Earnings volatility in Brazil (2012-2023)

Portella, Alysson; Gonçalves, Solange Ledi; de Souza, Pedro H. G. Ferreira; Firpo, Sergio Working Paper Earnings volatility in Brazil (2012-2023) Texto para Discussão, No. 3106 Provided in Cooperation with: Institute of Applied Economic Research (ipea), Brasília Suggested Citation: Portella, Alysson; Gonçalves, Solange Ledi; de Souza, Pedro H. G. Ferreira; Firpo, Sergio (2025) : Earnings volatility in Brazil (2012-2023), Texto para Discussão, No. 3106, Instituto de Pesquisa Econômica Aplicada (IPEA), Brasília, https://doi.org/10.38116/td3106-eng This Version is available at: https://hdl.handle.net/10419/316168 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/2.5/br/ 3106 EARNINGS VOLATILITY IN BRAZIL (2012-2023) ALYSSON PORTELLAALYSSON PORTELLA SOLANGE G. GONÇALVESSOLANGE G. GONÇALVES PEDRO H. G. FERREIRA DE SOUZAPEDRO H. G. FERREIRA DE SOUZA SERGIO FIRPOSERGIO FIRPO 3106 Brasilia, April 2025 EARNINGS VOLATILITY IN BRAZIL (2012-2023) ALYSSON PORTELLA1 SOLANGE L. GONÇALVES2 PEDRO H. G. FERREIRA DE SOUZA3 SERGIO FIRPO4 1. Researcher at Insper. E-mail: [email protected]. 2. Assistant professor at the University of São Paulo (USP). E-mail: solange.gon- [email protected]. 3. Researcher at the Institute for Applied Economic Research (Ipea). E-mail: pedro.ferreir[email protected].br. 4. Professor of Economics at Insper. E-mail: [email protected]. Discussion Paper A publication to disseminate the findings of research directly or indirectly conducted by the Institute for Applied Economic Research (Ipea). Due to their relevance, they provide information to specialists and encourage contributions. © Institute for Applied Economic Research – ipea 2025 Earnings volatility in Brazil (2012-2023) / Alysson Portella... [et al.]. Brasília: Ipea, Abr., 2025. 38 p. : il. – (Discussion Paper ; n. 3106). Inclui Bibliografia. 1. Desigualdade Salarial. 2. Distribuição de Renda. 3. Volatilidade dos Rendimentos. 4. Brasil. I. Portella, Alysson. II. Gonçalves, Solange L. III. Souza, Pedro H. G. Ferreira de. IV. Firpo, Sergio. V. Instituto de Pesquisa Econômica Aplicada. VI. Título. CDD 331.2 Ficha catalográfica elaborada por Elisangela da Silva Gomes de Macedo CRB-1/1670 How to cite: PORTELLA, Alysson et al. Earnings volatility in Brazil (2012-2023). Brasília: Ipea, Abr., 2025. 38 p. (Discussion Paper, n. 3106). DOI: http:// dx.doi.org/10.38116/td3106-eng JEL: D31; J31; O15; D63. DOI: https://dx.doi.org/10.38116/td3106-eng Ipea publications are available for free download in PDF (all) and ePUB (books and periodicals). Access: https://www.ipea.gov.br/portal/publicacoes The opinions expressed in this publication are of exclusive responsibility of the authors, not necessarily expressing the official views of the Institute for Applied Economic Research and the Ministry of Planning and Budget. Reproduction of this text and the data contained within is allowed as long as the source is cited. Reproduction for commercial purposes is prohibited. Federal Government of Brazil Ministry of Planning and Budget Officer Simone Nassar Tebet A public foundation affiliated to the Ministry of Planning and Budget, Ipea provides technical and institutional support to government actions – enabling the formulation of numerous public policies and programs for Brazilian development – and makes research and studies conducted by its staff available to society. President LUCIANA MENDES SANTOS SERVO Director of Institutional Development FERNANDO GAIGER SILVEIRA Director of Studies and Policies of the State, Institutions and Democracy LUSENI MARIA CORDEIRO DE AQUINO Director of Macroeconomic Studies and Policies CLÁUDIO ROBERTO AMITRANO Director of Regional, Urban and Environmental Studies and Policies ARISTIDES MONTEIRO NETO Director of Sectoral Studies and Policies, of Innovation, Regulation and Infrastructure FERNANDA DE NEGRI Director of Social Studies and Policies RAFAEL GUERREIRO OSÓRIO Director of International Studies KEITI DA ROCHA GOMES Chief of Staff ALEXANDRE DOS SANTOS CUNHA General Coordinator of Press and Social Communication GISELE AMARAL DE SOUZA Ombudsman: https://www.ipea.gov.br/Ouvidoria URL: https://www.ipea.gov.br CONTENTS ABSTRACT 1 INTRODUCTION ......................................................................6 2 DATA AND METHODS ............................................................8 2.1 Data sources ................................................................................8 2.2 Sample selection ...................................................................... 10 2.3 Measuring earnings volatility ................................................... 10 3 MAIN FINDINGS .................................................................... 13 3.1 Variance of arc percent changes ............................................. 13 3.2 Distribution of arc percent changes ........................................ 18 4 HETEROGENEITY ANALYSES .............................................20 4.1 Volatility by income level .......................................................... 20 4.2 Employment-nonemployment transitions ............................... 22 4.3 Type of employment ................................................................. 25 4.4 Demographic sub-groups ......................................................... 31 5 CONCLUSION ........................................................................35 REFERENCES ............................................................................ 35 ABSTRACT This article provides a comprehensive analysis of labor earnings volatility in Brazil between 2012 and 2023. During this period, Brazil’s economy experienced intense economic growth followed by large recessions, allowing us to assess changes in volatility over the business cycle. In addition, Brazil’s national household survey follows individuals throughout an entire year, covering both formal and informal sectors of the economy. This offers the perfect setting to study earnings volatility in developing countries, a subject that has been mostly investigated in the context of advanced economies. Apart from documenting the main volatility trends, we also investigate how they vary by income level, employment-nonemployment transitions, and transitions between formal and informal sectors. We also assess how volatility varies by gender, education, race, and marital status. Our results show that earnings volatility is much higher in Brazil than in rich countries, especially among low-wage, informal workers. Transitions into and out of employment account for a large share of wage volatility levels. Keywords: Brazil; wage inequality; income distribution; earnings volatility. DISCUSSION PAPER 6 3106 1 INTRODUCTION Earnings volatility – the level of fluctuations in pay over time – has been extensively studied in developed countries, particularly the United States, but also in other economies (Moffitt et al., 2023; Moffitt and Zhang, 2018; Cappellari and Jenkins, 2014; Jappelli and Pistaferri, 2010; Avram et al., 2022; Li, La and Sologon, 2021; OECD, 2011). Interest in this topic arises not only for its own sake, but also due to the links between earnings volatility and inequality and poverty dynamics, consumption patterns, social mobility, economic insecurity, and income risk (Meghir and Pistaferri, 2011; Attanasio and Weber, 2010; Shorrocks, 1978; Western et al., 2012). Despite the extensive literature on volatility in high-income countries, research on developing economies is limited. This article conducts a comprehensive analysis of earnings volatility in Brazil, a large and highly unequal developing country. There are several reasons to believe that volatility is higher in these countries. Income growth is more volatile in emerging economies (Aguiar and Gopinath, 2007), and the presence of large informal markets likely influences the level of earnings risks in these economies (La Porta and Shleifer, 2014; Ulyssea, 2020). High labor market turnover and unemployment are also prevalent (Gerard and Gonzaga, 2021). Additionally, high inequality and low mobility are characteristic features of developing countries, especially Latin America (Chancel and Piketty, 2021; Messina and Silva, 2021; Britto et al., 2022). Our analysis relies on panel data from the Continuous National Household Sample Survey (PNADC). We focus on year-over-year changes in earnings, using the variance of the arc percent change as our main measure of volatility. This measure can account for changes in earnings arising from nonemployment and has been extensively used to document trends in volatility in other countries (Shin and Solon, 2011; Celik et al., 2012; Ziliak, Hardy and Bollinger, 2011; Dynan, Elmendorf and Sichel, 2012; Moffitt et al., 2023). We calculate separate estimates for men and women, as these groups display different trajectories in high-income countries (Moffitt et al., 2023). We assess how volatility varies over the earnings distribution in the first period and the role of transitions into and out of employment. Our data allow us to investigate how volatility differs between individuals with formal, informal, and self-employed attachments to the labor market, which are important aspects of labor market in developing countries. We also investigate heterogeneity by education, race, and marital status. Our findings suggest that earnings volatility in Brazil is higher than in the United States and other advanced economies. The volatility of male earnings in Brazil was around 0.25 in 2015, compared to 0.10 in the United States before the Great Recession (Ziliak, Hokayem and Bollinger, 2023). Interestingly, in contrast to high-income countries, the volatility of female earnings in Brazil is greater than that of male earnings. DISCUSSION PAPER DISCUSSION PAPER 7 3106 However, this relationship was briefly reversed during the peak of the covid-19 pandemic. As documented in other countries, volatility is countercyclical in Brazil, which means that it increases during economic downturns. This effect is stronger among men. These results are robust to several changes in specification and sample definition. We find much higher earnings volatility among low-income individuals. Volatility reaches the lowest level among those who earn close to the minimum wage, slightly increasing thereafter. To the best of our knowledge, we are the first to document how volatility varies with earnings. We also decompose labor market volatility by transitions into and out of employment. This exercise reveals that Brazil’s high volatility is partly driven by more frequent periods with zero earnings. When considering the type of attachment to the labor market, we see that informal and self-employed workers experience the most volatile earnings. This pattern is primarily driven by the nature of informal work itself, rather than frequent sector transitions, as workers who hold informal jobs in both periods have similar volatility as those who change from formal to informal or vice versa. The heterogeneity analysis indicates that one reason for the greater earnings volatility among men is the higher prevalence of self-employment. Our heterogeneity analyses by educational levels, race, and marital status confirm that white individuals and those with higher educational attainment experience lower earnings volatility. However, the differences between married and single individuals are more subtle. We do not observe significant differences for men, while the results for married and single women depend on whether we include individuals with zero earnings. A substantial body of research has investigated the dynamics of earnings volatility in high-income countries, such as the United States, Europe, and Australia. Earlier research focused on estimating the variance of transitory income using parametric or nonparametric models (Gottschalk et al., 1994; Moffitt and Gottschalk, 2012), which requires long longitudinal panel surveys or administrative data. This literature has been complemented by more descriptive and transparent analyses based on the variance of earnings changes between two periods, often relying on shorter panels such as the Current Population Survey – CPS (Celik et al., 2012; Dahl, Deleire and Schwabish, 2011; Shin and Solon, 2011; Moffitt et al., 2023; Ziliak, Hokayem and Bollinger, 2023). Our paper follows the latter approach. More recently, a new line of research has focused on not only documenting the variance but also higher moments of earnings changes, highlighting how these innovations are far from normally distributed, showing asymmetry and high kurtosis (Guvenen, Ozkan and Song, 2014; Hoffmann and Malacrino, 2019; Guvenen et al., 2021; Arellano, Blundell and Bonhomme, 2017; Busch et al., 2022). Our paper complements a small literature on earnings volatility in developing countries. Beccaria et al. (2022) investigate income mobility in seven Latin American countries. DISCUSSION PAPER 8 3106 Contrary to us, they use household income and focus on individual-level mobility measures. Santos and Souza (2007) and Arabage and Souza (2019) use data restricted to the Brazilian formal labor market to estimate models for the variance of transitory and permanent incomes. Engbom et al. (2022) document long-term trends in earnings volatility using data on formal employment and data on formal and informal jobs, but only encompassing six Brazilian metropolitan regions. Martinez and Mello (2024) investigate how increased trade exposure affects the higher moments of earnings risk in Brazil using administrative records, which cover only the formal sector of the labor market. Gomes, Iachan and Santos (2020) use PNADC data to investigate earnings changes in the formal and informal labor markets but investigate higher moments among workers with positive earnings only. We complement their analysis by including individuals with zero earnings and calculating gross measures of volatility. Thus, we uncover how transitions into and out of employment are much more common in Brazil than in developed countries and contribute decisively to the overall higher levels of volatility. In addition, we show how earnings volatility is much higher at lower income levels, likely a combination of worse labor market attachment in the form of higher informality and more likely transitions out of employment. This finding has important policy implications, indicating that means-tested cash transfers might be inadequate to cover individuals with high risk of losing their employment and falling into poverty if they are targeted too narrowly. This paper proceeds in the following way. The next section discusses the data and methodology used. Section 3 presents the main results. Heterogeneity analyses based on employment transition, type of employment, education, race, and marital status are shown in section 4. Section 5 concludes. 2 DATA AND METHODS 2.1 Data sources We use data from the PNADC, Brazil’s flagship national household survey, conducted since 2012 by the Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e Estatística – IBGE).1 The PNADC is a nationally representative rotating panel survey that collects information on demographics, education, the labor market, and other topics. Households are interviewed once per quarter for five consecutive 1. PNADC replaced two previous surveys: the annual National Household Sample Survey (Pesquisa Nacional por Amostra de Domicílios – PNAD), a nationally representative, multi-purpose, cross-sectional household survey, and the Monthly Employment Survey (Pesquisa Mensal do Emprego – PME), a labor market survey with a rotating panel that covered only six metropolitan regions. Both surveys coexisted with the PNADC until 2015 and 2016, respectively, when they were discontinued. DISCUSSION PAPER DISCUSSION PAPER 15 3106 We also consider labor market volatility, which includes individuals with zero earnings in one or both periods. We differentiate between these two groups in figure 2. The solid lines refer only to individuals with earnings in at least one period, while the dashed lines include individuals with zero earnings in both. FIGURE 2 Labor market volatility – Brazil (2013-2023) 0.0 0.2 0.4 0.6 0.8 1.0 1.2 2013.1 2014.1 2015.1 2016.1 2017.1 2018.1 2019.1 2020.1 2021.1 2022.1 2023.1 2024.1 Period Variance of arc percent changes Men, zero earnings in one period Men, zero earnings in both periods Women, zero earnings in one period Women, zero earnings in both periods Authors’ elaboration. Obs.: Labor market volatility is measured using equation (1) for men and women separately between 2013 and 2023, using data from PNADC. Solid lines include individuals with at least one period of positive earnings. The dashed lines include individuals with zero earnings in both periods. We adjust volatility by age (quadratic) and quarter fixed effects and trim the bottom and top 1% positive earnings each quarter. Labels on the horizontal axes correspond to the final year of each panel cohort. Labor market volatility significantly surpasses earnings volatility for both genders in Brazil. The volatility for men was around 0.8 at the beginning of the period and began to increase during 2015. It peaked at the onset of the pandemic, which inaugurated a period of rapid fluctuations with swift declines and rises, presumably mirroring changes in employment rates, a topic we will explore further when decomposing earnings volatility across job transitions. In the last quarter of 2023, labor market volatility reached the same levels as at the beginning of the period. Similar to the findings in Ziliak, Hokayem and Bollinger (2023) for the United States, labor market volatility in Brazil exhibits a more pronounced countercyclical pattern compared to earnings volatility. Specifically, DISCUSSION PAPER 16 3106 including individuals with no earnings results in a stronger increase in volatility during economic downturns than when we exclude this group from our analysis. Contrary to earnings volatility, labor market volatility is higher for women than men in Brazil, a consequence of gender differences in labor market participation. There was a marked downward trend for women at the beginning of the period, probably reflecting this group’s increasing workforce participation rates. This downward trend continued after the 2014-2016 recession, while labor market volatility for men rose sharply. Both groups experienced a period of higher instability during the pandemic. By the end of 2023, labor market volatility for men was around the same level as in 2013, but women had a modest drop. Labor market volatility is similar for men and women if we include individuals with zero earnings in both periods. Because they have zero arc percent changes, volatility decreases when we include these individuals in the analysis. This is especially true for women, as their participation in the workforce is lower than men. Nevertheless, the overall pattern remains consistent regardless of the inclusion of these individuals in our analysis. Volatility declined between 2013 and 2015, then rose amidst the recession, plateauing during the late 2010s, followed by sudden jumps during the covid-19 pandemic. Such contrasting turns did not entail much cumulative change when we compare 2023 to 2013. Interestingly, women showed slightly higher volatility at the beginning of the period, but male volatility rose faster during the recession. Similarly, we observe a faster growth in earnings volatility for men if we consider only individuals with zero earnings in one period. Ziliak, Hokayem and Bollinger (2023) report much lower estimates of labor market volatility in the United States, which reached 0.4 in the aftermath of the Great Recession for men, much lower than 0.65, the lowest observed value in Brazil at the beginning of 2015. Still, the same caveats mentioned above apply to labor market volatility. Cappellari and Jenkins (2014) report estimates that suggest that volatility levels do not change significantly when imputed earnings are included. In both cases, labor market volatility in the United Kingdom ranges from 0.2 to 0.4, much lower than in Brazil. In any case, direct comparisons with figures for the United States are further complicated because we use monthly rather than annual earnings. By default, the share of individuals with zero earnings in both periods tends to be higher in our data. We conduct several robustness checks on these main results, all available upon request. The results do not change if we do not adjust for age. Trimming the bottom and top 5% of earnings marginally reduces volatility, while no trimming marginally increases it, but there are no significant changes in the results, especially in trends. Using inverse DISCUSSION PAPER DISCUSSION PAPER 17 3106 probability weighting to account for attrition increases slightly overall volatility, especially during the pandemic, but barely affects the main results. Changes in sample selection and the definition of earnings are more consequential, but the main results remain. Restricting the sample to workers with active participation in the workforce in both periods significantly reduces labor market volatility for men and women. Still, the effects on women are larger. Volatility becomes smaller for women than for men during most years in this scenario, dropping to around 0.5 at the beginning of the period, then rising to 0.7 and returning to 0.5 at the end of 2023. In any case, the countercyclical pattern is still noticeable. Either including one or two periods with zero earnings makes almost no difference if we exclude individuals with no participation in the workforce in at least one period. Earnings volatility is not affected because it requires participation in the labor market in both periods. Moreover, it is worth noting that the restriction to those participating in the labor market brings estimates of labor market volatility in Brazil closer to that observed in the United States, especially for women. The most significant deviation from our main results occurs when we change the definition of earnings and analyze usual rather than effective earnings. Volatility levels drop considerably, especially when we consider only individuals with positive earnings in both periods. Albeit not entirely unexpected, this discrepancy warrants further investigation, as it suggests workers are quite accurate at estimating their average long-run earnings even when facing considerable short-run fluctuations. This should be straightforward for formal wage workers, and indeed the PNADC questionnaire is designed to remind respondents to report effective wages net of bonuses and fines. Still, we would expect this task to be considerably more difficult for informal and self-employed workers. Finally, we also evaluate the effect of excluding self-employed workers from our estimates, as is done in some studies (Shin and Solon, 2011; Ziliak, Hardy and Bollinger, 2011; Ziliak, Hokayem and Bollinger, 2023). In this case, all volatility measures decline in magnitude, but the countercyclical trends do not change. The reduction in earnings volatility is larger for men than for women, and they become similar in magnitude throughout the period. This suggests that self-employment is more important for men as a source of volatility, something we explore in the next section. Labor market volatility also becomes smaller for men and women. If we consider only one period with zero earnings, the reduction in volatility is higher for women, which might suggest that self-employment is a more temporary placement for them. When zero earnings in both periods are considered, the reduction in volatility is similar between the two genders, and they continue to move together throughout the period. DISCUSSION PAPER 18 3106 3.2 Distribution of arc percent changes To better understand earnings volatility in Brazil, we analyze selected quantiles of the distribution of the arc percent changes for men and women, including and excluding zero earnings. Figure 3 shows a stable pattern throughout the period for both groups. About 90% of earnings innovations are within the [−1,1] interval, and around 50% are close to zero. In other words, 10% of the workforce typically experiences truly large year-over-year earnings fluctuations. As has already been characterized elsewhere and in Brazil, these earnings innovations are far from following a normal distribution (Gomes, Iachan and Santos, 2020; Guvenen et al., 2021; De Nardi et al., 2021). Deviations from this pattern occurred mainly during the pandemic. There was a brief spike in the left (negative) tail of the distribution of arc percent changes in 2020 and an equally short-lived uptick in the right (positive) tail in 2021. The distribution of arc percent changes is slightly more compressed for women, but the spikes observed during the pandemic were larger. FIGURE 3 Quantiles of the arc percent change among workers with positive earnings – Brazil a) Men b) Women 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Period Arc percentage change P5 P10 P25 P50 P75 P90 P95 Authors’ elaboration. Obs.: These figures plot the 5th, 10th, 25th, 50th, 75th, 90th, and 95th percentiles of the arc percent change for positive earnings, for men and women separately between 2013 and 2023, using data from PNADC. We adjust arc percent changes by age (quadratic) and quarter fixed effects and trim the bottom and top 1% positive earnings each quarter. Labels on the horizontal axes correspond to the final year of each panel cohort. DISCUSSION PAPER DISCUSSION PAPER 19 3106 Figure 4 displays the distribution of arc percent changes when we also include individuals with zero earnings in both periods. In this case, the distribution of innovations has much heavier tails. For both men and women, the top and bottom 5% of the distribution reached the maximum of ±2 throughout almost the entire period. During the pandemic, shocks leading to no earnings (hence, an arc percent change of -2) occurred for at least 10% of men and women. We can also observe that the interquartile range (p25-p75) is more compressed for women than men. FIGURE 4 Quantiles of the arc percent change including individuals with zero earnings in both periods – Brazil a) Men b) Women 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Period Arc percentage change P5 P10 P25 P50 P75 P90 P95 Authors’ elaboration. Obs.: These figures plot the 5th, 10th, 25th, 50th, 75th, 90th, and 95th percentiles of the arc percent change for all individuals, for men and women separately between 2013 and 2023, using data from PNADC. The sample includes individuals with zero earnings in both periods. We adjust arc percent changes by age (quadratic) and quarter fixed effects and trim the bottom and top 1% positive earnings each quarter. Labels on the horizontal axes correspond to the final year of each panel cohort. The results shown in figures 3 and 4 are more extreme than the estimates presented by Shin and Solon (2011) and Cappellari and Jenkins (2014). The latter report values around ± 0.5 for the P5 and P95 quantiles of changes in positive earnings in Britain, both for men and women, representing arc percent changes of 50% in earnings. Similar differences occur when we also include individuals with zero earnings. In Brazil, P10 DISCUSSION PAPER 20 3106 and P90 were around ± 1.0 most of the time, meanings that nearly 20% of the population experience arc percent changes larger than 100%, whereas this almost never happens in the United States or the United Kingdom. These contrasts show that the higher earnings volatility in Brazil does not arise only from more likely changes to and from nonemployment but also due to changes in earnings. 4 HETEROGENEITY ANALYSES 4.1 Volatility by income level We begin by investigating differences in volatility levels by earnings. We classify workers by ventiles of earnings in the first period, while adding an extra group for all individuals with zero earnings in the first period (percentile 0). We measure volatility in each group as the share of individuals with arc percent changes larger than 50% in either direction. Figure 5 shows the results for the aggregated samples of 2017 and 2018, for men and women separately.5 For both groups, we observe higher volatility in the lower tail of the earnings distribution, plateauing around the 30th percentile, with a slight upward trend as we move up the distribution. Overall, around 20% of the sample experienced arc percent changes larger than 50%.6 Results are qualitatively the same if we consider other periods and thresholds. 5. We present results for 2017 and 2018 to avoid noise introduced by the covid-19 pandemic, but results are qualitatively similar for other periods, as noted. 6. Nearly 20% of individuals with zero earnings get a job in the following year, hence their arc percent change is 200%. If we restrict the sample to individuals with zero earnings in one period only, then all of those with zero earnings in the first period will necessarily have positive earnings one year later. DISCUSSION PAPER DISCUSSION PAPER 21 3106 FIGURE 5 Share of individuals with arc percent changes higher than 50% by earnings in the first period – Brazil (2017 and 2018) a) Men b) Women 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Percentile Share of individuals (%) Positive earnings Including zero earnings in one period Including zero earnings in both periods Authors’ elaboration. Obs.: Each line shows the share of individuals with arc percent changes above 50%, negative or positive. We consider three groups: those with positive earnings in both periods; also including those with zero earnings in one period; and also including those with zero earnings in both periods. Individuals with positive in the first are grouped into 20 equally-size bins of approximately 5% of the weighted sample, while those with zero earnings are placed in the “zero” bin. We also investigate the volatility of earnings by analyzing the dispersion of arc percent changes for the same group of individuals using boxplots (figure 6). In this case, we only plot the results for measures including individuals with positive earnings and also individuals with zero earnings in one period. Again, we observe a much larger dispersion in the arc percent change among low-wage earners, for men and women alike. The minimum level of dispersion is observed between the 30th and 40th percentile. This is the position in the earnings distribution where the minimum wage becomes binding. After this period, dispersion increases, but to moderate levels when compared to the dispersion observed below the 30th percentile. The patterns observed for 2017 and 2018 do not change when compared with 2013 and 2014 (results not shown). DISCUSSION PAPER 22 3106 FIGURE 6 Distribution of arc percent changes – Brazil (2017 and 2018) a) Men b) Women 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95100 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95100 −2 −1 0 1 2 Percentile Arc percent changes Positive earnings Including zero earnings in one period Authors’ elaboration. Obs.: Boxes plot the median, interquantile range, and the whiskers correspond to 1.5 times the interquantile range. We consider individuals with positive earnings and zero earnings in one period. Individuals with positive in the first are grouped into 20 equally-size bins of approximately 5% of the weighted sample. 4.2 Employment-nonemployment transitions We follow Ziliak, Hardy and Bollinger (2011) and Cappellari and Jenkins (2014) and decompose the variance of arc percent change based on employment transitions. We classify workers into four groups: employed in both periods (11), nonemployed in both periods (00), employed in the first period but nonemployed in the second (10), and nonemployed in the first but employed in the second (01). Figure 7 shows the results of the decomposition in equation (2) for men. Panel (a) shows the total variance of arc percent changes, as well as how much each component contributes to it. Transitions into and out of employment and , respectively) account for the largest share of volatility, as well as for most changes in the period. These changes include the increase in volatility after 2015 and the fluctuations during the pandemics. Variance arising from those who remained employed accounts for less than 25% of total volatility throughout most of the period, even though most men are always employed – panel (b). DISCUSSION PAPER DISCUSSION PAPER 23 3106 FIGURE 7 Decomposition of labor market volatility for men – Brazil 0.0 0.2 0.4 0.6 0.8 1.0 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 Period Variance Total P01(2-M)2 P11V11 2 P10(2+M) P00M2 2 P11(M11-M) a) Variance decomposition 0 10 20 30 40 50 60 70 80 90 100 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 Period Percentage P01 P11 P10 P00 100M 11 b) Employment share Authors’ elaboration. Obs.: Panel (a) shows the total variance of arc percent changes and the respective components estimated using the decomposition in equation (2). Panel (b) shows the share of each employment transition group in each period. The groups are: employed in both periods (11), nonemployed in both periods (00), or transitioned from employment to non-employment (10) and from nonemployment to employment (01). We restrict the sample to men only. The sample includes individuals with zero earnings in both periods. We adjust arc percent changes by age (quadratic) and quarter fixed effects and trim the bottom and top 1% positive earnings each quarter. Labels on the horizontal axes correspond to the final year of each panel cohort. These results partially contrast with those found by Ziliak, Hardy and Bollinger (2011) and Cappellari and Jenkins (2014) for the United States and the United Kingdom, respectively. The former shows that volatility among continuously employed workers accounts for the majority of labor market volatility in the United States, especially up to the 1990s. The latter find that transitions out of and into employment account for a larger share of labor market volatility in the United Kingdom, but their importance decreased in more recent periods and converged to levels similar to those of continuous workers. These differences between Brazil and the United States and the United Kingdom are largely explained by the more common transitions out of and into employment in Brazil (panel b). While in the United States and the United Kingdom, between 80% and 90% of male workers remain employed in both periods, in Brazil these figures range from 60% to 80%. Approximately 10% of men transition to and out of employment in Brazil, while these proportions are around 5% in the United States and the United Kingdom. DISCUSSION PAPER 24 3106 Figure 8 shows the same results for women. They are somewhat similar to men if we consider the main contributors to labor market volatility. For women, transitions into and out of employment account for most of the volatility, and their share is larger than that observed for men. The main contrast is for participation in the labor market. The percentage of women employed in both periods is much lower than that of men, ranging between 40% and 50%. Moreover, nearly 40% of women remain nonemployed in both periods, due to their lower workforce participation rates. Women’s participation in the labor market is higher in the United States and United Kingdom, approximately 60% to 70%, while nonparticipation ranges from 20% to 25% (Ziliak, Hardy and Bollinger, 2011; Cappellari and Jenkins, 2014). Transitions into and out of employment are also higher for Brazilian women in contrast to the United States or the United Kingdom. In Brazil, it is around 10%, while in these countries it is always below 10%. FIGURE 8 Decomposition of labor market volatility for women – Brazil 0.0 0.2 0.4 0.6 0.8 1.0 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 Period Variance a) Variance decomposition 0 10 20 30 40 50 60 70 80 90 100 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 Period Percentage b) Employment share Total P 01 (2-M)2 P11V11 2 P 10 (2+M) P00M2 2 P 11 (M 11 -M) P01 P11 P 10 P00 100M 11 Authors’ elaboration. Obs.: Panel (a) shows the total variance of arc percent changes and the respective components estimated using the decomposition in equation (2). Panel (b) shows the share of each employment transition group in each period. The groups are: employed in both periods (11), nonemployed in both periods (00), or transitioned from employment to non-employment (10) and from nonemployment to employment (01). We restrict the sample to men only. The sample includes individuals with zero earnings in both periods. We adjust arc percent changes by age (quadratic) and quarter fixed effects and trim the bottom and top 1% positive earnings each quarter. Labels on the horizontal axes correspond to the final year of each panel cohort. DISCUSSION PAPER DISCUSSION PAPER 31 3106 4.4 Demographic sub-groups Figure 14 displays the labor market volatility among all workers and by three educational groups: individuals with less than high school, high school graduates, and college graduates. More educated individuals face smaller volatility in their earnings. The variance of the arc percent changes for male college-educated workers is around 0.5 throughout the period, although it became more volatile during the pandemic. Among workers who only completed high school, the corresponding figure was around 0.6 in the early 2010s, then increased over the decade, peaking at around 0.8 during the pandemic. Workers with less than high school started at a higher level (around 0.8), but followed a similar trend. In other words, college-educated workers fare better in income levels (as they typically earn more than less-educated workers in the cross-section) and income stability, especially during economic downturns. FIGURE 14 Labor market volatility by education level – Brazil a) Men b) Women 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Period Labor market volatility Total Less than HS High School College Authors’ elaboration. Obs.: We classify individuals based on their educational attainment. The black solid line presents results for all individuals. Labels on the horizontal axes correspond to the final year of each panel cohort. DISCUSSION PAPER 32 3106 For women, labor market volatility is also lower among more educated workers (figure 14), but differences between high school dropouts and high school graduates are small, although they widened a bit in more recent years. In this case, lower volatility among high school dropouts likely reflects decreasing labor market participation among the least educated women, as there is no labor market volatility for individuals who are never employed. College-educated women experienced labor market volatility around 0.6 for most of the period, while volatility fluctuated between 0.7 and 0.8 for the other educational groups. The relative stability of female labor market volatility is driven by an increase in the component explained by college and high school graduates and a decline in the share accounted by high school dropouts. This is to a large extent the result of a larger share of high school and college graduates in the population. Ziliak, Hardy and Bollinger (2011) also report differences in labor market volatility in the United States between educational groups. Their estimates indicate similar levels of volatility for high school and college graduates, while high school dropouts display larger volatility. This holds for both men and women. Hence, while a high school diploma seems to provide similar protection against earnings risk to a college degree in the United States, in Brazil the labor market volatility of high school graduates is larger compared to college graduates. Brazil is also marked by huge inequalities between White and non-White individuals, and the worse outcomes of non-White workers in the labor market are likely to manifest also in terms of higher earnings volatility. Ziliak, Hardy and Bollinger (2011) find that volatility levels for Black men and women in the United States are larger than their White counterparts, especially before the 2000s. Figure 15 presents estimates for labor market volatility by racial groups. Non-White workers experience more volatility than Whites, for men and women alike. For men, non-White volatility was around 0.7 and 0.8 in the early period, while it was around 0.6 for Whites. After the recession in 2015, volatility increases for both groups, but faster for non-Whites, reaching around 0.9 among them. For Whites, volatility remained below 0.7 before the pandemic. During the pandemic, both racial groups experiment similar swings, with non-White men always displaying higher volatility. DISCUSSION PAPER DISCUSSION PAPER 33 3106 FIGURE 15 Labor market volatility by racial groups – Brazil a) Men b) Women 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 0.0 0.2 0.4 0.6 0.8 1.0 Period Labor market volatility Whites Non-Whites Authors’ elaboration. Obs.: Panel (a) restricts the sample to men and panel (b) to women. The sample includes individuals with zero earnings in both periods. We adjust arc percent changes by age (quadratic) and quarter fixed effects and trim the bottom and top 1% positive earnings each quarter. Labels on the horizontal axes correspond to the final year of each panel cohort. Panel (b) in figure 15 shows that labor market volatility is also higher for non-White than White women, although the racial gap is smaller than for men. Both groups display roughly constant volatility levels prior to the pandemic, with small changes during the 2015 recession. The racial gap in volatility levels narrowed briefly during the pandemic but widened afterward. Finally, we investigate labor market volatility by marital status, as families are risk-pooling organizations that help buffer adverse earnings shocks. Marriage or cohabitation status8 matters because couples may make interdependent decisions to stabilize welfare, as illustrated by the “added worker effect” when spouses either enter the labor force or work longer hours to offset earnings losses resulting from the involuntary unemployment of the primary earner (Western et al., 2012). 8. In this section, we refer to marriage and cohabitation interchangeably, that is, we define couples regardless of the relationship’s legal status. DISCUSSION PAPER 34 3106 Figure 16 compares levels and trends in labor market volatility between groups defined by relationship status. We present volatility measures for individual earnings for singles and married individuals separately and add a series for couples based on the sum of earnings of both partners.9 Results are similar for single and married men, as well as couples. Estimates are noisier for single men – possibly due to smaller sample sizes – but the trend is the same as for married men and couples, with rising volatility between 2015 and the pandemic. Single and married women also have very similar volatility levels and trends. FIGURE 16 Labor market volatility by marital status – Brazil a) Men b) Women 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 2013.1 2015.1 2017.1 2019.1 2021.1 2023.1 0.0 0.2 0.4 0.6 0.8 1.0 Period Labor market volatility Single Married Couples Authors’ elaboration. Obs.: Panel (a) restricts the sample to men and panel (b) to women. The sample includes individuals with zero earnings in both periods. We adjust arc percent changes by age (quadratic) and quarter fixed effects and trim the bottom and top 1% positive earnings each quarter. Labels on the horizontal axes correspond to the final year of each panel cohort. There are some contrasts between our results and analyses for the United States in Ziliak, Hardy and Bollinger (2011). They show that labor market volatility is higher for unmarried men throughout the period they investigated. Unmarried women also display larger earnings fluctuations, but less so. In Brazil, we see that the earnings volatility of married and single men and women are generally similar. 9. By definition, the volatility of couples is the same for both men and women. DISCUSSION PAPER DISCUSSION PAPER 35 3106 5 CONCLUSION In this study, we have provided a comprehensive analysis of earnings volatility in Brazil, a highly unequal developing country. Our findings indicate that earnings volatility in Brazil is significantly higher than in the United States and other advanced economies, a pattern that is exacerbated by the large informal labor market and high labor turnover. In particular, the volatility of male earnings exceeds that of female earnings, diverging from the typical pattern observed in high-income countries, although this result reverses when we incorporate periods of zero earnings in the analysis. In addition, our results corroborate previous findings on the countercyclical nature of earnings volatility, with larger income fluctuations during recessions and the covid-19 pandemic. Our data allowed us not only to document the gross measures of earnings volatility for the whole population but also to explore the nuances introduced by different labor market attachments that are more common in emerging economies. We observed that informal and self-employed workers experience the most significant fluctuations in earnings, suggesting that the nature of informal work contributes substantially to the observed volatility. Moreover, we show how low-wage earners are significantly more exposed to labor market volatility. This is relevant information for the design of public safety nets in developing countries. Furthermore, our heterogeneity analysis shows how earnings volatility varies across educational, racial, and marital lines, revealing less volatility among individuals with higher educational attainment and White individuals, while marital status showed more nuanced effects. REFERENCES AGUIAR, M.; GOPINATH, G. Emerging market business cycles: the cycle is the trend. Journal of Political Economy, v. 115, n. 1, p. 69-102, Feb. 2007. ARABAGE, A. C.; SOUZA, A. P. Wage dynamics and inequality in the Brazilian formal labor market. EconomiA, v. 20, n. 3, p. 153-190, Sept.-Dec. 2019. ARELLANO, M.; BLUNDELL, R.; BONHOMME, S. Earnings and consumption dynamics: a nonlinear panel data framework. Econometrica, v. 85, n. 3, p. 693-734, May 2017. ATTANASIO, O. P.; WEBER. G. Consumption and saving: models of intertemporal allocation and their implications for public policy. Journal of Economic Literature, v. 48, n. 3, p. 693-751, Sept. 2010. AVRAM, S. et al. Household earnings and income volatility in the UK, 2009-2017. The Journal of Economic Inequality, v. 20, n. 2, p. 345-369, 2022. DISCUSSION PAPER 36 3106 BECCARIA, L. et al. Short-term income mobility in Latin America in the 2000s: intensity and characteristics. Socio-Economic Review, v. 20, n. 3, p. 1039-1067, July 2022. BOLLINGER, C. R. et al. Trouble in the tails? What we know about earnings nonresponse 30 years after Lillard, Smith, and Welch. Journal of Political Economy, v. 127, n. 5, p. 2143-2185, Oct. 2019. BRITTO, D. G. et al. Intergenerationalmobility in the land of inequality. Bonn: IZA, Sept. 2022. (Discussion Paper, n. 15611). BUSCH, C. et al. Skewed idiosyncratic income risk over the business cycle: sources and insurance. American Economic Journal: Macroeconomics, v. 14, n. 2, p. 207-242, Apr. 2022. CAPPELLARI, L.; JENKINS, S. P. Earnings and labour market volatility in Britain, with a transatlantic comparison. Labour Economics, v. 30, p. 201-211, Oct. 2014. CELIK, S. et al. Recent trends in earnings volatility: evidence from survey and administrative data. The BE Journal of Economic Analysis & Policy, v. 12, n. 2, June 2012. CHANCEL, L.; PIKETTY, T. Global income inequality, 1820-2020: the persistence andmutation of extreme inequality. Journal of the European Economic Association, v. 19, n. 6, p. 3025-3062, Dec. 2021. DAHL, M.; DELEIRE, T.; SCHWABISH, J. A. Estimates of year-to-year volatility in earningsand in household incomes from administrative, survey, and matched data. Journal of Human Resources, v. 46, n. 4, p. 750-774, 2011. DE NARDI, M. et al. Family and government insurance: wage, earnings, and income risks in the Netherlands and the U.S. Journal of Public Economics, v. 193, Jan. 2021. DYNAN, K.; ELMENDORF, D.; SICHEL, D. The evolution of household income volatility. The BE Journal of Economic Analysis & Policy, v. 12, n. 2, Dec. 2012. ENGBOM, N. et al. Earnings inequality and dynamics in the presence of informality: the case of Brazil. Quantitative Economics, v. 13, n. 4, p. 1405-1446, Nov. 2022. Retrieved from: https://onlinelibrary.wiley.com/doi/abs/10.3982/QE1855. FIRPO, S.; PORTELLA, A. L. The labor market in Brazil, 2001-2022. IZA World of Labor, Apr. 2024. GERARD, F.; GONZAGA, G. Informal labor and the efficiency cost of social programs: evidence from unemployment insurance in Brazil. American Economic Journal: Economic Policy, v. 13, n. 3, p. 167-206, Aug. 2021. DISCUSSION PAPER DISCUSSION PAPER 37 3106 GOMES, D. B.; IACHAN, F. S.; SANTOS, C. Labor earnings dynamics in a developing economy with a large informal sector. Journal of Economic Dynamics and Control, v. 113, Apr. 2020. GOTTSCHALK, P. et al. The growth of earnings instability in the US labor market. Brookings Papers on Economic Activity, v. 25, n. 2, p. 217-272, 1994. GUVENEN, F.; OZKAN, S.; SONG, J. The nature of countercyclical income risk. Journal of Political Economy, v. 122, n. 3, p. 621-660, June 2014. GUVENEN, F. et al. What do data on millions of U.S. workers reveal about lifecycle earnings dynamics? Econometrica, v. 89, n. 5, p. 2303-2339, Sept. 2021. HECKSHER, M. D. Inaudíveis: quem deixou de ser coberto pelas pesquisas telefônicas e diferenças em relação aos registros administrativos. Brasilia: Ipea, June 2021. (Nota Técnica Disoc, n. 95). HOFFMANN, E. B.; MALACRINO, D. Employment time and the cyclicality of earnings growth. Journal of Public Economics, v. 169, p. 160-171, Jan. 2019. JAPPELLI, T.; PISTAFERRI, L. Does consumption inequality track income inequality in italy? Review of Economic Dynamics, v. 13, n. 1, p. 133-153, Jan. 2010. LA PORTA, R.; SHLEIFER, A. Informality and development. Journal of Economic Perspectives, v. 28, n. 3, p. 109-126, 2014. LI, J.; LA, H. A.; SOLOGON, D. M. Policy, demography, and market income volatility: what shaped income distribution and inequality in Australia between 2002 and 2016? Review of Income and Wealth, v. 67, n. 1, p. 196-221, Mar. 2021. MARTINEZ, T. R.; MELLO, U. Trade shocks and higher-order earnings risk in local labor markets. International Economic Review, v. 65, n. 4, Nov. 2024. Retrieved from: https:// onlinelibrary.wiley.com/doi/10.1111/iere.12703. MEGHIR, C.; PISTAFERRI, L. Earnings, consumption and life cycle choices. In: CARD, D.; ASHENFELTER, O. (Ed.). Handbook of labor economics. Amsterdam: Elsevier, 2011, v. 4, p. 773-854. MESSINA, J.; SILVA, J. Twenty years of wage inequality in Latin America. The World Bank Economic Review, v. 35, n. 1, p. 117-147, Feb. 2021. MOFFITT, R. et al. Reconciling trends in US male earnings volatility: results from survey and administrative data. Journal of Business & Economic Statistics, v. 41, n. 1, p. 1-11, 2023. DISCUSSION PAPER 38 3106 MOFFITT, R.; GOTTSCHALK, P. Trends in the transitory variance of male earnings: methods and evidence. Journal of Human Resources, v. 47, n. 1, p. 204-236, 2012. MOFFITT, R.; ZHANG, S. Income volatility and the PSID: past research and new results. AEA Papers and Proceedings, v. 108, p. 277-280, May 2018. OECD – ORGANISATION FOR ECONOMIC CO-OPERATION AND DEVELOPMENT. Earnings volatility: causes and consequences. In: OECD – ORGANISATION FOR ECONOMIC CO-OPERATION AND DEVELOPMENT. OECD Employment Outlook 2011. Paris: OECD Publishing, 2011. Retrieved from: https://www.oecd.org/en/publications/oecd-employment-outlook-2011_empl_outlook-2011-en.html. OSORIO, R. G. Sobre a montagem e a identificação dos painéis da PNAD Contínua. Mercado de Trabalho: conjuntura e análise, Brasilia, n. 73, p. 69-81, Apr. 2022. SANTOS, A. L.; SOUZA, A. P. Earnings inequality in Brazil: is it permanent or transitory? Brazilian Review of Econometrics, v. 27, n. 2, p. 259-284, Nov. 2007. SHIN, D.; SOLON, G. Trends in men’s earnings volatility: what does the panel study of income dynamics show? Journal of Public Economics, v. 95, n. 7-8, p. 973-982, Aug. 2011. SHORROCKS, A. Income inequality and income mobility. Journal of Economic Theory, v. 19, n. 2, p. 376-393, Dec. 1978. ULYSSEA, G. Informality: causes and consequences for development. Annual Review of Economics, v. 12, p. 525-546, 2020. WESTERN, B. et al. Economic insecurity and social stratification. Annual Review of Sociology, v. 38, p. 341-359, 2012. ZILIAK, J. P.; HARDY, B.; BOLLINGER, C. Earnings volatility in America: evidence from matched CPS. Labour Economics, v. 18, n. 6, p. 742-754, Dec. 2011. ZILIAK, J. P.; HOKAYEM, C.; BOLLINGER, C. R. Trends in earnings volatility using linked administrative and survey data. Journal of Business & Economic Statistics, v. 41, n. 1, p. 12-19, 2023. Ipea – Institute for Applied Economic Research PUBLISHING DEPARTMENT Head of the Publishing Department Aeromilson Trajano de Mesquita Assistants to the Head of the Department Rafael Augusto Ferreira Cardoso Samuel Elias de Souza Supervision Ana Clara Escórcio Xavier Everson da Silva Moura Typesetting Anderson Silva Reis Augusto Lopes dos Santos Borges Cristiano Ferreira de Araújo Daniel Alves Tavares Danielle de Oliveira Ayres Leonardo Hideki Higa Vanessa Vieira Cover design Aline Cristine Torres da Silva Martins Graphic design Aline Cristine Torres da Silva Martins The manuscripts in languages other than Portuguese published herein have not been proofread. Ipea – Brasilia Setor de Edifícios Públicos Sul 702/902, Bloco C Centro Empresarial Brasília 50, Torre B CEP: 70390-025, Asa Sul, Brasilia-DF DISCUSSION PAPER Ipea´s mission To enhance state decision-making and public debate.