Inequality and earnings dynamics in France: National policies and local consequences

Kramarz, Francis; Nimier-David, Elio; Delemotte, Thomas Article Inequality and earnings dynamics in France: National policies and local consequences Quantitative Economics Provided in Cooperation with: The Econometric Society Suggested Citation: Kramarz, Francis; Nimier-David, Elio; Delemotte, Thomas (2022) : Inequality and earnings dynamics in France: National policies and local consequences, Quantitative Economics, ISSN 1759-7331, The Econometric Society, New Haven, CT, Vol. 13, Iss. 4, pp. 1527-1591, https://doi.org/10.3982/QE1876 This Version is available at: https://hdl.handle.net/10419/296318 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-nc/4.0/ Quantitative Economics 13 (2022), 1527–1591 1759-7331/20221527 Inequality and earnings dynamics in France: National policies and local consequences Francis Kramarz Department of Economics, CREST-ENSAE, Institut Polytechnique de Paris Elio Nimier-Dav i d Department of Economics, CREST-ENSAE, Institut Polytechnique de Paris Thomas Delemotte Department of Economics, CREST-ENSAE, Institut Polytechnique de Paris This paper provides new stylized facts about labor earnings inequality and dynamics in France for the period 1991–2016. Using linked employer–employee data, we show that (i) labor inequality in France is low compared to other developed countries and has been decreasing until the financial crisis of 2009 and increasing since then, (ii) women experienced high earnings growth, in particular at the bottom of the distribution, in contrast to the stability observed for men. Both result from a decrease in labor costs at the minimum wage and an increase in the hourly minimum in the aftermath of the 35h workweek policy, (iii) top earnings (top 5 and 1%) grew moderately while very top earnings (top 0.1 and 0.01%) experienced a much higher growth, (iv) inequality between and within cohorts follow the same U-shaped pattern as global inequality: it decreased before 2009 and then increased until 2016, (v) Individual earnings mobility is stable between 1991 and 2016, and very low at the top of the distribution, (vi) the distribution of earnings growth is negatively skewed, leptokurtic, and varies with age. Then, studying earnings dispersion both within and between territories, we document strong differences across cities as well as between urban and rural areas, even after controlling for observable characteristics. We also observe a continuous decrease in earnings inequality between territories. However, a larger inflation in rural territories mitigates this convergence. Finally, we document a strong reduction in Francis Kramarz: [email protected] Elio Nimier-David: [email protected] Thomas Delemotte: [email protected] We thank Serdar Ozkan and Sergio Salgado for their tremendous work on harmonizing the core statistics, as well as Pierre Boyer, Pauline Carry, Bertrand Garbinti, Gautier Lenfant, and seminar participants at the Global Repository of Income Dynamics conferences at Stanford SITE (2019) and held virtually (2020) for helpful comments and suggestions. We also thank two anonymous referees for their very insightful and constructive comments on a previous version of this article. This paper has been conducted in collaboration with the CASD with special thanks to Kamel Gadouche, Marie Vidal, and Raphaëlle Fleureux for their help and support. The access to the French administrative data has been made possible within a secure environment offered by the CASD (Ref. 10.34724/CASD). This paper has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme: Grant Agreement: 741467 FIRMNET. ©2022 The Authors. Licensed under the Creative Commons Attribution-NonCommercial License 4.0. Available at http://qeconomics.org.https://doi.org/10.3982/QE1876 1528 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) inequality within rural and remote territories, again driven by changes at the bottom of the wage distribution. Keywords. Earnings dynamics, top income inequality, mobility, geographic inequality. JEL classification. D31, E24, J31, J38, R00. 1. Introduction “How can you lead a country which has 258 sorts of cheeses?” is an often cited Général de Gaulle’s sentence that summarizes the tensions prevailing in France between a centralized government and its multiple localities with varied specific traits.1These tensions are a constituting feature of French history (see Weber (1976) for a historical perspective on some aspects of such tensions) and they regularly erupt, as evidenced by the “yellow vests” protests, which took place at the end of our study period. With this background in mind, we try to offer a systematic investigation of labor earnings inequality and dynamics over the 1991 to 2016 period, documenting the differences between men and women as well as on the differences between the national and the local levels. This paper is part of the Global Repository of Income Dynamics project, which two main objectives are: (i) to produce harmonized statistics to compare inequality and earnings dynamics between countries and over time, (ii) to zoom in on some specific features of each country, such as informal jobs in Latin America, social benefits in Sweden, and in our case, the role of geography when it interacts with National (central) labor market policies in shaping earnings inequality in France. We may start with the following question in mind: How is France different from other countries in the project? We believe that French labor market is characterized by a steady state with high unemployment and moderate growth. The large share of the GDP dedicated to “social shock absorbers” tends to smooth both ups and downs. Low-wage workers are “protected” by a high minimum wage, which has had adverse effects on their employment (see Kramarz and Philippon (2001)). Between 1991 and 2016, our period of interest, there were major reforms of the labor market institutions, which had a strong impact on inequality: massive increases in the minimum wage in association with the reduction of the workweek to 35 hours. At the same time, there was a strong reduction of the cost of low-wage workers (elimination of employer-paid payroll taxes) to attenuate the impact of the two previous reforms on unemployment. But, maybe counterintuitively, France is the only European country with a clear decrease in inequality over the period, in particular for women. As mentioned above, France has a highly centralized state implementing policies that apply to all territories. These policies are decided within a densely populated capital, Paris, which concentrates most economic and public decision-making centers. However, it also has a huge number of small municipalities with very heterogeneous economic conditions with little leeway in deciding their social and economic fates. As we argue below, it led to local tensions and many localized protests over our sample period. 1Charles de Gaulle, cited in Mignon (1962) Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1529 Because of these social tensions, it is important to understand the disparities in earnings across space as well as their evolution over time. In this paper, we combine employer–employee data on labor earnings to census data on educational attainment to study earnings inequality and dynamics between 1991 and 2016. We find that the strong increase in the minimum wage at the beginning of the 2000s, in the aftermath of the reduction of the workweek to 35h and the suppression of all employer-paid payroll taxes around the minimum wage, translates into a marked increase in bottom percentiles of the earnings distribution (the 10th percentile in particular) in the 2000s. This increase induces a decrease in inequality, defined as the differential between the 90th and the 10th percentile (P90-10), until the financial crisis when the bottom percentile stagnates and the other percentiles tend to increase. Upper-tail inequality explains only a small share of the variations observed as the differential between the 90th and the 50th percentile (P90-50) remains almost constant over the sample period. This extends to other 95th and 99th since their growth is comparable to that of lower ones. However, the very top percentiles (including P99.9 and above) display very high growth. The above changes hide extremely different trends for men versus women. For the former, wage growth is low, in particular at the bottom of the distribution implying increasing inequality since the financial crisis. For the latter, wage growth is higher at all percentiles but particularly so at the bottom. Hence, for females, inequality decreased until the financial crisis and increased only moderately since 2009. It resulted in a decrease in the (unconditional) gender pay gap by a third. Once we decompose earnings into hours worked and hourly wages, the 35-hour working week implied a mechanical reduction in hours for all except for women at the bottom of the wage distribution. The decrease in inequality for women described above is therefore driven both by an increase in hours and a higher increase in hourly wage for the bottom percentiles. Inequality between cohorts, defined as the P90-10 for individuals born a given year, follows a similar pattern. The increase at the bottom of the distribution until the end of the 2000s also caused a decrease in inequality for men and women aged 25 years old. However, the trend has reversed since the financial crisis. We observe increasing inequality both within and between cohorts since then. Turning to earnings changes, we find a moderate increase in the dispersion in individuals’ earnings growth rate for both men and women. Going beyond the first-order and second-order moments, the earnings changes distribution is characterized by negative skewness and high excess kurtosis. Skewness is clearly procyclical (i.e., the distribution is more negatively skewed during recessions), as observed in the U.S. (see Guvenen, Ozkan, and Song (2014)). We also find that wage mobility is pretty low in France, especially when compared to Scandinavian countries (see other articles in this issue, in particular, Friedrich, Laun, and Meghir (2022) for Sweden, Halvorsen, Ozkan, and Salgado (2022)forNorway,andLeth-Petersen and Sæverud (2022) for Denmark). Earnings mobility is very stable over the sample period as in other European countries (see, e.g., Bell, Bloom, and Blundell (2022), Drechsel-Grau, Peichl, Schmid, Schmieder, Walz, and 1530 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Wolter (2022), and Hoffmann, Malacrino, and Pistaferri (2022)). Other features are similar to those observed in the U.S.: the log-density of residual earnings growth has double Pareto tails, negative skewness and excess kurtosis are both increasing with age (see Guvenen, Karahan, Ozkan, and Song (2021)). We finally find huge differences in earnings between cities, especially at the top of the distribution. While bottom percentiles are pretty similar in most cities, top percentiles are much higher in Paris and strongly decrease with city size. These differences are only partially explained by observable characteristics such as age, education, occupations, firm size or the industry structure. These differences have been reduced over the period of interest for bottom percentiles but have increased for top percentiles. We also find large differences between and within territories (i.e., Paris, other centers, suburbs, rural, and remote municipalities). In particular, we observe a strong reduction in inequality within rural and remote municipalities over the sample period and a decrease of the gap in median earnings between these localities and the other urban territories. Related literature: Earnings inequality and mobility in France Verdugo (2014) provides a study of earnings inequality that spans 1950 to 2008, focusing mainly on men employed full-time, full-year in the private sector. Inequality increased between 1950 and 1965 both at the bottom and at the top of the earnings distribution. Then inequality tended to decrease both at the bottom of the wage distribution (the 1968 massive increases in the minimum wage) and at the top starting in the 1990s (the supply of educated individuals massively increased then). Charnoz, Coudin, and Gaini (2014) confirms the latter result, showing in particular the strong decrease in the returns to skills associated with increased supply and the lack of evidence for skilled biased technical change (in stark contrast with other countries). Guillot, Bozio, and Breda (2020) extend the previous analysis until 2015. In an interesting twist, the focus moves to a comparison between inequality in terms of labor cost and in terms of net earnings, the former including both employer and employee-paid payroll taxes when the latter excludes them. The results show that labor cost inequality increased by 8% between 1967 and 2015 while net wage inequalities decreased by 25%. These changes are directly caused by the continuous increase in the minimum wage as well as reforms of the structure of payroll taxes with taxes decreasing for low-wage workers and increasing at the top of the distribution. Godechot (2012) examines the 1975–2007 period and focuses on top percentiles, insisting on its changing composition: a decline in the number of CEOS but an increase in lower-rank managers, top athletes, with managers in the finance industry accounting for almost half of the rise at the top. Recently, Pora and Wilner (2020) decomposes earnings growth into the growth of wages and that of working time. They insist, as we will do here, on the role of working time in earnings volatility, in the aftermath of the various workweek reductions. All these papers used the French administrative data (DADS). A wave of papers, following Piketty’s lead, provides results on income inequality using tax data. Piketty (2003) shows that wage inequality has been very stable in the long run (1901–1998) when the secular decline in income inequality is essentially due to capital income and the two world wars. Furthermore, the top 1% and 10% wage shares are Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1531 stable since 1980. Garbinti, Goupille-Lebret, and Piketty (2018) complement this analysis for the period 1900 to 2014. The article presents “Distributional National Accounts” (using national accounts, tax, and survey data) with a focus on pretax income.2The top 10% share of total income has decreased between 1900 and the beginning of the 1980s, but increased since then. The middle 40% share (i.e., workers with earnings between the median and the 90th percentile) increases until WWII, but is almost stable until mid 1990s, and decreasing since. They also observe a continuous increase in the bottom 50% as well as an increase in the top 0.1% and the top 0.01% between the beginning of the 1980s and 2000s (with a decrease since). As for the labor income, the top 10% share has been decreasing since the mid-1960s when the top 1% decreased until the beginning of the 1990s but increased since then. Interestingly, women increased their presence within the top fractiles of the labor income. The present article does not examine the role of transfers on earnings inequality. However, Bozio, Garbinti, Guillot, Goupille-Lebret, and Piketty (2020) show that most of the long-term (1900–2018) decline in inequality in France is due to the fall in pretax inequality, resulting in low values of inequality as compared to the U.S. Focusing on gross earnings is therefore relevant to study inequality in France. Finally, our paper provides several measures of earnings mobility and studies their evolution since 1991. Buchinsky, Fields, Fougère, and Kramarz (2003) complement our analysis by examining measures of earning mobility based either on ranks in the income distribution or on Francs for periods of 3 years (using DADS between 1967 and 1999). Their results suggest a decrease in mobility over time. Using panel data allowing to track the tax returns of all French tax residents over time, Aghion, Ciornohuz, Gravoueille, and Stantcheva (2019) find a rank-rank correlation of total income of 0.83 between 2011 and 2015. Contrary to our results, they find no large differences between men and women, except at the bottom of the distribution where mobility is higher for men. The remainder of this paper is organized as follows. Section 2provides institutional details on the French labor market, describes the data and provides some descriptive statistics. We then present our main results on earnings inequality and dynamics in Section 3. Finally, Section 4provides complementary statistics on earnings inequality between and within various French territories. Section 5concludes. 2. Institutions and data 2.1 The French labor market institutions Figure 1presents, on the left (A), the unemployment rate in France and in the United States over the period 1985–2019 and, on the right (B), the GDP growth rate for the same countries. As can be seen from the figure, the unemployment rate in France has been staying consistently between 7.5% and 10% over the last 30 years (with one exception, 7%, in 2000). Expansions appear unable to decrease unemployment below this point, and recessions do not seem to have the same effect as, for instance, in the United States where unemployment between a trough and a peak can increase by 5 points when in 2Hence, before all taxes and transfers, except pensions and unemployment insurance. 1532 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure 1. The French business cycle. Note: Figure 1plots against time: (a) the unemployment rate, (b) the growth rate of the GDP, for France and the United States. Source: INSEE and BLS for the unemployment rates. The World Bank for the GDP. France the increase is 2.5 points. GDP growth rates, however, vary mostly in sync (with that of France being lower by one point during expansions). During our analysis period, France has witnessed two recessions, in 1993 and 2008 but one expansion, around 2000 when the US had more expansion years. The inability to decrease the unemployment rate in good times to, say, 5 or 6% as in other countries, together with the inability to generate as much growth as in the US, must be questioned. At least, these two questions will be in the back of our mind when exploring earnings dynamics and inequality. We then provide a brief description of the main institutional features of the French labor market. First, as in multiple other European countries, there are two main types of labor contracts: permanent and temporary. In 2017, 88% of the wage workers have a permanent contract while 87% of the new hires are made under temporary contracts. This rate has steadily increased since 1993 when it represented 76% of the new hires. The duration of these contracts tend to be short and the use of extremely short term contracts has strongly increased over the past decades. In 2017, almost one-third of the temporary contracts lasted only one day. Second, part-time jobs have been increasing for both women and men. In 2016, 30.1% of women and 8.2% of men had part-time jobs. It was respectively 23.4% and 4% in 1990.3 Finally, the participation rate strongly increased since the middle of the 1980s, mostly driven by the oldest age groups and women, as depicted in Appendix Figure B.1. Women’s participation rate managed to increase from the 1990s onward while men’s strongly decreased due to the low participation of older age groups. In particular, preretirement plans in the 1980s pushed down participation of people above 55, especially for men. Recent pension reforms in the 2000s and 2010s had a huge impact and partially reversed this last tendency. 3See “Emploi, chômage, revenus du travail,” INSEE 2020. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1533 For our analysis period (1991–2016), two policies had a huge effect on all labor market outcomes. We start by describing the changes in the minimum wage policy. Then we explain those on the workweek. The debates on minimum wages that took place in the 1990s in the United States tend to obscure why the French case was one of a dramatically high minimum wage with adverse employment effects (Kramarz and Philippon (2001)). For employers at least, the minimum wage by itself is only one part of the story. What really matters is the total labor cost, that is, the wage plus the payroll taxes (see Saez, Schoefer, and Seim (2019) and Huttunen, Pirttila, and Uusitalo (2013) for the employment effects of payroll taxes in Sweden and Finland). These payroll taxes comprise two components: one paid by the worker and one paid by the firm.4 Until the beginning of the 1990s, France was characterized by both a very high minimum wage and extremely high labor costs at the bottom of the wage distribution. It is still characterized by a very high minimum wage since the ratio of the net minimum wage to the median wage is equal to 0.63 in 2015, one of the highest in the OECD. Indeed, as Figure 2shows, the nominal and the real value of the minimum wage increased dramatically, by respectively 100% and 40%, between 1991 and 2016. However, as evidenced by Figure 3, the total labor cost barely budged thanks to the very strong decrease in employers’ contributions in the 1990s and the 2000s. These contributions are now virtually equal to zero. We will see in Section 3.1 that this strong increase in the minimum wage, especially since the middle of the 2000s, had a big impact on earnings inequality. We describe in Appendix Athe various steps of the workweek reduction and the path to 35 hours that took place mostly between 1998 and 2001. Importantly, wage compensation schemes and wage moderation agreements were implemented at the same time Figure 2. Nominal and real minimum wage over time. Note: Figure 2plots against time: (a) the nominal hourly minimum wage, (b) the real hourly minimum wage in France. All statistics are normalized to 0 in 1991. Source: INSEE. 4In the end, obviously everything is paid by the firm but these different payroll taxes may have different destinations. For minimum wage workers, every component is mandated by the law, with no possible trade off except on hours and employment. 1534 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure 3. Labor cost at minimum wage. Note: The labor cost is the sum of the gross wage and of the employer’s contributions. All statistics are normalized to 100 in 1980. Source: “Rapport à la Commission des comptes de la sécurité sociale de juin 2009.” so that monthly wages stayed constant in the short-term and did not increase too rapidly in the longer run. Labor costs for low-wage workers did not increase too strongly thanks to the payroll tax exemptions that were expanded in those years. Until 2005, when all minimum wages were unified, there existed a flurry of SMICs5 depending on the moment the firm reduced the workweek. As depicted in Appendix Figure A.1, the reduction of the workweek was followed by a strong increase in the minimum wage between 2002 and 2005, the highest over our period of interest. As can be seen on Figure 4, hours in France were decreasing at a brisk rate in the years preceding the implementation of the 35 hours workweek. Then they stabilized from 2002 onward. In the U.S. though, hours decreased too at a lower rate, and annual working time today is much higher than in France as it was in the 1980s. We will come back to this question of hours and to its impact on earnings inequality later in our analysis. We also detail in Online Appendix A (Kramarz, Nimier-David, and Delemotte (2022)) the sequence of social movements that took place in France over the last 20 years: from urban riots in 2005, red caps in Brittany, to the yellow vests recently. We attempt to show that each of these movements was an expression of disagreements with Paris, seen as the central administrative power, when it tried to impose a new carbon-style tax on trucks or increase taxes on gasoline to fight global warming. The government, legislating from a city benefiting from excellent infrastructure and well connected to the rest of France and the world, was seen as out of touch by most citizens living in the countryside without much infrastructure (public transport, in particular) and forced to use their cars every day. Paris imposed one-size-fits-all decisions are indeed less-well accepted than they once were. 5Salaire Minimum de Croissance Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1541 Figure 6. Earnings inequality. Note: Using real raw log earnings and the CS+TMax sample, Figure 6plots against time the following variables: (a) Men: P90-10 and 2.56*SD of log income, (b) Women: P90-10 and 2.56*SD of log income, (c) Men: P90-50 and P50-10, (d) Women: P90-50 and P50-10. Shaded areas represent recession years. 2.56*SD corresponds to P90-10 differential for a Gaussian distribution. Data set: Panel DADS. without taking into account neither workers’ nor jobs’ characteristics. First, we observe large differences in earnings between men and women. Mean and median earnings are respectively 43% and 29% higher for men than for women in 1991. We then observe a decrease of more than one-third in the gender pay gap over the period of interest due to the higher growth for women at all percentiles described in Figure 5.27 Using residual earnings yields essentially identical trends in earnings inequality, when controlling either for age or age and education (see Appendix Figures C.3 and C.4). The increase in bottom percentiles is even more pronounced compared to other percentiles when we account for compositional changes. In addition, there is no increase in inequality since the financial crisis once we control for the evolution of the age composition of the workforce and of educational attainment. 27We find a similar decrease in the gender pay gap when using the hourly wage instead of total labor earnings. 1542 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Other measures of inequality yield very similar conclusions. Appendix Figure C.5 displays the evolution of the income share (A) by quintile and (B) for top percentiles. The first quintile increases until 2009 before decreasing while the top quintile declines during the financial crisis and increases since then. Very top percentiles (above the 90th percentile) experience a very similar pattern. The Gini coefficient is almost stable until 2003. It then slightly decreases between 2003 and 2009 before increasing continuously until 2016 (see Appendix Figure C.6). Again, the trends differ by gender. For men, the gini coefficient is almost constant until 2009 and increasing since then. For women, we observe a decrease between 2002 and 2009 followed by a continuous increase. Finally, the distribution of earnings is (unsurprisingly) fat-tailed as we see from Appendix Figures C.7 and C.8. The linear relationship between (log) earnings and the log of the counter cumulative distribution28 suggests that the right-tail of the distribution has a Pareto shape. The strong increase in very top percentiles over time, depicted in Figure 5, translates into a thickening of the right-tail (associated with a decrease in the slope coefficient in absolute value) between 1995 and 2015. Similar results are obtained for the two cut-offs at 1 and 5%. Interestingly, the slope is smaller in absolute value for men; the top tail being thicker for men than for women. To conclude, variations in inequality seem to be only mildly related to the business cycle and, apparently, mainly driven by changes in institutions such as the strong increase in minimum wage over the period. Women, who are more likely to be minimum wage workers, have particularly benefited from these reforms, which translated into a strong reduction in inequality. 3.1.2 Inequality and the reduction of the working week Because hours are available since 1993, we are able to contrast wages, hourly wages, and hours on the subperiod 1993–2016. Hours and hourly wages are computed for people with nonmissing hours for all their jobs in the given year. Appendix Figures D.1 and D.2 plot respectively the annual number of hours worked, as a share of a full-time job, and the hourly wage for various percentiles of the earnings distribution. About hours worked, we observe notable differences between men and women at the bottom of the earnings distribution. Men with earnings close to the 25th percentile work full time (45% of a full time for the 10th percentile) while it is only 2/3 of a full time (resp., 1/3) for women. Looking at the trends, men with low earnings tend to work less over time while hours are increasing for women at the 10th and 25th percentiles. The overall decrease in hours between 1999 and 2002 is extensively discussed below. Turning to hourly wages, we observe that workers with earnings lower or equal to the median tend to earn similar hourly wages. This is especially the case for women where hourly wages are very close to the minimum wage. It results that women’s differences in earnings at the bottom of the distribution are mainly driven by the number of hours worked, not by their hourly wage. The dispersion of hourly wage by income percentiles is higher for men, especially at the top of the earnings distribution. 28The countercumulative distribution is defined as one minus the cumulative distribution function of labor earnings. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1543 We now discuss the impact of the reduction of the working week on labor earnings. Appendix Figure D.3 shows, on the left, the evolution of the number of hours worked for various percentiles of the earnings distribution and, on the right, the evolution of the hourly wage for the same percentiles. The first panel corresponds to the full population when the next two panels are for men and women, respectively. On Appendix Figure D.3a, we clearly see the impact of the reduction of the workweek for those percentiles at or above the median that worked 39 hours in 1993 and moved to 35 hours from 2002 on. The behavior of hours for the deciles below median are more erratic, but also follow this downward tendency. Contrasting men and women clearly shows that men with income below the median tend to work less hours in the aftermath of the workweek reduction, whereas it is the opposite for women with income below the median. The hourly wages increase very clearly and consistently across the earnings distribution over the period. For men, the increase is smaller than for women and starts because of the workweek reduction whereas women increase their hourly wages before, at, and after the workweek reduction. This results in a 20% increase in hourly wages for men but in a 30% increase for women between 1993 and 2016. Interestingly, the 10th percentile for women closely follows the evolution of the minimum wage. Hence, the moderate increase in labor earnings, especially for men, described in the previous section results from a 10% reduction in hours compensated by a stronger increase in hourly wage. Differences between men and women can be explained by both a higher increase in hourly wage and a smaller decrease in hours for women, especially at the bottom of the earnings distribution. Finally, we turn to the impact of hours and hourly wages on inequality. To do so, we decompose the variance of the log labor earnings (yit ) into the variance of the log number of hours worked (hit), the variance of the log hourly earnings (wit), and the covariance between the two using the following formula: Varlog(yit )=Varlog(hit )+Varlog(wit )+2Covlog(hit ),log(wit ). Appendix Figure D.4 plots the decomposition for men and women separately. The figures confirm the much lower variance in (log) hourly wages but higher variance in (log) hours for women. The decrease in the variance of earnings described in the previous section comes from a clear decrease of the variance of (log) hours, especially for women, a smaller decrease in (log) hourly earnings, together with a moderate increase in the covariance between hours and hourly wages. 3.1.3 Inequality between cohorts In this section, we study how inequality compares between cohorts and how it evolves over the life cycle of various cohorts. Indeed, the trends described in the previous two sections can be due to variations in initial conditions at labor market entry and/or variations in earnings dispersion over the life cycle. Figure 7 plots the P90-50 and the P50-10 every year for workers at age 25. Figures 7aand7bshow that the distributions are very similar for men and women at age 25, contrary to what was found in Figure 6. Indeed, lower-tail inequality is smaller and upper-tail inequality is higher when workers of all ages (25–55) are included when compared to what we see 1544 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure 7. Initial earnings inequality (at age 25). Note: Using real raw log earnings and the CS+TMax sample, Figure 7plots against time the following variables: (a) Men: P90-50 and P50-10 at age 25, (b) Women: P90-50 and P50-10 at age 25. Shaded areas represent recession years. Data set: Panel DADS. at age 25. This pattern is much more pronounced for men than for women. As in Section 3.1.1, we observe a decrease in inequality over time for workers entering the labor market, essentially driven by the bottom P50-10 (in particular for men). We also observe a moderate increase in the P90-50 for men since the financial crisis. We then study how the P90-10 differential evolved over the life cycle for several cohorts of workers. Figure 8shows that inequality was much higher at age 25 than at age 30 in the 1990s. This gap has been decreasing over time and has become small in the late 2000s, mainly because of the reduction in inequality at age 25. Furthermore, the life- Figure 8. Life-cycle inequality over cohorts. Note: Using real raw log earnings and the CS+TMax sample, Figure 8plots against time the following variables: (a) Men: P90-10 over the life cycle for all cohorts available, (b) Women: P90-10 over the life cycle for all cohorts available. The dashed grey lines plot the P90-10 for people aged 25, 30, and 35 in the year given in the x-axis. The four remaining lines plot the evolution of the P90-10 over time for specific cohorts. Data set: Panel DADS. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1545 cycle inequality of cohorts is decreasing for women and is U-shaped for men. Similar to what was found previously, within cohorts inequality has decreased until the financial crisis and increased since then for male workers (resp., stagnate for female workers). 3.2 Earnings changes 3.2.1 The distribution of earnings changes over time We study here the evolution over the sample period of the distribution of earnings changes after controlling for age. The results are presented in Figures 9for the 1-year growth and in Appendix Figure E.1 for the 5-year growth. We use the P90-10 as a measure of earnings growth volatility and decompose it into right tail dispersion (P90-50) and left tail dispersion (P50-10) separately for men and women. First, we find that the cross-sectional dispersion of the 1-year growth rate of residualized earnings is higher for women than for men. This gap is likely due, at least in part, to maternity leave and to the higher probability for women to work in part-time jobs. Second, the income volatility is slightly increasing over time, in particular for men. Using the above mentioned decomposition, we observe that most of the growth is due to the increase of the right-tail dispersion while the lefttail dispersion is almost constant over the period. Third, left- and right-tail dispersion tend to move in opposite directions over the business cycle. The P50-10 increases strongly before recessions while the P90-50 declines.29 As a result, recessions are associated with a higher probability of large downward movements and a smaller probability of large upward movements. Nevertheless, variations in left-tail dispersion are usually higher than variations in right-tail dispersion implying some countercyclicality of the P90-10 (i.e., more dispersion in periods of Figure 9. Dispersion of 1-year log earnings changes. Note: Using residual 1-year earnings changes (controlling for age) and the LX sample, Figure 9plots against time the following variables: (a) Men: P90-50 and P50-10, (b) Women: P90-50 and P50-10. Shaded areas represent recession years. Data set: Panel DADS. 29The 1 year delay between the variations of the P50-10 or the P90-50 and GDP growth is due to the fact that earnings growth is computed forward while GDP growth is computed backward. 1546 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) low GDP growth). Turning to the 5-year growth rate, we observe a moderate increase for men over the period and an inverted-U shape for women. In addition, comovements observed between left and right tail dispersion are even stronger using the 5-year growth rate. We then discuss the shape of the earnings growth distribution. Appendix Figures G.1 and G.2 display graphically (and very clearly) that the distribution of both 1 and 5-year residual earnings changes are very far from a normal distribution.30 The distribution is negatively skewed: the left tail of the distribution is longer than the right tail. As a result, the “bad” shocks (below the mean) are larger in absolute terms than the “good” shocks. In addition, the distribution is leptokurtic since the coefficient of kurtosis is much higher than 3, the level observed for a normal distribution. Hence, workers experience much more small and extreme changes than what would imply a normal distribution, and much less middling ones. Furthermore, dispersion is larger for women but kurtosis is larger for men. Very similar conclusions can be drawn from the study of the 5-year earnings growth distribution, except that the kurtosis, albeit high, is smaller for both men and women. Appendix Figures G.3 and G.4 plot the log-densities of 1 and 5-year earnings growth. We observe that the distribution has Pareto tails at the top and bottom of the distribution. This result is consistent with previous papers for the United States (see Guvenen et al. (2021)). In Table G.1, we report the main coefficients (slope, skewness, and kurtosis for the whole population) over several years. We observe a clear thinning of the (two) tails over time as evidenced by the increase of the coefficients in absolute value. The message is similar for both the 1-year and the 5-year growth measures. 3.2.2 Higher order moments of the distribution In this section, we study the second-, third-, and fourth-order moments of the distribution of earnings growth. Figure 10 depicts the evolution of (A) the Kelley skewness and (B) the excess Crow–Siddiqui kurtosis of the 1-year changes of residualized earnings over time. These two statistics provide measures of skewness and kurtosis based on percentiles, hence robust to extreme values. We find that skewness is procyclical: negative during recessions and positive in expansions. In addition, there is much less variation for women than for men. As for the kurtosis, no clear pattern emerges; it is low during our first recession but high during the second one and the nature of its fluctuations is hard to interpret.31 In Figure 11, we condition our various statistics using a measure of “permanent earnings.”32 We look at the P90-10 differential in the first panel, at the Kelley skewness in the second, and at the Crow–Siddiqui kurtosis in the third. Each time results for men are presented in the left column when those for women in the right one. Hence, Figure 11 allows to study uncertainty that workers with same gender, age, and permanent earnings face. 30The red dotted line plots the density of a normal distribution with similar variance as in our data. 31Results for the 5-year growth rate are displayed in Appendix Figure E.2. 32Permanent earnings are defined as the average of nonmissing earnings between t and t-2 net of age and year effects separately by gender. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1547 Figure 10. Skewness and kurtosis of 1-year log earnings changes. Note: Using residual 1-year earnings changes (controlling for age) and the LX sample, Figure 10 plots against time the following variables: (a) Men and Women: Kelley skewness, (b) Men and Women: Excess Crow–Siddiqui kurtosis calculated as P97.5−P2.5 P75−P25 −2.91 where the first term is the Crow–Siddiqui measure of kurtosis and 2.91 corresponds to the value of this measure for Normal distribution. Shaded areas represent recession years. Data set: Panel DADS. Our measure of dispersion (P90-10) follows a clear U-shaped pattern for men. It is highest at the bottom and very top of the permanent income distribution for both men and women. It is also mildly decreasing with age, especially for women at a young age. The measure of skewness is essentially flat for male, except at the very top. The pattern is identical at all ages. By contrast, skewness is very strongly negative for women in the younger age group (25–34) for almost all quantiles of the permanent income distribution. However, it is rather flat for the two other (older) age groups (35 and above). Finally, the Crow–Siddiqui kurtosis has an inverted U-shape across the quantiles of the permanent income distribution with a maximum below the median. This pattern holds for both men and women as well as for all age groups. We compute similar statistics using now the 5-year earnings changes. Results are presented in Appendix Figure E.3. The P90-10 differential looks pretty similar to that of the 1-year change. However, Kelley skewness is much more negative and much less flat (U-shaped in fact) than its 1-year equivalent. Finally, the kurtosis has again an inverted U-shape. However, the magnitude of the kurtosis is similar for men but smaller for women for the 5-year change than it is for the 1-year one. In Appendix Figures E.4 and E.5, we present (standardized) measures of the moments of the earnings changes using their standard deviation, their skewness, and their kurtosis. These measures are computed on the whole distribution rather than using percentiles. Hence, they are more likely to be sensitive to extreme values in contrast to those above. Nevertheless, these standardized measures yield results that are pretty consistent with the P90-10 differential, the Kelley skewness, and the excess Crow–Siddiqui kurtosis. Indeed, using either the P90-10 or the standard deviation give a similar U-shape pattern. Similarly, both skewness and Kelley skewness are negative for the 1-year and the 5-year 1548 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure 11. Dispersion, Kelley skewness, and excess Crow–Siddiqui kurtosis of 1-year log earnings changes by age group. Note: Using residual 1-year earnings changes and the H sample over the period 1994–2015, Figure 11 plots against permanent income quantile groups the following variables for the 3 age groups: (a) Men: P90-10, (b) Women: P90-10, (c) Men: Kelley Skewness, (d) Women: Kelley Skewness, (e) Men: Excess Crow–Siddiqui kurtosis, (f) Women: Excess Crow–Siddiqui kurtosis. Excess Crow–Siddiqui kurtosis calculated as P97.5−P2.5 P75−P25 −2.91 where the first term is the Crow–Siddiqui measure of kurtosis and 2.91 corresponds to the value of this measure for Normal distribution. Data set: Panel DADS. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1549 growth. Finally, both measures of our fourth-order moment display high excess kurtosis, the magnitude being even higher using kurtosis directly (as expected). 3.3 Earnings mobility We now focus on individual-level earnings mobility measured over 5 to 10 years, as well as its change over the sample period. Our first results are presented in Figure 12. They give the 10-year mobility rate, measured as the mean percentile in the “10 years after” distribution as a function of the position in the initial permanent earnings distribution.33 The use of a measure of permanent income allows us to mitigate concerns about a mechanical relationship due to a reversion to the mean. The 45-degree (dashed black) line shows what would be observed in a world without mobility. Results for men are presented on the left when those for women are shown on the right. For the 10-year mobility measure, we divide the population into two age groups. As expected, we observe upward mobility at the bottom of the distribution until the 40th percentile. As observed in most mobility studies, we also see that mobility is decreasing with income at the top, most particularly at the very top percentiles. Again, unsurprisingly, mobility is larger for the younger age group as well as for females (less so at the very top percentile though). The Appendix Figure F. 1 that focuses on 5-year mobility yields essentially similar conclusions. Figure 13 presents these 10-year mobility measures at two points in time, 1995 and 2005 (the starting years). Mobility is constant over time for both men and women. The Figure 12. Evolution of 10-year mobility over the life cycle. Note: Using permanent income and the H sample over the period 1993–2006, Figure 12 plots the average rank-rank mobility for two age groups separately for men and women over the period 1993–2006. The 45-degree (dashed black) line corresponds to the case of perfect immobility: workers stay on average at the same percentile after 10 years. Data set: Panel DADS. 33In this section, we use a slightly modified version of the permanent earnings. We keep individuals with earnings above the minimum threshold for at least 1 year (two in the previous definition). We hence also include in our sample workers with little attachment to the labor market. 1550 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure 13. Evolution of 10-year mobility over time. Note: Using permanent income and the Hsample,Figure13 plots the average rank-rank mobility in 1995 and 2005 separately for men and women. The 45-degree (dashed black) line corresponds to the case of perfect immobility: workers stay on average at the same percentile after 10 years. Data set: Panel DADS. Appendix Figure F. 2 shows a small increase in 5-year mobility for both men and women, including at the very top. In order to study more systematically the changes in (relative) mobility over time, we perform a rank-rank correlation analysis using the following regression: Rperm,i,t+k=β0+β1×Rperm,i,t+i,t, where Rperm,i,tis the rank in the permanent earnings distribution of individual iin year t. The coefficient β1provides the degree of relative labor earnings mobility. The Appendix Figure F. 3 plots coefficient β1for each sample year. The estimated value over the period is 0.8 (se: 0.0002). In addition, the coefficient barely moves over the period of interest with small ups around recessions and downs in periods of higher growth. We repeat the exercise by age group and by gender (see Appendix Figure F. 4 ). Consistent with our previous results, relative mobility is lower for older workers and higher for women. In anticipation of our analysis of French territories (just below), Appendix Figure F. 5 shows that relative earnings mobility is larger in Paris than in other territories in 1995, but all such mobility numbers converge across territories (decreasing markedly in Paris and slightly increasing everywhere else). 4. Geographic disparities in France In this section, we examine the spatial dimension of inequality. We start with an analysis of urban areas, our measure of cities. We then turn to territories, as defined in Section 2.3. 4.1 Inequality between urban areas Figure 14 presents our first assessment of inequality between cities. We rank the 759 urban areas by size (population in 2015), Paris being the first. Then we present vari- Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1557 change of establishment between tand t+1. As expected for a dense territory with more jobs and firms, mobility is highest in Paris’ urban unit. It is lowest in rural areas, again an unsurprising result in a less dense territory. For all territories though, the changes are strictly parallel and procyclical (with a maximum in 2000). Job-to-job mobility steadily decreases from 2001 on in all territories, a potential reflection of equalized opportunities within territories, as we will see in Section 4.4. Employees working full-time37 in t, experience similar trends but have a much lower mobility rate: around 14% in Paris and 9% in other territories. This large gap results from the inclusion in our main sample of many low earners, most likely to experience periods of unemployment and to be hired under very short-term contracts. Using firm-to-firm mobility does not alter the above results (mobility larger in Paris, smaller in rural areas) with between 13% and 20% of workers moving every year. 4.3 Geographic mobility In this subsection, we focus on geographic mobility in complement to our previous analysis of job-to-job mobility. Geographic mobility mostly takes place between municipalities (most likely within a commuting zone) rather than between commuting zones (see Online Appendix Figure C.1). We first describe the mobility between territories (BT-mobility hereafter): moves between, say, Paris and rural areas and their impact on earnings. In line with common perception, mobility from Paris is half as likely (2.5% per year) as from each other territory (5–6% per year in 1993, slightly increasing to 6–8% in 2015). And mobility to Paris, the largest destination in line again with common perception, mostly comes from central municipalities rather than from more remote places (see Online Appendix Figure C.2). Such geographic mobility takes place at a relatively young age. To better characterize those workers moving to different places, we plot the average percentile (rank of permanent earnings) of movers by area of destination (see Online Appendix Figure C.3). Indeed, workers moving to Paris have the largest average percentile (57th in 1993), followed by workers moving to suburban areas (54th in 1993), and other destinations (52th in 1993). The ranking is unchanged over the sample period but the average percentile decreases constantly (to 54th in Paris and 48–50th for other destinations).38 To give a fuller view of mobility, we contrast earnings growth for the stayers and for the movers. Online Appendix Figure C.4 shows the 5-year (log) earnings growth distribution for stayers (in Paris, rural territories, and in Province39). The distribution is very similar in Province and rural territories: leptokurtic with negative skewness. The distribution in Paris is less peaked and exhibits less negative skewness. Online Appendix Figure C.5 shows similar distributions for movers (again between these three origins/destinations). In contrast with stayers, moves to Paris induce a right shift in the 37We define full-time workers as worker working full-time at least 350 days during the year. 38A similar analysis using residual ranks yields similar results. 39To make graphics easier to read, we focus on three territories: the urban unit of Paris, rural territories, and the other territories that we denote “Province.” 1558 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) 5-year (log) earnings growth distribution with the associated thickening of the right tail (upwards mobility) whereas mobility from Paris has the opposite consequence (downwards mobility). For movers to and from destinations other than Paris, the earnings distributions do not shift much, very much similar to those observed for stayers, albeit with a decreased fraction of zero earnings growth (less peaked). 4.4 The decrease in within territories inequality In this section, we study how inequality evolved within the six territories considered. Figures 19 and 20 present changes for various percentiles of the log real earnings distribution, in difference with 1993, for men on the left side, and women on the right side of the figures. Results for Paris, central municipalities, and suburbs are shown in the first of the two figures when those for remote and rural municipalities are given in the second figure. Results are pretty striking. First, wage growth is quite low for men, and even negative for the bottom percentiles, for all territories. By contrast, wage growth is quite strong for women in all territories. The contrast between men and women is even stronger when looking at the bottom percentiles, especially the 10th percentile. Even more strikingly, this wage growth is largest for rural territories, then remote when the more urban territories look quite similar: strong growth at the bottom percentiles (but not as strong as that in rural and remote territories). Paris is once more different: wage growth is largest for women at the top percentile (P90). A direct consequence of these results (confirmed by Appendix Figures H.3 and H.4) is the decrease in inequality for women in all territories, except for Paris, at least until 2009. As seen above, the decrease is much larger in rural and remote territories. These figures also show the moderate decrease in inequality for men until 2009. Inequality increases since then, with inequality being back or even above its 1993 level. Turning to inequality levels, the difference between the 90th and 10th percentiles is much higher in Paris than in other territories where this difference looks pretty similar. However, inequality for men is pretty well ordered: much lower in rural and remote areas, intermediate in suburbs and central municipalities, and extremely high for Paris. The level of inequality is much lower for men in rural territories than it is for women but the two converge at the end of our sample period. Finally, inequality is higher in Paris for men than it is for women. Recent political events in France have placed earnings inequality but also on employment and its evolution in the public sector under the spotlight. In particular, almost always stemming from the recent protests was a demand to increase public services in remote and rural areas. Its origin was the perceived decrease in the supply of state-provided services, to the point of their disappearance, in these territories. As a complement to our results on earnings inequality by territory, Online Appendix Section D provides evidence on the evolution of public services across space and over time, proxied by public employment. We study its changes in the aggregate as well as Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1559 Figure 19. Change of percentiles of the log real earnings distributions by territory. Note: Using real raw log earnings separately for men and women, Figure 19 plots against time the P10, P25, P50, P75, P90 for: (a)–(b) the urban unit of Paris, (c)–(d) central municipalities, and (e)–(f) the suburbs. All statistics are normalized to 0 in the first available year. Territories and Paris are defined using urban units. Shaded areas represent recession years. Data set: Panel DADS. 1560 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure 20. Change of percentiles of the log real earnings distributions by territory (continuation). Note: Using real raw log earnings separately for men and women, Figure 20 plots against time the P10, P25, P50, P75, P90 for: (a)–(b) remote municipalities, (c)–(d) rural municipalities. All statistics are normalized to 0 in the first available year. Territories and Paris are defined using urban units. Shaded areas represent recession years. Data set: Panel DADS. for the three types of public sector employment: State civil servants, local civil servants, and hospital civil servants. Furthermore, we contrast these changes across territories. 4.5 Paris: The center, the suburbs, and its outskirts Because the Paris urban area is so large, comprising Paris municipality per se, Paris’s suburbs, and a large set of land that includes rural municipalities, we present in this subsection results for these three zones. These areas are presented on Figure 21.The central area (in dark) shows the municipality of Paris. The surrounding area (in mediumdark) corresponds to Paris’s urban unit (as defined in Section 2.3), a unit that includes 408 municipalities. Finally, the lighter area corresponds to the rest of Paris’s urban area and is composed of 1342 municipalities. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1561 Figure 21. Paris and its surroundings. Note: Figure 21 plots Paris and its suburbs. Numbers in parenthesis correspond to the number of municipalities. Figure 22. Paris center versus its surroundings. Note: Using raw log earnings for men and women, Figure 22 plots against time: (a) the median, (b) the mean, labor earnings differential between Paris and its suburbs. Paris corresponds to the municipality of Paris. Data set: panel DADS. 1562 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure 22 plots the median (A) and the mean (B) log earnings differential between the municipality of Paris and its suburbs (the urban unit without Paris) and between the municipality of Paris and Paris’s urban area (without Paris full urban unit) over the 1993 to 2015 period. For both measures, the differential increases implying that inequality between Paris and its suburbs or its outskirts increases steadily, potentially because Paris intramuros includes a top of the distribution that has very strongly increased over the period. While the gap with the municipality of Paris was low at the beginning of the period, the difference amounts to 25% for mean earnings (resp., 12% for median) in 2016. Interestingly, the trends and levels are very comparable in both suburbs and outskirts. As a result, previous results on between-areas inequality should be interpreted keeping in mind the strong divergence between the municipality of Paris and the rest of France. 5. Conclusion The French labor earnings inequality and dynamics over the last 25 years have been shaped by labor market institutions and their changes: strong increase in the minimum wage, sharp decrease in labor costs at and around the minimum wage, the implementation of the 35-hour workweek, resulting in the absence of a rising inequality. Even if the top 0.1% or 0.01% increased more than lower percentiles, as was observed in other countries, the lessons from France should not be centered on the top but on the bottom of the distribution, in particular for women who clearly benefited from the increase in the hourly minimum wage. Indeed, women often employed in part-time jobs increased hours (inducing a potential supply effect), in contrast to men, while labor costs at the minimum wage decreased (inducing a potential demand effect). A more complete analysis of the bottom of the French earnings distribution is clearly needed. The above changes seem to have had interesting and, again, counterintuitive consequences on the inequality between territories: the smaller urban areas have converged (in terms of median or mean earnings) to the larger ones. Furthermore, rural and remote territories have witnessed a clear decrease in inequality at the bottom of the earnings distribution. Finally, these remote or rural territories do not seem to have been “abandoned” by the central state, as often stated in these territories: public employment increased there over the period, both in local public jobs and public hospitals. All of these developments in these territories are a far cry from public perception. The tensions between centralized institutions deciding most policy changes and local entities (municipalities, constituencies) have generated reactions, even protests, to these changes that question policymaking and how resulting outcomes are perceived. In particular, those individuals closer to the middle of the earnings distribution do not seem to have benefited from these labor market policies. They also do not seem to have been aware of the positive changes in public employment, in a context where hospitals were closed and municipalities forced to regroup (see Tricaud (2020)). Were the Yellow Vests’ protests an echo from the populist movements that emerged in many other countries, in particular the United States? Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1563 Appendix A: The reduction of the workweek At the end of the 1990s, the Jospin government, with Martine Aubry as Minister of Labor, decided to fulfill an electoral promise and to go to 35 hours. Discussions between the government, which included the green party, and business unions were tense. Negotiations started within various industries and firms. But, at some point, Martine Aubry enacted a law essentially forcing firms above 20 employees to come up with some agreement with their workers’ unions or delegates. In addition, various incentives and subsidies were proposed at different moments in time. For instance, in June 1998, the socalled Aubry I laws gave establishments incentives to reduce their workweek and create or preserve employment in exchange for large subsidies. In order to receive these subsidies, firms had to reduce hours by at least 10% in order to attain an average weekly duration of 35 hours. In such a case, employment creation had to amount to 6% of total employment. A “defensive” aspect also allowed firms to receive subsidies to avoid economic separations or collective dismissals. The 2000 law, Aubry II, offered payroll tax subsidies for all firms that decided to go to 35 hours per week. Hence, among firms with more than 20 employees, at the beginning of the twenty-first century, various agreements prevailed. Some firms were still at 39 hours and had to pay overtime, others went to 35 hours between June 1998 and January 2000 and received incentives and subsidies, others refused the incentives (but received some “structural” subsidies) even though they went to 35 at similar dates (the so-called Aubry II forerunners). Firms also went to 35 hours after January 2000, receiving only the “structural” subsidies. Finally, remaining firms went to 35 hours and decided to receive no subsidies. Figure A.1. The several minimum wages in the 2000s. Note: Figure A.1 plots against time the five hourly minimum wages (GMR) for workers working 35 hours a week and the hourly minimum wage for workers working 39 hours a week (Smic 169h). Values are expressed in euros. GMR stands for “Garantie Mensuelle de Rémunération.” GMR 1-5 are applicable to firms, which started reducing their worker’s workweek respectively between: (1) 06/1998–06/1999, (2) 07/1999–06/2000, (3) 07/2000–06/2001, (4) 07/2001–06/2002, (5) in 07/2002. Source: Malik Koubi and Bertrand Lhommeau, “Les salaires en France,” 2007, Ministry of Labor. Go back to main text. 1564 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Appendix B: Data and descriptive statistics Figure B.1. Participation rate. Note: Figure B.1 plots against time the participation rate for: (a) people aged between 25 and 64, (b) people aged between 55 and 64, (c) men aged 25–64, (d) women aged 25–64. Source: OECD. Go back to main text. Table B.1. Minimum earnings threshold. Year 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Threshold 1,850 1,913 1,909 1,941 1,964 1,992 2,024 2,077 2,100 2,111 2,153 2,182 2,220 Year 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 Threshold 2,295 2,385 2,444 2,468 2,466 2,505 2,506 2,498 2,529 2,539 2,553 2,574 2,585 Note:TableB.1 displays for each year the minimum labor earnings threshold in euros 2018. Go back to main text. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1565 Figure B.2. Share of observations below the minimum earnings threshold. Note: Figure B.2 displays for each year the share of observations with annual earnings below the minimum labor income threshold displayed in Table B.1. Data set: Panel DADS. Go back to main text. Table B.2. Earnings distribution in France (Panel DADS). Year P1 P5 P10 P25 P50 P75 P90 P95 P99 P99.9 (a) Total population 1995 2,403 4,455 7,553 16,026 23,270 31,838 44,691 58,033 99,798 194,779 2005 2,933 5,386 8,873 17,595 24,382 33,178 47,000 60,675 106,531 236,190 2015 3,148 5,679 9,036 18,021 25,749 35,165 50,150 64,506 113,020 266,795 (b) Men 1995 2,565 5,423 9,516 18,612 25,415 35,139 51,483 67,704 115,618 227,547 2005 3,111 6,357 10,647 20,156 26,571 36,476 53,346 69,892 123,907 281,434 2015 3,244 6,172 9,992 20,293 27,879 38,578 56,254 72,921 131,998 327,918 (c) Women 1995 2,286 3,749 6,005 12,576 20,385 28,135 37,182 43,981 67,681 120,424 2005 2,809 4,667 7,398 14,595 21,773 29,419 39,734 48,360 77,769 147,850 2015 3,072 5,273 8,189 15,903 23,677 31,669 43,096 53,878 89,108 185,495 Note:TableB.2 shows summary statistics for CS sample separately for (a) total population, (b) men, and (c) women. Data set: Panel DADS. Go back to main text. 1566 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Table B.3. Descriptive statistics. Obs. Mean Income Women Age Shares % Education Shares % Year (Mill) Men Women % Share [25,35] [36,45] [46,55] <HS HS CD >CD 1995 0.58 29,896 21,772 43.2 37.7 35.1 27.2 67.8 13.9 10.7 7.7 2015 1.38 33,062 25,999 47.7 34.7 32.8 32.5 43.6 21.8 18.6 16.0 Note:TableB.3 shows descriptive statistics for CS sample. We construct four groups of workers based on their highest diploma. <HS are workers with less than a high school diploma, HS are workers with a high school degree, CD are workers with a 2-year college diploma and >CD are workers with an advanced university degree. Data set: panel DADS. Go back to main text. Appendix C: Inequality Figure C.1. Distribution of log real earnings in the population. Note: Using raw log earnings and the CS+TMax sample, Figure C.1 plots against time the following variables: (a) Men and women: P10, P25, P50, P75, P90, (b) Men and women: P90, P95, P99, P99.9, P99.99, (c) Men and women: P90-10 and 2.56*SD of log income, (d) Men and women: P90-50 and P50-10. All percentiles are normalized to 0 in the first available year. 2.56*SD corresponds to P90-10 differential for a Gaussian distribution. Shaded areas represent recession years. Data set: Panel DADS. Go back to main text. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1573 Appendix D: The reduction of the working week Figure D.1. Annual working time By earning percentile. Note: Figure D.1 plots against time, for 5 percentiles of the real raw earnings distribution, the following variables: (a) Men: median annual number of hours worked, (b) Women: median annual number of hours worked. For each of the 5 percentiles, we compute and plot the median number of hours worked as a share of a full time job in 1993. Data set: Panel DADS. Go back to main text. Figure D.2. Hourly wage By earning percentile. Note: Figure D.2 plots against time, for 5 percentiles of the real raw earnings distribution, the following variables: (a) Men: median hourly wage relative to the French minimum wage, (b) Women: median hourly wage relative to the French minimum wage. For each of the 5 percentiles, we compute and plot the median hourly wage divided by the national minimum wage. Data set: Panel DADS. Go back to main text. 1574 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure D.3. Evolution of hours and hourly wages by earning percentiles. Using real raw log earnings and the CS+TMax sample, Figure D.3 plots against time the following variables: (a) Men and Women: median number of hours worked for the P5, P10, P25, P50, P75, P90, P99 of the earnings distribution, (b) Men and Women: the median hourly wage for the P5 to P99 of the earnings distribution, (c) Men: median number of hours worked for the P5 to P99 of the earnings distribution, (d) Men: the median hourly wage for the P5 to P99 of the earnings distribution, (e) Women: median number of hours worked for the P5 to P99 of the earnings distribution, (f) Women: the median hourly wage for the P5 to P99 of the earnings distribution. All variables are normalized to 0 in 1993. Data set: Panel DADS. Go back to main text. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1575 Figure D.4. Decomposition of the variance of log earnings. Note: Figure D.4 plots against time, the decomposition of the variance of the real raw log earnings into the variance of the log hours, the log-hourly wage and the covariance between the two. Data set: Panel DADS. Go back to main text. 1576 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Appendix E: Earnings change Figure E.1. Dispersion of 5-Year earnings change. Note: Using residual 5-year earnings changes and the LX sample, Figure E.1 plots against time the following variables: (a) Men: P90-10 differential, (b) Women: P90-10 differential. Shaded areas represent recession years. Data set: Panel DADS. Go back to main text. Figure E.2. Skewness and kurtosis of 5-Year earnings changes. Note: Using residual 5-year earnings changes and the LX sample, Figure E.2 plots against time the following variables: (a) Men and Women: Kelley skewness, (b) Men and Women: Excess Crow–Siddiqui kurtosis calculated as P97.5−P2.5 P75−P25 −2.91 where the first term is the Crow–Siddiqui measure of kurtosis and 2.91 corresponds to the value of this measure for Normal distribution. Shaded areas represent recession years. Data set: Panel DADS. Go back to main text. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1577 Figure E.3. Dispersion, Kelley skewness and excess Crow–Siddiqui kurtosis of 5-year earnings changes. Note: Using residual 5-year earnings changes and the H sample over the period 1994–2011, Figure E.3 plots against permanent income quantile groups the following variables for the 3 age groups: (a) Men: P90-10, (b) Women: P90-10, (c) Men: Kelley skewness, (d) Women: Kelley skewness, (e) Men: Excess Crow–Siddiqui kurtosis, (f) Women: Excess Crow–Siddiqui kurtosis. Excess Crow–Siddiqui kurtosis calculated as P97.5−P2.5 P75−P25 −2.91 where the first term is the Crow–Siddiqui measure of kurtosis and 2.91 corresponds to the value of this measure for Normal distribution. Data set: Panel DADS. Go back to main text. 1578 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure E.4. Standardized moments of 1-year earnings changes. Note: Using residual 1-year earnings changes and the H sample over the period 1994–2015, Figure E.4 plot against permanent income quantile groups the following variables for the 3 age groups: (a) Men: Standard deviation, (b) Women: Standard deviation, (c) Men: Skewness, (d) Women: Skewness, (e) Men: Excess kurtosis, (f) Women: Excess kurtosis. Excess kurtosis is defined as the value of kurtosis minus 3, which is the corresponding value for a Normal distribution. Data set: Panel DADS. Go back to main text. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1579 Figure E.5. Standardized moments of 5-Year earnings changes. Note: Using residual 5-year earnings changes and the H sample over the period 1994–2011, Figure E.5 plot against permanent income quantile groups the following variables for the 3 age groups: (a) Men: Standard deviation, (b) Women: Standard deviation, (c) Men: Skewness, (d) Women: Skewness, (e) Men: Excess kurtosis, (f) Women: Excess kurtosis. Excess kurtosis is defined as the value of kurtosis minus 3, which is the corresponding value for a Normal distribution. Data set: Panel DADS. Go back to main text. 1580 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Appendix F: Earnings mobility Figure F.1. Evolution of 5-year mobility over the life cycle. Note: Figure F. 1 shows average rank-rank mobility for different age groups over the period 1993–2011 using the H sample. The 45-degree (dashed black) line corresponds to the case of perfect immobility: workers stay on average at the same percentile after 5 years. Data set: Panel DADS. Go back to main text. Figure F.2. Evolution of 5-year mobility over time. Note: Figure F. 2 shows average rank-rank mobility using the H sample over the period 1993–2011. The 45-degree (dashed black) line corresponds to the case of perfect immobility: workers stay on average at the same percentile after 5 years. Data set: Panel DADS. Go back to main text. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1581 Figure F.3. Evolution of the 5-year rank-rank correlation over time. Note: F. 3 shows the evolution over time of the permanent earnings 5-year rank-rank correlation coefficient. Data set: Panel DADS. Go back to main text. Figure F.4. Evolution of the 5-year rank-rank correlation by demographic group. Note: Figure F. 4 shows the evolution over time of the permanent earnings 5-year rank-rank correlation coefficient by age group and gender. Data set: Panel DADS. Go back to main text. 1582 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Figure F.5. Evolution of the 5-year rank-rank correlation by territory. Note: Figure F. 5 shows the evolution over time of the 5-year permanent earnings rank-rank correlation coefficient by territory based on the place of residence during the initial year. Data set: Panel DADS. Go back to main text. Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1589 References Abowd, John M., Francis Kramarz, and David N. Margolis (1999), “High wage workers and high wage firms.” Econometrica, 67, 251–333. [1535] Aghion, Philippe, Vlad Ciornohuz, Maxime Gravoueille, and Stefanie Stantcheva (2019), “Anatomy of inequality in France.” Working Paper. [1531] Bell, Brian, Nicholas Bloom, and Jack Blundell (2022), “Income dynamics in the United Kingdom and the impact of the Covid-19 recession.” Quantitative Economics,13(4), 1849–1878. [1529] Bilal, Adrien (2021), “The geography of unemployment.” Working Paper 29269, National Bureau of Economic Research. [1551] Bozio, Antoine, Bertrand Garbinti, Malka Guillot, Jonathan Goupille-Lebret, and Thomas Piketty (2020), “Predistribution vs. redistribution: Evidence from France and the U.S.” Working Papers 2020-24, Center for Research in Economics and Statistics. [1531] Buchinsky, Mosche, Gary S. Fields, Denis Fougère, and Francis Kramarz (2003), “Francs or ranks? Earnings mobility in France, 1967–1999.” CEPR Discussion Papers 3937, C.E.P.R. Discussion Papers. [1531] Charnoz, Pauline, Elise Coudin, and Mathilde Gaini (2014), “Forty years of decreasing wage inequality in France: The role of supply and hidden skill-biased technical change.” Working Papers 2014-20, Center for Research in Economics and Statistics. [1530,1537] Chauvet-Peyrard, Axelle (2013), Indices des prix à la consommation 1998–2012 selon la région d’habitation des ménages et selon la taille de la commune de résidence. Technical report. [1555] Cottet, Sophie (2020), “Payroll tax reductions for minimum wage workers: Relative labor cost or cash windfall effects?” Working papers, halshs-03010943. [1536] Diamond, Rebecca and Enrico Moretti (2021), “Where is standard of living the highest? Local prices and the geography of consumption.” Working Paper 29533, National Bureau of Economic Research. [1554] Drechsel-Grau, Moritz, Andreas Peichl, Kai D. Schmid, Johannes F. Schmieder, Hannes Walz, and Stefanie Wolter (2022), “Inequality and income dynamics in Germany.” Quantitative Economics, 13 (4), 1593–1635. [1529,1530] Friedrich, Benjamin, Lisa Laun, and Costas Meghir (2022), “Earnings dynamics of immigrants and natives in Sweden 1985–2016.” Quantitative Economics, 13 (4), 1803–1847. [1529] Garbinti, Bertrand, Jonathan Goupille-Lebret, and Thomas Piketty (2018), “Income inequality in France, 1900–2014: Evidence from Distributional National Accounts (DINA).” Journal of Public Economics, 162, 63–77. [1531] Godechot, Olivier (2012), “Is finance responsible for the rise in wage inequality in France?” Socio-Economic Review, 10, 447–470. [1530] 1590 Kramarz, Nimier-David, and Delemotte Quantitative Economics 13 (2022) Guillot, Malka, Antoine Bozio, and Thomas Breda (2020), “The contribution of payroll taxation to wage inequality in France.” IZA Discussion Paper No. 13317. [1530] Guvenen, Fatih, Fatih Karahan, Serdar Ozkan, and Jae Song (2021), “What do data on millions of u.s. workers reveal about lifecycle earnings dynamics?” Econometrica, 89, 2303–2339. [1530,1546] Guvenen, Fatih, Serdar Ozkan, and Jae Song (2014), “The nature of countercyclical income risk.” Journal of Political Economy, 122, 621–660. [1529] Halvorsen, Elin, Serdar Ozkan, and Sergio Salgado (2022), “Earnings dynamics and its intergenerational transmission: Evidence from Norway.” Quantitative Economics,13(4), 1707–1746. [1529,1540] Hoffmann, Eran B., Davide Malacrino, and Luigi Pistaferri (2022), “Earnings dynamics and labor market reforms: The Italian case.” Quantitative Economics, 13 (4), 1637–1667. [1530] Huttunen, Kristina, Benjamin Pirttila, and Roope Uusitalo (2013), “The employment effects of low-wage subsidies.” The Journal of Public Economics, 97, 49–60. [1533] Kramarz, Francis, Elio Nimier-David, and Thomas Delemotte (2022), “Supplement to ‘Inequality and earnings dynamics in France: National policies and local consequences’.” Quantitative Economics Supplemental Material, 13, https://doi.org/10.3982/ QE1876.[1534] Kramarz, Francis and Thomas Philippon (2001), “The impact of differential payroll tax subsidies on minimum wage employment.” Journal of Public Economics, 82, 115–146. [1528,1533] Leth-Petersen, Søren and Johan Sæverud (2022), “Inequality and dynamics of earnings and disposable income in Denmark 1987–2016.” Quantitative Economics, 13 (4), 1493– 1526. [1529] McKinney, Kevin L., John M. Abowd, and Hubert P. Janicki (2022), “U.S. long-term earnings outcomes by sex, race, ethnicity, and place of birth.” Quantitative Economics,13(4), 1879–1945. [1540] Mignon, Ernest (1962), Les mots du General de Gaulle. Fayard, Paris. [1528] Nimier-David, Elio (2022), “Local human capital and firm creation: Evidence from the massification of higher education in France.” Working Paper. [1537] Piketty, Thomas (2003), “Income inequality in France, 1901–1998.” Journal of Political Economy, 111, 1004–1042. [1530] Pora, Pierre and Lionel Wilner (2020), “A decomposition of labor earnings growth: Recovering Gaussianity?” Labour Economics, 63, 101807. [1530] Saez, Emmanuel, Benjamin Schoefer, and David Seim (2019), “Payroll taxes, firm behavior, and rent sharing: Evidence from a young workers’ tax cut in Sweden.” The American Economic Review, 109, 1717–1763. [1533] Quantitative Economics 13 (2022) Inequality and earnings dynamics in France 1591 Tricaud, Clémence (2020), “Better alone? Evidence on the costs of intermunicipal cooperation.” CESifo Working Paper Series 9000, CESifo. [1538,1562] Verdugo, Gregory (2014), “The great compression of the French wage structure, 1969– 2008.” Labour Economics, 28, 131–144. [1530,1537] Weber, Eugen (1976), Peasants Into Frenchmen: The Modernization of Rural France, 1870–1914. Stanford University Press. [1528] Co-editor Luigi Pistaferri handled this manuscript. Manuscript received 15 March, 2021; final version accepted 25 May, 2022; available online 26 July, 2022.