Job displacement and reallocation failure. Evidence from climate shocks in Morocco

Abstract

We investigate the impact of severe drought shocks in Morocco’s agricultural sector. Using a staggered difference-in-differences design, we estimate that climatic shocks resulted in a job displacement of approximately 6.5 percentage points for workers exposed to severe drought events. Additionally, we observe that, overall, approximately 39 per cent of these workers remained unemployed, leading to a partial reallocation failure. These effects are significant only for severe and extreme shocks, persist for at least 5 years, and are more pronounced among informal and female workers.

1. Introduction

Climate change is generating an unprecedented pressure on the agricultural sector as the increased variability of weather translates into extreme events that are largely responsible for short- and medium-term impacts (Day et al., 2019; Stevanović et al., 2016; Key and Sneeringer, 2014). The frequency, duration and intensity of these events, in particular droughts, are expected to dramatically intensify in the future (Fischer et al., 2021; Chiang et al., 2021). Recent estimates point at a reduction in the global agricultural total factor productivity by 22 per cent (Ortiz-Bobea et al., 2021). However, the regional distribution of these impacts is largely heterogeneous and climate conditions strongly interact with the local socioeconomic context; therefore, economies that are heavily dependent on the agricultural sector, with weaker institutions and poor safety nets, are those paying the highest toll (Hallegatte et al., 2018). These interactions are already manifesting with a greater intensity in the African continent, an area that is considered a climate hotspot (Blunden and Arndt, 2020). While the direct impacts of weather variability on land productivity are well documented (Schleussner et al., 2018; Dell et al., 2014; Deschênes and Greenstone, 2007; Di Falco et al., 2011: among others), we know much less about how climate extremes spill over into the labour markets. This gap is particularly evident in Middle East and North-African (MENA) countries, where the limited availability of data leaves policy makers with little guidance (Cramer et al., 2018).

In this paper we contribute to address this gap by investigating the effects of severe drought shocks on the Moroccan labour market using individual labour-force surveys aligned with granular weather data at the provincial level. In a staggered difference-in-difference setting, we test whether agricultural workers—who are the most exposed—faced a job displacement as a consequence of severe drought events that occurred in Morocco from 2000 to 2009, and—by looking at the employment rate of other economic sectors and the unemployment dynamic—a reallocation effect took place.¹

In industrialised countries, the employment loss in sectors largely exposed to climate shocks can be compensated by employment gains obtained in less exposed, or even new sectors of the ‘green economy’ (ILO, 2018b). However, in less developed countries, the risk of job displacement increases dramatically when climate-exposed activities account for a big portion of the GDP and labour force. Therefore, while industrialised countries are experiencing an adaptation process oriented towards transition to a green-economy, developing economies are struggling to adapt because they often lack the economic and institutional capacities needed to address these challenges; at the same time they are burdened with structural factors that increase the risk of adaptation deficit (Asfaw et al., 2018; Barrett et al., 2001). As a result, climate shocks can be accompanied by a reallocation failure and a dramatic increase in unemployment. Morocco is an interesting case in this regard, and is quite representative of the typical socioeconomic and climatic conditions of many other MENA countries (Clementi et al., 2023).

Morocco has experienced a sustained economic growth in the last two decades; however this has not been accompanied by the effective structural transformation needed to respond to a higher demand for high-skilled workers. Labour productivity remains low; an illiterate workforce accounts for about one third of the total and the agri-food sector, which is mostly low value added, still absorbs more than 30 per cent of the total workforce (Lopez-Acevedo et al., 2021). Over the years informality has remained over the years stubbornly high especially among women and youth. Finally, the labour market is not very inclusive: Morocco posts one of the lowest female labour force participation rates in the developing world (Belhaj et al., 2022).

From a climatic point of view, Morocco and most of the MENA region countries, 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 per cent in a scenario with the temperature increase of 4.8 that has been projected for 2100, while the Moroccan GDP projected impacts range from −3 per cent to +0.4 per cent by 2050 relative to 2003 (Pörtner et al., 2022). In the MENA region, it is projected that surface water availability will be reduced by 5–40 per cent from 2030 to 2065 compared to the availability from 1976 to 2005, with decreases of runoff from 10 to 63 per cent by mid-century in Morocco and Tunisia (Pörtner et al., 2022). In addition, the Morocco’s Ministry of Equipment and Water has declared that the country had experienced the fourth consecutive year of low rainfall and climatic disruptions and is currently facing the worst drought in 30 years.² Therefore, the agricultural sector will be continuously exposed to climate shocks that will make production outcomes highly uncertain (Moriondo et al., 2016).

The literature has largely documented how in developing economies international migration represents an important margin of adjustment for workers to cope with different types of climatic shocks (Kaczan and Orgill-Meyer, 2020; Gray and Mueller, 2012; Marchiori et al., 2012; Dillon et al., 2011). Nevertheless, the cost-opportunity offered by migration is largely heterogeneous (Cattaneo and Peri, 2016), and is more likely to take place with moderate but persistent shocks (Di Falco et al., 2022). Crop and labour diversification also constitute two well-known adaptation strategies in rural areas. However, while crop diversification is often adopted as an ex ante practice (Asfaw et al., 2018; Alfani et al., 2021; Aragón et al., 2021), labour diversification represents both a risk reducing ex ante strategy and ex post coping strategy to face unexpected and severe events such as droughts or floods. By and large, a negative shock in agricultural productivity is expected to drive farmers to allocate more time to economic sectors that are less climate-sensitive, generating an agricultural job displacement that may be associated with a rural-to-urban migration (for a recent review of this phenomenon, see (Cattaneo et al., 2020)). In such cases, workers could search for seasonal or new permanent local off-farm wages to compensate for their loss of agricultural income (Gröger and Zylberberg, 2016; Ito and Kurosaki, 2009). The switch to off-farm activities may be of a transient nature, particularly when motivated by the ex-post climate shock disinvestment in agricultural physical assets (Mueller and Osgood, 2009). The diminished value of these assets is expected to decrease the product of agricultural labour up to the time of their replenishment, which may span multiple periods (Kazianga and Udry, 2006). Still, when markets are not perfectly developed, a climate-induced reduction of the agricultural productivity depresses the aggregate demand, preventing the non-tradable economic sectors from absorbing the excess of rural labour supply (Foster and Rosenzweig, 2007). This effect is more likely to hit low-skilled workers, most of whom are employed in the agriculture sector (Emerick, 2018).

Despite there is an increasing debate among institutions and policy makers about the ‘winners’ and ‘losers’ of a changing climate era (ILO, 2018a), we have a growing evidence of the consequences of extreme climatic events on the job displacement. In Brazil, Albert et al. (2021) found that areas affected by abnormal dryness experienced a sharp reduction in population and employment in agriculture and services, with manufacturing absorbing only a small portion of the displaced workers. In West Africa, Elmallakh and Wodon (2021) have shown that climatic shocks led to an increase in female labour force participation, while the opposite results has been observed among women in India (Afridi et al., 2022). In China, Li and Pan (2021) and Minale (2018) found no impacts on the employment status caused by higher temperatures, even though they observed that workers do leave the agricultural sector and engage rural to urban migration as a response to abnormal temperatures or reduced precipitations. Jessoe et al. (2018) have found that hotter temperatures reduce labour opportunities in rural Mexico, and Branco and Feres (2021) have documented similar effects due to water scarcity in Brazil. Finally, in India, Emerick (2018) has estimated a modest increase in the non-agricultural labour share due to exogenous increases in agricultural output, while Colmer (2021) found that a temperature-driven decrease of agricultural labour demand is attenuated by a job reallocation in non-agricultural sectors. While we have immensely benefited from these studies, Jessoe et al. (2018) highlight that ‘apart from the channel of migration, little is known about the effect of rising temperatures on rural employment in less developed countries.’

Building on previous studies, our paper offers three main contributions. Firstly, we explore the effect of severe drought conditions on the labour market. Using the Standardised Evapo-transpiration Precipitation Index (SPEI), we identify the Moroccan provinces that have been hit by extreme drought events and frame the analysis in a staggered difference-in-differences setting in order to estimate the effects on three outcomes: the employment rate in the agricultural sector (including the most climate-sensitive activities), the employment rate in other sectors, and the unemployment rate. This provides an overall picture of the labour market dynamics at the province level for a period of up to 5 years after the drought’s shock.

Secondly, we conduct a rich analysis of the effects of heterogeneity by disentangling the impacts across age, gender, educational levels and formal/informal work participation for each of the three outcomes considered. Therefore, our analysis also speaks to the important strand of literature on climate injustice, which highlights how climatic shocks tend to have a significant distributional impact with unequal effects among the most vulnerable and most exposed population groups (Sovacool, 2013). Finally, we explore how drought events of different intensities affect the labour market; this provides a useful picture for understanding how the market responds to drought shocks of increasing severity, and how the policy response should be targeted accordingly.

Our estimates show that in Morocco severe drought events, measured by a 12-month SPEI lower than −2 s.d., caused a drop of up to 6.5 percentage points (p.p.) in the agricultural employment rate compared to provinces that were not affected by severe drought. In the same period, we also observe an non-significant increase of 4.8 p.p. in the employment rate of other economic sectors, and a significant increase in the unemployment rate of nearly 2.5 p.p. Even though our data do not allow observing the same workers over time, overall, those figures provide strong evidence of a large displacement of agricultural workers and a partial reallocation failure, considering that about 38 per cent of displaced workers remain unemployed.

We also find that these climate-induced impacts are unequally distributed, with informal workers and females the most affected. In addition, the sensitivity analysis based on an increasing treatment intensity, shows a sharp increase in the magnitude and significance of the impacts when drought shocks become severe, corresponding to SPEI values lower than −1.8 s.d. in all three labour market outcomes. Overall, our results are robust to various specifications and falsification tests.

The remainder of the paper proceeds as follows. Section 2 presents the data, our measures of drought shocks and some descriptive analysis. Section 3 outlines our empirical strategy while Section 4 presents the results and Section 5 presents a set of robustness checks. Section 6 offers a discussion of the findings and some policy implications. Additional information on the data employed and empirical analysis can be found in the  Appendix and Supplementary Material.

2. Data

We employed two data sources. First, we used the Enquête Nationale sur l’Emploi (ENE), which provides nationally representative socioeconomic information at the individual level from 2000 to 2009. The ENE is conducted by the Haut Commissariat au Plan (HCP) and consists of demographics and labour market data. The sampling follows a two-phase stratification approach with an urban–rural strata and a regional strata. Our analysis is restricted to the period from 2000 to 2009 since the provincial identifier was not provided in the survey rounds from 2010 to 2019.³ Despite the richness of available information, the survey consists of repeated cross-sections, which does not allow for tracking individuals over time. We started with a initial sample of nearly 2.7 million observations across 54 provinces. After restricting the sample to working age individuals (between 15 and 59 years old),⁴ since our treatment is at the provincial level we collapsed the data into province × year cells to obtain yearly outcomes of the labour markets. Considering the non-panel structure of our survey data, this aggregation does not come at the cost of losing information.⁵ Therefore, in order to identify the impact of a drought shock on labour markets we leveraged its within-province variation across years. Exploiting the details of employment status and the economic sector codes (NACE classification) for employed individuals, we focused on three labour market outcomes at the provincial level: the share of employment in the agricultural sector; the share of employment in the other economic sectors; and the unemployment rate. Other information at the provincial level includes the share of population across five age groups; the gender of the workforce; the share of educational levels across the population; the share of informal jobs⁶; and the share of the workforce population living in urban areas.

We aligned ENE socioeconomic data with detailed historical 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 agricultural sector. The data consists of gridded meteorological observations from weather stations interpolated on a |$25\times25$| km grid. They are available on a daily basis from 1979 in the European Union and its neighbouring countries, including Morocco. We selected variables for maximum, minimum, and mean air temperature (in Celsius degrees) and sum of precipitation (mm/day) to calculate the Standardised Precipitation Evapotranspiration Index (SPEI) from 1980 to 2009, our 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 collapsed the data to obtain medians of minimum SPEI values in each province × year cell. Considering the large density of point measures in each province, this procedure minimises the risk of assigning drought shocks to provinces that had only a negligible portion of territory exposed to extreme shocks. Moreover, to account for different population density in specific areas, we weight our SPEI measures by population density at 2005 using data from the Gridded Population of the World (GPW) dataset provided by the NASA Socioeconomic Data and Applications Center (SEDAC). Additional details on data construction are available in the  Appendix.

We considered SPEIs at two different time scales, 12 and 3 months, since SPEIs that are calculated at different accumulation periods capture different types of drought shocks. Specifically, an accumulation period of three months captures short- and medium-term moisture conditions and represents a good early warnings, since drought usually takes a season or more to develop. SPEI at longer duration scales, for example 12-months, are more appropriate to account for interseasonal precipitation patterns over a medium duration timescale (Svoboda et al., 2012). As mentioned in Section 1, the duration of drought also affects the response of the labour markets differently: longer and more severe drought periods are likely to generate more significant and persistent effects.

Unlike McGuirk and Burke (2020), who considered moderate and severe shocks, in order to identify our treatment group on both 12-month and 3-month SPEIs, we calculated binary variables with SPEI values |$\leq -$|2 s.d. According to Table B1, this threshold identifies extremely severe droughts, corresponding on average to an event probability of one in 50 years. The low probability of occurrence of these events makes them plausibly interpretable as as-good-as-random, allowing us to minimise the bias due to the sorting of workers into less exposed areas. Shocks of moderate intensity are expected to cause lower long-lasting displacement effects in the labour market; such shocks may, however, affect worker productivity, has been documented by Emerick (2018). We test this assumption in Section 4, where we show that weaker drought events, classified at least as ‘moderate,’ have no impact on either job displacement of agricultural workers or overall unemployment.

Figure 1 shows the national average (the dashed line) and the minimum value (the solid line) of the 12-month SPEI observed over the period 2000–2009. The national average SPEI ranges around −0.5; when we consider the extreme negative values, at least five periods of severe drought have taken place. To better characterise our empirical setting, we considered SPEIs at provincial level, and the associated classification of drought intensity reported in Table B1. Accordingly, we classified as treated, provinces in which the SPEI value assumed a value lower than −2 s.d. over the period 2000–2009; this threshold corresponds to ‘severe’ drought events. Provinces that never experience severe droughts constituted our control group. The distribution across years of these groups are reported in Table 1. Finally, Table 2 shows the summary statistics, for the treated and untreated provinces, for all of the socioeconomic and climatic variables described above.

3. Empirical strategy

Our goal is to estimate the causal effect of a drought shock on various labour market outcomes at the provincial level in Morocco. Since drought occurs at different points in time across provinces, we frame the analysis in a staggered difference-in-difference (DiD) settings that exploits two sources of variation: the cross-sectional variation in the probability of experiencing a SPEI lower than −2; and the variation in timing of the observed years when the drought hits a province. In a standard dynamic two-way fixed effects (TWFE), the following specification is estimated:

where y_pt is one of the three outcomes considered, i.e. the employment rate in the agricultural sector, the employment rate in other sectors and the unemployment rate in province p and year t, α_t are year fixed effects, γ_p are province fixed effects and |$\epsilon_{p,t}$| is an idiosyncratic error term. |$\mathbb{D}^{e}_{p,t}$| is a distance-to-event indicator being e periods away from the year when the province p experienced for the first time a severe drought event over the observed period. We consider absorbing treatment processes, assuming that once a province is shocked it remains treated for the remainder of the panel length. β_e are the parameters of interest that measure the marginal difference in the labour outcomes between treated and control provinces after e years of exposure to the treatment, relative to the same difference in |$e=-1$|⁠.

Recent studies have demonstrated that when units are treated at different points in time, as in our case, estimates from a TWFE can be biased because of the negative weight problem.⁷

Among the alternative methods used to address these limitations we follow Callaway and Sant’Anna (2021), who developed a disaggregated causal parameter, the group-time average treatments effect (⁠|$ATT(g,t)$|⁠), in which the group is defined by the provinces that receive the first shock in a common year. Under parallel trends and without anticipation effect, the |$ATT(g,t)$| is identified by comparing the expected change in outcome for group g between periods g − 1 and t to that for a control group, as follows:

where the control group space |$\mathcal{G}$| can include either the never-treated or the not-yet treated provinces. Building on the |$ATT(g,t)$|⁠, it is possible to obtain several aggregate parameters of interests such as the weighted average, by group size, of the |$ATT(g,t)$| across all groups and periods, or the simple average ATT for all groups across all periods. Further, the dynamic ATTs, by the length of exposure e, can be estimated in an event study setting with the proper weighting:

where w_g weights the groups equally or according to their relative frequencies in the sample of treated provinces. The associated event study plots can be used to see whether treated and control provinces were in parallel trends in the period before the shock.

All of our estimates were obtained using a doubly-robust inverse probability weighting (Sant’Anna and Zhao, 2020) over a balanced panel of provinces with a maximum length of time, e, of five years around the first drought shock. We used clustered bootstrapped standard errors at the provincial level and accounted for autocorrelation in the data (Kline and Santos, 2012).

4. Results

4.1. Main estimates

We begin by presenting plots of the dynamic ATTs in an event study setting. Our preferred estimation focuses on the 12-month SPEI shock as the treatment, and the never-treated provinces as the control group. Panel A in Figure 2 shows the effect on workers employed in the agricultural sector, the sector most exposed to climate change and therefore the sector in which most of the workers are at risk when there is a drought shock. The ATTs support our hypothesis. We observe a significant drop in the share of agricultural workers after the drought shock, which remains statistically significant at 5 per cent in all the post-event periods. The magnitude of the impacts increases over time up to the fourth year, when it reaches about 6.5 p.p, then it loses in intensity in the fifth year dropping back down to 5 p.p. Importantly, we also observe the absence of a significant trend in the pre-treatment period; this strengthens the causal interpretation of our results.

We now present the results on the employment rate in other sectors, and on the unemployment rate, which helps to explain the labour market dynamic concurrent with the displacement of workers employed in the agricultural sector. Panels (b) and (c) report, respectively, the effect of experiencing a drought shock on the employment share in less-climate sensitive sectors, and on the unemployment rate. We observe a larger, though barely significant, increase in the share of workers employed in non-agricultural sectors (up to 4.5 p.p. after 4 years) and a smaller but highly significant increase in the share of unemployed, which reaches a maximum and remains stable at about 2.5 p.p. after the third year.

Finally, for the sake of comparison, in Table 3, we present a few aggregated ATT measures obtained starting from the group-specific ATT and for several control groups, as explained in Section 3. In Panel A, we show the ATTs using the never treated provinces as control group, while in Panel B we report ATTs when the not-yet-treated provinces are included. In both cases, we show the simple group-weighted ATT observed over all groups and periods⁸ and an aggregation of all the point estimates for the post and pre-treatment periods obtained in the event study setting.⁹

The aggregated parameters in Panel A confirm what we observed above. Considering all periods and groups, we observe that the share of employment in agriculture in the treated provinces is about 4.3 p.p. lower, with respect to the difference with the never treated provinces in the pre-shock period, than it would have been if the drought had not occurred. At the same time, there is a non-significant larger share of 2.5 p.p in the employment rate in other economic sectors. We also show that the unemployment rate of shocked provinces is significantly larger by about 1.7 p.p. This means that about 39 per cent of the displaced agricultural workers remains unemployed.¹⁰ The aggregation of ATT parameters represented in the event study plots points to the same direction and magnitude as the simple ATT. Importantly, we observe an average of virtually-zero and non-significant effects over the entire pre-treatment period, which supports the parallel trend assumption. In the same manner, parameters in Panel B confirm that, even including the not yet treated provinces as the control group, the magnitude and direction as ATTs do not change, except for the aggregate ATTs of the event study, which are slightly larger for both employment in the agricultural sector and the unemployment rate.

4.2. Effects of heterogeneity

In this section we explore the effects of heterogeneity by analyzing the impacts along four important margins, i.e. age, gender, educational level and formal vs. informal jobs.

4.2.1. Age

To begin with, Figure 3 shows the heterogeneous effects across five different age groups. The drought impact appears much stronger and more significant among middle-aged workers (ages 18-50), among whom we observe a drop of about 2.5 p.p. in the agricultural employment rate in the second year, reaching about 7 p.p. in the fourth year. At the same time, the employment rate in other sectors moves in the opposite direction, though with a non-significant magnitude. We also observe an increase in the unemployment rate, with a coefficient that becomes significant and stable at about 2.5 p.p. after the third year, which signals a reallocation failure for only for a small share of middle-aged workers. We find non-significant effects both for workers under the legal age (15–17) and for workers close to retirement (51–60), with the exception of a sharp but temporary increase in the unemployment rate in the group of very young workers in the first year after the shock.

4.2.2. Gender

Figure 4 shows the heterogeneous effects by gender. While overall we find significant effects in both females and males, the temporal dynamic appears very different. The drop in the agricultural employment rate is U-shaped for female workers, reaching 9.8 p.p. in the fourth year, vis-á-vis 6 p.p. for males. However, among female workers the effect decreases to about 5.1 p.p. after 5 years, while it steadily increases for males. As a consequence of the lack of significant evidence for reallocation in other sectors, the unemployment rate for females increases up to 4.9 p.p., more than twice what is observed for males.

4.2.3. Education

Figure 5 reveals the existence of heterogeneous effects in relation to educational level. Our empirical analysis uncovers significant impacts on the unemployment rate concerning individuals lacking a formal educational qualification or possessing a low educational level. Specifically, the latter group experiences an increase in unemployment of nearly 5 p.p. in the second year since the shock. Conversely, the former group exhibits a negligible effect that dissipates within 1 year. This finding suggests that individuals with higher education encounter greater challenges in achieving labour market mobility within shorter timeframes, compared to their counterparts with no educational attainment. Nonetheless, the investigation of other labour outcomes does not yield statistically significant findings or displays unreliable delayed effects. The imprecision of results extends to the examination of high education levels, where observed pre-trends appear less aligned with the parallel trend assumption and prevent a causal interpretation of the estimated coefficients.

4.2.4. Formal vs. informal employment

One of the most heterogeneous effects emerges when we look at formal and informal workers, reported in Figure 6. We observe that formal agricultural workers are only weakly and non-significantly affected by drought shocks. Consequently, we do not find any evidence of reallocation of these workers to other sectors. On the contrary, we observe a strong and highly significant drop of workers informally employed in agriculture, and a reallocation effect of similar magnitude and temporal dynamic toward the other economic sectors. More specifically, the drop in the agriculture employment rate is about −2.5 p.p. already during the first year after the shock and reaches a maximum of about −8 p.p. after 4 years, with a symmetric dynamic of the employment rate in other sectors.

4.3. Treatment intensity

A relevant issue for policy makers and scholars is whether there is a threshold in the intensity of the shock that significantly affects the labour market. Additional knowledge about such a potential threshold could indeed help better set and target policy response, and reducing the underlying uncertainty involved in addressing climate impacts. In this section we address this important point by looking at how province labour markets in Morocco have responded to droughts of different severity as measured by varying SPEI values. We calculate dummy indicators for SPEI bins of 0.05 s.d. in a range −2.3 to 1.3 s.d. According to the classification reported in Table B1, this range captures droughts from very moderate to extreme intensity.

Figure 7 plots the effects by treatment intensity in the three outcomes of interest: the employment rate in agriculture (Panel A); the employment rate in other sectors (Panel B); and the overall unemployment rate (Panel C). The first important evidence emerging from these figures is the absence of effects for drought classified as ‘moderate’ (SPEI ranging from −1.5 to −1 s.d.). As the shock increases in intensity, we observe a rapid and significant decline in the agricultural employment rate (Panel A). Specifically, for SPEI values lower than about −1.8 s.d. we observe a drop of about 5 p.p., which remains stable and significant for shocks of greater intensity. At the same time, for similar SPEI values, we observe a corresponding increase in the unemployment rate of about 2 p.p., which then steadily increases up to about 3.5 p.p. as the shock becomes more severe. With SPEI values lower than −2 s.d.¹¹ the number of shocks becomes lower or even absent in many provinces, and we lose precision in the estimates: both the drop in the agricultural employment rate and the unemployment rate show larger standard errors.

5. Robustness checks

5.1.1. Effects of short-term drought

The impacts of drought can be different depending on the timescale considered. In the previous section, we present the effects of mid-term drought shocks using a SPEI calculated at 12 months. In this section, we replicate the main analysis using a 3-month SPEI to test the robustness of our findings when considering drought shocks of shorter duration. Figure A1 shows that estimates with a treatment of a 3-month SPEI produces a very similar pattern of results in magnitude, but is much less precise. We observe a maximum decrease of about −0.8 p.p. in the agricultural employment rate after 4 years (significant at 10 per cent) and an increase in the employment rate of 0.75 p.p. after 6 years (significant at 1 per cent), while the employment rate in other sectors reach a maximum increases of about 0.04 p.p. in the fourth year. Overall, the analysis of drought shocks of the same magnitude but shorter duration confirms the direction of findings we obtained using a 12-month SPEI, but without significant impacts.

5.1.2. Reshuffling of shocks

An additional check we present is a falsification test that alters the combinations of shocks across years. We therefore run our preferred regression specification on a set of randomised shock years, maintaining the same composition of treated and control provinces that were used in the original sample. Since drought occurs in specific years, by altering the time of onset one would expect to find non-statistically significant effects. In Figure 8 we report event study results for the three outcomes considered, from which we observe the absence of significant effects after the shock occurs. In addition, estimates of aggregate ATTs, reported in Table B3, confirm that the placebo results using both 3- and 12-month SPEIs, do not statistically differ from zero. Altogether, these tests provide evidence that our main estimates are not driven by some underlying systematic trend in the data, and that the validity of our identification strategy is warranted.

5.1.3. Conditional parallel trend estimates

Our estimates in Section 4 assume that the parallel trend assumption holds unconditionally. We therefore present conditional parallel trend parameters, controlling for a set of observed provincial characteristics of the working age population, that is age, female population, educational level and urban population. Comparing those ATTs with the parameters in Table 3, we observe no differences in the ATT of the employment in the agricultural sectors, while slightly less evidence is detectable for the unemployment rate. To strengthen the causal interpretation of our conditional estimates, Table B4 shows a balancing test of the covariates adopted in the conditional estimates, in which we regress each control variable on the treatment; that is, the drought shocks measured with the 12-month SPEI values lower than −2 s.d., controlling for year and province fixed effects, and regional-specific time trends. With the exception of educational level, which, however, is only weakly statistically significant, none of the estimated coefficients are significant, confirming that the characteristics for which we controlled are predetermined and are only contributing to increasing the precision of our estimates.

5.1.4. Additional robustness checks

As additional robustness checks, we address two potential concerns: the effect of multiple shocks within a province and, the mirror effect due to positive SPEI shocks (severe wet conditions). To address the former, we adopt the recent de Chaisemartin and D’Haultfoeuille (2020) estimator which allows for binary switch on/off multiple treatments. Estimates are shown in the  Appendix Figure A2 and confirm our main results.¹² For what concerns potential mirror effects, our data exclude that in the sample there are provinces that experience both negative and positive severe shocks in the same year. Then, we estimate the model using SPEI values |$\gt= 2$| s.d. as treatment measure and find non-significant effects on the three outcomes considered (see Supplementary Material).

6. Discussion and conclusions

In this paper we explore the impact of severe drought in the labour markets of Moroccan provinces during the period 2000–2009. We use nationally representative labour force surveys and the SPEI, a state-of-the-art climate indicator, to estimate the causal effects of severe drought shocks on labour market dynamics by observing the employment rate in agriculture, in other less-exposed sectors and in the unemployment rate. Our staggered diff-in-diff estimates shed a light on the effects occurring in Morocco, an important MENA country that is considered a climate hotspot and for which little evidence is available. Importantly, Morocco is a good case study to investigate the interacted effects of climate extremes on the labour markets in a socioeconomic context characterised by high demographic expansion, a growing labour force and structural market frictions.

We find that in provinces that experienced a severe and prolonged drought, measured by a 12-month SPEI value lower than −2 s.d., the share of employment in agriculture drop of about 4.3 p.p., on average. The effect takes place already in the aftermath of the shock and is long-lasting: the coefficients remain significant and negative up to 5 years after the drought occurred. Moreover, we observe a contemporaneous and significant increase in the unemployment rate, which signals that those shocked workers found difficulty in accessing alternative jobs in less climate-exposed sectors. These figures point to general reallocation failure as a consequence of severe drought shocks, a failure which is more pronounced for the most vulnerable workers.

The effects we find are strongly heterogeneous and penalise informal workers and, to a lesser extent, women and younger individuals. This evidence signals the presence of pronounced climate injustice in the labour markets of Morocco. In particular, women represent the population group for which we observe the largest displacement effect and the weakest adjustment mechanism in terms of reallocation to other economic sectors. Middle-aged workers (ages 18–50) also face significant displacement but, at the same time, they seem to experience limited market frictions when looking for a job in other sectors. Considering the ‘premature de-industrialisation’, industry is not able to absorb the excess of labour supply from the agricultural sector. Therefore, displaced workers are more likely to relocate in the tertiary sector, mainly in tourism or services, which does not rely on local demand. On the contrary, we observe an effective relocation process in informal employment, in which the dynamic between the employment rate in the agricultural and the other sectors is perfectly symmetric. This evidence is in line with empirical studies that show that informal workers react counter-cyclically during market shocks (Johannes et al., 2009; Loayza and Rigolini, 2011); this represents a safety net for displaced workers, who benefit from a greater flexibility in finding jobs in alternative economic activities. Still, even though informal employment may represent a temporary asset for displaced workers in agriculture, in the mid- and long-run perspective this translates into a weaker labour market attachment and a considerable obstacle to the necessary structural transformation of the labour market (Lopez-Acevedo et al., 2021). Overall, the heterogeneous effects we find generate concerns about an exacerbation of climate injustice in the future, if we also consider the historic trend in MENA countries characterised by a large unemployment among youth and females (Alfani et al., 2023).

Another consideration relates to the analysis of shock intensity and the associated effects. This set of results can help policy makers address the impact of droughts on the basis of a standard classification of shock intensity, that has been observed to produce significant effects in the labour market due to specific intensity thresholds. In particular, we find significant and negative effects in the labour market only for severe and extreme drought shocks of long duration—those corresponding to SPEI 12 values lower than about −1.8 s.d. At these intensities, we observe a substantial drop in the share of workers employed in agriculture and a simultaneous increase in the unemployment rate. On the contrary, we do not detect significant negative effects for drought shocks classified as moderate (SPEI values greater than 1.7 s.d.). Further, short duration shocks, such as those related to SPEI 3 months, do not cause a significant change in the share of labour allocation between climate sensitive and non-sensitive sectors. In this context, the more prolonged and intense the shock, the larger the dis-investment of agricultural assets as ex-post coping strategy and the time span needed to replenish them (Mueller and Osgood, 2009). The persistence of the effects up to the fourth year after the shock, seems to confirm this hypothesis. In light of this evidence and considering the increasing trend in drought events in MENA, policy makers should consider interventions to mitigate the transient displacement effect of the most extreme events subsidising insurance for the agricultural activities or providing structural safety net measures (Bjerge and Trifkovic, 2018); it should be noted however, that previous studies have documented that milder events do not come without costs in terms of loss of productivity (Bedi et al., 2022).

Our analysis needs, however, to be hedged with some caveats. First, results rely on a time period (2000–2009) that is only relatively recent, because of missing information on workers’ location in the subsequent labour force surveys. Therefore, we cannot extend the validity of the results to more recent years. In this respect, the obtained results should be interpreted as a lower bound of the stronger displacement effect and reallocation failure, accompanied with higher unemployment, that may occur in the future as a combination of demographic expansion, greater climate variability and a large dependence on climate-exposed sectors. This for two main reasons. First, our study considers a period during which climate variability has been more limited, both in Morocco and, generally in the MENA region, than it has been in the subsequent decade, and compared to recent estimates of future trends (Pörtner et al., 2022). Secondly, Morocco—like other MENA countries—has not experienced a gradual shift of labour and capital from agriculture to manufacturing and services, and the industry value added has remained largely unchanged over the past 20 years (Moussir and Chatri, 2020).

Another limitation is that the individual data employed in this study do not allow for tracking individuals over time. This hinders the possibility to directly observe individual out-migration as a response to shock; nevertheless our outcome variables are normalised to the province labour force and therefore represents the net of migration effects.

Despite these limitations, our study offers new evidence to help policy makers properly tackle and assess the expected climate impacts by planning and developing proper safety-net measures and reforms. The latter should be targeted towards the most vulnerable population groups, i.e. women and informal workers, in order to create the conditions for a more equitable distribution of the impacts and costs of drought.

Acknowledgements

We wish to thank the participants at the IAERE 2022, IFAD 2022 and SIE 2023 conferences, invited seminar at the University of Milano Statale for their suggestions, and two anonymous referees. We also benefited from useful discussions with Luca Citino, Daniele Curzi, Salvatore Di Falco, and Giuseppe Maggio.

Footnotes

References

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 |$25\times25$| km, we first calculate SPEIs at 12 and 3 months in each grid point in order to reduced noise due to spatial interpolation. Then we collapse data at province-year cells calculating the median of each SPEI value. Moreover, to account for different population density in specific areas, in collapsing the data we weight our SPEI measures by population density. We proceed as follows. First, we collected data on population density at 2005 (the year in between our period of analysis) from the Gridded Population of the World (GPW) dataset provided by the NASA Socioeconomic Data and Applications Center (SEDAC). This data is very granular and provides information on population count on a grid of |$4\times4$| km. Secondly, by means of GIS, for each grid point with weather information (available on a grid of |$25\times25$| km), we calculate a circular buffer with a radius of about 10km and then calculate population density for each area delimited by the buffered grid point. Lastly, we use this population density measure to weight our SPEI indexes when collapsing the data from grid-point values to median values at the province level. To obtain province-year shock indicators, we calculate dummy variables to detect SPEI values |$\leq -$|2 s.d. in each province, corresponding to severe drought events. The final climate dataset is composed of a balanced panel of 54 provinces observed over 10 years with dummy indicators equal to one whether a province experiences a severe drought shock in each year. Since shocks occur at different points in time, in our empirical setting we account for variation in the timing of treatment by considering a staggered design, but we do not consider that a province can be exposed to the same shock for multiple periods. This implies that we only consider the first shock.