Policy Research Working Paper 10766 When the Rain Stops Falling Effects of Droughts on the Tunisian Labor Market Federica Alfani Giacomo Pallante Alessandro Palma Abdelkader Talhaoui Poverty and Equity Global Practice May 2024 Policy Research Working Paper 10766 Abstract This paper investigates the effects of severe drought shocks percentage points in agricultural employment with respect on Tunisia’s agriculture sector during 2000–19. Using the untreated or not-yet-treated governorates. There is a labor force surveys aligned with granular weather data, contemporaneous opposite dynamic in the employment it calculates the Standardized Potential Evapotranspira- rate of low-skill and less climate-sensitive sectors, as well tion Index to detect moderate-to-severe drought shocks as a modest and transient increase in unemployment. The at the governorate level and frames the analysis in a effects are largely heterogeneous across groups of workers, staggered difference-in-differences setting. The findings with very young individuals, women, and low-educated show that shocked areas experience a drop of 7.4 to 10.6 workers paying the highest toll. This paper is a product of the Poverty and Equity Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at falfani@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team When the Rain Stops Falling: Effects of Droughts on the Tunisian Labor Market ∗ Federica Alfani†, Giacomo Pallante§, Alessandro Palma¶, Abdelkader Talhaoui# J.E.L. codes: Q1, Q54, J21 Keywords: drought, agriculture, employment, gender gap, Tunisia * The authors want to thank the National Institute of Statistics (INS) of Tunisia for the ongoing collaboration and Gabriele Restelli for excellent assistance with data preparation. We also acknowledge participants to the authors’ Workshop of the MENA Chief Economist's Research Programs on Gender and Labor Markets and the SITES 2023 conference in Naples for useful comments. † World Bank, E-mail: falfani@worldbank.org # Tunisian National Institute of Statistics. Email: talhaoui.abdelkader@ins.tn § School of International Studies and DSRS, University of Trento, E-mail: giacomo.pallante@unitn.it ¶ Gran Sasso Science Institute (GSSI), E-mail: alessandro.palma@gssi.it 1. Introduction The impacts of climate change are expected to become more severe, especially on the agriculture sector. Changes in the weather pattern are more and more associated with extreme events that have short- and medium-term impacts (Day et al., 2019; Stevanović et al., 2016; Key and Sneeringer, 2014). However, the distribution of these impacts varies widely by region. Southeast Asia and Sub-Saharan Africa are more prone to floods, while the Middle East and North Africa (MENA) regions are experiencing recurrent droughts. The MENA region is already one of the driest areas on the planet and drought shocks are expected to become more frequent, longer-lasting, and more intense in the future (Fischer et al., 2021; Chiang et al., 2021, Pörtner et al., 2022). Recently, Alfani et al. (2022) have documented that severe drought shocks significantly and persistently reduced the agricultural employment in Morocco during the period 2000-2009, with a limited reemployment of displaced workers and residual unemployment. However, while the impact of climate change on the agriculture sector in Africa is well documented (Abidoye and Odusola, 2015; Asafu-Adjaye, 2014), we know much less about the effects of climate extremes in the MENA region. In this paper we investigate the effects of drought shocks on the labor market in Tunisia using rich labor- force survey data aligned with granular weather information at the governorate level. Building on Alfani et al. (2022), we test whether agricultural workers - who are the most exposed to climate variability - faced a job displacement as a consequence of severe drought events that occurred in Tunisia from 2000 to 2019. We also look at the employment rate of other economic sectors and at the unemployment rate to investigate whether a drop in agricultural employment is offset by job reallocation or translates into higher unemployment. We frame the analysis in a difference-in-difference setting by considering multiple and non-absorbing shocks at different points in time to account for differential treatment dynamics (Chaisemartin and D'Haultfœuille, 2020). Tunisia is a representative case of other MENA countries. Despite a decline from 20% to 14% in the employment rate during the last decade, the agriculture sector still absorbs a relevant fraction of the labor force. Overall, the labor market is characterized by several frictions, mainly due to the gender gap in labor market participation, low education returns and informal employment. According with the recent Tunisia’s Jobs Landscape (World Bank, 2022), less than half of the working-age population is actively 2 participating in the labor market, either through employment or job seeking. This indicates a significant underutilization of the country's human capital, despite past public investments in education that have led to notable improvements in educational attainment. Labor force participation is very low among women (26.5% in 2017) as well as the employment rate among women (20.5% in 2017) and youth (28.2% in 2017). Despite some progress led by highly educated young people, average participation of women remains scant due to weak labor demand, assigned gender roles, and a lack of affordable childcare services. Moreover, a significant gender wage gap in the private sector effectively translates into almost three months of unpaid labor per year for women, further discouraging their participation in the labor force. Women with higher level of education are more likely than men to be employed in the public sector than in the private sector. The wage gap plays a role also in reinforcing traditional gender roles. In addition, in 2019 a significant proportion of workers in Tunisia were employed informally (43.9%), and 55.7% among private sector workers (World Bank, 2022). Informal workers lack access to social insurance and operate unregistered, unincorporated businesses. This is especially prevalent among male wage workers, youth, and those with limited education in rural and inland regions (World Bank, 2022). Although these workers face difficulties in accessing formal jobs in the public or private sector and are not covered by social insurance, they do not experience wage penalties. However, informal workers face significant risks, such as health events, unemployment, old age, and disability, and they lack protection from social insurance policies. From a climatic point of view, over the years Tunisia has had to confront temperature increases, less intense and more variable rainfall, rising sea levels, saltwater intrusion and extreme weather events such as droughts, floods and forest fires. Many areas of the country are largely prone to recurrent arid conditions and drought events are dramatically increasing. The recent Sixth IPCC Assessment Report points to an average GDP loss of about 11 percent in a scenario with the temperature increase of 4.8 °C that has been projected for 2100. The reduced water availability will in turn affect agricultural production, exacerbating expected agricultural losses due to extreme weather events and further straining people’s livelihoods (Tunisia CCDR, forthcoming). This increasing water scarcity, combined with a poor drought management system (Zouabi and Peridy, 2015; Bahri et al., 2019) and the frictions in the labor market, pose unprecedented challenges to the agriculture sector, which is the most exposed to the climate variability. 3 Our estimates show that the occurrence of severe drought events, measured by an average 12-month SPEI lower than -1 s.d., caused a share of employment in the agriculture sector that is up to 10.6 percentage points (p.p.) lower compared to governorates that were not affected by these shocks. We also observe that, with an opposite dynamic, the employment rate in other sectors of the economy increased in the same periods, suggesting that the displacement effect in the agriculture sector was to a large extent offset by reemployment in other less climate-exposed sectors. These effects are temporary, as they last for only two years from the shock. Severe drought episodes also led to a slight increase in unemployment, which, however, disappears after one year. We also conducted a comprehensive analysis of the differential effects on important individual characteristics such as age, gender, educational level, and geographical residence. These findings reveal a significant heterogeneity in the impacts, with the greatest burden falling on very young workers, women, and low-skilled workers. We conclude that drought shocks in Tunisia, expected to intensify in the future, significantly contribute to exacerbating socio-economic inequality. The remainder of the paper is as follows. Section 2 introduces the Tunisian case study, and Section 3 presents the data, our measures of drought shocks and some descriptive analysis. Section 4 outlines the empirical strategy, Section 5 presents the results, and Section 6 discusses heterogeneity. Section 7 concludes. 2. Weather shocks and the labor market in Tunisia The Tunisian agriculture sector represents the permanent income of 470,000 farmers in rural areas, which is 35% of the population (World Bank, 2021). The share of employment in the agriculture sector has decreased in the past 20 years with a rate similar to the other MENA countries, as illustrated in Figure 1. Figure 2 shows as the decrease in agricultural employment was more pronounced over the female population that passed from 26% in the 1998 to less than 10% in 2021. In line with the stylized facts about the labor markets in the Arab countries, employment in the bureaucracy and security forces represents a large fraction of formal sector employment (Assaad, 2014). In Tunisia, after the revolution, the number of public sector workers increased from 444,905 to 669,300 4 between 2011 and 2021. 1 The social and security context are the main ways of public sector growth by the recruitment of staff and the regularization of contract workers (Brockmeyer et al., 2015). The service sector absorbs 58% of the female employment and 49% of the male employment. On the other hand, despite the rapidly falling gender gaps in educational attainment and the positive cultural effect caused by the Arab spring, the female labor force participation rates have been stagnating and the cultural effect is heterogenous across the country (Bargain et al., 2019). In rural areas too, most jobs, including informal employment for the youth population, are in the service sector (David et al., 2023). This trend is partly explained by the low wages in the traditional labor- intensive construction and agriculture, despite that the minimum nominal wage was recently equalized to other sectors. Further, internal migration from low-earnings regions to higher-earnings regions is not common with around 2% of workers in the southwest region and 5% from the southeast region moving to Grand Tunis each year (World Bank, 2015). In this context, climate change and weather shocks are increasingly affecting households’ income stability. According to a recent survey by the European Investment Bank, 2 the rising occurrence of droughts, coastal erosion and floods had a negative impact on the livelihoods of Tunisians, with 52% claiming that their income has been affected. In 2023, Tunisia introduced water rationing and suspended water distribution in some areas of the capital and other cities for seven hours each night, as the country suffered its fourth year of severe drought. In addition, the Agriculture Ministry imposed a temporary quota system on water for irrigation. 3 Therefore, Tunisia represents an important case study for assessing the severe impacts of drought on the labor market in a country that, like many others, is increasingly destined to face the impacts of a changing climate. Future scenarios project rising temperatures, decreasing precipitation, increasing evapotranspiration and decreasing availability of water resources with detrimental effects on agricultural productivity (Abdelmalek and Nouiri, 2020; Verner et al., 2018). 3. Data 1 INS statistics at https://www.ins.tn/enquetes/caracteristiques-des-agents-de-la-fonction-publique-et-leurs-salaires-2010-2021 2 The survey is available at: https://www.eib.org/en/press/all/2022-561-84-of-tunisians-say-climate-change-is-already-affecting-their-everyday- life#:~:text=The%20survey%20results%20confirm%20that,such%20as%20floods%20and%20hurricanes. 3 https://www.reuters.com/world/africa/tunisia-introduces-water-quota-system-2023-03-31/ 5 We employed two data sources. First, thanks to the collaboration with the Tunisian National Institute of Statistics (INS), we rely on the Enquête Nationale sur la Population et l’Emploi (ENPE), which provides nationally representative socioeconomic information at the individual level from 2000 to 2019. The ENPE is conducted by the INS and consists of demographics and labor market data. The survey consists of repeated cross-sections and does not allow to track individuals over time. However, it includes rich information and represents a valuable source of data to analyze the labor market in Tunisia. We started with an initial sample at the individual level across 24 governorates. After restricting the sample to working age individuals (between 15 and 65 years old), we collapsed the data into governorate x year cells to obtain a balanced panel of yearly outcomes of the labor markets. Exploiting the details of employment status and the economic sector codes (NACE classification) for employed individuals, we focused on four labor market outcomes at the governorate level: the share of employment in the agriculture sector, the share of employment in less climate sensitive sectors (namely, manufacturing and construction), the share of employment in all the other economic sectors (excluding the public sector), and the unemployment rate. Other information includes the shares of the population across five age groups; the gender of the workforce; the share of educational levels across the population; and the share of the workforce population living in urban areas. We aligned ENPE socioeconomic data with detailed weather information that we obtained from the Agri-4-Cast database. These data are provided by the Food Security Unit of the Joint Research Center (JRC.D.5) and were specifically employed for identifying the climate change impacts in the agriculture sector. The data consists of gridded meteorological observations from weather stations interpolated on a 25 × 25 km grid. They are available on a daily basis from 1979 in the European Union and its neighboring countries, including Tunisia. We selected variables for minimum and maximum air temperatures (in Celsius degrees) and sum of precipitation (mm/day) to calculate the Standardized Precipitation Evapotranspiration Index (SPEI), our main treatment measure. The SPEI represents a state- of-the-science indicator for measuring the impact of increased temperatures on water demand (Vicente- Serrano et al., 2010; Chiang et al., 2021). Since climate grids come at a finer spatial resolution than our administrative unit of analysis, we calculated SPEI values for each grid point falling within each administrative unit. We then weighted the SPEI value of each grid data point with the resident population. Finally, we collapsed the data to obtain medians of population-weighted SPEI values in each governorate 6 x year cell; considering the large density of point measures in each governorate, this procedure minimizes the risk of assigning drought shocks to governorates that had only a negligible portion of territory exposed to drought shocks and are scarcely populated. We considered SPEIs at 12 months. Specifically, an accumulation period of 12 months is more appropriate to account for interseasonal precipitation patterns over a medium duration timescale (Svoboda et al., 2012). We classified as treated governorates in which the average 12-month SPEI value assumed a value equal or lower than -1 s.d. at least one time over the period 2000-2019. According to the classification provided by the World Bank and reported in Appendix Table A1, we build a discrete and ordered treatment with 4 categories, where the untreated governorate never experienced an average SPEI lower than -1. This threshold identifies droughts classified as at least moderate. Figure 3 shows the national average (the black line) and the minimum value (the red line) of the 12-month SPEI observed over the period 2000-2019. The national average SPEI ranges around -0; when we consider the minimum SPEI, we detect several episodes of moderate drought (SPEI ≤ 1 s.d.), with most of them falling in the second half of the period. Figure 4 shows the distribution of the minimum 12-month SPEI averaged over the period of analysis (2000-2019) across Tunisian governorates. The driest conditions are mostly located in the Northern and Eastern areas, including several coastal governorates. Table 1 shows summary statistics for all of the socioeconomic and climatic variables described above. 4. Empirical strategy We estimate dynamic treatment effects assuming that each governorate can be exposed to drought shocks of different intensity and at different points in time t. Moreover, the treatment can switch on and off, implying that after the first drought shock, a governorate could experience years with standard weather conditions or additional drought shocks, from severe to extremely severe. In this discrete non- absorbing difference-in-differences design, , is the period-t SPEI value in the governorate and , is the observed labor market outcome. To estimate the treatment effect in this setting, we follow the approach by De Chaisemartin and D’Haultfoeuille (2020) that proposes the following estimator , : , = ( , + ()|) , + �, � − 7 In an event study setting with a discrete and ordered treatment, , represents the expected difference + , that is years after it was treated for between governorate ’s actual labor market outcome at year the first time, and the counterfactual outcome that would have obtained at that period if it had remained equal to its period one value from year 1 to + . In our design, it corresponds to the average effect of being treated for + 1 years rather than being left with the treatment value of period 1. Dynamic effects can be aggregated in different ways that allow to compute estimators for initially untreated groups or a weighted average of initially treated and untreated groups. Beyond the dynamic effect, one might be interested in the average treatment effect per unit increase of drought intensity, + . For equal to the last period when there is still a group untreated since the first year, + is given by the weighted average of the event study reduced form effect , , for all the ∈ }, scaled by the sum of all the treatments , + . In this case, showing a first stage event {0, … , − study is recommended to detect the average treatment intensity after a first treatment switch. To be identified, , and + require no anticipation and strong parallel trend assumptions. Placebo estimators, + − 2 to δ, , allows to compare ’s outcome evolution to that of untreated governorate from period − 1, which is before governorate g is treated for the first time. The placebo estimator measures if treated and not-yet treated governorate at + are on parallel trends when untreated for + 1 years, the number of years over which the parallel trends should hold for having unbiased , . 5. Results Our preferred model specification focuses on the 12-month average SPEI shock as discrete ordered treatment, and initially untreated governorates as control group. Table 2 shows the main results for both the dynamic estimators, , and the aggregate average treatment effect + . The average effect on the agricultural employment is -5.2 p.p. when we aggregate estimators up to 4 years after the drought shock. This means that, on average, governorates that experienced a drought show a share of agricultural employment about 5 p.p. lower than the untreated governorates. To better visualize the dynamic of the impacts after the shock and test pre-trends, we present the results in an event study setting. In all estimates we cluster the standard errors at the governorate and cohort levels. In the Appendix Figure 1, we present the first stage treatment effect on the average drought 8 + for governorates that were treated for the first year. Governorate experiencing both a intensity in single or multiple drought shocks over time, had an average SPEI intensity of -1.25σ (standard deviations) in the first year and -0.8σ in the second year. This means that all the average effect on the labor outcomes must be interpreted as a drought intensity between moderate and severe. Two years after the first drought shock, there are no governorates showing additional droughts. Figure 5 shows the effect on the agricultural employment rate. Agriculture is the most exposed sector to climate variability and therefore the one in which most of workers are at risk during a drought shock. We find that drought events caused a loss of agricultural employment in governorates that experience such shocks. The magnitude of the effects reaches its maximum in the year of shock (about -10.6 p.p.), it remains large and significant after one year (about -7.4 p.p.) and then disappears in the third year after the shock. The absence of a significant trend in the placebo estimators strengthens the causal interpretation of this result. Figures 6 and 7 show the effect of drought shocks on the employment dynamic of less climate sensitive sectors, namely manufacturing and construction, and all the other sectors. While the latter do not highlight job replacement, we observe a simultaneous and slightly lower increase in the employment rate in manufacturing and construction that follows a reverse pattern with respect to agricultural employment. The increase reaches a maximum of about 5.1 p.p. in the year of shock, decreases to about 5.4 p.p. after one year and drops to zero in the third year. This supports the hypothesis that agricultural workers may have relocated to other sectors of the economy that are less exposed to climate variability. In particular, those who worked in the agriculture sector, with low skills, could find a temporary job in other low-skilled sectors such as construction. To confirm this assumption, in Section 5 we present results disaggregated by education level, as proxy for the workers’ skill. However, the cumulate drop in agricultural occupation does not seem to be completely offset by the cumulate simultaneous increase in non-agricultural employment in the first two years. To shed further light, we also report in Figure 8 the effect of drought on unemployment, from which we observe an absence of significant trend in the pre-shock period and a significant increase in the unemployment rate of about 6.3 p.p, which lasts only one year. 9 Additional estimates provide results when estimators of initially treated governorates are computed. When we include in our computation the early-treated governorates, the average treatment effect is reduced in terms of magnitude, as reported in Figures 9 to 12. Including governorates that face a drought shock since the first year of observation reduces the magnitude of the effects of the aggregate estimates, likely because the adjusting mechanism between climate-sensitive and less climate-sensitive sectors was already at work. In this setting, both the instantaneous shock and the subsequent one cause a drop in the agricultural employment around 7.3 p.p., while the higher employment in less climate sensitive sectors is 3.7 p.p. in the first year. The effect on unemployment is 6 p.p, in the year of the shock. 6. Heterogeneous effects It is known that climate shocks can have differential effects among more or less vulnerable population groups, leading to climate inequality (Islam and Winkel, 2017; Diffenbaugh and Burke, 2019; Fruttero et al., 2023; Afridi et al., 2022). To test if this characteristic is also present in the impacts on the labor market in Tunisia, in this section of results, we present the findings on the differential effects on certain individual characteristics. Specifically, we show the effects divided by different age groups, by gender, by educational level, and by geographic area. Effects across age classes Figures 13, 14, 15 and 16 show the results obtained for workers in the agriculture sector, manufacturing and construction, other sectors and unemployment, across five age classes: 15-17, 18-25, 26-45, 46-55 and over 55. Each of these categories reflects different work experiences and conditions. For example, workers aged 15 to 17 are individuals below the legal age of majority, often employed informally, and therefore subject to greater precariousness and lack of social safety nets. On the other hand, the age category between 46 and 55 likely includes workers with more experience and a stronger labor market attachment, which implies larger social guarantees and a higher contractual stability. Finally, workers over the age of 55 are nearing retirement and may have less flexibility in potential reemployment in the labor market following a sectoral shock. Consistent with these conjectures, the graphs in Figure 13 show that the greatest impacts in the agriculture sector occur for very young and young workers, aged 15 to 25, where displacement reaches large negative peaks in the first year, decreases only marginally in the second 10 year, and nearly disappears in the third year. However, those in the age group between 46 and 55 also experience a significant and relatively wide impact of about 10 p.p., lasting for a short duration of two years. Although the impacts in the agriculture sector appear to be very high, their duration over the control group numbers does not exceed two years and, most importantly, they are absorbed by the manufacturing and construction sectors. Figure 14 shows indeed that estimate of the employment rate in these sectors, in fact, mimics those of the agricultural jobs loss, while in Figure 15 we observe no impacts in all the other sectors of the economy. Although we cannot directly infer it, this dynamic probably implies that the displacement occurred in the agriculture sector due to climatic shocks is largely absorbed by the demand for labor in low-skill sectors and, importantly, this process does not penalize any particular age group of workers. On the other hand, Figure 16 shows that the main impacts in terms of unemployment increase are on the youngest groups. Effects by gender Like age groups, gender also represents a proxy for different working conditions. This is certainly the case in Tunisia, where only one-third of the workforce is comprised of women, often employed without a formal contract and experiencing significant wage penalties and adverse working conditions. Figure 17 shows that women absorb almost all the effects of the decline in agricultural employment due to drought shocks, reaching approximately a poorly significant 20 p.p. in the first year and a significant 17 p.p. in the second year, while in the case of men, the decline in employment is reduced. The decline in female employment is only partially absorbed: approximately 10.5 p.p. of female workers are reallocated to low- skilled sectors (manufacturing and construction), about 4 p.p. to other non-agriculture sectors, while the remaining portion contributes to an increase in the unemployment rate, although this latter result is weakly statistically significant. With a much smaller magnitude, the results for men follow a similar dynamic, with the only difference being that we observe no reallocation of male workers in other non-agriculture sectors. This is reasonable as we presume that the construction sector, which is typically male-dominated, played a major role in absorbing the displaced workers from agriculture. Effects across education levels By disaggregating the impacts by educational level, we are able to provide more precise information on 11 how drought shocks are distributed among groups of workers with different skills and functions. Since agriculture is a sector that traditionally employs low-skilled labor, based on theory and previous empirical evidence, we can hypothesize that low-skill workers will have a greater chance of being reallocated to sectors with similar characteristics. Conversely, we do not expect significant and wide-ranging effects on workers with higher levels of education and therefore more complex skills. The results presented in Figure 18 appear to be fully in line with our predictions. The impact of drought shocks is concentrated mainly on workers with the lowest levels of education, where we observe a decline in employment of approximately 11 p.p. in the year of the shock, and simultaneously, an absorption of an equal magnitude only in the manufacturing and construction sectors. Importantly, we do not observe significant increases in the unemployment rate. Once again, the duration of the impact is concentrated within just two years, and subsequently, it is completely absorbed. The effects we find on workers with higher levels of education are lower than the low education group. Effects in urban and rural areas Finally, we present the results obtained by differentiating the sample between workers in urban areas and rural areas. This distinction is important because a large part of climate shocks affects the agriculture sector, which is geographically concentrated in rural areas. Figure 19 confirms that the impacts on the agriculture sector are significant and economically meaningful only in rural areas. However, it is important to note that the labor market dynamics in urban areas highlight a slight decrease in the other sectors and an increase in the unemployment rates. 7. Discussion and conclusions In this paper we explore the impact of severe drought shocks in the labor market of Tunisia during the period 2000-2019. Tunisia is an important MENA country that is considered a climate hotspot and for which little evidence is available. Importantly, Tunisia is a good case study to investigate the interacted effects of climate change on the local labor markets in a socioeconomic context characterized by a narrow demographic window, a labor force participation rate higher than the average in the Middle East and North Africa region, and regionally clustered economic activities and employment opportunities. We align nationally representative labor force surveys and weather data that allow to calculate SPEI, a 12 state-of-the-art climate indicator, to estimate the causal effects of severe drought shocks on labor market dynamics. We look at three outcomes: employment rate in agriculture, a climate-dependent sector; employment rate in low-skill sectors (manufacturing and construction); and employment rate in other non-agriculture sectors, and unemployment. We found that in governorates that experienced a severe drought the share of employment in agriculture drops by about 5.2 p.p., on average in the mid run (over 4 years after the shock). However, the largest and significant negative effects are detectable just in the aftermath of the shock and up to one year later, with a magnitude ranging between 7.4 and 10.5 percentage points. We also find a contemporaneous increase of employment in the other non-agriculture sectors of the economy. These sectors are manufacturing and construction, while no positive effect is found on other sectors where skilled work could be demanded. The magnitude of the employment gain almost fully offsets the loss of agricultural employment. We also observe a significant and modest increase in unemployment only in the year of shock. These dynamics signal that the temporary displacement of agricultural workers was absorbed to a large extent by reemployment in other less climate exposed sectors, with only a temporary and limited impact on climate- induced unemployment. When estimating impacts for disaggregated groups based on different characteristics used as proxies of exposure and vulnerability (age, gender, level of education, and geographical location in urban or rural areas), we observe that the effects are concentrated on the most fragile groups of workers: young individuals with little experience and limited job security, women, and individuals with low levels of education. The quality of jobs in the agriculture sector remains low, as evidenced by the high rate of informality, precariousness and poverty (Nasri and Belhadj 2022). Large impacts of the drought in the younger groups could reflect the difficult transition from school to work, for which young workers, once displaced from a precarious job in the agriculture sector decide to search for better opportunities at a higher wage, while adult workers, with the poorest instruction, will accept a job in other low-wage sectors, such as construction. This explanation seems to be confirmed by the fact that highly educated displaced workers are not able to relocate themselves, on the contrary of the less educated displaced workers. In the same way, the magnitude of the effect of the weather shock on women could be motivated by the precariousness of their jobs in the agriculture sector, that achieves 58% (ILO, 2018). Women are more likely to have precarious employment conditions than men. Moreover, as highlighted by the literature 13 (Assaad, 2018), for cultural motivations women in Tunisia have high reservation working conditions with a preference for formal or public/government jobs. If they cannot relocate in these types of jobs, they will leave the labor force or remain unemployed, as our results show. This body of evidence documents a strong heterogeneity in the labor market impacts of drought shocks in Tunisia, which likely exacerbates the existing social inequality. In light of these findings, policy makers should consider that, in order to mitigate the labor market impacts of drought shocks which are projected to increase significantly in this region in the future and adapt to the new norm of the climate, it would be vital to invest in water savings, new irrigation infrastructures and agricultural technologies. Mobilizing climate finance is key in achieving this objective. On the other hand, severe and prolonged drought conditions call for an enhancement of the social safety nets and other welfare measures for the most vulnerable groups. The reduction of the cost of the formal economy should facilitate the transition of workers from informal to formal jobs, thereby enlarging the potential beneficiaries of these measures. 14 References Abdelmalek, M. B., & Nouiri, I. (2020). Study of trends and mapping of drought events in Tunisia and their impacts on agricultural production. Science of the Total Environment, 734, 139311. Abidoye, B. O., & Odusola, A. F. (2015). Climate change and economic growth in Africa: an econometric analysis. Journal of African Economies, 24(2), 277-301. Afridi, F., Mahajan, K., & Sangwan, N. (2022). The gendered effects of droughts: Production shocks and labor response in agriculture. Labour Economics, 78, 102227. Alfani, F., Molini, V., Pallante, G., & Palma, A. (2023). Job Displacement and Reallocation Failure: Evidence from Climate Shocks in Morocco. The World Bank Policy Research Working Paper No. 10279. Washington, D.C.: World Bank Group. Asafu-Adjaye, J. (2014). The economic impacts of climate change on agriculture in Africa. Journal of African Economies, 23(suppl_2), ii17-ii49. Assaad, R. (2014). Making sense of Arab labor markets: the enduring legacy of dualism. IZA Journal of Labor & Development, 3(1), 1-25. Ragui A., Ghazouani S., Krafft C. (2018). ‘The Composition of Labor Supply and Unemployment in Tunisia.’ In The Tunisian Labor Market in an Era of Transition, edited by Ragui Assaad et Mongi Boughzala, 1–38. Oxford, UK: Oxford University Press. Bahri, H., Annabi, M., M'Hamed, H. C., & Frija, A. (2019). Assessing the long-term impact of conservation agriculture on wheat-based systems in Tunisia using APSIM simulations under a climate change context. Science of the Total Environment, 692, 1223-1233. Bargain, O., Boutin, D., & Champeaux, H. (2019). Women's political participation and intrahousehold empowerment: Evidence from the Egyptian Arab Spring. Journal of Development Economics, 141, 102379. Brockmeyer, A., Khatrouch, M., Raballand, G. (2015). Public sector size and performance management: a case- study of post-revolution Tunisia. World Bank Policy Research working paper No. 7159. Washington, D.C.: World Bank Group. Chiang, F., Mazdiyasni, O., and AghaKouchak, A. (2021). Evidence of anthropogenic impacts on global drought frequency, duration, and intensity. Nature communications, 12(1):1–10. David, A., Diallo, Y., & Nilsson, B. (2023). Informality and Inequality: The African Case. Journal of African Economies, 32(Supplement_2), ii273-ii295. Day, E., Fankhauser, S., Kingsmill, N., Costa, H., and Mavrogianni, A. (2019). Upholding labour productivity under climate change: an assessment of adaptation options. Climate policy, 19(3):367–385. De Chaisemartin, C., d’Haultfoeuille, X. (2020). Two-way fixed effects estimators with heterogeneous treatment effects. American Economic Review, 110(9), 2964-2996. Diffenbaugh N. S., Burke M. (2019). Global warming has increased global economic inequality. Proceedings of the National Academy of Sciences, 116(20), 9808-9813. Fischer, E., Sippel, S., and Knutti, R. (2021). Increasing probability of record-shattering climate extremes. Nature Climate Change, 11(8):689–695. Fruttero, A., Halim, D., Broccolini, C., Coelho, B., Gninafon, H., & Muller, N. (2023). Gendered Impacts of Climate Change. ILO. (2018). Women’s and Youth Empowerment in Rural Tunisia – An assessment using the Women’s Empowerment in Agriculture Index (WEAI) / International Labour Office, Taqeem Impact Report Series, Issue 11. Geneva: 2018 15 Islam N., Winkel J. (2017). Climate change and social inequality. DESA Working Paper No. 152 Key, N. and Sneeringer, S. (2014). Potential effects of climate change on the productivity of us dairies. American Journal of Agricultural Economics, 96(4):1136–1156. Nasri, K., & Belhadj, B. (2022). Household Vulnerability and Resilience in Tunisia: Evidence Using Fuzzy Sets and Multidimensional Approach. Studies in Microeconomics, 23210222221098836. Pörtner, H.-O., Roberts, D. C., Adams, H., Adler, C., Aldunce, P., Ali, E., Begum, R. A., Betts, R., Kerr, R. B., Biesbroek, R., et al. (2022). Climate change 2022: Impacts, adaptation and vulnerability. IPCC Sixth Assessment Report. Stevanović, M., Popp, A., Lotze-Campen, H., Dietrich, J. P., Müller, C., Bonsch, M., Schmitz, C., Bodirsky, B. L., Humpenöder, F., and Weindl, I. (2016). The impact of high-end climate change on agricultural welfare. Science advances, 2(8):e1501452. Svoboda, M., Hayes, M., Wood, D. A., et al. (2012). Standardized precipitation index user guide. Verner, D., Treguer, D., Redwood, J., Christensen, J., McDonnell, R., Elbert, C., & Konishi, Y. (2018). Climate variability, drought, and drought management in Tunisia’s agricultural sector. World Bank, Washington, DC, World Bank. Vicente-Serrano, S. M., Beguería, S., and López-Moreno, J. I. (2010). A multiscalar drought index sensitive to global warming: the standardized precipitation evapotranspiration index. Journal of climate, 23(7):1696–1718. World Bank (2022). Tunisia’s Jobs Landscape. Available at: https://www.worldbank.org/en/country/tunisia/publication/tunisia-s-jobs-landscape World Bank (forthcoming). Tunisia Country Climate and Development Report (CCDR). Washington, D.C. : World Bank Group World Bank. (2015). Labor Policy to Promote Good Jobs in Tunisia: Revisiting Labor Regulation, Social Security, and Active Market Programs, 2015. World, Bank, 2021. Climate Risk Profile: Tunisia (2021): The World Bank Group. Zouabi, O., & Peridy, N. (2015). Direct and indirect effects of climate on agriculture: an application of a spatial panel data analysis to Tunisia. Climatic change, 133, 301-320. 16 Figures Figure 1: Employment in agriculture (share of total employment) – Tunisia and MENA countries Figure 2: Female and Male employment in agriculture (share of total employment) 17 Figure 3: Average and minimum 12-month SPEI in Tunisia between 2000 and 2019 Notes: The figure shows average and minimum SPEI values averaged at the national level from 2000 to 2019 at monthly frequency. Figure 4: Minimum 12-month SPEI in Tunisian governorates averaged over the period 2000-2019 18 Figure 5 – Effect of Drought Shocks on Agricultural Employment Notes: The figure shows event study estimates of the effect of drought shocks (12-month SPEI <= -1 s.d.) on the employment rate in agriculture over the period 2000-2019. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) confidence intervals are at 95 percent. Governorate linear trends are included. Figure 6 – Effect of Drought Shocks on Manufacturing and Construction Notes: The figure shows event study estimates of the effect of drought shocks (12-month SPEI <= -1 s.d.) on the employment rate in manufacturing and construction over the period 2000-2019. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) confidence intervals are at 95 percent. Governorate linear trends are included. 19 Figure 7 – Effect of Drought Shocks on Other Sectors Notes: The figure shows event study estimates of the effect of drought shocks (12-month SPEI <= -1 s.d.) on the employment rate in other sectors over the period 2000-2019. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) confidence intervals are at 95 percent. Governorate linear trends are included. Figure 8 – Effect of Drought Shocks on Unemployment Notes: The figure shows event study estimates of the effect of drought shocks (12-month SPEI <= -1 s.d.) on the unemployment rate over the period 2000-2019. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) confidence intervals are at 95 percent. Governorate linear trends are included. 20 Figure 9 – Effect of Drought Shocks on Agricultural Employment, early-treated Governorates Notes: The figure shows event study estimates of the effect of drought shocks (12-month SPEI <= -1 s.d.) on the employment rate in agriculture over the period 2000-2019. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) confidence intervals are at 95 percent. Governorate linear trends are included. Figure 10 – Effect of Drought Shocks on Manufacturing and Construction, early-treated Governorates Notes: The figure shows event study estimates of the effect of drought shocks (12-month SPEI <= -1 s.d.) on the employment rate in manufacturing and construction over the period 2000-2019. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) confidence intervals are at 95 percent. Governorate linear trends are included 21 Figure 11– Effect of Drought Shocks on other sectors, with early-treated Governorates Notes: The figure shows event study estimates of the effect of drought shocks (12-month SPEI <= -1 s.d.) on the employment rate in non-agriculture sectors (manufacturing and construction) over the period 2000-2019. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) confidence intervals are at 95 percent. Governorate linear trends are included. Figure 12 – Effect of Drought Shocks on Unemployment, early-treated Governorates Notes: The figure shows event study estimates of the effect of drought shocks (12-month SPEI <= -1 s.d.) on the unemployment rate over the period 2000-2019. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) confidence intervals are at 95 percent. Governorate linear trends are included. 22 Figure 13 – Effect across age classes, Agriculture Age 15-17 Age 18-25 Age 26-40 Age 41-55 Age 55+ 23 Figure 14 – Effect across age classes, Manufacturing and Construction Age 15-17 Age 18-25 Age 26-40 Age 41-55 Age 55+ 24 Figure 15 – Effect across age classes, other sectors Age 15-17 Age 18-25 Age 26-40 Age 41-55 Age 55+ 25 Figure 16 – Effect across age classes, unemployment Age 15-17 Age 18-25 Age 26-40 Age 41-55 Age 26-40 Age 41-55 Age 55+ 26 Figure 17 – Effect by gender Females Males Agriculture Manufacturing & Construction Other sectors Unemployment 27 Figure 18 – Effect across educational levels Low Education High Education Agriculture Manufacturing & Construction Other sectors Unemployment 28 Figure 19 – Effect in urban and rural areas Rural Urban Agriculture Manufacturing & Construction Other sectors Unemployment 29 Tables Table 1: Summary Statistics Variable Description Treated Control Agriculture Employment rate in Agriculture 0.19 0.18 (0.13) (0.03) Other Sectors Employment rate in other sectors (manufacturing and construction) 0.54 0.52 (0.11) (0.12) Unemployment Unemployment rate 0.16 0.11 (0.006) (0.04) Females Share of active females 0.26 0.27 (0.04) (0.04) Age 15-17 Share of population ages 15-17 0.01 0.01 (0.03) (0.01) Age 18-50 Share of population ages 18-50 0.26 0.27 (0.04) (0.02) Age 51-65 Share of population ages 51-65 0.04 0.01 (0.01) (0.01) No education Share of working age population with no education 0.08 0.07 (0.06) (0.02) Low education Share of working age population with at least 0.33 0.35 secondary education (0.07) (0.04) High education Share of working age population with at 0.57 0.57 least tertiary education (0.11) (0.06) Urban Share of working age population in urban areas 0.61 0.62 (0.22) (0.04) SPEI at 12 months Average value of SPEI at 12 months -0.21 0.37 (0.68) (0.58) Notes: Sample size is 480 (24 governorate × 20 year cells). Standard deviation in parenthesis. 30 Table 2 – Dynamic and average treatment effect on labor outcomes Panel a: Agricultural employment Estimate SE Effect (t0) -0.106*** 0.031 Effect (t1) -0.074** 0.040 Effect (t2) 0.020 0.035 Effect (t3) 0.028 0.044 Effect (t4) 0.018 0.036 Average -0.052 0.050 Effect (t-2) -0.001*** 0.000 Effect (t-3) -0.023*** 0.000 Panel b: Manufacturing and construction Estimate SE Effect (t0) 0.051** 0.025 Effect (t1) 0.054*** 0.020 Effect (t2) 0.005 0.030 Effect (t3) -0.035 0.027 Effect (t4) -0.006 0.031 Average 0.032 0.053 Effect (t-2) -0.040 0.000 Effect (t-3) 0.039 0.000 Panel c: Other sectors Estimate SE Effect (t0) -0.006** 0.018 Effect (t1) -0.007 0.007 Effect (t2) -0.003 0.007 Effect (t3) 0.013 0.021 Effect (t4) 0.018 0.018 Average 0.007 0.032 Effect (t-2) 0.032*** 0.000 Effect (t-3) 0.045*** 0.000 Panel d: Unemployment Estimate SE Effect (t0) 0.063*** 0.023 Effect (t1) 0.014 0.028 Effect (t2) -0.014 0.038 Effect (t3) 0.014 0.019 Effect (t4) -0.016 0.019 Average 0.028 0.038 Effect (t-2) 0.002*** 0.000 Effect (t-3) -0.061*** 0.000 N 480 Notes: 95 percent CI: ***. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) t-1 is assumed as omitted category. Confidence intervals are at 95 percent. Governorate linear trends are included. Dynamic effects computed over -2 and 4 years. 31 Appendix Table A1: Drought Classification Based on the SPEI Drought category SPEI range Normal -1.0 < SPEI < 1.0 Moderate drought -1.5 < SPEI ≤ -1.0 Severe drought -2.0 < SPEI ≤ -1.5 Extreme drought SPEI ≤ -2.0 Notes: Source: https://olc.worldbank.org/ system/files/Drought%20Indices-EO4SD%20CR% 20SPEI%20index%20products.pdf Figure A1: First stage – average drought intensity, before and after the first drought episode Notes: The figure shows event study estimates of the effect of the first drought shocks on the drought intensity over the period 2000- 2019. Estimates are obtained using the did_multiplegt Stata command by D'Haultfoeuille and de Chaisemartin (2020). Standard errors are clustered on governorates and cohorts. Bootstrapped (250 reps.) confidence intervals are at 95 percent. 32 Data construction Drought shocks – To build our climate shock measures, we use daily weather data from the Agri-4-Cast (A4C) database described in Section 2. Total rainfall precipitation, minimum and maximum temperatures are input variables to calculate SPEIs at different time scales using the SPEI R package. As A4C data come on a grid of 25×25 km, we first calculate SPEIs at 12 months in each grid point in order to reduce noise due to spatial interpolation. This implies that each governorate initially has multiple SPEI measures. To obtain governorate -year shock discrete indicators, we calculate the average SPEI values in each grid point and collapse data on governorate-year cells and weighting SPEI values by local population. The discrete treatment is obtained assigning for each governorate-year cell a value from 0 to 3 according to table A1, where the untreated governorates show an average SPEI higher than -1. The final climate dataset is composed of a balanced panel of 24 governorates observed over 20 years with dummy indicators equal to one whether a governorate experiences a severe drought shock in each year. 33