## 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: 20 January 2019

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

Chapter:
(p.55) 4 Cross-Country Analysis of the Growth–Employment–Poverty Nexus
Source:
Growth, Employment, and Poverty in Latin America
Publisher:
Oxford University Press
DOI:10.1093/oso/9780198801085.003.0004

# Abstract and Keywords

This chapter looks at the cross-country link between growth, employment, and poverty. Findings indicate that: (i) faster growth is associated with larger improvements in labour market indicators, but the relationships are weak; (ii) there is no substantial relationship between the changes in labour market indicators and initial gross domestic product per capita or the initial level of the labour market indicator; (iii) some macroeconomic factors are related to changes in labour market indicators, some in the welfare-improving direction and some in the welfare-reducing direction; (iv) labour market indicators tend to move together, with no indicator improving while another is worsening; (v) there is a strong cross-country association between reductions in poverty and improvements in earnings and employment indicators.

# 4.1 Cross-Country Patterns: Economic Growth Rate and Changes in Labour Market Indicators

Chapter 3 showed that the improvements in labour market indicators during the 2000s were remarkably widespread in the Latin American countries. In this chapter, we analyse whether the improvements in labour market indicators were directly related to the rate of economic growth across countries.

Cross-country analysis of the role of the rate of economic growth in determining improvements in labour market indicators is presented in two complementary parts. First, we analyse how the rate of economic growth is related to the percentage of labour market indicators that moved in the welfare-improving direction during the 2000s in each of the sixteen countries. Second, we perform cross-country analysis separately for each labour market indicator.

Previous literature has covered this topic only partially, presenting the correlation between cross-country changes in labour market indicators and the GDP per capita growth rate for a limited subset of indicators. Pagés, Pierre, and Scarpetta (2009) provide an example of this type of analysis linking the annual growth rate of employment for some Latin American and other developing countries during 1980–2004 to the annual GDP per capita growth rate. For this time period (longer but out of date compared to our analysis), the authors find that Latin American countries outperformed many of the comparator countries, as they had higher rates of employment growth for a given GDP per capita growth rate. However, the authors claim that jobs were created at the same rate that labour supply increased, leading to small increases in the employment rate and even unemployment rate rises for some countries.

Weller and Kaldewei (2013) and Weller (2014) perform a similar exercise, correlating changes in the shares of wage/salaried employees and self-employed workers with the GDP growth rate for the Latin American and the Caribbean (p.56) region overall during the 1995–2012 period. The analysis is performed on a year-by-year basis and indicates a strong positive correlation between changes in the share of wage/salaried employees and the GDP growth rate. For the share of self-employed workers, the authors find a less clear relationship with the GDP growth rate; in several years of the analysed period self-employment behaved countercyclically, while in some other years it behaved procyclically.

Finally, ILO (2013) analyses the cross-country relation between annual growth in both GDP and employment before the international crisis (1997–2007) and after that episode (2008–11). The correlation was positive in both subperiods, but stronger before the crisis. This last result is explained by the fact that in Central American and the Caribbean countries, whose economies are strongly connected to the North American market, the recovery was slower than in South American countries. We turn now to our analysis.

## 4.1.1 Analysis of the Percentage of Labour Market Indicators that Changed in the Welfare-Improving Direction (Z)

What is the relationship between improvements in labour market indicators and the rate of economic growth? Figure 4.1 presents a scatterplot. We see in the figure that over the 2000s, GDP per capita increased in every country and that more than 60 per cent of the labour market indicators increased in every country except for Honduras, which suffered a generalized worsening of labour market conditions.

Figure 4.1 Cross-country relationship between the percentage of labour market indicators moving in the welfare-improving direction and growth rate of GDP per capita during the 2000s

Note: This figure displays the percentage of labour market indicators that changed in the welfare-improving direction according to Table 3.2 and the annualized growth rate during the period under study according to Table 3.1. 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).

Across these countries, does a higher economic growth rate result in a higher percentage of labour market indicators improving? Let Zi be the percentage of labour market indicators with a statistically significant improvement in country i, and %Δ‎GDPpci be the annualized percentage change of GDP per capita in country i. To quantify the association between the two variables in the figure, we estimate the following regression:

$Display mathematics$
(2)

We observe a positive but weak relationship (R-squared of 0.112 and statistically insignificant) between the percentage of labour market indicators that improved during the 2000s and the rate of economic growth. Upon removing Honduras, which is the only country in our sample with a generalized worsening in labour market indicators over the period, the R-squared increases slightly to 0.120, but the slope coefficient is smaller and still not statistically significant. The reason for the lack of relationship between the percentage of improving indicators and the rate of economic growth is the limited variation in the evolution of labour market indicators, since for most countries in our sample and regardless of their annualized rates of economic growth we (p.57) observe that 75 per cent or more of these indicators improved during the period under study.

## 4.1.2 Analysis of the Labour Market Indicators One by One (Yk)

The weak relationship between the percentage of labour market indicators that improved in each country (Zi) and the rate of economic growth (%Δ‎GDPpci) may be due to the type of aggregation implicit in our index of the percentage of labour market indicators that improved over the period. Rather than constructing alternative indices, which would also be arbitrary in terms of the indicators included, the weight assigned to each one, etc., we can instead extend this analysis beyond our aggregate measure of improvement of labour markets and study the relationship between economic growth and each of the underlying indicators one by one.

Our results indicate that faster growth is associated with larger improvements in labour market indicators, but the goodness of fit of most of the relationships analysed is generally low. This conclusion is based on Figure 4.2, which displays the scatterplots for each country’s annualized change in the (p.58) (p.59) (p.60) k’th labour market indicator and its rate of economic growth (one plot for each labour market indicator). Let GDPpci be GDP per capita in country i, Yik be the labour market indicator k for country i, Δ‎ be the annualized change in percentage points, and %Δ‎ be the annualized percentage change. We quantify the underlying relationship between the variables in the plots by estimating one of the following regressions, depending on the units of the indicators:

$Display mathematics$
(3)

Figure 4.2 Cross-country relationship between the annualized changes in labour market indicators and annualized growth rate of GDP per capita during the 2000s

Note: The vertical axes display the annualized change in each labour market indicator. Δ‎ denotes changes in percentage points and % Δ‎ denotes percentage changes. The line in each figure represents the linear regression specified at the bottom. Robust standard error of the slope coefficient between parentheses. R-squared of the regression indicated along the title of each figure.

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

We consider a relationship to be tight if the R-squared is above the arbitrary threshold of 0.15. The R-squared was chosen instead of other commonly used statistics as the slope or an F test of statistical significance, since we wanted to capture how much of the variation in Yk can be explained by changes in GDP per capita.

Among the employment and earnings indicators, only three exhibited a relatively tight relationship between their changes during the 2000s and the rate of economic growth. These indicators were the share of registered workers, the share of high-earnings occupations, and the share of low-earnings occupations. There thus seems to be a significant relationship between the rate of economic growth and different aspects of the occupational mix. More specifically, countries that grew faster experienced larger declines in the share of low-earnings occupations, and higher increases in the share of highly paid occupations in total employment (R-squareds of 0.15 and 0.33, respectively). Moreover, the share of workers registered with social security tended to increase more in countries with stronger economic growth, and this is the tightest of the relationships we computed (R-squared of 0.44). The increase in the share of registered workers is a manifestation of the procyclicality of registered employment, which has been extensively documented and discussed before for the region as a whole, and for most countries in the region over time (Gasparini and Tornarolli 2009).

For the remaining employment and earnings indicators, as well as for the poverty and inequality indicators, we find no statistically significant relationship or only a weak relationship between the annualized change in the labour market indicator and the rate of economic growth (R-squared lower than 0.15). For instance, there is a weak positive relationship between growth and the change in the share of wage/salaried employees (R-squared of 0.09). There are also weak negative relationships between the rate of economic growth and the changes in the unemployment rate, in the moderate poverty rate, and in the shares of unpaid workers and of low-earnings sectors.

These mostly weak relationships between the rate of economic growth and the substantial majority of indicators of labour market performance seem to be driven by the experiences of the countries which grew at moderate rates by Latin American standards. The two fastest-growing economies (Panama and Peru) (p.61) exhibited widespread and large improvements in their labour market indicators, and the two slowest-growing economies (Mexico and El Salvador) showed among the smallest improvements (and even some deteriorations) in labour market indicators over the 2000s. However, these changes and deteriorations were not extreme, which accounts partially for the modest slopes of the aggregate relationships across all sixteen countries. Moreover, the other twelve countries in the middle of the growth scale exhibited a large degree of variability in the magnitudes of the changes in labour market indicators despite having similar economic growth rates. For instance, Bolivia, Brazil, and Honduras had nearly the same economic growth, and while in Bolivia and Brazil all labour market indicators improved and in some cases the improvements were larger than for Panama or Peru (the two fastest-growing economies), Honduras had by far the worst performance among the sixteen countries (Table 3.2 in Chapter 3). Some other countries exhibited larger economic growth rates when compared to Bolivia and Brazil, but smaller improvement in labour market indicators. That was the case of the Dominican Republic.

# 4.2 Cross-Country Patterns: Beyond Economic Growth

The analysis in sections 4.1.1 and 4.1.2 revealed that labour market conditions improved substantially in all but one of the sixteen Latin American countries covered in this study. These improvements, though widespread, occurred in countries with high and low rates of economic growth. This lack of a systematic cross-country relationship between economic growth and improvements in the labour market as measured either by the aggregate index Z or by the individual labour market indicators Yk motivates the analysis in this chapter, in which we attempt to move beyond aggregated indicators such as economic growth and delve into more detailed macroeconomic variables.

The analysis of the role of macroeconomic variables other than the rate of economic growth in determining changes in labour market indicators proceeds as follows. To determine whether the richer Latin American countries differed from the poorer ones in terms of their labour market trajectories, we first study the relationship between countries’ changes in labour market conditions and their initial level of GDP. Next, we study the changes in each labour market indicator as a function of the country’s initial level of this indicator, to uncover any potential convergence effect in these indicators. Then we analyse a number of other macroeconomic variables which might be significant correlates of changes in labour market conditions. These variables are changes in: agriculture as a percentage of GDP; industry as a percentage of GDP; services as a percentage of GDP; domestic consumption expenditure as a percentage of GDP; exports as a percentage of GDP; terms of trade; foreign direct investment (p.62) as a percentage of GDP; revenues from natural resources as a percentage of GDP; expenditure in education and health as a percentage of GDP; public expenditure in social security as a percentage of GDP; and the stock of public debt as a percentage of GDP. Finally, we look to see whether the changes in certain labour market indicators are linked systematically to the changes in others: for example, whether countries with more rapidly rising real wages are those with more rapidly rising unemployment or whether real earnings and employment move together.

Some of these topics have been studied in the previous literature although in a more limited way—typically, for a smaller group of labour market indicators, for a smaller group of countries, or for a different time period. For instance, World Bank (2012) highlights the role of the increasing terms of trade for the net commodity-exporting countries in the region to explain the increase in the relative demand for low-skilled workers and then the reduction in labour income inequality (measured by the returns to secondary and higher education) between 2000 and 2009–10. Similarly, World Bank (2015) relates real wage increases, reductions in household income inequality (measured by the Gini coefficient), and reductions in the poverty rate (measured by the 4 dollars-a-day international line) in the Latin American region during 2003–12 to improving terms of trade, because countries experiencing a commodity boom did much better in these labour market indicators than non-commodity-boom countries.

Tsounta and Osueke (2014) analyse the determinants of falling income inequality in Latin America from 1990 to 2012. The authors present evidence of convergence in the Gini coefficient of household per capita income across Latin American countries (eighteen countries) over the period. They also find a negative cross-country correlation between changes in tax revenues as a share of GDP and changes in household income inequality (where changes correspond to the difference between the years 2000 and 2012), and a negative cross-country correlation between changes in government spending on education as a share of GDP and changes in income inequality.

Finally, Damill and Frenkel (2014) estimate the effect of annual changes in the real exchange rate on annual changes in the unemployment rate for a panel of eighteen Latin American countries from 1990 to 2010. They find that a more depreciated real exchange rate tends to reduce unemployment. In a similar exercise, they relate the poverty rate to unemployment and the inflation rate and find that increases in the unemployment rate and in the inflation rate tend to increase the poverty rate.

## 4.2.1 Initial GDP per Capita

An ongoing debate in the modern theory of economic growth is whether there is convergence or divergence in growth rates: that is, whether poorer countries (p.63) tend to grow at higher rates than richer ones (thus tending to converge in terms of GDP) or not. We start our analysis with the related question of whether the improvement in labour market indicators over the period under study was correlated with each country’s initial GDP per capita. This relationship could be either positive or negative: poorer economies could have more room to improve in the labour market, so that these countries might exhibit larger improvements in related indicators, or alternatively initially richer economies may have better conditions to channel the economic growth during the period under study in the direction of improved conditions in the labour market.

Examining these competing views empirically, we find that there is no important cross-country relationship between initial GDP per capita and aggregate changes in labour market conditions. Figure 4.3 plots initial GDP per capita in 2005 PPP dollars and the percentage of improving labour market indicators for each country. Let $GDPit0$ be the GDP per capita at 2005 PPP in the first period under study for country i, and Zi be the percentage of labour market indicators that experienced an improvement in the period under study. To quantify the cross-country relationship, we estimate the following regression:

$Display mathematics$
(4)

Figure 4.3 Cross-country relationship between the percentage of labour market indicators moving in the welfare-improving direction during the 2000s and initial GDP per capita

Note: This figure displays the percentage of labour market indicators that changed in the welfare-improving direction according to Table 3.2 and GDP per capita of the initial year at PPP 2005. 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 relationship is positive, indicating that initially richer countries enjoyed larger improvements in labour market indicators measured by Z, but weak (R-squared of 0.11). However, even this low association is entirely driven by Honduras, which is a clear outlier: without Honduras, the R-squared and slope of the fitted line are virtually equal to zero.

Our finding of a lack of relationship between initial GDP per capita level and labour market conditions across countries means that there were substantial improvements in labour markets both in initially poorer and in initially richer countries, and that countries with similar initial levels of GDP per capita exhibited very different patterns in the number of labour market indicators that improved over the period under study. For instance, Peru and the Dominican Republic had almost the same level of initial GDP per capita, but the Peruvian experience was markedly more successful: all sixteen labour market indicators improved in Peru, but only ten improved in the Dominican Republic.

While there does not seem to be a relationship between initial GDP per capita and the percentage of indicators that improved, there could still be a relationship between the magnitude of changes in some of the individual labour market indicators and the initial level of GDP per capita. In Figure 4.4, we present this relationship for each of the sixteen labour market (p.64) indicators. Let $GDPit0$ be the GDP per capita at 2005 PPP in the initial year under study for country i, Yik be the labour market indicator k for country i, Δ‎ be the annualized change in percentage points, and let %Δ‎ be the annualized percentage change. We quantify these relationships estimating regressions of the form:

$Display mathematics$
(5)

Figure 4.4 Cross-country relationship between the annualized changes in labour market indicators during the 2000s and initial GDP per capita

Note: The vertical axes display the annualized change in each labour market indicator. Δ‎ denotes changes in percentage points; % Δ‎ denotes percentage changes. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses. R-squared of the regression indicated along the title of each figure.

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

Using equation (5), we also fail to find a relationship between initial GDP per capita and changes in individual labour market indicators. The results displayed in Figure 4.4 indicate that we can reject the hypothesis of an association between the initial level of GDP per capita and the changes in each of the labour market indicators. All the R-squareds are lower than 0.06, and the slopes are practically (p.65) (p.66) (p.67) equal to 0. In brief, initial GDP per capita does not make an important difference to the rate of change of any of the labour market indicators.

## 4.2.2 Convergence/Divergence Patterns in Labour Market Indicators

In this section, we study how, across countries, the change in each of the sixteen labour market indicators is related to the initial level of that indicator. In order to do that, let $Yikt0$ be the value of the labour market indicator k in the initial year under study for country i, Yik be the labour market indicator k for country i, and let Δ‎ be the annualized change in percentage points, and %Δ‎ be the annualized percentage change. We estimate regressions of the form:

$Display mathematics$
(6)

We define convergence and divergence as follows: given the initial value of the k’th labour market indicator, a convergent (divergent) relationship is one where the countries with worse (better) initial values tend to have larger subsequent improvements. Convergent patterns would reflect some sort of decreasing marginal returns to growth or to improvements in a given indicator, i.e. it is harder to achieve large reductions when the labour market indicator is already high (in a welfare-increasing direction). Alternatively, divergent patterns would signal the presence of ‘traps’ or absorbent states in that once the labour market indicator is at a low level, it is hard for the country to bring it up.

Figure 4.5 presents the relationship between the changes in each labour market indicator and its initial value. There seems to be convergence for about a third (five out of sixteen) of our selected indicators, namely: the unemployment rate, the share of unpaid family workers, the poverty and extreme poverty rates, and the inequality of household per capita income. The relationships are especially tight for the unemployment rate, and for the share of unpaid family workers (R-squareds of about 0.73 and about 0.5, respectively). That is, countries with higher initial unemployment rates and higher shares of unpaid family workers exhibited much larger reductions in these indicators than other countries; these countries are not stuck with high unemployment rates or high shares of workers in unpaid family jobs. The results in Figure 4.5 also reveal some weak convergent patterns: for example, the share of low-earnings occupations and the share of workers with low levels of education converged, but not as much as the unemployment rate and the share of unpaid family workers did (R-squareds of 0.06 and 0.09). For the other indicators, no discernible convergence/divergence patterns appeared.

Figure 4.5 Cross-country relationship between the annualized changes in labour market indicators during the 2000s and the initial value of labour market indicators

Note: The vertical axes display the annualized change in each labour market indicator. Δ‎ denotes changes in percentage points; % Δ‎ denotes percentage changes. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses. R-squared of the regression indicated along the title of each figure.

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

(p.68) (p.69)

## (p.70) 4.2.3 Other Potential Macroeconomic Correlates of Changing Labour Market Indicators

In this section, we turn to other macroeconomic variables besides the rate of economic growth and the initial level of national income, and study which, if any, are significantly correlated with improvements in the labour market. The macroeconomic variables analysed here fall into two categories. Most have to do with the composition of GDP. These variables, expressed as changing percentages of GDP, include the share of agriculture, the share of industry, the share of service, the share of domestic consumption expenditure, the share of expenditure in education and health, the share of expenditure in social security, the share of exports, the share of foreign direct investments, the share of revenues of natural resources, and the share of the stock of public debt. We also consider the changes in the country’s terms of trade; this variable is not a share of GDP (Appendix 3 provides the macroeconomic variables time series for each country). Let Zi be the share of improving labour market indicators for country i, and Xij be the macroeconomic variable j in country i. To quantify the association between the two variables we estimate the following regression:

$Display mathematics$
(7)

These bivariate tests yield several strong relationships. Most notably, the share of labour market indicators that improved was larger in countries with larger increases in exports as a percentage of GDP, larger reductions in domestic consumption expenditure as a percentage of GDP, and larger falls in the stock of public debt as a percentage of GDP (when excluding Honduras, an outlier as discussed in section 4.1.1) (Figure 4.6). There appear to be some weak positive relationships also between the share of labour market indicators that improved and the change in terms of trade and in revenues from natural resources as a percentage of GDP.

Figure 4.6 Cross-country relationship between the percentage of improving labour market indicators and the annualized changes in macroeconomic variables during the 2000s

Note: The vertical axes display the annualized change in each labour market indicator. Δ‎ denotes changes in percentage points; % Δ‎ denotes percentage changes. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses. R-squared of the regression indicated along the title of each figure.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014), World Development Indicators (World Bank 2014), and CEPALSTAT (UN-ECLAC 2015).

Besides these relationships between changes in these macroeconomic aggregates and the share of labour market indicators that improved over the period under study, we can also study the relationship between these macroeconomic variables and the sixteen individual labour market indicators. To gauge their importance, we perform a series of regressions between the change in labour market indicator and the changes in the macroeconomic variables. Let Yik be the labour market indicators k for country i, and Xij be the macroeconomic variable j in country i. To quantify the association between the two variables we estimate the following regression:

$Display mathematics$
(8)

and (p.71) (p.72) (p.73)

$Display mathematics$
(9)

With sixteen indicators and eleven macroeconomic variables, we have 176 regressions to estimate. The results are summarized in Table 4.1. In Table 4.1, Positive indicates that the R-squared is above 0.15 and that an increase in the macroeconomic variable is associated with an improvement in the labour indicator; and similarly Negative indicates that the relationship is also significant, but an increase in the macro variable is related to a deterioration in the indicator;NR (No relationship) indicates a regression with an R-squared of less than 0.15. In Table 4.2, we present the R-squared for each regression, and Appendix 2 presents the figures corresponding to each of these individual regressions.

Table 4.1 Direction of the cross-country relationship between annualized changes in macroeconomic variables and annualized changes in labour market indicators and GDP per capita growth during the 2000s

Indicator

Share of agriculture in GDP

Share of industry in GDP

Share of services in GDP

Domestic expenditure (% of GDP)

Public expend. in education and health (% of GDP)

Public expend. in social security (% of GDP)

Exports (% of GDP)

Foreign direct investment (% of GDP)

Revenues from natural resources (% of GDP)

Stock of public debt (% of GDP)

Unemployment

Decrease in the unemployment rate

NR

Positive

Negative

Negative

NR

NR

NR

Negative

NR

NR

NR

Occupations

Decrease in the share of low-earnings occupations

NR

NR

NR

Negative

NR

NR

Positive

NR

NR

Positive

Negative

Increase in the share of high-earnings occupations

NR

NR

NR

NR

NR

NR

NR

NR

NR

NR

Negative

Occupational position

Increase in the share of wage/salaried employees

NR

NR

NR

NR

NR

NR

Positive

NR

NR

NR

Negative

Decrease in the share of self-employment

NR

NR

NR

NR

NR

NR

Positive

NR

NR

NR

Negative

Decrease in the share of unpaid family workers

NR

NR

NR

Negative

NR

NR

Positive

Positive

NR

Positive

Negative

Economic sector

Decrease in the share of workers in low-earnings sectors

NR

NR

NR

NR

NR

NR

NR

NR

NR

NR

Negative

Increase in the share of workers in high-earnings sectors

NR

NR

NR

NR

NR

Positive

NR

NR

NR

NR

Negative

Education

Decrease in the share of low-educated workers

NR

NR

NR

NR

NR

Positive

NR

NR

Negative

NR

Negative

Increase in the share of high-educated workers

NR

NR

NR

NR

NR

Positive

NR

NR

NR

NR

Negative

Workers registered with SS

Increase in the share of workers registered with SS

NR

NR

NR

NR

NR

NR

NR

NR

Positive

NR

NR

(p.75) Earnings

Increase in mean labour earnings

NR

Positive

Negative

Negative

NR

NR

Positive

Positive

NR

Positive

Negative

Poverty

Decrease in 4 dollars-a-day poverty

NR

Positive

Negative

Negative

NR

NR

Positive

Positive

NR

Positive

Negative

Decrease in 2.5 dollars-a-day poverty

NR

Positive

Negative

Negative

NR

NR

Positive

Positive

NR

Positive

Negative

Inequality

Decrease in GINI of household per capita income

NR

Positive

Negative

Negative

NR

NR

Positive

Positive

Negative

Positive

NR

Decrease in GINI of labour earnings

NR

Positive

Negative

Negative

NR

NR

Positive

Positive

Negative

Positive

NR

Economic growth

Increase in GDPpc at PPP 2005

NR

NR

NR

Negative

NR

Negative

NR

NR

Positive

NR

Negative

Number of relationships with labour market indicators

0

6

6

8

0

3

9

7

4

7

12

Percentage of total indicators

0.0

37.5

37.5

50.0

0.0

18.8

56.3

43.8

25.0

43.8

75.0

Note: Positive denotes an increase of the macroeconomic variable is associated with a change in the labour market indicator in the welfare-improving direction. Negative denotes an increase of the macroeconomic variable is associated with a change in the labour market indicator in the welfare-worsening direction. NR denotes no relationship, that is the R-squared of a linear regression is smaller than 0.15.

Grey shading implies that the R-squared is higher than 0.15 and the relationship is Positive. Bold typeface with no shading implies that the R-squared is larger than 0.15 and the relationship is Negative.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014), World Development Indicators (World Bank 2014), and CEPALSTAT (UN-ECLAC 2015).

Table 4.2 Tightness of the cross-country relationship (R-squared) between annualized changes in macroeconomic variables and annualized changes in labour market indicators and GDP per capita growth during the 2000s

Indicator

Share of agriculture in GDP

Share of industry in GDP

Share of services in GDP

Domestic expenditure (% of GDP)

Public expend. in education and health (% of GDP)

Public expend. in social security (% of GDP)

Exports (% of GDP)

Foreign direct investment (% of GDP)

Revenues from natural resources (% of GDP)

Stock of public debt (% of GDP)

Unemployment

Decrease in the unemployment rate

0.00

0.21

0.32

0.20

0.02

0.02

0.11

0.15

0.01

0.03

0.02

Occupations

Decrease in the share of low-earnings occupations

0.00

0.06

0.09

0.19

0.01

0.01

0.18

0.10

0.00

0.01

0.55

Increase in the share of high-earnings occupations

0.09

0.02

0.01

0.11

0.08

0.03

0.06

0.03

0.01

0.04

0.27

Occupational position

Increase in the share of wage/salaried employees

0.08

0.01

0.00

0.12

0.00

0.01

0.23

0.02

0.00

0.04

0.28

Decrease in the share of self-employment

0.11

0.08

0.02

0.06

0.00

0.01

0.24

0.04

0.00

0.02

0.19

Decrease in the share of unpaid family workers

0.00

0.07

0.10

0.27

0.00

0.01

0.20

0.18

0.00

0.29

0.28

Economic sector

Decrease in the share of workers in low-earnings sectors

0.01

0.00

0.00

0.11

0.02

0.04

0.03

0.02

0.00

0.03

0.15

Increase in the share of workers in high-earnings sectors

0.05

0.00

0.00

0.01

0.00

0.36

0.02

0.12

0.11

0.01

0.25

Education

Decrease in the share of low-educated workers

0.05

0.02

0.00

0.00

0.00

0.37

0.01

0.02

0.20

0.01

0.18

Increase in the share of high-educated workers

0.07

0.00

0.04

0.06

0.02

0.17

0.07

0.14

0.04

0.00

0.22

(p.77) Workers registered with SS

Increase in the share of workers registered with SS

0.12

0.05

0.00

0.06

0.00

0.06

0.00

0.00

0.16

0.04

0.06

Earnings

Increase in mean labour earnings

0.06

0.30

0.25

0.29

0.00

0.06

0.21

0.40

0.09

0.24

0.53

Poverty

Decrease in 4 dollars-a-day poverty

0.07

0.34

0.26

0.39

0.09

0.02

0.33

0.41

0.12

0.38

0.31

Decrease in 2.5 dollars-a-day poverty

0.05

0.36

0.31

0.41

0.05

0.02

0.31

0.46

0.07

0.38

0.38

Inequality

Decrease in GINI of household per capita income

0.00

0.16

0.23

0.27

0.09

0.01

0.23

0.16

0.19

0.24

0.03

Decrease in GINI of labour earnings

0.00

0.29

0.39

0.28

0.00

0.08

0.15

0.36

0.30

0.19

0.09

Economic growth

Increase in GDPpc at PPP 2005

0.08

0.02

0.00

0.25

0.08

0.15

0.04

0.00

0.42

0.06

0.21

Note: Grey shading implies that the R-squared is higher than 0.15 and the relationship is Positive according to Table 4.1. Bold typeface with no shading implies that the R-squared is larger than 0.15 and the relationship is Negative according to Table 4.1.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014), World Development Indicators (World Bank 2014), and CEPALSTAT (UN-ECLAC 2015).

The results are mixed, with some robust positive and negative relationships and several instances of no clear pattern of association. The change in the share of industry in GDP has a positive association with a number of indicators—an increase in labour earnings, a decline in the unemployment rate, and better distributional indicators (i.e. lower levels of poverty, extreme poverty, and inequality of household per capita income and labour earnings)—and no statistically discernible association with other labour market indicators. The change in exports as a percentage of GDP is positively associated with an increase in mean earnings and in improvements in the labour mix (decline in the share of low-earnings occupations, increase in the share of wage/salaried employees, fall in the share of self-employment and unpaid family workers), as well as improved distributional indicators. The change in terms of trade and the change in revenues from natural resources as a percentage of GDP have a similar pattern of relationships with labour market indicators as the change in exports.

Other macroeconomic variables appear to have a negative association with some of our selected labour market indicators (i.e. increases in the macroeconomic variables seem related to worsenings in these indicators). This is the case for the change in the share of services in GDP, the change in domestic expenditure as a percentage of GDP, and the change in the stock of public debt as a percentage of GDP. Increases in the share of services in GDP are associated with smaller increases/declines in mean labour earnings, smaller declines/increases in the unemployment rate, and a worsening in distributional indicators (i.e. higher levels of poverty and inequality). Similarly, increases in domestic expenditure as a percentage of GDP are associated with smaller increases/declines in mean labour earnings, smaller declines/increases in unemployment, and a worsening in distributive indicators. Increases in the stock of public debt are associated with a general worsening in labour market outcomes (with the exception of the unemployment rate, the share of registered workers, and levels of inequality).

(p.74) (p.76) (p.78) We find little or no consistent pattern of association of labour market indicators with the following macroeconomic variables: change in the share of agriculture in GDP, change in public expenditure on education and health as a percentage of GDP, change in public expenditure on social security as a percentage of GDP, and change in foreign direct investment as a percentage of GDP.

Looking at the experiences of countries with widespread labour market improvements in Latin America, we find that there is no unique configuration of macroeconomic factors associated with the number of welfare-improving changes in labour market indicators. On the one hand, there is a group of countries which benefited from better external conditions: higher terms of trade, increased exports, and related to that, increasing revenues from natural resources, and increasing share of industry in GDP. That was the case, for example, for Bolivia and Peru. For these countries, increases in exports seem to have resulted in a shift to the right of the labour demand for high-earnings occupations and wage/salaried employees (improving the mix of jobs), raising labour earnings, and reducing poverty. Some of these countries took advantage of the favourable external conditions, and translated them into higher levels of investment (proxied by the reduction in consumption’s share of GDP) and to an improved fiscal balance (as indicated by the fall in the stock of public debt as a percentage of GDP). On the other hand, there is a group of countries where the better external conditions were not present, but the labour market conditions also improved. That was the case of Panama and Costa Rica, which exhibited some of the largest increases in the share of services in GDP and some of the largest reductions in terms of trade and in the stock of public debt as a percentage of GDP. These countries were successful in increasing the labour demand in the service sector, the driving force of these economies.

Our next step is to add the GDP per capita growth rate as a second explanatory variable in the previous models. Our objective is to test the robustness of some of the results obtained in this section: (1) faster growth is associated with larger improvements in labour market indicators, but the relationship is weak; and (2) some macroeconomic variables were associated with changes in labour market conditions always in the welfare-improving direction and some others always in the welfare-reducing direction. The reason for adding the GDP per capita growth rate as an additional regressor to the bivariate models where the explanatory factor is a macroeconomic variable is that the two variables (GDP per capita growth rate and macroeconomic variable) could be correlated, e.g. countries with larger increases in terms of trade enjoy larger increases in GDP per capita. Including the two of them as regressors allows us to separate, at least partially, the effect of the GDP per capita growth rate on the change in labour market indicators from the effect of macroeconomic factors.

(p.79) We perform a series of regressions for the change in labour market indicators on the changes in the macroeconomic variables and the change in GDP per capita. Let Yik be the labour market indicators k for country i, Xij be the macroeconomic variable j in country i, and GDPpci be GDP per capita in country i. We estimate the following regression for two employment and earnings indicators (the hange in the unemployment rate and the change in mean labour income), and two poverty indicators (changes in the 2.5 and 4 dollars-a-day poverty rates):

$Display mathematics$
(10)

and

Table 4.3 Cross-country relationship between annualized changes in labour market indicators and annualized changes in macroeconomic variables and in GDP per capita during the 2000s

Part A

∆ Unemployment rate

∆% Mean labour earnings

(1)

(2)

(3)

(1)

(2)

(3)

GDPpc growth rate coefficient

R-squared

∆X variable coeff.

R-squared

∆X variable coeff.

GDPpc growth rate coefficient

R-squared

GDP pc growth rate coefficient

R-squared

∆X variable coeff.

R-squared

∆X variable coeff.

GDP pc growth rate coefficient

R-squared

∆% GDP per capita

−0.075

0.090

0.228

0.032

(0.05)

(0.254)

∆ Exports (% of GDP)

−0.143

0.105

−0.121

−0.061

0.162

1.033

0.209

0.992

0.114

0.217

(0.108)

(0.108)

(0.058)

(0.505)**

(0.546)*

(0.259)

−0.039

0.151

−0.040

−0.078

0.248

0.000

0.000

0.322

0.398

0.325

0.254

0.437

(0.022)

(0.022)

(0.063)

(0.109)**

(0.107)**

(0.19)

∆ Share of services in GDP

0.467

0.316

0.466

−0.075

0.405

0.000

−2.124

0.249

−2.123

0.228

0.281

(0.197)*

(0.187)*

(0.068)

(1.019)*

(1.056)*

(0.217)

∆ Share of industry in GDP

−0.313

0.205

−0.287

−0.058

0.259

0.000

1.938

0.300

1.885

0.122

0.309

(0.166)

(0.176)

(0.059)

(0.908)*

(0.954)*

(0.215)

∆ Share of agriculture in GDP

−0.070

0.003

−0.208

−0.086

0.110

0.000

−1.665

0.055

−1.417

0.154

0.068

(0.316)

(0.318)

(0.057)

(1.374)

(1.39)

(0.27)

∆ Public expend. in education and health (% of GDP)

−0.313

0.024

−0.522

−0.093

0.150

0.000

0.389

0.001

0.983

0.262

0.040

(0.651)

(0.625)

(0.057)

(2.341)

(2.68)

(0.289)

∆ Public expend. in social security (% of GDP)

−0.223

0.018

−0.493

−0.104

0.165

0.000

2.013

0.056

3.082

0.411

0.144

(0.298)

(0.395)

(0.045)*

(1.394)

(1.533)**

(0.274)

∆ Domestic expenditure (% of GDP)

0.204

0.198

0.180

−0.025

0.205

0.000

−1.255

0.286

−1.398

−0.155

0.297

(0.12)

(0.148)

(0.073)

(0.547)*

(0.602)*

(0.26)

∆ Foreign direct investment (% of GDP)

0.108

0.008

0.599

−0.152

0.223

0.000

−1.945

0.092

−4.596

0.820

0.332

(0.327)

(0.517)

(0.076)*

(1.312)

(1.216)**

(0.292)**

(p.81) ∆ Revenues from natural resources (% of GDP)

−0.138

0.029

−0.086

−0.068

0.100

0.000

2.025

0.238

1.959

0.087

0.242

(0.215)

(0.243)

(0.057)

(0.859)*

(0.926)*

(0.268)

∆ Stock of public debt (% of GDP)

0.033

0.022

0.003

−0.073

0.090

0.000

−0.813

0.527

−0.914

−0.252

0.558

(0.055)

(0.069)

(0.064)

(0.177)**

(0.21)**

(0.235)

(p.82) Part B

∆ 2.5 dollars-a-day poverty

∆ 4 dollars-a-day poverty

(1)

(2)

(3)

(1)

(2)

(3)

GDP pc growth rate coefficient

R-squared

∆X variable coeff.

R-squared

∆X variable coeff.

GDP pc growth rate coefficient

R-squared

GDP pc growth rate coefficient

R-squared

∆X variable coeff.

R-squared

∆X variable coeff.

GDP pc growth rate coefficient

R-squared

∆% GDP per capita

−0.138

0.056

−0.249

0.105

(0.097)

(0.138)

∆ Exports (% of GDP)

−0.589

0.325

−0.563

−0.073

0.340

−0.753

0.308

−0.692

−0.170

0.355

(0.191)**

(0.193)**

(0.081)

(0.26)**

(0.266)***

(0.123)

−0.149

0.410

−0.151

−0.150

0.475

−0.208

0.462

−0.212

−0.266

0.582

(0.051)**

(0.045)**

(0.095)

(0.065)**

(0.053)**

(0.105)*

∆ Share of services in GDP

0.991

0.260

0.991

−0.138

0.315

1.409

0.305

1.408

−0.249

0.410

(0.423)*

(0.427)*

(0.095)

(0.527)**

(0.51)**

(0.116)*

∆ Share of industry in GDP

−0.947

0.343

−0.910

−0.087

0.365

−1.276

0.361

−1.197

−0.182

0.416

(0.391)*

(0.413)*

(0.088)

(0.499)*

(0.524)*

(0.113)

∆ Share of agriculture in GDP

0.826

0.065

0.659

−0.103

0.093

0.936

0.048

0.583

−0.219

0.123

(0.601)

(0.707)

(0.122)

(0.739)

(0.819)

(0.163)

∆ Public expend. in education and health (% of GDP)

1.400

0.086

1.178

−0.098

0.112

1.380

0.049

0.884

−0.219

0.124

(1.063)

(1.202)

(0.114)

(1.354)

(1.523)

(0.158)

∆ Public expend. in social security (% of GDP)

−0.579

0.022

−1.109

−0.204

0.125

−0.748

0.022

−1.650

−0.347

0.194

(0.783)

(0.877)

(0.126)

(1.099)

(1.096)

(0.162)*

∆ Domestic expenditure (% of GDP)

0.668

0.388

0.724

0.060

0.396

0.899

0.407

0.896

−0.004

0.408

(0.226)**

(0.23)**

(0.081)

(0.3)**

(0.314)**

(0.109)

∆ Foreign direct investment (% of GDP)

1.003

0.117

2.481

−0.457

0.475

1.045

0.074

3.169

−0.657

0.502

(0.767)

(0.639)**

(0.141)**

(1.019)

(0.708)**

(0.166)**

∆ Revenues from natural resources (% of GDP)

−1.176

0.384

−1.133

−0.056

0.393

−1.543

0.383

−1.432

−0.146

0.418

(0.36)**

(0.401)**

(0.115)

(0.458)**

(0.484)**

(0.137)

∆ Stock of public debt (% of GDP)

0.287

0.314

0.293

0.016

0.314

0.413

0.378

0.397

−0.041

0.381

(0.089)**

(0.136)*

(0.141)

(0.105)**

(0.162)*

(0.179)

Note: ** significant at 1% level,

* significant at 5% level.

Source: Authors’ calculations based on SEDLAC (CEDLAS and World Bank 2014), World Development Indicators (World Bank 2014), and CEPALSTAT (UN-ECLAC 2015).

$Display mathematics$
(11)

Our results are presented in Table 4.3. Model 1 uses GDP per capita growth rate as the only regressor and replicates the results obtained in section 4.1. Model 2 uses the changes in macroeconomic variables as regressors (one at a time) and replicates the results obtained previously in this section. Finally, Model 3 includes both the GDP per capita growth rate and the changes in macroeconomic variables as explanatory factors. In general, the magnitudes of the coefficients and standard errors of the estimations in the multivariate model (Model 3) are similar to those obtained in the bivariate models (Models 1 and 2). The details of these findings are as follows. First, from the forty-four regressions (eleven macroeconomic variables × four labour market indicators), in only four cases did the macroeconomic variables move from being not statistically significant in the bivariate model (Model 2) to being significant at the 5 per cent level in the multivariate model (Model 3). In all four cases, the sign of the relationship remained the same when moving from the bivariate model to the model that also includes the change in GDP per capita as a control variable. Second, in no case did a macroeconomic variable that was significant in statistical terms in the bivariate model (Model 2) turn to insignificance in the multivariate model (Model 3). Third, out of the forty-four regressions, in only six cases was the GDP per capita growth rate a significant factor explaining changes in labour market indicators across countries in the multivariate model (Model 3) when it was not in the bivariate model (Model 1), and the sign of the relationship was always the same as the one obtained in the bivariate regression. In conclusion, the weakness of the relationship between changes in labour market indicators and the GDP per capita growth rate across countries is not related to the effect (p.80) (p.83) of macroeconomic variables added one at a time. Similarly, the finding of a tight relationship between changes in labour market indicators and changes in some macroeconomic factors is not related to the rate of GDP per capita growth.

In summary, increases in some macroeconomic variables were associated with changes in labour market conditions in Latin America during the 2000s, some of them always in the welfare-improving direction and some others always in the welfare-reducing direction. There is no unique configuration of macroeconomic variables that was associated with the several successful experiences among our sample of sixteen countries. Finally, the correlation between the change in GDP per capita and the change in macroeconomic variables seemed to be small enough so as not to affect in general the magnitudes of the coefficients and standard errors in the estimations of the relationships between changes in labour market indicators and the rate of GDP per capita growth on the one hand, and changes in macroeconomic variables on the other hand.

## 4.2.4 Relationship between Labour Market Indicators

Another question is whether the labour market indicators tend to improve or worsen together, or whether there are pairs of indicators such that a higher rate of improvement in one is associated with a lower rate of improvement or a worsening of the other. For example, a higher rate of earnings growth could be associated with a higher increase in unemployment due to employers moving up along a single downward-sloping labour demand curve.

Our findings indicate that labour market indicators either improved jointly or worsened jointly. Table 4.4 displays the cross-country correlations between the changes or percentage changes in each of our sixteen labour market indicators. In particular, we estimate the following sets of correlations:

$Display mathematics$

$Display mathematics$

for

$Display mathematics$
(12)

Table 4.4 Cross-country correlation matrix between the annualized changes in labour market indicators during the 2000s

Part A

Occupations

Occupational position

Economic sector

Decline in unemployment

Declined in share of low-earnings occupations

Increase in share of high-earnings occupations

Increase in the share of wage/salaried employees

Decrease in the share of self-employment

Decrease in the share of unpaid family workers

Decline in the share of workers in low-earnings sectors

Increase in the share of workers in high-earnings sectors

Unemployment

Decrease in the unemployment rate

1.00

0.33

0.20

−0.06

−0.03

−0.15

0.07

0.01

Occupations

Decrease in the share of low-earnings occupations

0.33

1.00

0.43

0.57

0.37

0.53

0.52

0.44

Increase in the share of high-earnings occupations

0.20

0.43

1.00

0.35

0.42

0.23

0.22

0.21

Occupational position

Increase in the share of wage/salaried employees

−0.06

0.57

0.35

1.00

0.82

0.75

0.57

0.67

Decrease in the share of self-employment

−0.03

0.37

0.42

0.82

1.00

0.39

0.32

0.50

Decrease in the share of unpaid family workers

−0.15

0.53

0.23

0.75

0.39

1.00

0.66

0.58

Economic sector

Decrease in the share of workers in low-earnings sectors

0.07

0.52

0.22

0.57

0.32

0.66

1.00

0.66

Increase in the share of workers in high-earnings sectors

0.01

0.44

0.21

0.67

0.50

0.58

0.66

1.00

Education

Decrease in the share of low-educated workers

−0.02

0.35

0.31

0.64

0.52

0.53

0.68

0.89

Increase in the share of high-educated workers

0.32

0.61

0.60

0.68

0.58

0.48

0.50

0.75

Workers registered with SS

Increase in the share of workers registered with SS

0.23

−0.08

0.10

−0.03

−0.16

0.18

0.46

0.05

Earnings

Increase in mean labour earnings

0.42

0.72

0.51

0.48

0.22

0.52

0.44

0.67

Poverty

Decrease in 4 dollars-a-day poverty

0.36

0.65

0.44

0.55

0.28

0.58

0.62

0.61

Decrease in 2.5 dollars-a-day poverty

0.41

0.73

0.47

0.59

0.28

0.61

0.58

0.61

Inequality

Decrease in GINI of household per capita income

0.23

0.32

0.07

0.38

0.24

0.48

0.55

0.29

Decrease in GINI of labour earnings

0.47

0.43

0.29

0.33

0.13

0.50

0.58

0.46

Part B

Education

Poverty

Inequality

Decrease in the share of low-educated workers

Increase in the share of high-educated workers

Workers registered with SS

Increase in mean labour earnings

Decline in 4-dollar-a-day poverty

Decline in 4-dollar-a-day poverty

Decline in GINI of household per capita income

Decline in GINI of labour earnings

Unemployment Decrease in the unemployment rate

−0.02

0.32

0.23

0.42

0.36

0.41

0.23

0.47

Occupations

Decrease in the share of low-earnings occupations

0.35

0.61

−0.08

0.72

0.65

0.73

0.32

0.43

Increase in the share of high-earnings occupations

0.31

0.60

0.10

0.51

0.44

0.47

0.07

0.29

Occupational position

Increase in the share of wage/salaried employees

0.64

0.68

−0.03

0.48

0.55

0.59

0.38

0.33

Decrease in the share of self-employment

0.52

0.58

−0.16

0.22

0.28

0.28

0.24

0.13

Decrease in the share of unpaid family workers

0.53

0.48

0.18

0.52

0.58

0.61

0.48

0.50

Economic sector

Decrease in the share of workers in low-earnings sectors

0.68

0.50

0.46

0.44

0.62

0.58

0.55

0.58

Increase in the share of workers in high-earnings sectors

0.89

0.75

0.05

0.67

0.61

0.61

0.29

0.46

Education

Decrease in the share of low-educated workers

1.00

0.67

0.01

0.60

0.53

0.51

0.28

0.46

Increase in the share of high-educated workers

0.67

1.00

0.07

0.68

0.56

0.63

0.20

0.49

Workers registered with SS

Increase in the share of workers registered with SS

0.01

0.07

1.00

0.04

0.17

0.17

0.30

0.40

Earnings

Increase in mean labour earnings

0.60

0.68

0.04

1.00

0.83

0.88

0.28

0.58

Poverty

Decrease in 4 dollars-a-day poverty

0.53

0.56

0.17

0.83

1.00

0.98

0.70

0.77

Decrease in 2.5 dollars-a-day poverty

0.51

0.63

0.17

0.88

0.98

1.00

0.62

0.75

Inequality

Decrease in GINI of household per capita income

0.28

0.20

0.30

0.28

0.70

0.62

1.00

0.85

Decrease in GINI of labour earnings

0.46

0.49

0.40

0.58

0.77

0.75

0.85

1.00

Note: The grey shading indicates a positive correlation larger than 0.4. Correlations for occupations do not include Argentina for which we do not have data.

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

A correlation coefficient between 0.4 and 1 implies that, in a regression of the annualized changes in two labour market indicators, the R-squared is larger than 0.15 (which corresponds with the cut-off value we used previously), and that the association between the two variables is positive. Conversely, a coefficient between −1 and −0.4 indicates a negative relationship. (p.84) (p.85) (p.86) (p.87) The shaded cells in Table 4.4 indicate a strong relationship between two labour market indicators.

We find that most of our labour market indicators tend to move together, with not even one instance of a substantial trade-off between changes in our selected labour market indicators, i.e. improvements in one do not come at the cost of worsening in others. Specifically, of the 120 correlations we computed, we find that seventy-one (59 per cent of the total) of the pairs of indicators have a positive and significant association, while for the forty-nine remaining pairs we found only weak but generally positive associations. Finally, there is not even a single value in the matrix with a negative sign and above (in absolute value) our cut-off value equal to −0.4, which will indicate a trade-off between two labour market indicators: the lower value is equal to −0.16.

Several labour market indicators are highly correlated among them, with a few exceptions (Table 4.4). On the one hand, labour earnings, the sectoral and educational composition of employment, and the distributive indicators have a significant correlation with at least ten other labour market indicators. On the other hand, the unemployment rate, the share of self-employed, the share of registered workers, the Gini of household per capita income, and the share of high-earnings occupations do not co-vary as much with other indicators (they are significantly correlated with six or fewer of the others).

Some clear patterns of correlations appear from this evidence. The results from Table 4.4 and Figure 4.7 indicate that changes in labour earnings tend to be highly correlated with changes in the job mix (i.e. the occupational, positional, sectoral, and educational composition of the employed population). There may be a simple explanation for these relationships: a rightward shift of the labour demand curve, such that in order to attract more workers into the better job categories, employers must raise wages. Average earnings may also increase just by a composition effect: in a context of high unemployment, a rightward shift of the labour demand curve may lead to an increase in the share of better paying occupations, and thus in average earnings, with fixed hourly wages. As expected, increases in labour earnings are also highly correlated with reductions in poverty: countries in which labour earnings increased were generally ones in which poverty fell, which indicates the importance of labour earnings in the total income of the household. Increases in labour earnings are also related to reductions in the inequality of their distribution, indicating that the process of growth was also inequality-reducing. The evidence of improvements in the job mix, of increases in labour earnings, and of reductions in earnings inequality suggests that workers moved on average to better-paying jobs.

Figure 4.7 Cross-country relationship between annualized changes in labour market indicators and annualized changes in mean labour earnings during the 2000s

Note: The vertical axes display the annualized changes in each labour market indicator.Δ‎denotes changes in percentage points; % Δ‎ denotes percentage changes. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses. R-squared of the regression indicated along the title of each figure.

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

(p.88) (p.89) (p.90) (p.91) We now turn to analyse the relationship between the share of wage/salaried employees and some selected indicators, illustrated in Figure 4.8.1 An increase in the share of wage/salaried employees is associated with a general improvement in the labour market. Not only is the share of wage/salaried employees related to reductions in moderate and extreme poverty, but also with increases in the shares of high-earnings occupations and high-earnings sectors, as well as reductions in the shares of low-earnings occupations and sectors. These findings are also consistent with a rightward shift of labour demand in wage/salaried jobs, which seem to have a high incidence in high-earnings occupations and sectors, increasing their shares of employment and reducing poverty.

Figure 4.8 Cross-country relationship between annualized changes in selected labour market indicators and annualized changes in the share of wage/salaried employees in total employment during the 2000s

Note: The vertical axes display the annualized changes in each labour market indicator. Δ‎ denotes changes in percentage points; % Δ‎ denotes percentage changes. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses. R-squared of the regression indicated along the title of each figure.

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

# 4.3 Cross-Country Patterns: Changing Employment, Earnings, and Inequality Indicators and Changes in Poverty

Our results in Chapter 3 indicated that real GDP per capita grew substantially in all Latin American countries in the 2000s, with an average per capita growth rate of approximately 3 per cent a year. We also reported that poverty, extreme poverty, and inequality also fell substantially in all but one of the sixteen countries in the region in the 2000s. At the same time, while employment and earnings indicators also improved in most countries, they did so more in some countries than in others. In this section, we analyse in more detail the relationship between changes in employment and earnings indicators, and changes in poverty indicators. We aim to establish whether larger improvements in employment and earnings are associated with larger reductions in poverty, over and above the rate of economic growth. We present here a cross-country analysis of the employment and earnings–poverty relationships based on sixteen data points (one for each country) representing the annualized changes between the initial and the final years for each country.

Although poverty reduction in the region is a well-documented fact, the studies that analyse its labour market determinants are scarcer and they only focus on a small group of labour market variables to explain the fall in the poverty rates during the 2000s. An example is ECLAC-ILO (2015), which relates the remarkable progress in reducing poverty during the period 2002–12 in the Latin America region to labour market trends. The main factors mentioned in this study are: the strong job creation, especially in wage/salaried positions, and public policies, such as minimum wages increases, formalization of workers, and expanding coverage of social protection systems and education. Other papers rely on decomposition approaches to disentangle (p.92) the importance of different labour market outcomes in reducing poverty. This is the case of World Bank (2015), which uses different decomposition approaches to investigate the role played by different income sources in reducing poverty. The study finds that changes in labour earnings were the most important factor in explaining poverty reduction in the region during 2003–8 and 2008–13, but its importance was lower in the post-international-crisis period than before. ECLAC (2014) points out that the most important factor explaining the decline in poverty during the 2000s was the combined increase in employment and wages, although in general, labour earnings increases had a greater impact than employment growth on household income changes. This is consistent with Beccaria et al. (2011), World Bank (2013), and Inchauste et al. (2014), who also decompose changes in the poverty rate and report increases in labour income as the main channel to poverty reductions.

## 4.3.1 Response of Poverty to Employment and Earnings Indicators

Our evidence reveals a strong and consistent cross-country pattern of association between reductions in poverty and extreme poverty, and improvements in earnings and employment indicators. These relationships are illustrated in the scatter plots presented in Figure 4.9 (for poverty based on the 2.5 dollars-a-day poverty line) and Figure 4.10 (for poverty based on the 4 dollars-a-day poverty line). We find that eleven out of fourteen of the associations in Figure 4.9 and twelve out of fourteen of the associations in Figure 4.10 (excluding the relationship between the two poverty indicators in both cases) present an R-squared above a 0.15 threshold, and in almost all cases, whether the relationships are above this threshold or not, the sign of correlation is in the expected direction, i.e. improvements in earnings and employment indicators are associated with reductions in poverty rates.

Figure 4.9 Cross-country relationship between annualized changes in labour market indicators and annualized changes in the 2.5 dollars-a-day poverty rate during the 2000s

Note: The vertical axes display the annualized changes in each labour market indicator. Δ‎ denotes changes in percentage points; % Δ‎ denotes percentage changes. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses. R-squared of the regression indicated along the title of each figure.

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

Figure 4.10 Cross-country relationship between annualized changes in labour market indicators and annualized changes in the 4 dollars-a-day poverty rate during the 2000s

Note: The vertical axes display the annualized changes in each labour market indicator. Δ‎ denotes changes in percentage points; % Δ‎ denotes percentage changes. The line represents the linear regression specified at the bottom of the figure. Robust standard error of the slope coefficient between parentheses. R-squared of the regression indicated along the title of each figure.

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

Among employment and earnings indicators, there is a very strong negative cross-country correlation between changes in mean earnings and changes in moderate and extreme poverty rates during the period under study, with a stronger relationship for moderate poverty: that is, mean earnings rose faster while poverty fell faster. The relationships between changes in the two poverty rates and the percentage change in mean labour earnings are the strongest in both Figures 4.9 and 4.10. In both cases, larger increases in labour earnings are associated with larger reductions in poverty levels, with a somewhat stronger relationship for moderate poverty in Figure 4.9 (R-squared of 0.78), than for extreme poverty in Figure 4.10 (R-squared of 0.68). The correlations between percentage changes in mean labour earnings and changes in the two poverty measures, however, are both very strong, and the difference between the two is only a matter of degree. This result is consistent with the discussion (p.93) (p.94) (p.95) (p.96) (p.97) in the literature for Latin America stressing that the extreme poor do not benefit as much as those closer to the moderate poverty line from the trickle-down of economic growth (and the subsequent increase in labour earnings), which implies that improving the living conditions of those harder to reach needs more government-based redistribution than those relatively better-off among the poor (see, for instance, Cruces and Gasparini 2013, and references therein). This is also apparent in the weaker relationship between poverty rates and unemployment that we analyse in the following paragraph.

There is a consistent and relatively strong cross-country pattern of association between reductions in poverty and extreme poverty, and improvements in the job mix (distributions of workers among occupations, occupational positions, sectors, and educational levels). The correlations in Figures 4.9 and 4.10 are qualitatively and quantitatively similar for moderate and extreme poverty, although slightly tighter for the moderate poverty rate. We thus report them together, citing the R-squared for extreme poverty (Figure 4.9) first and then that for moderate poverty (Figure 4.10). Specifically, we find a clear pattern of a positive correlation between changes in poverty and changes in the share of low-earnings occupations (R-squared of 0.36 for extreme poverty and of 0.43 for moderate poverty), and a corresponding negative correlation between changes in poverty and changes in the share of (p.98) high-earnings occupations (R-squared of 0.26 and 0.29). Similarly, reductions in the share of low-earnings sectors are associated with reductions in the poverty rates (R-squared of 0.38 and 0.34), whereas increases in the share of high-earnings sectors over the period are correlated negatively with changes in the poverty rates (R-squared of 0.37 and 0.38). The share of workers with low educational levels tended to fall over this period, while that of workers with high educational levels tended to increase, and both changes were associated with reductions in the poverty rates (R-squared of 0.26 and 0.28 for the share of workers with low educational levels and R-squared of 0.40 and 0.31 for the share of workers with high educational levels). Finally, the pattern for occupational position is not as clear as in the cases of occupations, sectors, and education. We observe a negative correlation between poverty changes and changes in the share of wage/salaried employees over the period (R-squared of 0.31 and 0.35), and also a relatively strong positive correlation between poverty changes and changes in the share of unpaid workers (R-squared of 0.33 and 0.37). However, we do not find a meaningful pattern between poverty changes and changes in the share of self-employed workers, with positive but weak correlations (R-squared of 0.08 in both cases). The same is true, perhaps surprisingly, for the changes in the share of workers registered with social security. While the correlations between changes in this indicator and changes in poverty measures are negative, as expected, the relationships are relatively flat and not very tight (R-squared of 0.11 in both cases).

The negative cross-country correlation between percentage changes in mean earnings and changes in moderate and extreme poverty rates in Latin America in the 2000s is robust: it is still present after controlling for changes in unemployment and changes in GDP per capita. We check the robustness of the bivariate relationship between changes in mean earnings and changes in poverty by also performing multivariate regressions. We regress the percentage changes in extreme and moderate poverty rates on the percentage changes in labour earnings, GDP per capita, and unemployment. The analysis is limited since we only have sixteen observations when studying cross-country correlations over the 2000s, but we can still probe whether the correlation between changes in the poverty rates and in mean earnings is present conditional on one or two other relevant variables.

The top panel of Table 4.5 presents the results of these regressions for the extreme poverty rate. In line with the previous discussion about the lack of trickle-down effects of growth at the very bottom of the income distribution and the results in section 4.1, the relationship between changes in GDP per capita and changes in extreme poverty is not statistically significant. According to the results in column 2 of Table 4.5, there seems to be a negative and statistically significant elasticity between extreme poverty and unemployment (in contrast with the regression in changes instead of percentage (p.99) (p.100) changes in unemployment in Figure 4.9) of about 0.32, with a relatively low R-squared of 0.17. However, these relationships do not seem to be very robust: when including both variables in the same regression (column 4), the two are not statistically significant. Finally, and as expected from previous results, the labour earnings extreme-poverty elasticity is strongly significant, with a regression coefficient of −1.55 and R-squared of about 0.64 (column 3). The results in columns 5 to 7 in the top panel of Table 4.5 confirm the robustness of this elasticity: controlling for percentage changes in GDP per capita (column 5), for percentage changes in unemployment (column 6), or for both, none of the additional variables is statistically significant, and the labour earnings elasticity remains virtually unchanged around −1.5, and still strongly significant (which is all the more remarkable again with the limited number of observations available).

Table 4.5 Cross-country poverty elasticities with respect to GDP per capita, unemployment rate, and labour earnings during the 2000s

Dependent variable: %∆ 2.5 dollars-a-day poverty

(1)

(2)

(3)

(4)

(5)

(6)

(7)

%∆ GDP per capita

−0.778

−0.381

−0.438

−0.469

(0.407)

(0.616)

(0.254)

(0.306)

%∆ Unemployment rate

0.320

0.265

0.046

−0.024

(0.14)*

(0.196)

(0.076)

(0.091)

%∆ Labour earnings

−1.550

−1.489

−1.496

−1.513

(0.286)**

(0.294)**

(0.342)**

(0.376)**

R-squared

0.098

0.173

0.638

0.191

0.668

0.640

0.668

Observations

16

16

16

16

16

16

16

Dependent variable: %∆ 4 dollars-a-day poverty

(1)

(2)

(3)

(4)

(5)

(6)

(7)

%∆ GDP per capita

−0.890

−0.553

−0.608

−0.625

(0.338)**

(0.505)

(0.181)**

(0.189)**

%∆ Unemployment rate

0.305

0.225

0.080

−0.014

(0.113)**

(0.156)

(0.08)

(0.071)

%∆ Labour earnings

−1.319

−1.234

−1.225

−1.247

(0.185)**

(0.198)**

(0.233)**

(0.249)**

R-squared

0.183

0.223

0.656

0.278

0.739

0.669

0.739

Observations

16

16

16

16

16

16

16

Note: Poverty elasticities are calculated using the percentage change in the poverty rates, GDP per capita, the unemployment rate, and mean labour earnings between the initial and the final years in each country. Robust standard errors in parentheses.

** significant at 1% level, * significant at 5% level.

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

The corresponding results for the moderate poverty elasticities are presented in the bottom panel of Table 4.5. The elasticity with respect to labour earnings is again strongly significant but somewhat lower in absolute value (between −1.22 and −1.32), and also robust to the inclusion of percentage changes in GDP per capita and unemployment as conditioning variables. The elasticity between moderate poverty and unemployment is again significant when unconditional (column 2), but not statistically significant when either change in GDP per capita or change in labour earnings or both are included (columns 4, 6, and 7). The main difference with respect to the results for the extreme-poverty elasticities is the elasticity coefficient between moderate poverty and GDP per capita: the coefficient for this variable is significant when included on its own (column 1), but also when controlling for labour earnings (column 5) and labour earnings and unemployment (column 7). The unconditional elasticity is −0.890, and it is reduced to −0.625 when including the additional controls. The elasticity with respect to labour earnings also falls (although only slightly) when including the additional controls. The fact that the two variables are jointly statistically significant in the conditional regression presented in column 7 suggests that while related, the two operate also through separate channels. In other words, poverty seems to fall when labour earnings increase over and above the effect of GDP per capita growth, and vice versa. Besides the robustness of the effect of the percentage change in labour earnings on moderate and extreme poverty, the pattern of results suggests that GDP per capita growth reaches the bottom of the distribution through its effect on mean labour earnings but not through other channels.

## 4.3.2 Response of Poverty to Inequality Indicators

Moving now to the inequality indicators, there is a strong positive cross-country correlation between changes in poverty rates and changes in income (p.101) and labour earnings inequality. Figures 4.9 and 4.10 present the scatter plots of changes in extreme and moderate poverty and percentage changes in the Gini coefficient of household per capita income and in the Gini coefficient of labour earnings. Both correlations appear to be stronger for the Gini of labour earnings (R-squared of 0.60 for extreme poverty and 0.57 for moderate poverty) than for the Gini of household per capita income (R-squared of 0.49 and 0.38, respectively). While there is a mechanical component, which implies that other incomes remaining equal, reductions in poverty imply reductions in inequality, the strong associations illustrate the overall improvement in the income distribution (besides poverty only) in Latin America during the 2000s.

# 4.4 In Summary

In this chapter, we looked at the cross-country link between growth, employment, and poverty. First, we found that faster growth is associated with larger improvements in employment and earnings indicators and in poverty and inequality indicators, but the relationships were in general weak. For only three out of sixteen indicators did we obtain a strong relationship between the rate of improvement of the indicator and the rate of economic growth. These three indicators were the share of low-earnings occupations, the share of high-earnings occupations, and the share of registered workers, all of which moved in the welfare-improving direction significantly more in countries that experienced higher rates of growth.

Second, we looked at four correlates of cross-country changes in labour market indicators beyond economic growth. The first question was, were the changes in labour market indicators across countries related to initial GDP per capita? We found no substantial relationship between either the share of labour market indicators that improved or the change in individual labour market indicators on the one hand and initial GDP per capita on the other. The second was whether other macroeconomic factors could help explain the differences across countries in labour market indicators. We found that increases in seven macroeconomic factors were related to changes in labour market indicators, some in the welfare-improving direction (exports as a percentage of GDP, terms of trade, revenues from natural resources as a percentage of GDP, and the share of industry in GDP) and some in the welfare-reducing direction (stock of public debt as a percentage of GDP, domestic consumption as a percentage of GDP, and the share of services in GDP). The third issue was whether changes in individual labour market indicators were related to their initial level. For five indicators (the unemployment rate, the share of unpaid family workers, the poverty and extreme poverty rates, and the inequality of household per capita income), we found that worse initial levels were (p.102) associated with larger improvements. For the other indicators, no relationship surfaced. The fourth issue was whether some labour market indicators tended to move together with others and, if so, in which direction. We found that 59 per cent of the pairs improved significantly together and no significant relationship appeared between the other 41 per cent of the pairs; no indicator improved while another one worsened.

Finally, we studied the cross-country relationship between improvements in employment and earnings indicators and poverty changes. Our evidence revealed a generally strong and consistent cross-country pattern of association between reductions in poverty and extreme poverty on the one hand, and improvements in earnings and employment indicators on the other. From a multivariate analysis we concluded: (1) poverty, measured by the 4 dollars-a-day poverty line, fell when labour earnings increased over and above the effect of GDP per capita growth, and vice versa; and (2) GDP per capita growth did not reach the bottom of the distribution beyond its effects on labour earnings.

References

Bibliography references:

Beccaria, L., R. Maurizio, A. L. Fernández, P. Monsalvo, and M. Álvarez (2011). ‘Dynamics of Poverty, Labor Market and Public Policies in Latin America’. PMMA Working Paper 2011–05. Nairobi: Poverty and Economic Policy Research Network.

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.

Cruces, G. and L. Gasparini (2013). ‘Políticas sociales para la reducción de la desigualdad y la pobreza en América Latina y el Caribe: Diagnóstico, propuesta y proyecciones en base a la experiencia reciente’. CEDLAS Working Paper 142. La Plata: Centro de Estudios Distributivos, Laborales y Sociales.

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.

ECLAC (2014). ‘Social Panorama of Latin America, 2014’. LC/G 2635-P. Santiago de Chile: Economic Commission for Latin America and the Caribbean.

ECLAC-ILO (2015). ‘Employment Situation in Latin America and the Caribbean: Universal Social Protection in Labour Markets with High Levels of Informality’. Report prepared jointly by the Economic Division of the Economic Commission for Latin America and the Caribbean and the Office for the Southern Cone of Latin America of the International Labour Organization. Santiago de Chile.

Gasparini, L. and L. Tornarolli (2009). ‘Labour Informality in Latin America and the Caribbean: Patterns and Trends from Household Survey Microdata’, Revista Desarrollo y Sociedad 63 (1): 13–80.

(p.103) ILO (2013). ‘Global Wage Report 2012/13: Wages and Equitable Growth’. Geneva: International Labour Office.

Inchauste, G., J. P. Azevedo, B. Essama-Nssah, S. Olivieri, T. Van Nguyen, J. Saavedra-Chanduvi, and H. Winkler (2014). ‘Understanding Changes in Poverty: Directions in Development’. Washington, DC: World Bank.

Pagés, C., G. Pierre, and S. Scarpetta (2009). Job Creation in Latin America and the Caribbean: Recent Trends and Policy Challenges. Basingstoke: Palgrave Macmillan.

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.

UN-ECLAC (2015). CEPALSTAT, United Nations, Economic Commission for Latin America and the Caribbean. Available at <http://estadisticas.cepal.org/cepalstat/WEB_CEPALSTAT/Portada.asp?idioma=i>, accessed April 2015.

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

Weller, J. and C. Kaldewei (2013). ‘Empleo, crecimiento sostenible e igualdad’. Serie Macroeconomía del Desarrollo 145. Santiago de Chile: United Nations.

World Bank (2012). ‘The Labour Market Story behind Latin America’s Transformation’. Washington, DC: World Bank.

World Bank (2013). ‘Shifting Gears to Accelerated Shared Prosperity in Latin America and the Caribbean’. Document of the World Bank 78507. Washington, DC: World Bank.

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

World Bank (2015). ‘Working to End Poverty in Latin America and the Caribbean: Workers, Jobs, and Wages’. Document of the World Bank. Washington, DC: World Bank.

## Notes:

(1) We provide a detailed analysis of the cross-country relationship between poverty indicators and employment and earnings indicators in section 4.3.1.