## Guillermo Cruces, Gary S. Fields, David Jaume, and Mariana Viollaz

Print publication date: 2017

Print ISBN-13: 9780198801085

Published to Oxford Scholarship Online: June 2017

DOI: 10.1093/oso/9780198801085.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in OSO for personal use (for details see www.oxfordscholarship.com/page/privacy-policy).date: 18 January 2019

# Within-Country Analysis of the Growth–Employment–Poverty Nexus

Chapter:
(p.104) 5 Within-Country Analysis of the Growth–Employment–Poverty Nexus
Source:
Growth, Employment, and Poverty in Latin America
Publisher:
Oxford University Press
DOI:10.1093/oso/9780198801085.003.0005

# Abstract and Keywords

This chapter analyses the within-country growth–employment–poverty nexus. First, it calculates labour market indicators’ elasticities with respect to gross domestic product per capita growth. It finds that in the Latin America region and in most countries, labour market indicators improved with percentage increases in GDP per capita. Second, it estimates poverty elasticities with respect to employment and earnings indicators and finds that in the region and in most of the countries, poverty measures were related in the welfare-improving direction with percentage changes in some employment and earnings indicators. Finally, it analyses the patterns of earnings changes across deciles of the earnings distributions in each country and finds that 70 per cent of the country-decile cells exhibited positive earnings changes.

# 5.1 Assessing the Response of Labour Market Indicators to Growth using Growth Elasticities

During the 2000s, there was a clear correlation over time between poverty and GDP per capita, labour earnings, and unemployment in the Latin American region: in general, poverty fell when GDP per capita increased, labour earnings increased, and unemployment decreased. This is clearly apparent in Figure 3.1 in Chapter 3, which shows the evolution of the unweighted averages for sixteen Latin American countries of the sixteen labour market indicators and GDP per capita over the period 2000–12. Average GDP per capita in the region was stagnant from 2000 to 2003, but then increased every year afterwards except for the 2008 international crisis. Mean labour earnings among the employed decreased from 2000 to 2003 but then increased every year after that, even during the international crisis, ending about 10 per cent higher in 2012 than in 2000. Unemployment increased from 2000 to 2002 and then fell every year afterwards except for an increase during the international crisis. Other employment indicators follow a similar improving pattern over the period. The 4 dollars-a-day poverty rate at first increased from 40.4 per cent in 2000 to 43.0 per cent in 2002, but then poverty declined in every year, reaching 25.4 per cent in 2012. Notably, the average poverty rate among Latin American countries did not increase during the international crisis of 2008, which is consistent with our previous finding of an increase in poverty in only five out of the sixteen countries during the crisis, while poverty fell during the crisis in eight countries (Table 3.3 in Chapter 3).

(p.105) In this section, we analyse in more detail the nexus between growth, on the one hand, and labour market indicators—employment and earnings indicators and poverty and inequality indicators—on the other. Our analysis is based on the estimation of labour market indicators’ elasticities with respect to GDP per capita growth for each country in our sample and for the Latin American region as a whole.

We use year-by-year data for each country in contrast to Chapters 2–4 where we used the annualized changes between the initial and the final years for each country. This procedure means moving from using sixteen data points (one for each country) to 169 data points (an average of eleven per country) when we compute the average year-by-year elasticities for the region. This calls for a note on interpretation of the results from these different procedures. For instance, we might find with the year-by-year results a negative and statistically significant poverty–growth elasticity, which might seem to contradict our previous evidence of weak cross-country association between GDP growth rates and changes in the poverty rate between the initial and the final year. However, the two results are complementary. In our calculations in Chapter 4, the question we answered was: across countries, were differences in progress in reducing poverty between 2000 and 2012–13 linked to differences in economic growth rates? Our answer, according to the evidence in that chapter, was that countries with higher economic growth rates experienced larger reductions in poverty but the relationship was weak. On the other hand, the calculation of poverty–growth elasticities in this section answers a different question: if a country grows faster, what is the effect of faster growth on the change in its poverty rate? Our answer, based on the year-by-year regressions presented in section 5.1.2, is that economic growth reduces poverty but at different rates in different countries.

The literature on this topic has mainly focused on the effect of economic growth on poverty by estimating poverty–growth elasticities. In a recent review of poverty–growth elasticities, Alvaredo and Gasparini (2015: 784) present evidence on these elasticities for 114 developing countries over the period 1981–2010. They find that the change in poverty is closely and negatively related to economic growth, either in per capita gross national income (from national accounts) or per capita consumption/income growth (as measured in household surveys). Their estimations of the poverty elasticities with respect to per capita gross national income over the period 1999–2010 are (in absolute value) 1.2 and 1.9 for moderate and extreme poverty respectively. In other recent papers, Fosu (2011) and the World Bank (2011) estimate poverty–growth elasticities for Latin America for extreme and moderate poverty, based on regressions over the period 1980–2007 in the case of Fosu (2011) and 2003–10 in the case of the World Bank (2011). Estimates from Fosu’s study are 0.8 and 1.2 for moderate and extreme poverty rate, while the World Bank’s estimates are 1.6 and 2.0 respectively. These findings are (p.106) consistent with previous studies, which find that poverty generally falls when economic growth takes place, and that poverty tends not to fall in countries where economic growth has not taken place (Fields 2001).

Regarding the elasticity of other labour market indicators with respect to growth, the previous evidence for the region is more scattered. One example is Weller (2014) who estimates elasticities of the share of wage/salaried workers and share of self-employed with respect to GDP growth during the period 1995–2012 for fourteen Latin American countries. The study finds that a 1 per cent increase in GDP is associated with an increase in the share of wage/salaried employees of 0.5 per cent and a reduction in the share of self-employed by 0.27 per cent.

We compute the elasticities by regressing the year-by-year percentage change in the relevant dependent variable on the year-by-year percentage change in GDP per capita. Let %Δ‎Yikt be the year–by-year percentage change in indicator k for country i in period t. Let GDPit be GDP per capita for country i at time t. Let Ci be country-fixed effects which are included only in aggregate regressions η‎k for the region, but not in country-specific regressions; we call these aggregate regressions ‘stacked regressions’ which means that all the observations for all the countries are stacked. And let eit be the error term. We estimate the growth elasticity η‎k for indicator k in the stacked regressions as follows:

$Display mathematics$
(13)

with i = {AR, BO,…, VE}

k = {labour earnings, unemployment rate, etc.}

t = 2001,…, 2012/2013.

For country i (i = {AR, BO,…, VE}) we estimate the country-specific growth elasticity η‎k for indicator k as:

$Display mathematics$
(14)

with k = {labour earnings, unemployment rate, etc.}

t = 2001,…, 2012/2013.

We present the results from these growth elasticities in Table 5.1, with the aggregate elasticity from the stacked regression in the first column (for a total of 169 country–year observations from sixteen countries), and then in the remaining columns, we present the time series regression for each country, with a more limited number of observations (eleven on average for each country).

Table 5.1 Labour market indicators’ elasticities with respect to GDP per capita during the 2000s by country and for the Latin American region

Indicator

Stacked regression

AR

BO

BR

CL

CO

CR

DO

EC

HN

MX

PA

PE

PY

SV

UY

VE

Unemployment

Elasticity coefficient

−1.953

−1.340

−1.921

−2.790

−6.659

−0.385

−6.694

−3.790

−1.284

−2.270

−5.609

−1.893

−0.396

−1.001

−3.301

−1.704

−1.663

(0.331)**

(0.786)

(2.507)

(0.755)**

(2.718)*

(0.654)

(1.858)**

(1.704)*

(2.388)

(1.609)

(1.732)**

(1.435)

(0.434)

(0.979)

(1.356)*

(0.312)**

(0.404)**

Share of low-earnings occupations

Elasticity coefficient

−0.118

0.086

0.102

0.517

−1.311

−0.163

0.213

−0.007

−0.130

0.020

−0.014

−0.041

−0.083

−0.075

0.003

−0.195

(0.111)

(0.517)

(0.096)

(0.719)

(0.203)**

(0.161)

(0.218)

(0.285)

(0.184)

(0.471)

(0.237)

(0.33)

(0.181)

(0.128)

(0.116)

(0.146)

Share of high-earnings occupations

Elasticity coefficient

0.208

0.995

−0.493

−1.047

0.299

−0.005

−0.421

−0.431

2.077

2.352

−0.053

0.134

2.039

−0.043

0.097

(0.231)

(4.026)

(0.321)

(2.339)

(0.482)

(0.271)

(0.487)

(1.205)

(0.265)**

(0.761)**

(0.375)

(0.655)

(0.982)*

(0.172)

(0.488)

Share of wage/salaried employees

Elasticity coefficient

0.156

0.055

−1.601

0.226

0.179

−0.469

0.257

−0.105

−0.485

0.474

0.267

0.407

0.074

0.594

0.495

0.195

0.237

(0.055)**

(0.062)

(0.92)

(0.049)**

(0.081)*

(0.135)**

(0.225)

(0.266)

(0.647)

(0.205)*

(0.647)

(0.177)*

(0.172)

(0.095)**

(0.467)

(0.05)**

(0.026)**

Share of self-employment

Elasticity coefficient

−0.337

−0.328

1.190

0.087

−0.013

0.953

−0.851

0.322

−0.298

−1.547

−1.036

−0.814

0.353

−0.755

−0.823

−0.633

−0.343

(0.096)**

(0.24)

(0.714)

(0.3)

(0.224)

(0.153)**

(0.585)

(0.306)

(0.832)

(0.435)**

(1.072)

(0.358)*

(0.225)

(0.179)**

(0.672)

(0.151)**

(0.071)**

Share of unpaid family workers

Elasticity coefficient

−0.399

−0.012

0.478

−0.962

−0.065

−1.504

−1.464

3.434

0.783

−0.641

−0.442

0.285

−1.046

−0.467

−0.542

−1.094

−0.865

(0.309)

(0.787)

(2.15)

(0.572)

(3.549)

(0.447)**

(0.81)

(3.13)

(2.659)

(0.906)

(2.035)

(0.999)

(0.411)*

(0.631)

(1.107)

(0.512)*

(0.593)

Share of workers in low-earnings sectors

Elasticity coefficient

−0.019

0.468

0.299

−0.661

0.098

−0.633

−0.081

0.318

0.061

−0.348

0.591

0.047

−0.145

−0.395

−0.617

0.266

−0.208

(0.091)

(0.12)**

(0.738)

(0.299)*

(0.318)

(0.101)**

(0.278)

(0.364)

(0.233)

(0.354)

(0.727)

(0.321)

(0.248)

(0.101)**

(0.219)**

(0.121)*

(0.048)**

Share of workers in high-earnings sectors

Elasticity coefficient

−0.005

−0.570

1.890

−0.005

−1.217

0.696

−0.200

0.167

0.663

1.520

−0.333

0.503

−0.475

0.394

0.228

−0.303

0.004

(0.11)

(0.168)**

(0.99)

(0.197)

(0.85)

(0.104)**

(0.178)

(0.492)

(0.794)

(0.216)**

(0.554)

(0.34)

(0.24)*

(0.214)

(0.324)

(0.112)**

(0.114)

Share of low-educated workers

Elasticity coefficient

−0.046

−0.072

−0.749

−0.142

0.824

−0.612

0.140

−0.174

−0.032

−0.346

0.187

−0.010

−0.236

0.004

−0.196

0.364

−0.094

(0.057)

(0.047)

(0.369)*

(0.098)

(0.204)**

(0.167)**

(0.091)

(0.139)

(0.285)

(0.093)**

(0.384)

(0.264)

(0.205)

(0.27)

(0.112)

(0.29)

(0.079)

Share of high-educated workers

Elasticity coefficient

0.250

0.103

2.758

1.188

−1.242

1.007

−0.167

0.030

−0.394

2.144

−0.003

0.090

0.080

1.142

0.258

−0.764

0.108

(0.152)

(0.169)

(1.998)

(0.756)

(0.411)**

(0.255)**

(0.299)

(0.345)

(0.207)

(0.561)**

(1.659)

(0.446)

(0.423)

(0.72)

(0.641)

(0.487)

(0.138)

Share of workers registered with SS

Elasticity coefficient

0.541

0.402

6.716

0.574

0.625

1.592

0.096

2.582

−0.053

3.625

0.757

0.061

−1.124

0.655

1.175

0.307

0.180

(0.157)**

(0.181)*

(4.752)

(0.311)

(0.143)**

(0.438)**

(0.193)

(0.687)**

(1.084)

(3.125)

(0.513)

(0.282)

(1.325)

(0.358)

(0.171)**

(0.106)**

(0.128)

Mean labour earnings

Elasticity coefficient

1.133

1.597

1.521

0.616

−1.176

0.912

0.181

1.741

0.319

1.361

1.238

0.555

0.128

1.265

0.306

1.055

1.232

(0.155)**

(0.43)**

(0.634)*

(0.409)

(1.109)

(0.331)**

(0.673)

(0.716)*

(1.093)

(0.494)**

(0.526)*

(0.41)

(0.728)

(0.251)**

(0.528)

(0.256)**

(0.258)**

2.5 dollars-a-day poverty

Elasticity coefficient

−2.100

−3.866

0.036

−0.904

−1.910

0.233

−2.329

−0.436

−0.703

−0.480

−0.209

0.551

−0.006

−1.758

−1.623

−3.576

−2.030

(0.354)**

(0.167)**

(2.898)

(0.597)

(0.605)**

(0.332)

(1.404)

(1.11)

(0.772)

(1.739)

(0.946)

(0.962)

(0.907)

(0.933)

(2.139)

(0.549)**

(0.613)**

4 dollars-a-day poverty

Elasticity coefficient

−1.427

−2.578

−0.655

−0.603

−0.210

−0.430

−1.471

−0.175

−1.014

−0.344

−0.004

−0.289

−0.106

−0.719

−0.341

−2.954

−1.315

(0.261)**

(0.234)**

(1.583)

(0.41)

(1.292)

(0.221)

(1.006)

(0.762)

(0.534)

(0.979)

(0.555)

(0.699)

(0.44)

(0.69)

(0.999)

(0.483)**

(0.419)**

Gini of household per capita income

Elasticity coefficient

−0.082

−0.253

−0.233

−0.093

−0.594

0.191

−0.096

0.144

−0.588

0.516

0.947

0.036

0.107

0.317

−0.485

−0.292

−0.058

(0.074)

(0.074)**

(1.268)

(0.031)**

(0.075)**

(0.116)

(0.381)

(0.669)

(0.446)

(0.625)

(0.241)**

(0.223)

(0.423)

(0.205)

(0.124)**

(0.12)*

(0.127)

Gini of labour earnings

Elasticity coefficient

−0.123

−0.363

−0.387

−0.251

−0.059

−0.247

0.202

−0.004

−0.595

0.418

0.965

−0.443

0.234

0.053

−0.247

−0.383

−0.001

(0.069)

(0.058)**

(0.774)

(0.064)**

(0.036)

(0.159)

(0.271)

(0.174)

(0.768)

(0.471)

(0.56)

(0.485)

(0.358)

(0.193)

(0.188)

(0.131)**

(0.167)

Note: Labour market indicators’ elasticities are calculated using the year-by-year percentage change in labour market indicators and GDP per capita within each country. The first column shows the results of the regression for the sample of all countries including country-fixed effects. The country-specific regressions do not include extra controls. Robust standard errors in parentheses.

(**) significant at 1% level,

(*) significant at 5% level.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014).

## 5.1.1 Response of Employment and Earnings to Growth

We start by analysing the aggregate elasticity of labour earnings with respect to GDP per capita (stacked regression column in Table 5.1 and mean labour (p.107) (p.108) (p.109) earnings row). We find that mean labour earnings increased more than proportionately as GDP per capita grew. The labour earnings elasticity with respect to GDP per capita is 1.13: a 1 per cent increase in GDP per capita from one year to the next is associated with an average increase of 1.13 per cent in mean labour earnings. This relationship is also statistically significant for nine countries in the region (columns 2 to 17 in Table 5.1 and mean labour earnings row): Argentina, Bolivia, Colombia, Dominican Republic, Honduras, Mexico, Paraguay, Uruguay, and Venezuela. For all but one of these nine countries the elasticities are higher than 1, whereas the elasticities for the countries for which we find no significant coefficients are all below 0.6 but still positive (with the exception of Chile, with a negative coefficient).

We find a strong negative and significant aggregate year-by-year elasticity of unemployment with respect to GDP per capita of around −2 (stacked regression column in Table 5.1 and unemployment row). We find again, however, a high degree of heterogeneity when looking at the country-specific elasticities (columns 2 to 17 in Table 5.1 and unemployment row). While the estimated coefficients are all negative, they are significant and about −3 or larger in absolute value in Brazil, Chile, Costa Rica, Dominican Republic, Mexico, and El Salvador, substantially closer to the aggregate elasticity and significant for Uruguay and Venezuela, and not significant but still negative and large for the remaining countries except for Colombia and Peru where the estimates are closer to zero.

We also find significant aggregate year-by-year elasticities of labour market indicators broadly associated with the job mix and quality of employment with respect to GDP per capita (stacked regression column in Table 5.1 and corresponding indicator row): the share of workers registered with social security, the share of wage/salaried employees (both positive), and the share of self-employment (negative). Specifically, these results indicate that the share of registered workers increased by 0.54 per cent for each 1 per cent increase in GDP per capita, whereas the elasticity for the share of wage/salaried employees is substantially smaller (0.16). At the same time, an increase of 1 per cent in GDP is related to a decrease in the share of self-employment of about 0.34 per cent. As with the previously discussed indicators, there is a large degree of heterogeneity when looking at the estimates by country (columns 2 to 17 in Table 5.1 and corresponding indicator row).

We found insignificant aggregate year-by-year growth elasticities for a series of labour market indicators (stacked regression column in Table 5.1 and corresponding indicator row). Some of these results were not as expected ex ante: for instance, the lack of a significant aggregate relationship between percentage changes in GDP per capita and percentage changes in the share of high- and low-earnings occupations, in the share of workers in low- and high-earning sectors, and in the share of unpaid family workers.

## (p.110) 5.1.2 Response of Poverty and Inequality to Growth

Now we turn to the analysis of the poverty and inequality indicators’ elasticities with respect to GDP per capita. The aggregate year-by-year changes in poverty and in extreme poverty are found to be strongly negatively correlated with changes in GDP per capita (stacked regression column in Table 5.1, and 2.5 and 4 dollars-a-day poverty rows). This means that for each 1 per cent increase in GDP per capita from one year to the next, poverty decreases, on average, by 1.43 per cent, and extreme poverty decreases by 2.1 per cent. Expressing these estimates in terms of percentage points rather than percentages, we find that an increase of GDP per capita of 1 per cent implies a fall of about 0.58 percentage points in moderate poverty, and of about 0.50 percentage points in extreme poverty (with respect to the unweighted average of the moderate and extreme poverty rates of the year 2000 in Figure 3.1 in Chapter 3). These values are in line with those obtained in the literature for developing countries.1

Table 5.1 also includes poverty–growth elasticities country by country (columns 2 to 17 in Table 5.1, and 2.5 and 4 dollars-a-day poverty rows). We find a large degree of heterogeneity across countries. In only four of the sixteen countries in our sample do we find a statistically significant (at 5 per cent level) moderate poverty–growth or extreme poverty–growth elasticity (Argentina, Chile, Uruguay, and Venezuela). It should be noted that all but three of the estimated elasticities for all countries are negative, and that the country analysis is less robust since we only have between six and thirteen observations in each case. The larger (in absolute value) and most significant elasticities are those found for countries which suffered domestic crisis at the beginning of the 2000s, and then have larger variability in their year-by-year data. That was the case for Argentina (elasticities of −3.87 for extreme poverty and −2.58 for poverty), Uruguay (−3.58 and −2.95 respectively), and Venezuela (−2.03 and −1.32 respectively).

With respect to inequality indicators, we find small negative and not significant aggregate growth elasticities for the Gini of household per capita income (HPCI) and the Gini of labour earnings (LI) (stacked regression column in Table 5.1 and Gini of household per capita income and Gini of labour earnings rows) because of great heterogeneity in country experiences (columns 2 to 17 in Table 5.1 and Gini of household per capita income and Gini of labour earnings rows). The country elasticities are negative and significant for Argentina (HPCI (p.111) and LI), Brazil (HPCI and LI), Chile (HPCI), El Salvador (HPCI), and Uruguay (HPCI and LI), and positive and significant only for Mexico (HPCI).

To finalize this section, we illustrate in Figure 5.1 some of the country-specific elasticities with respect to GDP per capita showing the year-by-year changes for some selected labour market indicators (mean labour earnings, extreme and moderate poverty rates) for some illustrative countries: Honduras, the Dominican Republic, Bolivia, and Brazil. The four countries experienced positive GDP per capita growth rates in most of the years and had relatively similar annualized growth rates: 2.1 per cent for Honduras, 3.6 for the Dominican Republic, 2.2 for Bolivia, and 2.4 for Brazil. However, their labour market experiences were dissimilar. Honduras and the Dominican Republic are relatively bad performers in terms of the evolution of poverty and labour market indicators in the 2000s, while Bolivia and Brazil present much better patterns for these variables over time (see section 5.3). With this exercise we want to look deeper into the year-by-year changes that underlie our elasticities estimations.

Figure 5.1 Mean labour earnings, 2.5 and 4 dollars-a-day poverty rates elasticity with respect to GDP per capita for illustrative countries

Note: The points in each figure represent year-by-year percentage changes in the labour market indicator indicated in the vertical axes and GDP per capita. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014) and World Development Indicators (World Bank 2014).

The top row in Figure 5.1 presents the relationship between annual percentage changes in mean labour earnings and annual percentage changes in GDP per capita. The elasticities of labour earnings with respect to GDP per capita (slope coefficient of the regression line in the bottom of each figure) are quite similar in Honduras, the Dominican Republic, and Bolivia (between 1.36 and 1.74). However, the figure allows us to discern the different evolution of labour earnings over time in each country. For both Honduras and the Dominican Republic, we observe negative percentage changes in mean labour earnings with respect to the previous year for most of the years (evidenced by the fact that many of the points are below the zero horizontal line). Moreover, these losses in average earnings occurred even in years with positive GDP per capita growth rates. On the contrary, the figure indicates that average earnings in Bolivia and Brazil increased with respect to the previous year for most of the years we analyse (e.g. most of the points are above the zero horizontal line), even in periods with no growth in GDP per capita. In conclusion, Bolivia and Brazil were more effective than Honduras and the Dominican Republic in translating GDP per capita growth into labour earnings increases. This can be clearly seen by comparing the height of the regression lines in the top row of Figure 5.1 which we reproduce for Honduras and Bolivia in the first graph of Figure 5.2.

Figure 5.2 Relationship between percentage changes in mean labour earnings, 2.5 and 4 dollars-a-day poverty rates, and percentage changes in GDP per capita for illustrative countries

Note: Linear regression of the year-by-year percentage changes in each labour market indicator on year-by-year percentage changes in GDP per capita.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014) and World Development Indicators (World Bank 2014).

Turning now from labour earnings to poverty, in the second row of Figure 5.1, we observe that the extreme poverty–growth elasticities were very similar in Honduras and the Dominican Republic (slope coefficient of the regression lines at the bottom of each figure: −0.48 and −0.44 respectively), but while in the Dominican Republic the poverty rate measured by the 2.5 dollars-a-day line fell in most of the years (most of the points are below the zero horizontal line), in Honduras it increased most of the time, even in times (p.112) (p.113) (p.114) of positive GDP per capita growth rates. In Bolivia, the percentage changes in the extreme poverty rate were not significantly associated with GDP per capita growth. Bolivia exhibited similar reductions in the extreme poverty rate in years of low and high GDP per capita growth. Finally, in Brazil the extreme poverty rate fell most of the time and the reductions were larger the higher the GDP per capita growth, producing an estimated elasticity of −0.9. The Dominican Republic, Bolivia, and Brazil were more successful than Honduras in reducing extreme poverty (the regression lines were always below the zero horizontal line for the Dominican Republic, Bolivia, and Brazil, and always above zero for Honduras). While the Dominican Republic and Brazil seem to have translated GDP per capita growth into lower extreme poverty rates, Bolivia managed to reduce extreme poverty in both high-growth and low-growth years. The second graph in Figure 5.2 provides a clear comparison of the regression lines for Honduras and Bolivia.

Turning to moderate poverty (third row of Figure 5.1), the figures for moderate poverty elasticity are very similar to the ones of extreme poverty–growth, Bolivia being the only exception. In Bolivia, the moderate poverty rate fell most of the time and the reductions were larger when the GDP per capita grew the most.

To sum up, in the Latin American region, the year-by-year percentage changes in some employment and earnings indicators (unemployment, share of wage/salaried employees, share of self-employed, mean earnings) and poverty indicators (2.5 and 4 dollars-a-day poverty rates) were related in the welfare-improving direction to GDP per capita growth (stacked regression column of Table 5.1). The same was true for most of the countries, but the (p.115) magnitudes of the effect and the patterns over time varied substantially from country to country (remaining columns of Table 5.1). The heterogeneity among countries explains why in section 4.1 we found a weak relationship across countries between improvements in the labour indicators and the rate of economic growth. It is not the case that economic growth was unimportant for improvements in labour market indicators. It is the case that more rapid economic growth improved labour market indicators in all the countries, but at a different rate in each one of them.

# 5.2 Response of Poverty to Employment, Earnings, and Inequality Changes

In this section, we analyse in more detail the link between employment and earnings indicators and poverty. Our analysis is based on the estimation of moderate and extreme poverty elasticities with respect to employment and earnings indicators. We compute these elasticities using year-by-year data for each country in our sample as in section 5.1 and in contrast to Chapter 4 where we used the annualized changes between the initial and the final years for each country.

There is a limited literature on the elasticities of poverty with respect to employment, earnings, and inequality indicators. Fosu (2011), Olinto et al. (2014) and Alvaredo and Gasparini (2015) have focused on the poverty elasticity with respect to inequality in household per capita income. Both papers use data from Povcalnet, which is the World Bank’s computational tool available online to estimate absolute poverty and other distributional indicators (e.g. the Gini coefficient) in developing countries. Povcalnet provides an estimate of poverty and inequality indicators for each country every three years since 1981. Fosu (2011) finds a positive poverty elasticity with respect to inequality (measured by the Gini of per capita household income) for the world as a whole and in particular for the stacked data from countries of Latin American and the Caribbean for the period 1981–2007. Similar results are obtained by Olinto et al. (2014) and Alvaredo and Gasparini (2015) for a group of 100 developing countries, including those from the Latin American region for the period 1981–2010. These papers use a similar data structure to ours (stacked data for each country with many countries included), but they implement different econometric techniques than ours (e.g. instrumenting for the increase on inequality) and also use a different period of time.

An example of a study analysing the response of poverty to employment indicators is Damill and Frenkel (2014), who estimate that poverty tends to increase 0.7 percentage points for each one percentage point increase in the unemployment rate in Latin America for the period 1990–2010.

(p.116) Since we are focused on the employment and poverty nexus, below we estimate poverty elasticities with respect to the mean of individuals’ labour earnings. However, the previous literature—for example, Fosu (2011), Olinto et al. (2014), and Alvaredo and Gasparini (2015)—has mainly estimated the elasticity of poverty with respect to mean household per capita income, finding that poverty tends to fall with increases in mean household per capita income.

Let P(l)it be the poverty rate measured using the poverty line l, for country i in period t. Let Kit be either labour earnings, the unemployment rate, or any other employment and earnings indicator for country i at time t. Let Ci be country-fixed effects which are included only in aggregate regressions for the region, but not in country-specific regressions; as in section 5.1, we call these aggregate regressions ‘stacked regressions’ which means that observations of each country are stacked. And let eit be the error terms. We estimate the elasticity of poverty with respect to the labour market indicator k (δ‎k) in the stacked regression as follows:

$Display mathematics$
(15)

with l = 2.5 or 4 dollars-a-day poverty lines

i = {AR, BO,…, VE}

t = 2001,…, 2012/2013

k = {labour earnings, unemployment rate, etc.}.

For country i (i = {AR, BO,…, VE}) we estimate the country-specific elasticity of poverty with respect to the labour market indicator k (δ‎k) as follows:

$Display mathematics$
(16)

with l = 2.5 or 4 dollars-a-day poverty lines

t = 2001,…, 2012/2013

k = {labour earnings, unemployment rate, etc.}.

We present the results from these estimations in Tables 5.2 (for extreme poverty) and 5.3 (for moderate poverty).

Table 5.2 2.5 dollars-a-day elasticity with respect to employment and earnings indicators and inequality indicators during the 2000s by country and for the Latin American region

Indicator

Stacked regression

AR

BO

BR

CL

CO

CR

DO

EC

HN

MX

PA

PE

PY

SV

UY

VE

Unemployment

Elasticity coefficient

0.332

1.164

0.391

0.425

0.171

0.314

0.363

−0.077

0.206

0.073

0.158

0.192

0.743

0.867

−0.077

1.217

1.138

(0.096)**

(0.212)**

(0.266)

(0.137)**

(0.095)

(0.445)

(0.143)*

(0.04)

(0.118)

(0.166)

(0.186)

(0.261)

(0.418)

(0.164)**

(0.281)

(0.848)

(0.227)**

Share of low-earnings occupations

Elasticity coefficient

0.318

−0.565

3.808

−0.900

−0.185

1.102

−3.172

0.919

−0.373

1.232

1.028

1.449

4.292

2.986

−0.217

2.063

(0.419)

(1.235)

(2.235)

(1.041)

(0.173)

(1.492)

(0.706)**

(1.464)

(2.292)

(1.564)

(1.279)

(1.573)

(3.503)

(5.64)

(3.202)

(1.472)

Share of high-earnings occupations

Elasticity coefficient

−0.132

0.093

−1.035

−0.001

−0.226

0.050

1.096

0.223

−0.288

0.308

−1.456

−0.420

−0.758

0.725

0.335

(0.177)

(0.271)

(0.396)**

(0.336)

(0.486)

(0.441)

(0.393)**

(0.254)

(0.671)

(0.565)

(1.088)

(0.57)

(0.401)

(1.424)

(1.239)

Share of wage/salaried employees

Elasticity coefficient

−1.501

−7.447

−0.966

−4.035

−9.594

−0.772

−3.990

−0.217

0.117

−0.411

−0.347

−1.081

−1.443

−2.716

−1.105

−12.378

−6.124

(0.368)**

(4.677)

(0.786)

(1.715)*

(0.761)**

(0.386)*

(1.976)*

(1.502)

(0.437)

(0.624)

(1.079)

(1.909)

(1.416)

(1.371)*

(0.933)

(2.673)**

(1.842)**

Share of self-employment

Elasticity coefficient

1.115

2.492

0.765

0.951

2.723

0.368

1.127

−0.878

0.089

0.321

−0.156

0.504

1.367

1.724

0.655

4.174

3.349

(0.259)**

(0.777)**

(0.706)

(1.193)

(4.172)

(0.182)*

(0.712)

(0.576)

(0.594)

(0.445)

(0.561)

(1.089)

(1.15)

(1.042)

(0.713)

(0.896)**

(1.172)**

Share of unpaid family workers

Elasticity coefficient

0.227

−0.115

−0.087

0.775

0.149

0.233

1.031

0.136

−0.040

0.567

0.258

0.093

0.103

1.234

0.217

0.703

0.274

(0.079)**

(0.292)

(0.424)

(0.257)**

(0.278)

(0.202)

(0.556)

(0.033)**

(0.157)

(0.337)

(0.33)

(0.187)

(0.438)

(0.302)**

(0.197)

(0.419)

(0.154)

Share of workers in low-earnings sectors

Elasticity coefficient

0.005

−4.864

−0.914

1.134

−1.224

−0.592

−0.018

−1.558

2.193

1.292

0.425

0.870

1.950

2.034

2.288

−2.968

7.115

(0.581)

(1.457)**

(1.012)

(0.373)**

(1.632)

(0.259)*

(0.858)

(1.518)

(1.427)

(0.697)

(0.446)

(0.724)

(2.126)

(1.606)

(2.071)

(1.556)

(1.6)**

Share of workers in high-earnings sectors

Elasticity coefficient

0.183

4.131

−0.222

−0.390

0.553

0.404

0.386

0.549

0.179

−0.522

0.536

−0.738

−1.151

0.031

−1.164

1.738

−1.772

(0.252)

(0.899)**

(0.361)

(0.792)

(0.321)

(0.427)

(0.954)

(0.844)

(0.495)

(0.766)

(1.077)

(0.564)

(0.863)

(0.315)

(1.004)

(1.908)

(1.628)

Share of low-educated workers

elasticity coefficient

0.264

4.290

−0.151

3.614

−2.323

0.645

−1.972

−0.441

2.334

2.669

0.307

1.099

1.159

1.910

1.211

−2.020

4.051

(0.558)

(4.61)

(1.606)

(0.9)**

(0.376)**

(0.781)

(1.057)

(2.013)

(1.16)*

(2.261)

(1.024)

(0.87)

(1.536)

(1.176)

(3.828)

(1.249)

(2.7)

Share of high-educated workers

Elasticity coefficient

−0.065

−2.869

0.058

0.150

1.208

−0.207

2.738

0.912

1.114

0.147

−0.046

0.264

−0.245

−0.823

0.698

1.211

−3.655

(0.187)

(5.79)

(0.348)

(0.189)

(0.562)*

(0.394)

(1.719)

(0.152)**

(1.398)

(0.292)

(0.29)

(0.953)

(0.883)

(0.353)*

(0.541)

(0.66)

(2.084)

Share of workers registered with SS

Elasticity coefficient

−0.114

−2.457

0.138

−0.829

−2.576

−0.112

−3.503

0.112

−0.132

0.144

−0.616

−2.810

−0.208

−0.744

−1.027

−4.944

−1.513

(0.186)

(2.265)

(0.141)

(0.193)**

(1.202)*

(0.505)

(2.799)

(0.338)

(0.365)

(0.185)

(1.031)

(1.388)*

(0.26)

(0.401)

(1.294)

(1.405)**

(1.321)

Mean labour earnings

Elasticity coefficient

−1.236

−1.835

0.231

−1.298

0.265

−0.452

−0.544

−0.654

−0.267

−0.905

0.342

−0.413

−0.534

−1.427

−0.960

−2.184

−1.536

(0.171)**

(0.139)**

(0.732)

(0.156)**

(0.432)

(0.274)

(1.213)

(0.424)

(0.337)

(0.497)

(0.505)

(0.283)

(0.621)

(0.718)*

(1.347)

(0.487)**

(0.29)**

Gini of household per capita income

Elasticity coefficient

2.083

8.333

2.095

−0.717

2.848

−0.618

1.669

−1.041

0.061

2.334

1.235

3.517

1.186

2.818

2.867

3.573

1.394

(0.378)**

(2.631)**

(0.291)**

(2.908)

(0.812)**

(0.991)

(1.271)

(0.611)

(0.829)

(0.719)**

(1.175)

(1.632)*

(0.847)

(1.414)*

(1.435)*

(2.263)

(1.411)

Gini of labour earnings

Elasticity coefficient

1.266

7.405

3.053

0.397

6.059

−0.836

−0.257

−2.060

0.360

1.481

0.395

0.354

0.860

2.014

0.409

3.529

0.519

(0.391)**

(1.55)**

(0.428)**

(2.043)

(13.147)

(0.573)

(1.239)

(0.703)**

(0.5)

(0.338)**

(0.54)

(1.362)

(0.903)

(1.614)

(0.507)

(1.702)*

(0.846)

Note: Labour market indicators’ elasticities are calculated using the year-by-year percentage change in the 2.5 dollars-a-day poverty rate and in employment and earnings indicators and inequality indicators within each country. The first column shows the results of the regression for the sample of all countries including country-fixed effects. The country-specific regressions do not include extra controls. Robust standard errors in parentheses.

(**) significant at 1% level,

(*) significant at 5% level.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014).

Table 5.3 4 dollars-a-day elasticity with respect to employment and earnings indicators and inequality indicators during the 2000s by country and for the Latin American region

Indicator

Stacked regression

AR

BO

BR

CL

CO

CR

DO

EC

HN

MX

PA

PE

PY

SV

UY

VE

Unemployment

Elasticity coefficient

0.193

0.673

0.214

0.224

−0.100

0.168

0.188

−0.072

0.120

0.063

0.033

0.260

0.633

0.546

−0.091

1.181

0.724

(0.066)**

(0.191)**

(0.137)

(0.115)

(0.11)

(0.158)

(0.116)

(0.027)**

(0.076)

(0.106)

(0.098)

(0.135)

(0.284)*

(0.156)**

(0.138)

(0.465)*

(0.14)**

Share of low-earnings occupations

Elasticity coefficient

0.587

0.080

2.080

−1.360

0.249

0.325

−1.939

0.823

0.115

0.766

0.699

1.514

3.276

1.512

3.260

1.608

(0.295)*

(0.605)

(1.533)

(0.474)**

(0.142)

(0.911)

(0.381)**

(0.916)

(1.451)

(0.617)

(0.734)

(0.918)

(2.281)

(2.487)

(2.478)

(1.08)

Share of high-earnings occupations

Elasticity coefficient

−0.159

−0.012

−0.578

0.173

−0.325

−0.368

0.434

0.108

−0.304

0.075

−1.311

−0.534

−0.491

−0.901

−0.017

(0.112)

(0.134)

(0.308)

(0.094)

(0.304)

(0.488)

(0.303)

(0.187)

(0.457)

(0.299)

(0.743)

(0.357)

(0.291)

(1.089)

(0.852)

Share of wage/salaried employees

Elasticity coefficient

−0.972

−4.215

−0.375

−3.289

−4.521

−0.227

−3.637

−0.500

0.001

−0.321

−0.320

−1.961

−1.026

−1.488

−0.438

−11.149

−4.449

(0.259)**

(3.59)

(0.498)

(1.055)**

(3.059)

(0.458)

(1.201)**

(0.894)

(0.347)

(0.375)

(0.316)

(1.025)

(0.863)

(1.006)

(0.435)

(1.642)**

(1.175)**

Share of self-employment

Elasticity coefficient

0.802

1.415

0.176

0.306

5.546

−0.082

1.190

−0.317

0.089

0.228

0.066

0.960

0.581

0.906

0.329

3.835

2.301

(0.181)**

(0.648)*

(0.477)

(0.689)

(0.552)**

(0.263)

(0.355)**

(0.475)

(0.407)

(0.245)

(0.178)

(0.581)

(0.841)

(0.747)

(0.32)

(0.601)**

(0.735)**

Share of unpaid family workers

Elasticity coefficient

0.169

−0.059

−0.099

0.570

0.361

0.230

0.710

0.106

0.016

0.295

0.135

0.111

0.242

0.814

0.060

0.388

0.207

(0.053)**

(0.216)

(0.192)

(0.193)**

(0.028)**

(0.157)

(0.363)

(0.015)**

(0.089)

(0.184)

(0.088)

(0.075)

(0.197)

(0.246)**

(0.089)

(0.334)

(0.098)*

Share of workers in low-earnings sectors

Elasticity coefficient

0.098

−3.596

−0.262

0.746

1.631

0.095

0.064

−0.864

1.313

0.621

0.481

1.240

1.710

1.046

1.110

−0.458

4.181

(0.379)

(0.979)**

(0.597)

(0.261)**

(1.059)

(0.334)

(0.801)

(1.175)

(1.145)

(0.392)

(0.174)**

(0.28)**

(1.273)

(0.997)

(0.898)

(1.398)

(1.151)**

Share of workers in high-earnings sectors

Elasticity coefficient

0.024

2.846

−0.207

0.004

−0.676

0.092

−0.481

0.781

−0.093

−0.409

−0.004

−0.666

−0.891

0.016

−0.305

0.066

−1.028

(0.173)

(0.69)**

(0.206)

(0.533)

(0.224)**

(0.375)

(1.08)

(0.439)

(0.332)

(0.473)

(0.523)

(0.252)**

(0.567)

(0.204)

(0.415)

(1.648)

(1.142)

Share of low-educated workers

Elasticity coefficient

0.521

2.314

−0.031

2.262

−0.388

0.745

−0.088

−0.238

1.822

1.505

0.067

1.038

1.305

1.309

1.067

−0.489

2.141

(0.409)

(3.323)

(0.998)

(0.65)**

(1.405)

(0.39)

(1.691)

(1.8)

(1.048)

(1.226)

(0.741)

(0.315)**

(0.968)

(0.729)

(1.564)

(1.163)

(2.109)

Share of high-educated workers

Elasticity coefficient

−0.119

−2.044

−0.006

−0.026

−0.346

−0.264

1.396

0.466

0.778

0.047

−0.024

0.276

−0.557

−0.503

0.254

0.440

−2.382

(0.125)

(4.064)

(0.212)

(0.15)

(0.745)

(0.163)

(1.566)

(0.166)**

(0.792)

(0.17)

(0.218)

(0.792)

(0.564)

(0.241)*

(0.213)

(0.652)

(1.304)

Share of workers registered with SS

Elasticity coefficient

−0.104

−1.618

0.039

−0.760

0.548

−0.064

−1.745

0.051

−0.065

0.054

−0.343

−2.154

−0.074

−0.484

−0.203

−4.358

−0.738

(0.116)

(1.638)

(0.094)

(0.151)**

(1.784)

(0.239)

(2.378)

(0.181)

(0.27)

(0.135)

(0.444)

(0.921)*

(0.199)

(0.242)*

(0.646)

(1.327)**

(0.87)

Mean labour earnings

Elasticity coefficient

−0.950

−1.250

−0.203

−0.955

−0.694

−0.445

−0.791

−0.626

−0.180

−0.538

−0.086

−0.323

−0.626

−0.825

−0.611

−2.015

−1.078

(0.111)**

(0.123)**

(0.415)

(0.094)**

(0.227)**

(0.166)**

(0.751)

(0.211)**

(0.261)

(0.268)*

(0.256)

(0.227)

(0.384)

(0.489)

(0.578)

(0.275)**

(0.137)**

Gini of household per capita income

Elasticity coefficient

1.244

5.885

1.140

0.762

0.547

−0.704

0.684

−0.495

0.149

1.301

0.574

2.283

0.298

1.976

1.090

3.416

0.685

(0.261)**

(2.016)**

(0.217)**

(2.027)

(1.966)

(0.809)

(1.047)

(0.419)

(0.539)

(0.386)**

(0.551)

(1.441)

(0.55)

(0.977)*

(0.698)

(1.306)**

(1.016)

Gini of labour earnings

Elasticity coefficient

0.891

5.457

1.753

0.783

−7.159

−0.499

−1.111

−1.179

0.319

0.967

0.266

0.875

0.099

1.511

−0.025

3.613

0.192

(0.288)**

(0.953)**

(0.325)**

(1.171)

(13.986)

(0.615)

(1.309)

(0.449)**

(0.312)

(0.257)**

(0.314)

(0.933)

(0.691)

(1.128)

(0.235)

(1.091)**

(0.636)

Note: Labour market indicators’ elasticities are calculated using the year-by-year percentage change in the 4 dollars-a-day poverty rate and in employment and earnings indicators and inequality indicators within each country. The first column shows the results of the regression for the sample of all countries including country-fixed effects. The country-specific regressions do not include extra controls. Robust standard errors in parentheses.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014).

We start by analysing poverty–labour earnings elasticities. We see in the stacked regressions (stacked regression column in Tables 5.2 and 5.3 and mean labour earnings row) that the percentage changes in poverty and in extreme poverty are strongly correlated with the evolution of labour earnings in the expected direction, i.e. higher increases in labour earnings being associated with larger poverty reductions. The aggregate extreme poverty–labour earnings elasticity is −1.23, and the elasticity for moderate poverty is −0.95 (both significant at the 1 per cent level). These poverty–labour earnings elasticities are substantially smaller in absolute magnitude than the poverty–growth elasticities we estimated in section 5.1.2 (see Table 5.1). This could be expected (p.117) (p.118) (p.119) (p.120) (p.121) from the trends observed in Figure 3.1 in Chapter 3, which shows that labour earnings and GDP per capita followed similar trends, but changes in labour earnings were more attenuated than those in GDP per capita.

The poverty–labour earnings elasticities differ between countries (columns 2 to 17 in Table 5.2 and Table 5.3 and mean labour earnings row). The magnitudes of the moderate poverty–earnings elasticities go from −2.0 in Uruguay to −0.09 in Mexico, while the values of the extreme poverty–earnings elasticities vary from −2.2 in Uruguay to 0.23 in Bolivia. At least one of the two elasticities (poverty or extreme poverty) is statistically significant at standard levels for nine out of sixteen countries (Argentina, Brazil, Chile, Colombia, Dominican Republic, Honduras, Paraguay, Uruguay, and Venezuela). As in the case of the poverty–growth elasticities, the poverty–labour earnings elasticities are large and highly significant for Argentina (−1.10 for moderate poverty and −1.55 for extreme poverty), Uruguay (−1.92 and −2.14 respectively), and Venezuela (−1.20 and −1.75 respectively). The results are also large and significant for Brazil (−1.07 for moderate poverty and −1.74 for extreme poverty), for which the poverty–growth elasticities were not statistically different from zero.

Turning now to the elasticity of poverty with respect to unemployment in the stacked regression, we find a strong and significant correlation between reductions in the unemployment rate and reductions in poverty and extreme poverty. Earlier we found a clear positive correlation between the unweighted averages of the unemployment rate and the poverty rates (Figure 3.1 in Chapter 3). Consistent with this, we find here significant and positive aggregate elasticities of moderate and extreme poverty rates with respect to unemployment (stacked regression column in Tables 5.2 and 5.3 and unemployment row) of 0.19 for moderate poverty (Table 5.3) and 0.33 for extreme poverty (Table 5.2), both significant at the 1 per cent level. This implies that, on average, for each 10 per cent reduction in the unemployment rate (for example, from approximately 9 per cent, the average for all sixteen countries at the beginning of the period, to 8.1 per cent), poverty falls by 1.9 per cent and extreme poverty by 3.3 per cent. Looking at the country level, as with the other elasticities discussed above, the poverty–unemployment elasticities are highly variable between countries (columns 2 to 17 in Table 5.2 and Table 5.3 and unemployment row). One or both of these elasticities (poverty or extreme poverty) are significant for Argentina, Brazil, Costa Rica, Dominican Republic, Peru, Paraguay, Uruguay, and Venezuela. The magnitudes of the elasticities are large and strongly significant for Argentina (0.67 for moderate poverty and 1.16 for extreme poverty), Paraguay (0.55 and 0.87 respectively), and Venezuela (0.72 and 1.14 respectively).

We also find a strong correlation between percentage changes in moderate and extreme poverty and percentage changes in the three labour market indicators related to the occupational position in the stacked regression (p.122) (stacked regression column in Tables 5.2 and 5.3 and the corresponding indicator row). First, we find a negative and significant aggregate elasticity between extreme and moderate poverty and the share of wage/salaried employees, with a substantially higher coefficient (in absolute terms) for extreme poverty (−1.50) than for moderate poverty (about 0.97). The elasticities of poverty with respect to the occupational positions that we identified as signals of worse labour market outcomes, the share of self-employment (second) and the share of unpaid family workers (third), are positive, and substantially larger for the share of self-employment (1.12 for extreme poverty and 0.80 for moderate poverty), than for the share of unpaid workers (0.23 for extreme poverty and 0.17 for moderate poverty). As with the previous indicators, there is a high degree of heterogeneity in the magnitude of the elasticities between countries, although the signs seem to be mostly consistent among them (columns 2 to 17 in Table 5.2 and Table 5.3 and the corresponding indicator row). We did not find a significant average year-by-year poverty elasticity for the remaining employment and earnings indicators and inequality indicators, such as the share of high- and low-earnings occupations, the share of workers registered with social security, the share of workers in low- and high-earnings sectors, and the Gini coefficient of household per capita income and labour earnings.

As in section 5.1, we present in Figures 5.3 and 5.4 some of the elasticities of poverty with respect to mean labour earnings and unemployment for four countries in our sample: Honduras, the Dominican Republic, Bolivia, and Brazil. In Honduras, the extreme and moderate poverty rates increased in about half of the years under study, and the increases took place even with reductions in the unemployment rate (top row of Figure 5.3 for the extreme poverty rate and Figure 5.4 for the moderate poverty rate). That determines very small positive elasticities of moderate and extreme poverty (0.06 and 0.07 respectively) with respect to the unemployment rate, and very small R-squareds (0.02 and 0.01 respectively) (regression details in the bottom of each figure). The Dominican Republic is the only country among the four where the poverty–unemployment elasticities are negative (slope coefficient of the regression line in the bottom of each figure: −0.08 for moderate poverty and −0.07 for extreme poverty). This result is determined mainly by one year that had a large increase in the unemployment rate jointly with a large reduction in the poverty rates. In Bolivia and Brazil, both poverty rates fell most of the time, and continued to decline when the unemployment rate increased (most of the points are below the zero horizontal line). The poverty–unemployment elasticities are similar in magnitude in both countries (about 0.4 for moderate poverty and 0.2 for extreme poverty).

Figure 5.3 2.5 dollars-a-day poverty rates elasticity with respect to unemployment and mean earnings for illustrative countries

Note: The points in each figure represent year-by-year percentage changes in the 2.5 dollars-a-day poverty rate, and the labour market indicator indicated in the horizontal axes. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014).

Figure 5.4 4 dollars-a-day poverty rates elasticity with respect to unemployment and mean earnings for illustrative countries

Note: The points in each figure represent year-by-year percentage changes in the 2.5 dollars-a-day poverty rate, and the labour market indicator indicated in the horizontal axes. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014).

The analysis of the relationship between percentage changes in poverty and percentage changes in mean earnings (second row of Figure 5.3 for the (p.123) (p.124) (p.125) (p.126) (p.127) extreme poverty rate and Figure 5.4 for the moderate poverty rate) reveals that in Honduras and the Dominican Republic mean earnings fell most of the time (most of the points are to the left of the zero vertical line). In Honduras, the moderate and extreme poverty rates tended to increase when mean earnings fell and to decrease when mean earnings grew, determining a negative elasticity (slope coefficient of the regression line in the bottom of each figure: −0.54 for moderate poverty and −0.91 for extreme poverty). In the Dominican Republic, the poverty–earnings elasticities were also negative, but in this country the poverty rates continued to decrease when labour earnings fell. This specificity of the Dominican Republic case determined a regression line that is below the one for Honduras. In Bolivia, mean earnings increased most of the time, but in some of the years the poverty rates increased. This determined a negative and small moderate poverty–earnings elasticity (−0.2) and a very small R-squared (0.02). The extreme poverty–earnings elasticity was positive (0.2) with an R-squared of zero. Finally, in Brazil both poverty rates fell most of the time and mean earnings increased. The poverty reductions were larger the larger the increases in mean labour earnings. Thus, the poverty–earnings elasticities are negative (−0.96 for moderate poverty and −1.3 for extreme poverty) and the relationships very tight (R-squareds of 0.73 for extreme poverty and 0.89 for moderate poverty).

To sum up, in the Latin American region and in most of the countries, the year-by-year percentage changes in both poverty measures (2.5 and 4 dollars-a-day poverty rates) were related in the welfare-improving direction with percentage changes in some employment and earnings indicators (unemployment, share of wage/salaried employees, share of self-employed, share of unpaid workers, mean earnings), but the magnitude of the effect and the pattern over time varied substantially from country to country.

# 5.3 Assessing Changes of Labour Earnings across the Earnings Distribution within each Country Using Growth Incidence Curves

In this section we extend the analysis of the within-country growth–employment–poverty nexus, focusing on proportional and dollar changes in labour earnings along the earnings distribution in each country. The reason for having a section completely devoted to the analysis of labour earnings changes is that earnings are the main source of income for Latin American households, and increases in the earnings at the bottom of the income distribution have been shown to be the most important contributor to the observed decline in household per capita income inequality in the region (Azevedo et al. 2013).

(p.128) We base our analysis on the construction of growth incidence curves (GICs) for labour earnings. GICs show the change in an income variable (labour earnings in our case) in percentage terms or in dollars, between two years (initial and final year in our case) by quantiles of the distribution of that income variable (deciles in our case). The purpose of this section to show the changes in labour earnings over all deciles of each country’s income distribution during the 2000s.2

There is a limited literature on GICs of monthly labour earnings in Latin America during the 2000s. One example is Brambilla and Tortarolo (2015), who look at GICs of hourly wages and monthly incomes for six countries in Latin America (Brazil, Chile, Costa Rica, Ecuador, Honduras, and Mexico) for the period 1998–2009. They find that there was growth in labour earnings across the whole income distribution for each country and the percentage growth rate was larger for lower-income percentiles.

Most of the previous literature has looked into GICs of household per capita income. Lustig, Lopez-Calva, and Ortiz-Juarez (2013) calculated GICs for household per capita income during the period 2000–10 for Argentina, Brazil, and Mexico, while Tsounta and Osueke (2014) estimated GICs for the same sixteen Latin American countries and the same period (2000–12) studied in this book. The authors find that the incomes of all deciles increased during this period, and the poorer the decile the greater the percentage increase in household per capita income. If the reader is interested in GICs, the World Bank offers an online tool to calculate the growth incidence curves for each country in Latin America (and for the region as a whole) between any two years from 2000 to 2013.3

We found earlier that mean real earnings grew in most of the countries in our sample. Here, we uncover two additional findings: that the percentage gain tended to be larger for the poorer deciles, while the gain in dollars tended to be larger for the richest deciles.

Figures 5.5 and 5.6 display, for each country, the GICs for employed workers with positive earnings between the initial year and the final year. Figure 5.5 presents the percentage changes of labour earnings, while Figure 5.6 shows the dollar changes. Four main results emerge from these figures. First, as observed in section 5.3, comparing the earliest survey year with the latest, mean real labour earnings (the change in this variable is displayed as the (p.129) (p.130) (p.131) (p.132) (p.133) dashed horizontal line in the figures) increased in eleven countries and decreased in five (with very similar patterns for median labour earnings). Second, for more than half the countries (Brazil, Chile, Colombia, Costa Rica, Ecuador, Panama, Peru, Paraguay, and Venezuela), the GICs based on percentage earnings change are always above zero: that is, all deciles register positive earnings changes. For Argentina and Bolivia, all deciles except the top ones in each case are above zero. For Mexico and Uruguay, most deciles did not experience changes in average incomes, with reductions in the top and bottom deciles in both countries. For the remaining three countries—Dominican Republic, Honduras, El Salvador—all or nearly all of the deciles are below zero. Overall, then, most deciles in most countries experienced an increase in labour earnings. From the 160 deciles under study (ten deciles by sixteen countries), 113 (70 per cent) presented increases in labour earnings from the initial year to the final year. Note that forty-seven (30 per cent) of the country-decile cells did not experience positive earnings growth, of which forty-five belong to the five countries where mean labour earnings fell, and the remaining two to the top decile in Argentina and Bolivia. Labour earnings did not fall for the first nine deciles in any country that experienced increases in mean labour earnings. Third, in more than half of the countries, the changes in labour earnings in percentage terms were largest for the poorer deciles. In most of the remaining countries, the changes in labour earnings benefited the middle deciles the most. In only one case (Costa Rica), the percentage changes in labour earnings were largest for the richest deciles. Finally, in ten out of sixteen countries, the largest dollar increases in labour earnings took place either in the ninth or the tenth decile (i.e. the two richest). In five of the sixteen, there were losses in dollars overall, and the largest losses were in the richest decile. In one country (Argentina), the largest increase in dollars took place in the middle of the distribution. At the low end of the earnings distribution, earnings were essentially unchanged in dollars for the poorest decile in all sixteen countries. What makes these minimal dollar changes for the poor consistent with the higher percentage changes for the poor than for others is that the poor have so few dollars of earnings to begin with.

Figure 5.5 Growth incidence curves of labour earnings by country. Percentage changes for the sample of employed workers with positive labour earnings

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014).

Figure 5.6 Growth incidence curves of labour earnings by country. Dollar changes for the sample of employed workers with positive labour earnings

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014).

# 5.4 In Summary

In this chapter, we analysed the within-country growth–employment–poverty nexus in three parts. First, we studied the response of labour market indicators to economic growth. Second, we investigated the response of poverty to employment and earnings changes. Finally, we presented evidence on changes of labour earnings across the earnings distribution within each country.

(p.134) The first part of the section used year-by-year data to examine whether employment and earnings indicators and poverty and inequality indicators changed in the welfare-improving direction when GDP per capita grows. We found that in the Latin American region as a whole and in most of the countries, the year-by-year percentage changes in some employment and earnings indicators (unemployment, share of wage/salaried employees, share of self-employed, mean earnings) and poverty indicators (2.5 and 4 dollars-a-day poverty rates) improved with increases in GDP per capita, but the magnitude of the effect and the pattern over time varied substantially from country to country.

In the second part of the section, we examined the year-by-year response of the moderate and extreme poverty rates to changes in employment and earnings indicators and to changes in inequality indicators. We found that in the Latin American region and in most of the countries, the year-by-year percentage changes in both poverty measures (2.5 and 4 dollars-a-day poverty rates) were related in the welfare-improving direction to percentage changes in some employment and earnings indicators (unemployment, share of wage/salaried employees, share of self-employed, share of unpaid workers, mean earnings). Again, the poverty rates were differentially responsive to changes in employment and earnings indicators in different countries. The pattern of poverty changes over time was also different across countries.

Finally, we analysed the patterns of earnings changes across different deciles of the earnings distributions in each of the countries. We used anonymous GICs to compare initial earnings (typically 2000) with final earnings (typically 2012) by decile, calculating both percentage changes and dollar changes. We found that 70 per cent of the country-decile cells exhibited positive earnings changes while the other 30 per cent either stagnated or decreased. The largest percentage increases were for the lowest deciles but the highest increases in dollars took place in the richest deciles.

References

Bibliography references:

Alvaredo, F. and L. Gasparini (2015). ‘Recent Trends in Inequality and Poverty in Developing Countries’, in A. Atkinson and F. Bourguignon (eds), Handbook of Income Distribution. Amsterdam: North-Holland, 697–805.

Azevedo, J. P., M. E. Dávalos, C. Díaz-Bonilla, B. Atuesta, and R. A. Castañeda (2013). ‘Fifteen Years of Inequality in Latin America: How Have Labor Markets Helped?’. Policy Research Working Paper 6384. Washington, DC: World Bank.

Bourguignon, F. (2011). ‘Non-Anonymous Growth Incidence Curves, Income Mobility and Social Welfare Dominance’, Journal of Economic Inequality 9: 605–27.

(p.135) Brambilla, I. and D. Tortarolo (2015). ‘Growth in Labor Earnings across the Income Distribution: Latin America during the 2000s’. Technical Note 766. Washington, DC: Inter-American Development Bank.

CEDLAS and World Bank (2014). SEDLAC—Socio-Economic Database for Latin America and the Caribbean. Centro de Estudios Distributivos, Laborales y Sociales, Facultad de Ciencias Económicas, Universidad Nacional de La Plata, and World Bank Poverty Group LCR. Available at <http://sedlac.econo.unlp.edu.ar/eng/index.php>, accessed 2014.

Damill, M. and R. Frenkel (2014). ‘Macroeconomic Policies, Growth, Employment, Poverty, and Inequality in Latin America’, in G. A. Cornia (ed.), Falling Inequality in Latin America: Policy Changes and Lessons. New York: Oxford University Press, 213–33.

Fields, G. S. (2001). Distribution and Development: A New Look at the Developing World. Cambridge, MA: MIT Press and New York: Russell Sage Foundation.

Fosu, A. K. (2011). ‘Growth, Inequality, and Poverty Reduction in Developing Countries: Recent Global Evidence’. WIDER Working Paper 2011/01. Helsinki: UNU-WIDER.

Lustig, N., L. F. Lopez-Calva, and E. Ortiz-Juarez (2013). ‘Deconstructing the Decline in Inequality in Latin America’. Policy Research Working Paper 6552. Washington, DC: World Bank.

Olinto, P., G. Lara Ibarra, and J. Saavedra-Chanduvi (2014). ‘Accelerating Poverty Reduction in a Less Poor World: The Roles of Growth and Inequality’. Policy Research Working Paper 6855. Washington, DC: World Bank.

Ravallion, M. and S. Chen (2003). ‘Measuring Pro-Poor Growth’, Economics Letters 78 (1): 93–9.

Tsounta, E. and A. I. Osueke (2014). ‘What Is Behind Latin America’s Declining Income Inequality?’. IMF Working Paper 124. Washington, DC: International Monetary Fund.

Weller, J. (2014). ‘Aspects of Recent Developments in the Latin American and Caribbean Labour Markets’, CEPAL Review 114: 8–28.

World Bank (2011). ‘The Edge of Uncertainty: Poverty Reduction in Latin America and the Caribbean during the Great Recession and Beyond’. LAC Poverty and Labor Brief. Washington, DC: World Bank.

World Bank (2014). World Development Indicators. Available at <http://data.worldbank.org/data-catalog/world-development-indicators>, accessed April 2014.

## Notes:

(1) These values are similar to the poverty–growth elasticities estimated by Alvaredo and Gasparini (2015) for the period 1999–2010, and World Bank (2011) for the period 2003–10. Estimations from Fosu (2011) for the period 1980–2007 are smaller than our estimates, which is consistent with the increase in the level of inequality that took place in the 1980s but especially in the 1990s in the majority of the countries in the region.

(2) For more on the GIC approach, see Ravallion and Chen (2003), Bourguignon (2011), and the references cited therein. As is most common in the literature, we are presenting here anonymous GICs: that is, changes in earnings for whichever individuals are in the bottom 10 per cent of the earnings distribution, next 10 per cent, and so on. We do this because so-called non-anonymous GICs require panel data, which we do not have.