run, where the change in the income going to the top 10 share of population
after the �nancial crisis will be regressed on the average in�ation during the same period.
The change in the top 10 share after the crisis will be measured as the average annual change
between 2007 and the last data point (2010 or 2011). In�ation will be measured as the
average annual consumer price in�ation during 2008-2010. To gain some insights about the
underlying mechanisms driving the relationship, the same regression will be run for di¤erent
groups of countries.
To assess the underlying mechanisms in greater detail, data from household surveys will
be used. More precisely, it will be compared how income from the three main sources -
labour, capital and social transfers, changed after the crisis (2010 vs. 2007) for households
with di¤erent levels of income.
4
The macro analysis will cover the following 42 countries: Argentina, Austria, Belgium,
Bulgaria, Canada, Colombia, Cyprus, Czech Republic, Denmark, Dominican Republic, Ecuador,
Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Japan, Latvia,
Lithuania, Luxembourg, Malta, Moldova, Netherlands, New Zealand, Norway, Paraguay,
Peru, Poland, Portugal, Romania, Singapore, Slovakia, Slovenia, Spain, Sweden, United
Kingdom, United States, and Uruguay.
We chose to work with the top 10 percent (instead of the top 1 percent, for instance) since
data on the top 10 percent are available through several di¤erent sources, which increases the
number of observations. Three di¤erent sources will be used: the World Top Income Database
(WTID) of Alvaredo et al. (2013), Eurostat, and the World Development Indicators (WDI)
of the World Bank. For the countries that have data in more than one database, the primary
source is the WTID, and then Eurostat.1 The sources di¤er in their methodology: the
WTID compiles the data from tax records, the other two, from surveys. As a result, there
are signi�cant di¤erences across databases, for countries that are present in two of them. For
instance, 10 countries are accounted for both in the WTID and the Eurostat. For all of them,
the top 10 share is much higher in the former than in the latter (see Figure 1). However, it
can be also noted that the di¤erence between the two is relatively stable.2 Since our focus is
on the within-country change in inequality (i.e. we do not compare levels of inequality across
countries, but changes in inequality), the use of di¤erent methodologies should not represent
a signi�cant problem for our analysis.
1. Exceptions are countries on which the WTID has data only until 2009, and Eurostat until 2010 or 2011.These countries are represented by the Eurostat data. These are: Finland, France, Ireland, Italy, Norway,Switzerland, and the U.K.
2. Exception is Norway, for which the two data sources display some divergent movements in 2006 and2007.
5
Figure 1: Comparison of the top 10 percent share of income
in the two databases (%)
Table 1 shows average annual change in the top 10 share after the crisis, the last data
point and the data source for each country.
The micro analysis will be done using household survey data from the Luxembourg In-
come Study (LIS). Of the 42 countries analyzed in the macro part, these 15 countries are
present in the LIS both before the crisis (2007) and after it (2010): Colombia, Estonia, Fin-
land, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Poland, Slovakia, Slovenia,
Spain, United Kingdom, United States.
6
Table1:Changeinthetop10shareafter
thecrisisacrosscountries
Country
Change
Lastpoint
Source
Country
Change
Lastpoint
Source
Country
Change
Lastpoint
Source
Ecuador
-1.64
2010
WDI
Cyprus
-0.27
2011
Eurostat
CzechR.
0.00
2011
Eurostat
Peru
-1.08
2010
WDI
Spain
-0.27
2010
WTID
Latvia
0.00
2011
Eurostat
Iceland
-1.00
2011
Eurostat
Greece
-0.25
2011
Eurostat
Slovakia
0.03
2011
Eurostat
Romania
-1.00
2011
Eurostat
Canada
-0.21
2010
WTID
Bulgaria
0.05
2011
Eurostat
Argentina
-0.80
2010
WDI
Japan
-0.18
2010
WTID
Hungary
0.08
2011
Eurostat
Uruguay
-0.65
2010
WDI
Finland
-0.18
2011
Eurostat
Slovenia
0.08
2011
Eurostat
DominicanR.
-0.65
2010
WDI
Poland
-0.18
2011
Eurostat
Sweden
0.14
2011
WTID
Estonia
-0.50
2011
Eurostat
Luxembourg
-0.15
2011
Eurostat
UK
0.15
2011
Eurostat
Netherlands
-0.50
2011
Eurostat
New
Zealand
-0.11
2010
WTID
US
0.22
2011
WTID
Moldova
-0.48
2010
WDI
Colombia
-0.11
2010
WDI
Malta
0.25
2011
Eurostat
Lithuania
-0.38
2011
Eurostat
Italy
-0.05
2011
Eurostat
Denmark
0.29
2010
WTID
Portugal
-0.38
2011
Eurostat
Norway
-0.02
2011
Eurostat
Ireland
0.57
2010
Eurostat
Paraguay
-0.37
2010
WDI
Austria
0.00
2011
Eurostat
Singapore
0.60
2010
WTID
Germany
-0.30
2011
Eurostat
Belgium
0.00
2011
Eurostat
France
1.00
2011
Eurostat
7
III. Analysis
If one compares the change in the top 10 share of income after the global �nancial crisis
with the in�ation during the same period for these 42 countries, one can observe that there
is a clear negative correlation between the two: higher in�ation in the post-crisis period is
associated with a higher decline in inequality (see Figure 2 and Equation 1).
Figure 2: Inflation and change in inequality after the crisis
(1)
top10_ch = 0:21 �0:12���inflation
(0:10) (0:02)
n = 42;R2= 0:32
This association, however, cannot be interpreted as causal straightforwardly. It may
happen that some omitted variables, a¤ecting both the change in the top 10 share and the
in�ation, are driving the results. It may also happen that the causality �ows from inequality
8
to in�ation, as Beetsma and van der Ploeg (1996), Dolmas, Hu¤man and Wynne (2000),
Crowe (2006) and Albanesi (2007) have argued.
The literature on determinants of inequality identi�es several groups of factors that may
a¤ect inequality (see Roine, Vlachos and Waldenström (2009), Scheve and Stasavage (2009),
Afonso, Schuknecht and Tanzi (2010), Jovanovic (2014)). The �rst group refers to economic
activity. GDP growth may a¤ect inequality both positively and negatively, depending on the
conditions. Since GDP growth may also a¤ect in�ation, failure to control for its in�uence
is likely to bias the results. Economic activity in certain sectors may also a¤ect inequality,
like �nancial shocks. The second group refers to developments in inequality itself, like trends
or initial conditions. For example, if inequality is on a downward trend for a longer period
of time, for whatever reasons, and if the trend is somehow correlated with the in�ation,
the omission of the trend may bias the results. The third groups of factors is related to
public policy, like taxes, or social transfers, or government size. For instance, increase in
taxes may reduce inequality and lead to higher in�ation at the same time. The next group
comprises factors related to labour market characteristics, like labour marker regulations or
minimum wages. Increase in labour market regulations, or minimum wages, may reduce
inequality, and if they are related to in�ation, their omission may lead to wrong inference.
Monetary policy may also a¤ect both inequality and in�ation - monetary expansion may lead
to higher in�ation and to higher inequality, through higher �nancial income for the better-o¤
individuals. If one fails to control for this, one may wrongly conclude that in�ation a¤ects
inequality. The next group of factors refers to institutions and politics - political parties of
certain ideology may pursue in�ationary policies and push for higher or lower redistribution.
Finally, both in�ation and inequality may be a¤ected by productivity and openness.
We evaluate the e¤ect of these factors in turn, in Table 2. Each column presents the results
of one regression, with the column headline indicating which group of factors is included.
The exact de�nitions of the variables and the data sources are displayed in Table A1 in the
Appendix. As can be seen from Table 2, in all the cases the coe¢ cient on in�ation remains
intact, implying that the negative association between in�ation and the change in the top 10
9
share is not likely to be due to omitted variables.
Table 2: Adding variables to the basic regression-1 -2 -3 -4 -5 -6 -7 -8
Baseline Crisisandrecovery
Initialineq.andtrend
Govern-ment
Labour Monetary Institutions Productivityandopenness
in�ation -0.12*** -0.11*** -0.10*** -0.10*** -0.08*** -0.10*** -0.10*** -0.10***(0.02) (0.030) (0.021) (0.029) (0.028) (0.023) (0.027) (0.022)
fall_gdp -0.02(0.015)
recov_gdp 0.00(0.025)
fall_se -0.00(0.001)
recov_se 0.00(0.003)
top10_init -0.02(0.010)
top10_trend -0.11(0.116)
gov_size 0.01(0.026)
tax_top -0.00(0.009)
tax_top_ch -0.00(0.016)
bene�ts -0.00(0.006)
labour 0.09*(0.047)
participation -0.02**(0.009)
labour_ch -0.16(0.151)
min_wage -0.00(0.003)
ir 0.00(0.019)
m2 -0.00(0.003)
corrupt 0.07(0.068)
corrupt_ch -0.74(0.475)
left -0.15(0.113)
prod_ch -0.01(0.013)
open 0.00**(0.001)
Constant 0.21 0.04 0.63** 0.22 0.82 0.19 0.14 -0.03(0.10) (0.170) (0.278) (0.507) (0.583) (0.115) (0.150) (0.150)
Observ. 42 39 41 39 42 41 42 41R2 0.32 0.345 0.379 0.273 0.472 0.326 0.388 0.366
Dependent variable in all regressions is change in top 10 share. *** p<0.01, ** p<0.05, * p<0.1Robust standard errors in parentheses
10
To further assess the arguments for omitted variables, we conduct a Bayesian model av-
eraging exercise (BMA), including all these variables as potential explanatory variables for
the change in the top 10 share after the crisis. In this way, we basically investigate which of
the variables are most likely to be signi�cant determinants of the post-crisis change in the
top 10 share. BMA has gained prominence in recent years in applied research when there
is uncertainty regarding the appropriate theoretical model. BMA addresses the problem of
uncertainty by considering information from all available models (i.e. combinations of the
variables), instead of selecting only one model. The models are estimated using Bayesian
techniques, and then weighted by their goodness of �t, to obtain the desired statistics. Infer-
ences are usually based on the posterior inclusion probability (PIP), which acts as a measure
of the signi�cance of the variable, with PIP above 0.5 implying signi�cance, and on the
weighted averages of the posterior means and standard errors of the candidate variables.
It should be noted, though, that these are not directly comparable to the OLS coe¢ cients
and standard errors, because they include also models in which some of the variables are
excluded. A comprehensive explanation of BMA can be found in Hoeting et al. (1999). For
a brief introduction, see Jovanovic (2014).
The application of BMA usually requires the setting of priors for the model parameters,
the setting of priors for the models and the determination of how to choose from all the
available models. In our case, since the number of explanatory variables is rather low, 21,
and the number of potential models is therefore only 221 = 2097 152, instead of choosing only
a subset of models by Markov Chain Monte Carlo methods, we will estimate all the potential
models. Regarding the model prior, we will use two priors - the uniform prior, which assumes
that all the models have an equal prior probability of being correct, and the ��xed�prior of
Ley and Steel (2009), which sets the expected model size equal to 10 variables. Regarding
the prior for the model parameters, we will use the "hyper g" prior of Liang et al. (2008) and
the empirical Bayes (EBL) prior of Hansen and Yu (2001). We will report the PIP and the
posterior means for the coe¢ cients from the 100 best models. The results are presented in
Table 3, with each column heading denoting which combination the results refer to.
As can be seen, in�ation is always signi�cant, alongside the labour force participation
rate, the pre-crisis trend in the top 10 share, the change in corruption and the recovery of
11
the GDP and the stock exchange. We interpret these results as an additional support to the
claim that the identi�ed negative association between the change in the top 10 share after
the crisis and the average in�ation during the crisis is not likely to be driven by omitted
variables.
Table 3: BMA analysis of determinants of the change in
the top 10 percent share
Variables Coef. prior: hyper g Coef. prior: EBL Coef. prior: hyper g Coef. prior: EBL
Model prior: uniform Model prior: uniform Model prior: �xed Model prior: �xed
Post. Mean PIP Post. Mean PIP Post. Mean PIP Post. Mean PIP
top10_trend 0.67 0.99 0.69 1.00 0.67 0.99 0.69 1.00
participation -0.08 0.99 -0.09 1.00 -0.08 0.99 -0.09 1.00
corrupt_ch -3.91 0.95 -4.29 0.98 -3.28 0.84 -3.90 0.94
recov_se 0.02 0.88 0.02 0.89 0.02 0.89 0.02 0.88
recov_gdp 0.08 0.76 0.10 0.88 0.06 0.59 0.08 0.72
in�ation -0.12 0.70 -0.14 0.83 -0.09 0.53 -0.11 0.64
left -0.11 0.27 -0.12 0.30 -0.08 0.20 -0.10 0.25
tax_top_ch 0.01 0.16 0.01 0.22 0.00 0.09 0.00 0.12
m2 0.00 0.10 0.00 0.15 0.00 0.07 0.00 0.10
fall_se 0.00 0.10 0.00 0.13 0.00 0.08 0.00 0.08
fall_gdp 0.00 0.09 0.00 0.08 0.00 0.05 0.00 0.05
top10_init 0.00 0.09 0.00 0.08 0.00 0.05 0.00 0.07
labour 0.01 0.07 0.01 0.06 0.01 0.06 0.01 0.08
gov_size 0.00 0.06 0.00 0.07 0.00 0.04 0.00 0.06
bene�ts 0.00 0.06 0.00 0.07 0.00 0.06 0.00 0.06
tax_top 0.00 0.06 0.00 0.07 0.00 0.06 0.00 0.06
IR 0.00 0.05 0.00 0.07 0.00 0.06 0.00 0.05
labour_ch 0.01 0.05 0.01 0.06 0.02 0.05 0.01 0.05
corrupt 0.00 0.05 0.00 0.04 0.00 0.04 0.00 0.04
open 0.00 0.05 0.00 0.05 0.00 0.06 0.00 0.06
prod_ch 0.00 0.04 0.00 0.04 0.00 0.04 0.00 0.05
12
We next explore the arguments for reverse causality, i.e. we assess whether the observed
correlation between in�ation and the change in the top 10 share may be due to the latter
causing the former. The arguments for inequality causing in�ation are related to political
factors. As Beetsma and van der Ploeg (1996) argue, in democratic societies with high
inequality, the median voter would prefer pro-poor policies, i.e. policies that redistribute
income from the rich to the poor. Unanticipated in�ation may be one of those policies, which
would imply that higher inequality may result in higher in�ation. A similar mechanism is
proposed by Dolmas, Hu¤man and Wynne (2000), who argue that with higher inequality, the
median voter prefers higher in�ation as a way of �nancing higher government expenditures.
Crowe (2006) and Albanesi (2007) also provide models in which higher inequality leads to
higher in�ation, although for di¤erent reasons. In their models, greater income inequality
leads to greater inequality in political in�uence, with the rich having greater political power.
If the rich perceive that in�ation will favour them, because they may be insulated from the
in�ation tax, di¤erently from the poor, they may push for in�ationary policies.
One corollary of the explanation that the causality goes from inequality to in�ation is
that one would expect to see that the association between in�ation and the change in the
top 10 share is signi�cant only in countries with high inequality, or in countries with certain
political parties in power. Hence, if the relationship is present irrespectively of the level of
inequality or the political party in power, that could be treated as an argument against the
hypothesis that the causality goes from inequality to in�ation. We next test if there are
structural breaks related to the initial inequality and the political party in power.
Table 4 presents these results. In the �rst column, we allow for a structural break for
countries with high initial top 10 share, setting the median of the variable as a threshold
for high initial top 10 share. The coe¢ cient for the countries with high initial top 10 share
is virtually identical to the one for countries with low initial top 10 share. In the second
column, we use an alternative measure for the initial inequality - the Gini coe¢ cient in 2005-
2007 (data on Gini are from the Standardized World Income Inequality Database of Solt
(2009)). Results are very similar to the previous ones. Finally, we check whether in�ation
has reduced the top 10 share only in countries with leftist governments during the Great
Recession, which are considered to be more inclined to increases in public spending (column
13
3). It turns out that there are no di¤erences, again.
We interpret these results as arguments against the hypothesis that the causality in the
observed relationship between in�ation and the change in the top 10 share goes from inequal-
ity to in�ation.
Table 4: Reverse causality-1 -2 -3
High initial High initial Leftisttop 10 share Gini governments
in�ation -0.11*** -0.09** -0.11***(0.032) (0.037) (0.028)
hi_top10_init -0.19(0.197)
hi_top10_init*in�ation 0.00(0.043)
hi_gini_init -0.12(0.214)
hi_gini_init*in�ation -0.02(0.049)
left -0.12(0.199)
left*in�ation -0.01(0.051)
Constant 0.27** 0.22 0.25*(0.133) (0.135) (0.133)
in�ation+hi_top10_init*in�ation -0.10***(p value) 0.000in�ation+hi_gini_init*in�ation -0.11***(p value) 0.001in�ation+left_in�ation -0.12***(p value) 0.006Observations 42 42 42R-squared 0.354 0.360 0.352
Dependent variable in all regressions is change in top 10 share.*** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses
IV. Channels through which inflation has reduced
inequality
The literature identi�es several potential channels through which in�ation can reduce in-
equality (see Kane and Morisett (1993), Crowe (2006), Doepke and Schneider (2006), Coibion
et al. (2012)). The �rst one is through the di¤erent sensitivity to in�ation of the main source
of income of low-income and high-income individuals. Namely, the main source of income of
14
low income individuals is wages, which are sensitive to in�ation (unless they are indexed),
while high-income individuals realize big part of their income from capital, which is less sen-
sitive to in�ation. The second channel is through the asset holding. Low-income individuals
usually hold their assets in cash, whose real value is eroded by in�ation, while high-income
individuals hold most of their savings in in�ation-proof assets, such as equity, real assets,
commodities, or certain �nancial products. Another channel is through the redistribution
from lenders (i.e. the rich) to borrowers (i.e. the poor), when the in�ation in unanticipated,
which would make the poor bene�t from higher in�ation. The �nal channel is through the
government expenditure. Usually, in�ation serves as a windfall for the government, due to the
higher revenues and the erosion of the public debt, which may lead to higher social transfers,
bene�ting the poor.3
Thus, the �nding that in�ation has reduced the top 10 share after the global �nancial
crisis can be due to: 1) lower sensitivity of the lower-income individuals to in�ation, due to
increases in their wages; 2) low-income individuals holding more of their savings in in�ation-
proof assets, due to �nancial development; 3) redistribution from net-lenders to net-borrowers,
due to the low interest rates during the Great Recession; 4) higher government support to
the low-income individuals, due to the windfall from the higher in�ation. We next evaluate
these potential explanations, �rst, using macroeconomic, cross-country data, and then using
microeconomic data from household surveys.
IV.A. Cross-country data
If low-income individuals have been protected from in�ation, through wage-indexation or
wage increases due to the in�ation, one would expect to observe that the negative relationship
between in�ation and the top 10 share would be present only in countries which are more
likely to experience minimum wage increases, such as countries with strong trade unions, or
countries that have experienced higher increases in the minimum wages. Therefore, we next
estimate the in�ation-inequality regression, allowing for a di¤erential impact in countries with
strong trade unions, and in countries with increases in the minimum wages. We measure the
3. It should be noted that this relationship may turn around at very high levels of in�ation. If in�ation isvery high, government revenues may fall, due to time lags in collection, reduce real government spending (theOlivera-Tanzi e¤ect). This e¤ect is likely to bene�t the net tax payers (i.e. the rich). However, this e¤ect isunlikely to appear in our sample of countries, where the highest rate of in�ation is 10 percent.
15
strength of the trade unions in two ways. First, we use data on the trade union membership,
from the New Unionism Network. Second, we use data on the share of employees covered by
collective bargaining, from the International Labour Organization (ILO). Data on minimum
wages are obtained from the Doing Business Indicators of the World Bank. In each case,
we set the median of the variables as a cut-o¤ point between strong and weak characteristic.
The estimations are presented in Table 5, columns 1, 2 and 3..
It can be seen that the negative relationship between in�ation and the top 10 share is
present only in countries with strong trade unions and high minimum wage increases - the
coe¢ cient on in�ation is insigni�cant, but its sum with the coe¢ cient of its cross-product is
signi�cant at 1%. This supports the �rst hypothesis, that the observed relationship between
in�ation and the top 10 share is due to stronger trade unions protecting the low-income
individuals from the adverse e¤ects of in�ation through minimum wage increases.
The second channel, through asset holding, is more di¢ cult to assess, because there are
no data on investment in in�ation-proof assets. One proxy variable can be the degree of
inclusion of the �nancial system - in countries with more inclusive �nancial systems, low-
income individuals will be also able to hedge from in�ation. One often-used measure of the
inclusion of the �nancial system is the share of bank deposits to the GDP. The problem with
this variable is that it is determined by many variables, such as the level of development, or
the nature of the �nancial system (banking vs. stock-exchange). A better measure would,
therefore, be the rate of growth of the deposits - countries with higher deposit growth are
experiencing a wave of �nancial development, i.e. �nancial inclusion. This is the variable
that we will use, i.e. we will use the growth of the bank deposits before the crisis (2004-2007),
from the Financial Access Survey of the International Monetary Fund. As previously, we will
divide the countries in two groups - with high and low inclusiveness, using the median of the
variable as a cut-o¤ point, and will estimate the regression for the two groups of countries.
The results, shown in column 4 of Table 5, point out that there are no di¤erences in the
in�ation-change in inequality relationship for the two groups of countries, arguing against
the second channel.
To investigate the third hypothesis, that the relationship is due to redistribution from
net-lenders (i.e. the rich) to net-borrowers (i.e. the poor), we estimate the basic regression
16
for countries with low and high interest rates during the crisis. With low nominal interest
rates, higher in�ation results in low (sometimes even negative) real interest rates. If the
relationship is found only in countries with low nominal interest rates during the crisis, then
this would support the hypothesis that the negative relationship is due to redistribution.
We classify countries into �low-interest rates�and �high-interest rates�on the grounds of the
average money-market interest rate during 2008-2010, using the median as a cut-o¤ point.
These results are presented in Table 5, column 6. Somewhat surprisingly, the relationship is
signi�cant only for the countries with high-interest rates. The explanation for this may be
that countries with low interest rates actually had very low levels of in�ation. In any case,
the results do not support this hypothesis.
Finally, to assess whether the identi�ed relationship between in�ation and the change in
the top 10 share is due to the stronger government support to the low-income individuals,
due to the higher revenues from the in�ation, we estimate the basic equation separately for
countries with high and low government support through the bene�ts during the crisis. The
results are presented in Table 5, column 5. The relationship holds for the two groups of
countries, which implies that the underlying channel is the government support.
17
Table 5: Channels-1 -2 -3 -4 -5 -6
Union Collective Minimum Financial Government Low interestdensity bargaining wage inclusion bene�ts rates
in�ation -0.08 -0.03 -0.10 -0.10** -0.10*** -0.03(0.054) (0.040) (0.061) (0.040) (0.029) (0.057)
hi_union 0.25(0.228)
hi_union*in�ation -0.06(0.056)
hi_col_barg 0.23(0.225)
hi_col_barg*in�ation -0.11**(0.048)
hi_min_wage -0.16(0.192)
hi_min_wage*in�ation 0.00(0.065)
hi_inclusion 0.19(0.199)
hi_inclusion*in�ation -0.02(0.046)
hi_bene�ts 0.05(0.214)
hi_bene�ts*in�ation -0.03(0.050)
hi_ir_level 0.13(0.212)
hi_ir_level*in�ation -0.09(0.061)
Constant 0.06 0.00 0.23 0.11 0.19 0.07(0.217) (0.191) (0.161) (0.177) (0.119) (0.171)
in�ation+hi_union*in�ation -0.14***(p value) 0.000in�ation+hi_col_barg*in�ation -0.14***(p value) 0.000in�ation+hi_min_wage_in�ation -0.10***(p value) 0.000in�ation+hi_inclusion_in�ation -0.13***(p value) 0.000in�ation+hi_bene�ts*in�ation -0.13***(p value) 0.002in�ation+hi_ir_level*in�ation -0.12***(p value) 0.000Observations 42 42 42 42 42 42R-squared 0.342 0.393 0.340 0.337 0.332 0.366
Dependent variable in all regressions is change in top 10 share. *** p<0.01, ** p<0.05, * p<0.1Robust standard errors in parentheses
18
IV.B. Household-level data
The household-level analysis will be done using data from the Luxembourg Income Study.
Of the 42 analyzed countries, 15 are present in the LIS dataset with data both before and
after the crisis (Colombia, Estonia, Finland, Germany, Greece, Ireland, Italy, Luxembourg,
Netherlands, Poland, Slovakia, Slovenia, Spain, United Kingdom, United States). Ten of
them experienced a decline in the top 10 income share after the crisis - Colombia, Estonia,
Finland, Germany, Greece, Italy, Luxembourg, Netherlands, Poland and Spain. Only four of
these had at the same time above-average in�ation (i.e. average annual in�ation during 2008-
2010 above than 2.5 percent): Colombia, Estonia, Greece, Poland. These are the countries
on which we will focus. We will try to assess to what extent the decline in the top 10 share
that these countries experienced can be explained by the in�ation, through the alternative
channels that were explained above.
The pre-crisis data refer to 2007, while the post-crisis data to 2010. The data are from
the household-level surveys, not individual-, due to the better coverage. They are expressed
in per-capita terms and are bottom-coded (i.e. the negative values are replaced with zeros).
They are weighted by the provided weights, so the statistics can be interpreted as referring
to the total population. The values for 2010 are de�ated by the cumulative in�ation in
2008-2010.
The assessment of the relative merit of the alternative channels using micro data will
be done by comparing the income the low- and high-income households have realized from
di¤erent sources before and after the crisis. Households will be grouped in ten groups (deciles),
depending on the per-capita disposable household income (i.e. gross household income, minus
taxes and contributions). Three main sources of income will be analyzed: labour income,
capital income and income from social security transfers. The data on the di¤erent sources of
income are in gross terms, not net. Because of this, we will not try to assess the contributions
of the changes in the alternative sources of income to the change in total disposable income,
because the taxes for each category of income are not known, but will just compare the
changes in the di¤erent incomes before and after the crisis.
We begin with Colombia. Colombia had a decline in the top 10 share of income in 2010,
with respect to 2005 (no data for 2006 and 2007), of 0.54 percentage points. Average annual
19
in�ation in 2008-2010 was 4.5 percent. Table 6 presents the mean per capita income from the
three sources for households belonging to di¤erent income deciles before and after the crisis.
Table 6: Mean per capita household income
for different households, Colombia
Income decile Labour income Capital income Social transfers
2007
1 172,240 12,267 1,207
2 666,806 18,942 2,839
3 1,149,795 26,879 75,252
4 1,422,576 21,358 107,363
5 1,725,213 44,865 121,271
6 2,290,361 88,556 191,905
7 2,609,996 91,548 343,572
8 3,366,113 143,173 430,565
9 4,903,635 184,249 790,569
10 12,600,000 1,419,970 2,111,646
2010
1 257,199 17,609 553
2 841,822 35,816 21,701
3 1,280,106 34,737 111,161
4 1,567,708 34,328 121,756
5 1,877,156 70,596 148,222
6 2,363,258 77,773 187,907
7 2,704,567 123,265 347,103
8 3,573,100 139,300 505,333
9 4,964,984 234,313 761,473
10 11,455,950 1,335,398 2,503,636
In national currency. 2010 �gures are in 2007 prices (i.e. de�ated).
Bolded �gures point main changes.
Several things should be noted. First, poorer households (1st and 2nd decile) experi-
enced sizeable increase in their labour income between 2010 and 2007. Second, the top decile
households experienced a decline in both their labour and capital income. Hence, the pre-
vious explanation - that the poor households have been insulated from the adverse e¤ects
20
of in�ation, through minimum wage increases, di¤erently from the rich - seems to hold for
Colombia. Although trade union coverage is actually very low in Colombia, with only 2
percent of the workers being members, minimum wage has been increased between 2008 and
2010 by 18.7 percent in total. Regarding the decline in the capital income, it has been mostly
due to the decline in the rental income, not income from interest. It remains unclear why
the rental income declined.
Overall, it can be said that the micro data for Colombia support the previous explanation,
at least to some extent - in�ation has redistributed income from rich to the poor, mainly
through the labour income, as the former have been protected by the minimum wage increases,
di¤erently from the latter.
We next turn to Estonia. Estonia has seen a decline in its top 10 share of income of 2
percentage points in 2011, compared to 2007. Average in�ation in 2008-2010 has been 4.4
percent per annum. Table 7 presents its micro data developments.
21
Table 7: Mean per capita household income
for different households, Estonia
Income decile Labour income Capital income Social transfers
2007
1 4,351 111 26,130
2 15,744 36 26,723
3 26,542 171 23,066
4 34,858 141 20,646
5 48,848 313 13,984
6 62,951 320 10,448
7 68,523 303 9,642
8 75,140 188 8,266
9 98,970 538 8,505
10 148,012 3,563 7,855
2010
1 6,705 86 18,201
2 8,415 145 34,711
3 26,372 155 18,414
4 25,552 475 24,065
5 31,398 430 23,471
6 48,073 489 14,696
7 58,352 185 13,847
8 68,859 568 11,058
9 78,290 1,065 10,814
10 132,487 1,603 12,671
In national currency. 2010 �gures are in 2007 prices (i.e. de�ated).
Bolded �gures point main changes.
The situation in Estonia seems very similar to that in Colombia - the labour income of
the poorest (1st and 2nd decile) increased after the crisis, while labour income of the top
decile households declined. The top 10 percent also su¤ered a decline in their capital income,
but the magnitude of this decline is smaller. Union membership in Estonia is also rather low
(11 percent), but this did not preclude rise in the minimum wage, which was increased by 21
percent in 2008. At the same time, the labour income of the top 10 percent was not covered
22
by this, and has declined after the crisis. Regarding the capital income of the top decile, it is
hard to assess the reasons for its decline, because there are no data on income from interests
and dividends.
Table 8: Mean per capita household income
for different households, Greece
Income decile Labour income Capital income Social transfers
2007
1 1,649 65 1,785
2 1,645 122 3,052
3 2,789 144 2,580
4 3,695 123 2,684
5 4,735 237 2,675
6 6,061 333 2,293
7 7,345 309 2,295
8 8,855 515 2,371
9 12,194 530 2,176
10 20,449 1,790 2,939
2010
1 601 60 1,350
2 1,334 108 2,663
3 1,928 43 3,300
4 2,843 146 2,603
5 3,729 110 2,973
6 4,268 254 2,874
7 5,139 255 3,155
8 7,280 278 2,496
9 9,026 478 2,791
10 16,547 1,234 3,638
In national currency. 2010 �gures are in 2007 prices (i.e. de�ated).
Bolded �gures point main changes.
Next, we turn to Greece. In Greece, the top 10 share of income in 2011 declined by
1 percentage points vis-a-vis 2007, while average annual in�ation has been 3.4 percent in
23
2008-2010. The decline in the top 10 share, as can be seen from Table 8, was mostly due to
the decline in the labour income, which has been probably due to the crisis. Di¤erently from
Colombia and Estonia, the labour income in Greece declined for households from all deciles,
likely due to the severity of crisis. However, the decline for the lower-income households has
been milder (in absolute terms), possibly due to the minimum wage increase, which has been
increased by 13 percent in 2008-2010. It can also be noted that the capital income has also
declined for the richest households, but, again, it is not possible to assess if this is due to
lower interest income, due to unavailability of disaggregated data.
24
Table 9: Mean per capita household income
for different households, Poland
Income decile Labour income Capital income Social transfers
2007
1 867 8 4,313
2 2,060 13 5,464
3 3,093 18 4,600
4 3,877 8 4,108
5 4,738 15 3,761
6 5,745 21 3,545
7 6,783 8 3,194
8 8,273 26 3,164
9 10,524 40 2,838
10 19,856 259 2,325
2010
1 836 13 5,108
2 2,269 13 6,358
3 3,632 18 4,785
4 4,420 18 4,512
5 5,573 20 4,207
6 6,889 24 3,739
7 8,210 26 3,451
8 10,021 31 3,067
9 12,781 59 2,756
10 22,947 192 2,274
In national currency. 2010 �gures are in 2007 prices (i.e. de�ated).
Bolded �gures point main changes.
Poland, the last country we analyze, is somewhat di¤erent. In Poland, the top 10 share
of income declined in 2011, relative to 2007, by 0.7 percentage points, with average in�ation
in 2008-2010 of 3.6 percent per annum. The main reason behind the decline in the top 10
share in Poland seems to be the higher labour income for the other deciles, as well as the
higher social transfers for the households from the �rst two deciles of income. Regarding the
latter, it is possible that the increase in the social transfers is due to the higher government
25
revenues, due to the higher in�ation - both in�ation and government expenditure on social
bene�ts have been higher in 2008-2010 than in the previous four years, the former by 1.3
percentage points, the latter by 2.7 percentage points.
To summarize, the micro analysis supports, to some extent, the �ndings from the cross-
country analysis, that higher in�ation has some role to play in the decline in the top 10
share of income after the crisis. While the case for redistribution from lenders to borrowers
seems weak for the analyzed countries, there seems to have been some redistribution through
the labour income, since higher income earners have not been protected from the higher
in�ation, while lower income earners have been, through higher minimum wages. There is
also some evidence in favour of the hypothesis for higher government support through the
social transfers.
V. Conclusions
We have investigated the role of in�ation for the decline in the income of the rich (the top
10 percent) after the recent global �nancial crisis. Using both cross-country and household-
survey data, we �nd that higher in�ation has reduced the share of income going to the top
10 percent through several channels. First, in�ation has eroded the labour income of the
rich, but not of the poor, which have been protected by the trade unions, through minimum
wage increases. Second, higher in�ation has increased government revenues, and this has
been canalized towards higher social transfers.
The size of the e¤ect may seem small at �rst - increase in annual in�ation of one percentage
point decreases the share of income going to the top 10 percent by only 0.1 percentage points
per year. Still, the e¤ect can be sizeable if observed through a longer period of time - if
in�ation is increased from 2 to 5 percent, for a period of 15 years, that would reduce the top
10 share of income by 5 percentage points.
Given that the main channel through which in�ation has a¤ected inequality is the di¤erent
response of low- and high-income individuals�labour income to in�ation, the reduction in the
above scenario is likely to be smaller, because it seems unlikely that the labour income of the
rich would remain unchanged if in�ation averages 5 pecent for 15 years. Since the e¤ect of
in�ation on inequality reduction is modest, and since there are many uncertainties involved,
26
raising in�ation may not be a viable measure for decreasing income inequality. Perhaps a
better reading of the results is through the measures that should be taken during periods of
mild in�ation, in order to bene�t from it. In that sense, our results suggest that measures
aimed at protecting the most vulnerable individuals, through minimum wage increases, seem
to be the most e¤ective way to reduce inequality during in�ationary periods.
More research is needed in order to derive sounder conclusions. The �rst avenue for future
extension is to try to �nd household level data for the countries which had higher rates of
in�ation and higher decreases in the top 10 share during the Great Recession, like the Latin
American countries. This would allow us to gauge the underlying mechanisms more reliably.
Second area for improvement is to see if the results extend to other periods as well, or hold
only for the recent post-crisis period.
References
Afonso, António, Ludger Schuknecht, and Vito Tanzi. 2010. �Income distributiondeterminants and public spending e¢ ciency.�Journal of Economic Inequality, 8(3): 367�389.
Aizenman, Joshua, and Nancy Marion. 2011. �Using in�ation to erode the US publicdebt.�Journal of Macroeconomics, 33(4): 524�541.
Albanesi, Stefania. 2007. �In�ation and inequality.� Journal of Monetary Economics,54(4): 1088�1114.
Alvaredo, Facundo, Anthony B. Atkinson, Thomas Piketty, and EmmanuelSaez. 2013. �The World Top Incomes Database.� http: // topincomes. g-mond.parisschoolofeconomics. eu/ .
Bach, G. L., and Albert Ando. 1957. �The Redistributional E¤ects of In�ation.�TheReview of Economics and Statistics, 39(1): 1�13.
Bach, G. L., and James B. Stephenson. 1974. �In�ation and the Redistribution ofWealth.�The Review of Economics and Statistics, 56(1): 1�13.
Beck, Thorsten, George Clarke, Alberto Gro¤, Philip Keefer, and PatrickWalsh. 2001. �New tools in comparative political economy: The Database of Politi-cal Institutions.�World Bank Economic Review, 15(1): 165�176.
Beetsma, Roel M W J, and Frederick van der Ploeg. 1996. �Does Inequality CauseIn�ation?: The Political Economy of In�ation, Taxation and Government Debt.�PublicChoice, 87(1-2): 143�62.
Blanchard, Olivier, Giovanni Dell�Ariccia, and Paolo Mauro. 2010. �RethinkingMacroeconomic Policy.�Journal of Money, Credit and Banking, 42: 199�215.
Bulir, Ales. 2001. �Income Inequality: Does In�ation Matter?� IMF Sta¤ Papers, 48(1): 5.
27
Coibion, Olivier, Mary Yuriy Gorodnichenko, and Johannes Wieland. 2012. �TheOptimal In�ation Rate in New Keynesian Models: Should Central Banks Raise TheirIn�ation Targets in Light of the Zero Lower Bound?� The Review of Economic Studies.
Coibion, Olivier, Yuriy Gorodnichenko, Lorenz Kueng, and John Silvia. 2012.�Innocent Bystanders? Monetary Policy and Inequality in the U.S.�National Bureau ofEconomic Research, Inc NBER Working Papers 18170.
Crowe, Christopher W. 2006. �In�ation, Inequality, and Social Con�ict.�InternationalMonetary Fund IMF Working Papers 06/158.
Di Bartolomeo, Giovanni, Nicola Acocella, and Patrizio Tirelli. 2013. �The come-back of in�ation as an optimal public �nance tool.� Department of Communication,University of Teramo wp.comunite 0100.
Doepke, Matthias, and Martin Schneider. 2006. �In�ation and the Redistribution ofNominal Wealth.�Journal of Political Economy, 114(6): 1069�1097.
Dolmas, Jim, Gregory W. Hu¤man, and Mark A. Wynne. 2000. �Inequality, in�a-tion, and central bank independence.�Canadian Journal of Economics, 33(1): 271�287.
Fitoussi, Jean-Paul, and Francesco Saraceno. 2010. �Inequality and MacroeconomicPerformance.� Observatoire Francais des Conjonctures Economiques (OFCE) Docu-ments de Travail de l�OFCE 2010-13, Paris.
Goodfriend, Marvin. 2007. �How the World Achieved Consensus on Monetary Policy.�Journal of Economic Perspectives, 21(4): 47�68.
Gwartney, James, Robert Lawson, and Joshua Hall. 2013. �2013 Economic Free-dom Dataset.�Fraser Institute Economic Freedom of the World: 2013 Annual Report.available at: http://www.freetheworld.com/datasets_efw.html.
Hansen, M., and B. Yu. 2001. �Model selection and the principle of minimum descriptionlength.�Journal of the American Statistical Association, 96(454): 746�774.
Hoeting, J. A., D. Madigan, A. E. Raftery, and C. T. Volinsky. 1999. �BayesianModel Averaging: A Tutorial.�Statistical Science, 14(4): 382�417.
Jovanovic, Branimir. 2014. �The Top 10 Percent After the Global Financial Crisis.�available at: http://branimir.ws/wp-content/uploads/2013/07/paper_ineq3.pdf.
Kane, Cheikh, and Jacques Morisett. 1993. �Who would vote for in�ation in Brazil?:an integrated framework approach to in�ation and income distribution.� The WorldBank Policy Research Working Paper Series 1183.
Ley, E., and M. F. Steel. 2009. �On the E¤ect of Prior Assumptions in Bayesian ModelAveraging with Applications to Growth Regressions.�Journal of Applied Econometrics,24: 651�674.
Liang, F., R. Paulo, G. Molina, M. A. Clyde, and J. O. Berger. 2008. �Mixtures of gPriors for Bayesian Variable Selection.�Journal of the American Statistical Association,103: 410�423.
Milanovic, Branko. 2009. �Two Views on the Cause of the Global Crisis - Part I.�Yale Global Online, 4 May 2009. available at: http://yaleglobal.yale.edu/content/two-views-global-crisis.
Nordhaus, William D. 1973. �The E¤ects of In�ation on the Distribution of EconomicWelfare.�Journal of Money, Credit and Banking, 5(1): 465�504.
Rajan, Raghuram. 2010. Fault Lines: How Hidden Fractures Still Threaten the WorldEconomy. Princeton:Princeton University Press.
28
Ranciere, Romain, and Michael Kumhof. 2010. �Inequality, Leverage and Crises.�International Monetary Fund IMF Working Paper 10/268, Washington DC.
Rogo¤, K. S. 2008. �In?ation is now the lesser evil.� Project Syndicate,December 2008. available at: http://www.project-syndicate.org/commentary/inflation-is-now-the-lesser-evil.
Roine, Jesper, Jonas Vlachos, and Daniel Waldenström. 2009. �The long-run de-terminants of inequality: What can we learn from top income data?� Journal of PublicEconomics, 93(7-8): 974�988.
Scheve, Kenneth F., and David Stasavage. 2009. �Institutions, Partisanship, andInequality in the Long Run.�World Politics, 61(2): 215½U253.
Schmitt-Grohé, Stephanie, and Martín Uribe. 2010. �The Optimal Rate of In�ation.�In Handbook of Monetary Economics. Vol. 3 of Handbook of Monetary Economics, , ed.Benjamin M. Friedman and Michael Woodford, Chapter 13, 653�722. Elsevier.
Solt, Frederick. 2009. �Standardizing the World Income Inequality Database.�Social Sci-ence Quarterly, 90(2): 231�242. SWIID Version 4.0, September 2013.
Stiglitz, Joseph. 2009. �The Rocky Road to Recovery.� Project Syndicate, 9January, 2009. available at: http://www.project-syndicate.org/commentary/the-rocky-road-to-recovery.
Wade, Robert. 2009. �The global slump. Deeper causes and harder lessons.�Challenge,52(5): 5�24.
Woodford, Michael. 2009. �Convergence in Macroeconomics: Elements of the New Syn-thesis.�American Economic Journal: Macroeconomics, 1(1): 267�79.
29
Table A1: VariablesVariable name Variable de�nition Sourcetop10_ch Change in the top 10 share in 2011 (or 2010, analogously to the
last available data point) and the average for 2005-2007.WTID,Eurostat,WDI.
in�ation CPI in�ation, average for 2008-2010. WDIfall_gdp Annual change in real GDP per capita in 2009. WDIrecov_gdp Annual change in real GDP per capita in 2010, or average for
2010-2011, analogously to the last data point.WDI
fall_se Annual change in the market capitalization of listed companies(as a % of GDP) in 2008. The 2008 was selected instead of2009 because the �nancial shock happened in 2009.
WDI
recov_se Change in the market capitalization of listed companies, as a% of GDP, in 2010 (or 2011) with respect to 2009.
WDI
top10_init Initial top 10 share, i.e. top 10 share in 2005-2007. WTID,Eurostat,WDI.
top10_trend Pre-crisis trend in the top 10 share, i.e. average change in theperiod 2003-2007.
WTID,Eurostat,WDI.
gov_size General government consumption in 2010 (or 2010-2011), as apercent of GDP.
WDI
tax_top Top marginal tax rate in 2010 (or 2010-2011). EFWtax_top_ch Change in the top marginal tax rate in 2010 (or 2010-2011),
with respect to 2009.EFW
bene�ts Real change in subsidies and other government transfers in 2010(or 2010-2011), with respect to 2009.
WDI
labour Labor market regulation index in 2010 or 2010-2010. Higherindex=less regulated.
EFW
participation Labor participation rate, total (% of total population ages15+), 2010 or 2010-11
WDI
labour_ch Change in labor market regulation index in 2010 or 2010-2011,with respect to 2009. Higher value = made less regulated.
EFW
min_wage Minimum wage, change in 2010 with respect to 2007. DBIir Money market rate, 2010 (or 2010-2011) minus 2009. In real
terms. For the euro area countries, it�s the euro area data.IFS
m2 M2, % of GDP, 2010 (or 2011) vs. 2009. WDIcorrupt Control of corruption index, in 2010 (or 2010-2011). Higher
value = higher control.WGI
corrupt_ch Change in the control of corruption index, in 2010 (or 2010-2011) with respect to 2009. Higher value = improvement inthe control.
WGI
left Dummy if the country had a leftist government in 2010 or 2011 DPIprod_ch Di¤erence between productivity growth in 2000�s and 1990�s ILOopen Exports of goods and services, as a percent of GDP, 2010 or
2010-2011.WDI
WTID stands for the World Top Income Database of Alvaredo et al. (2013).WDI stands for the World Development Indicators of the World Bank.WGI stands for the Worldwide Governance Indicators of the World Bank.EFW stands for the Economic Freedom of the World dataset of Gwartney, Lawson and Hall (2013).DPI stands for the Database on Political institutions of Beck et al. (2001).IFS stands for the International Financial Statistics of the IMF.DBI stands for the Doing Business Indicators of the World Bank.30