© 2020 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowl- edge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. COVID-19 in South Asia: An unequal shock, an uncertain recovery Findings on Labor Market impacts from round 1 of the SAR COVID Phone Monitoring Surveys Key Takeaways ▪ The imposition of lockdowns aimed to curb the rising infection rates of COVID-19 in the South Asia Region (SAR) in early 2020 led to a drastic, abrupt disruption of physical mobility and economic activities in all eight SAR countries. Additionally, the continuous promulgation of different forms of local and national restrictions meant that, in most SAR countries, day-to-day mobility remained consistently below pre-COVID levels for an extended period of time between March 2020 and March 2021. ▪ The immediate labor market impacts of these stringent lockdowns and associated contractions in economic activity have been profound and widespread. In every SAR country, more than one in three workers employed in January 2020 (pre-COVID) experienced a negative labor market shock either in the form of a job loss or an earnings loss. For example, in Nepal, which had among the most prolonged and severe lockdowns in the SAR, 30 percent of workers employed in January 2020 lost their pre-COVID job, and an additional 22 percent had reduced earnings. ▪ Job losses were relatively more concentrated in the non-agricultural sector, which raises serious concerns for the long-term sustainability of economic growth and development, given that these sectors have been a key driver of poverty reduction in many countries in the region over the last decade and employ an increasing share of the workforce. ▪ Female workers were also disproportionately affected compared to male workers, which can further worsen the preexisting gender inequalities in labor market outcomes in SAR countries. ▪ The data also shows high closure rates of firms, including larger firms, with a significant loss of entrepreneurial capital. In addition, high job losses among the oldest cohort in many SAR countries raise the risk of early retirements and a permanent loss of human capital, which can impede the recovery process. ▪ Spatial and sectoral differences between firm closures and new firm openings within each country indicate a significant geographic and structural realignment of economic activities that further risks the recovery process. ▪ While a recovery of jobs also followed these observed job losses, the rate of recovery is extremely low across the region. This early stage of recovery is also characterized by employment volatility and multiple job transitions that are likely to continue going forward. ▪ Within-country differences in the recovery rate reveal that females are disproportionately left out of the recovery process. This underscores a worrying trend that female workers could be doubly cursed: they were more likely to have lost their pre-COVID employment immediately following the lockdown, and they have greater difficulty recovering a job in the early stages of the recovery process relative to their male counterparts. ▪ The data also provides strong evidence that less educated and low-skilled workers, who are often among the most disadvantaged and vulnerable in the population, face relatively greater difficulty in finding new employment opportunities during the early stage of recovery. ▪ More importantly, the recovery of jobs has not been accompanied by the recovery in labor earnings, highlighting the continued risks to worker welfare, including for those who are currently employed. ▪ Across all SAR countries, high rates of sectoral and industry transitions have characterized the early recovery process. Such workers’ possible skill misallocation during rehiring is also associated with worsening job quality and lowered labor earnings. ▪ Geographic mobility of labor has played an important role in recovery, with domestic migration significantly improving the likelihood of finding a new job among those who lost their pre-COVID employment in all SAR countries. 2 Migration-induced recovery is, however, associated with a greater likelihood of misallocation of worker skills, along with a decline in job quality and labor earnings. ▪ Migration patterns observed within each SAR country indicate significant costs associated with geographic mobility that increase with distance and constrain specific demographic subgroups, including females and older cohorts, from moving that have hampered their recovery. ▪ While formal social safety net programs can be important for protecting worker welfare during unemployment spells, the high incidence of international migration observed in most SAR countries meant that the share of remittance recipients far exceeded the share of unemployed individuals with access to formal assistance. This highlights the importance of international migration and remittances from absent household members for recovery in the region. ▪ Informal mechanisms like subsistence agriculture have been a critical safety net for many unemployed workers in SAR countries. Recent engagement in subsistence activity is also correlated with longer unemployment spells, thereby providing the most affected with a valuable source of non-monetary income and potentially allowing individuals to prolong their unemployment spell for better future labor market outcomes. 3 Contents Key Takeaways .................................................................................................................................................................... 2 Introduction ........................................................................................................................................................................ 5 Section 1: Labor market impacts of COVID-19 ................................................................................................................. 10 Section 2: The early stage of the recovery process .......................................................................................................... 20 Section 3: Worker skills and the risks to recovery ............................................................................................................ 28 Section 4: The role of migration in the recovery process ................................................................................................. 32 Section 5: Unemployment spells, discouragement, and the role of social safety nets .................................................... 38 Conclusion ........................................................................................................................................................................ 44 References ........................................................................................................................................................................ 46 Annex A: Sampling weights .............................................................................................................................................. 47 Annex B: Additional Tables and Figures ........................................................................................................................... 49 4 Introduction All countries across South Asia, faced with the rising risks of COVID-19 infection rates, implemented severe economic lockdowns in early 2020 with varying frequencies and time periods. While the exact nature and duration of these lockdowns varied across countries in the South Asia Region (SAR), almost all SAR countries imposed their first economic lockdown in late March 2020 in response to the growing health threat of COVID-19 infections. In India, for instance, the national lockdown was first introduced in late March 2020, which coincided with the imposition of similar lockdowns in Bangladesh, Nepal, and Sri Lanka, followed by a national lockdown in Pakistan on April 1, 2020. By April 17, 2020, the population of all SAR countries was under severe lockdown with varying rules and conditions based on national or local directives.1 The introduction of these lockdowns led to a drastic, abrupt disruption in all forms of physical mobility and economic activities. Trends from the Google COVID-19 Community Mobility data reveal this sharp drop in day-to-day mobility related to four different types of economic activity across 6 out of 8 SAR countries for which this data was available.2 Figure 1 plots the daily change in the Google Mobility index, which is constructed by taking an equally weighted mean across the four dimensions of economic activity for the five weeks before March 2020. In the six SAR countries, the average mobility remained approximately, on average, 58 percent below their respective pre-COVID levels during the first week of the lockdown. For example, in Nepal, where the lockdown was first introduced on March 24, 2020, mobility (as measured by the Google Mobility index) was 66 percent below pre-COVID levels on the first day of the lockdown; and it remained, on average, 71.5 percent below per-COVID levels between March 24, 2020, and March 30, 2020. We observe a similar pattern of immediate and large disruptions in mobility in all SAR countries, except in Afghanistan (22.5 percent below pre-COVID levels), where restrictions were more localized. In addition, in many SAR countries, day-to-day mobility remained consistently below the pre-COVID levels for an extended period. This is perhaps not surprising given that in many SAR countries, the initial lockdown period was extended multiple times; additional rules on restrictions and, in some cases, full lockdowns were enforced at a later date, after the easing of the initial lockdown. For example, in Nepal, which had among the most prolonged and severe lockdowns within SAR, a combination of national and local lockdowns was effectively continuously in place until the end of 2020. Similarly, in Bangladesh, a strict national lockdown imposed in late March 2020 for an initial 10-day period was extended multiple times. However, these restrictions gradually eased starting in June 2020, and all restrictions were subsequently lifted by September 2020. The Google Mobility index closely follows these changes in rules and conditions in SAR countries, which varied over time within each country as well as across countries. In countries like Nepal, India, and Sri Lanka, with an extended period of restrictions imposed through national or local directives at different points in time, mobility had not returned to pre-COVID levels even as late as April 2021. In Nepal and Sri Lanka, where the second lockdown was introduced in 1 More details on the imposition of various restrictions and national lockdowns in SAR countries are provided in Table B1. 2 Google COVID-19 Community Mobility data is not available for Bhutan and Maldives. The four types of mobility measures include mobility related to: (i) retail and recreation, (ii) grocery and pharmacy, (iii) transit stations, and (iv) workplaces. 5 August and November 20203, respectively, we observe a sharp drop again in mobility after a gradual recovery following the easing of the first lockdown. In other SAR countries like Afghanistan, Bangladesh, and Pakistan, mobility only returned to pre-COVID levels between September and October 2020. These results underscore the dramatic and prolonged impact that COVID-19 induced lockdowns have had on mobility and economic activity, which is perhaps unprecedented in the region, at least in recent history4. These lockdowns are likely to have important implications on various socio-economic dimensions of welfare, including labor market outcomes, both immediately and in the medium- to long-term. More importantly, the long-term impacts will also be determined by the nature and the pace of recovery observed in these countries in the months and years after the initial phase of lockdown. Moreover, the emergence of new mutants leaves open the possibility of future lockdowns as a policy response to mitigate the health effects of the virus, which could impact economic activity and reverse observed recoveries. Considering this, the first round of the SAR COVID-19 Phone Monitoring Survey (SAR-CPMS) was designed to monitor and analyze the immediate labor market impacts and to assess the ability of the individuals to meet basic needs, and their access to coping strategies when met with an income shock. In this paper, we focus on individual labor market outcomes and the changes to their labor market conditions, along with challenges faced by workers and firms that pose risks to labor earnings during and after the lockdown periods. In addition, we also focus on employment transitions and coping strategies that provide key insights into the early recovery process, highlight mechanisms such as labor mobility that underlie the pace of the ongoing recovery, and determine the nature of the recovery process in the long run. The survey administered a detailed retrospective labor module to consistently track labor market outcomes within and across SAR countries, which measures changes in employment status, job type, sector of employment, earnings losses, and various other coping strategies in response to the negative labor market shock. In particular, the labor module tracks job- and, and to a limited degree, earnings-related outcomes across three separate periods: (i) January 2020 (before lockdowns), (ii) late March 2020 (at the time lockdowns were imposed), and (iii) at the time of the survey, which spans between 5 to 12 months after the imposition of the first lockdown (depending on the survey timeline in each country). The three reference periods allow us to build a retrospective panel for individuals with pre-COVID baseline outcomes in January 2020. In this paper, the changes to the labor market outcomes are studied relative to this baseline for individuals who were employed in January 2020. 3 Sri Lanka did not impose a country-wide lockdown during the second wave (October -November 2020) for fear of bankruptcies and rising unemployment. Zonal and district lockdowns were implemented in places with high infection and doubling rates. By the end of October, many districts were under quarantine curfews. 4 There is little doubt that the lockdowns and mobility restrictions are likely to have had positive public health impacts, especially in curtailing the spread of disease in densely populated environments with limited public health response capacity. However, it is beyond the scope of this work to measure the likely beneficial effects of the lockdowns. Round 2 analysis, which will focus on the impacts of the second Delta wave in SAR which was not accompanied by such stringent mobility restrictions, can provide a comparative analysis on the differing severity of economic impacts. 6 Figure 1: Google COVID-19 Community Mobility trends in South Asia Afghanistan (AFG) Bangladesh (BGD) 40 20 -10 -30 -60 -80 India (IND) Nepal (NPL) 0 20 -20 -40 -30 -60 -80 -80 Pakistan (PAK) Sri Lanka (LKA) 40 10 20 -10 0 -30 -20 -50 -40 -70 -60 -90 Note: The y-axis measures the average day-to-day % change in the Google Mobility index, calculated by taking an unweighted mean of the % change in four different dimensions of mobility related to: (i) retail and recreation, (ii) grocery and pharmacy, (iii) transit stations, and (iv) workplaces. For each measure, day-to-day % change is calculated with respect to the medium value for the corresponding day of the week, in the 5-week period between Jan 3rd – Feb 6th, 2020. Data plotted here are percentage change in mobility from baseline value for the day of the week. The data is plotted between February 15th,2020 (before lockdown) to April 30th, 2020, which is the last day of data collection for the SAR COVID-19 Phone Monitoring Survey. Google mobility data is not available for Bhutan and the Maldives. The blue shaded region shows the survey period in the respective countries. Source: Google COVID-19 Community Mobility Report (https://www.google.com/covid19/mobility/). 7 The SAR-CPMS covered 44,880 individuals across all eight SAR countries (Afghanistan, Bangladesh, Bhutan, the Maldives, Nepal, Pakistan, Sri Lanka, and India). The survey instrument, including the labor module, was harmonized across all countries and was administered in 19 commonly spoken languages in the region. In each country, the survey was implemented using random digit dialing (RDD) methodology, in which respondents are contacted using randomly generated plausible phone numbers.5 To the extent that a valid number is reached and the respondent consents to be interviewed, the survey was designed to roughly represent a mobile phone-owning population in each country.6 To ensure the survey adequately covers key groups within each country, geographical quotas on sub-national sample sizes were used in all eight countries and a minimum 30 percent respondents were required to be females.7 Table 1 presents these key details on the design and implementation of SAR-CPMS, including the sample size in each country and the share of female respondents within each country-sample in Columns 2 and 3, respectively. In addition, Column 1 of Table 1 reports the minimum intended sample size for a nationally representative sample in each country calculated based on the most recent national survey in the respective country. More details on the methodology used to calculate sampling weights used in the survey is presented in Annex A. Table 1: Timeline, sample size, and additional survey details Minimum Realized % female No. of Survey timeline intended sample (3) languages (5) sample size* size available (1) (2) (4) Afghanistan 5000 5053 29% 3 August 20 – November 25, 2020 Bangladesh 7500 7685 31% 2 August 24 – November 23, 2020 Bhutan 1000 1502 47% 7 September 14 – October 24, 2020 India 10000 10281 27% 10 October 27, 2020 – April 28, 2021 Maldives 1000 1943 46% 2 September 18, 2020 – February 27, 2021 Nepal 5000 5170 39% 2 August 16 – November 9, 2020 Pakistan 7500 8153 28% 2 December 1, 2020 – March 31, 2021 Sri Lanka 5000 5072 49% 3 September 21 – December 23, 2020 Note: *Minimum intended sample size is the target sample size for a nationally representative sample in each country, calculated based on the most recent national survey in the respective country. 5 An operator dials numbers from a list of randomly generated numbers, until she/he reaches a working number. 6 The survey excludes households that do not own a mobile phone, or are out of network coverage, or choose not to pick up unknown phone numbers. 7 To ensure a minimum of 30% female respondents, we randomized to ask 50 percent of the respondents to pass the phone to an adult female household member during the call introduction. In cases where the pass is successful, consent is re- registered to the new female respondent. In cases where the pass is not successful, the interview is continued with the original respondent. 8 The data from the first round of SAR-CPMS allows us to examine the immediate to short term labor market impacts of COVID-19 induced lockdowns across all SAR countries with direct implications for immediate worker welfare, as well as potential structural and geographic realignments that can have long-term impacts on the labor market and impede the recovery process. These results are presented in Section 1. In Section 2, we provide a detailed description of the early stage of recovery, along with the distributional impacts that could disproportionately leave certain subgroups out of the recovery process and widen inequality in the region. The analysis in Section 3 documents the nature of employment transitions observed during the early stage of recovery and highlights the risk to both individual wage growth and the country’s long-term recovery prospects due to possible misallocation of worker skills during rehiring. In Section 4, we examine the role of domestic migration for recovery, especially in the context of significant geographic realignments of new employment opportunities within each SAR country and highlight the challenges associated with migration-led recovery for a proper allocation of worker skills during rehiring. Lastly, in section 5, we analyze the role of formal and informal safety nets, including formal assistance, remittances and subsistence agriculture, in protecting worker welfare, and potentially allowing workers to better invest in job search and improve their future labor market outcomes. In addition, future rounds of SAR-CPMS or comparable country-level panel surveys, which can continue to track the evolution of labor market trajectories across different stages of, what is most likely to be, a long recovery process, will be critical to addressing some of these challenges and risks to a robust and inclusive recovery in the region. First and foremost, a longitudinal survey with multiple rounds of data, which tracks the same individuals over time and across their multiple employment stints, will provide a more accurate and detailed description of the recovery process. This will help to analyze the roles of external factors and policies, like cultural norms and assistance programs, in influencing key inter-temporal decisions on firm hiring and worker skill investment that directly affect wage growth trajectory and economic growth prospects in the long-term. In addition, high-frequency data that can be used to closely monitor individuals’ recovery paths and job search strategies over time, can play an important role in terms of better understanding the evolving labor market conditions, including wage adjustments or the lack thereof, along with informing labor market policies to facilitate the recovery process. Moreover, any such future rounds of data can complement the analyses presented in this paper and help examine in greater detail the underlying mechanisms and factors that underlie the risks and challenges highlighted in this paper. 9 Section 1: Labor market impacts of COVID-19 The immediate impact of COVID-19 induced restrictions and lockdowns on labor market outcomes has been widespread. In all eight SAR countries, at least one-third of workers employed in January 2020 (pre-COVID) experienced a negative labor market shock following the lockdown. This share of the negatively affected is the highest in Nepal, where more than half of its working population either lost a pre-COVID job or experienced a loss in earnings (Figure 2). Likewise, the share of workers who experienced similar negative labor market impacts in Bangladesh and Sri Lanka is 49 and 48 percent, respectively. Compared to these countries, workers in India, Pakistan, Afghanistan, the Maldives, and Bhutan that make up the rest of the SAR, experienced relatively muted effects. But even across these latter countries, the share of negatively affected workers (either in the form of job loss or earnings loss) is high, ranging between 33 and 44 percent. Figure 2: Labor market impacts of COVID-19 Job loss Earnings Loss Unaffected 51% 48% 52% 57% 61% 59% 63% 67% 22% 41% 26% 38% 38% 20% 32% 30% 30% 9% 13% 15% 10% 6% 7% 7% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020. The job loss category includes individuals who lost or changed the job, which they were employed in January 2020, between March 2020 and the time of the survey. The earnings loss category includes individuals who, at the time of the survey, were employed in the same job that they had in January 2020 but reported earning less than that before March 2020. In addition, cross-country comparisons show key differences in the severity and type of labor market impacts experienced by workers across the region. In Nepal, 57 percent of the affected workers lost their job, whereas, in Bangladesh and Sri Lanka, most of the affected labor force experienced only a loss in their earnings (83 and 79 percent, respectively), as shown in Figure 3. Job losses were particularly striking in Nepal, where 30 percent of workers employed in January 2020 lost their pre-COVID employment. Nevertheless, job losses were also not trivial in Bangladesh and Sri Lanka either, or in other relatively unaffected countries like the Maldives and Bhutan. In the latter two countries, around 15 percent of workers employed in January 2020 lost their pre-COVID job, and this group makes up almost 40 percent of the affected population in each country. 10 Even in countries like India, Pakistan, and Afghanistan, with relatively fewer job losses, a sizeable proportion of workers still experienced an earnings loss. In these countries, between 30 to 38 percent of workers who were employed in January 2020 and did not lose their pre-COVID job experienced earnings loss measured based on the difference in their earnings at the time of the survey relative to that prior to March 2020.8 If earnings for some workers had already recovered by the time they were interviewed, after an initial loss, the above estimates on earnings are likely to underestimate the negative impacts in these countries. Figure 3: Types of labor market impacts of COVID-19 Earnings loss Job loss 13% 17% 18% 19% 21% 39% 37% 57% 87% 83% 82% 81% 79% 61% 63% 43% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020 and were negatively affected, either in the form of a job loss or earnings loss. The job loss category includes individuals who lost or changed the job, which they were employed in January 2020, between March 2020 and the time of the survey. The earnings loss category includes individuals who, at the time of the survey, were employed in the same job that they had in January 2020 but reported earning less than that prior to March 2020. High job losses in the non-agriculture sector are particularly alarming given that these sectors have been a key driver of poverty reduction in SAR over the last decade, and have accounted for a growing share of employment and domestic output. In 7 out of 8 SAR countries, job losses were relatively higher among individuals employed in the two non-agriculture sectors (manufacturing and services) than those in the agriculture sector (Figure 4).9 For example, in Nepal, which saw the highest rate of job losses in the region, the job loss rate in the agriculture sector is 26 percent compared to 32 percent in the non-agriculture sector (34 and 26 percent in manufacturing and services, respectively).10 8 This subgroup includes individuals who, at the time of the survey, were employed in the same job that they had in January 2020 but reported earning less than that in January 2020. In the survey, respondents were asked to report the salary (or wages) that they earned in the month (or week) preceding the survey date, along with the salary (or wage) that they earned in a typical month (week) before March 2020. 9 One exception is the Maldives, where job losses were higher among agricultural workers (24 percent relative to 16 percent among non-agricultural workers). However, the agriculture sector in the Maldives accounted for only 6 percent of the overall employment in January 2020. 10 Differential job loss rates by sector of employment (and other attributes like age cohort, gender, education, and occupation quality that are examined later in the paper) could be confounded by potential rural-urban differences in the 11 This is particularly worrying because non-agricultural employment accounted for 75 to 95 percent of the workforce in January 2020 across all SAR countries (Figure B1). More importantly, the expansion of jobs in the non-agriculture sector has been a crucial driver of poverty reduction in the region, raising concerns for the long-term sustainability of economic growth and development in the region (Balcázar, et al, 2016; World Bank, 2002, 2016a, 2016b, 2017, 2019a, 2019b, 2019c, 2021a, 2021b). Figure 4: Job loss rates, by sector of employment Agriculture Manufacturing Services 34% 29% 26% 24% 21% 16% 14% 12% 12% 11% 11% 10% 9% 9% 8% 8% 8% 8% 8% 6% 6% 5% 5% 4% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020. The job loss category includes individuals who lost or changed the job, which they were employed in January 2020, between March 2020 and the time of the survey. Figure 5: Job loss rates, by gender Male Female 40% 27% 16% 14% 13% 12% 11% 10% 10% 10% 9% 8% 7% 7% 6% 6% AFG BGD BTN IND MDV NPL PAK LKA distribution of such attributes. Annex B, Table B2 estimates the difference in job loss rates by age cohort, gender, education, sector of employment, and occupation quality (in columns 1-5, respectively), while also controlling for the urban location dummy (in the pooled sample with all SAR countries), and the results are robust to controlling for the location of residence. 12 Note: The sample includes individuals who were employed in January 2020. The job loss category includes individuals who lost or changed the job, which they were employed in January 2020, between March 2020 and the time of the survey. In addition, female workers were also disproportionately affected compared to male workers, which is likely to worsen the existing gender inequalities in labor markets in the South Asia region. In 6 out of 8 SAR countries, job loss rates were higher among female workers than their male counterparts (Figure 5). This difference in job losses by gender was most stark in Nepal: 40 percent of female workers in Nepal lost their pre-COVID job, while only 27 percent of male workers lost the job that they had in January 2020. Similarly, in other countries like Afghanistan, Bangladesh, India, and the Maldives, the job loss rate was higher among female workers than men; however, this difference was relatively more muted (between 2 to 4 percentage points). The differential rate of job losses by gender cannot be explained by differences in the sector of employment and in the education levels between male and female workers. For instance, across all sectors in Nepal, we observe relatively higher job losses among female workers relative to their male counterparts employed within the same sector (Annex B, Figure B2, Panel A). We also observe a similar pattern of within-sector gender differences in job loss rates in countries like Afghanistan and the Maldives. Moreover, in 5 out of 6 countries where female workers were, on average, more likely to lose a job compared to men, we find a relatively higher job loss rate across both educated as well as the uneducated female workforce as compared to men with similar levels of education (Annex B, Figure B2, Panel B).11 Figure 6: Job loss rates, by age cohort 15-25 years 26-35 years 36-45 years 46 or older 43% 28% 27% 23% 22% 20% 17% 17% 15% 14% 12% 11% 11% 11% 10% 10% 9% 8% 8% 8% 8% 8% 8% 7% 7% 6% 6% 6% 6% 6% 5% 3% AFG BGD BTN IND MDV NPL PAK LKA 11 Annex B, Table B3 estimates the differential rate of job losses by gender using a regression with gender dummy variable along with additional controls for demographic and employment characteristics such as age cohort, education, sector of employment, occupation quality, and location of residence. The coefficient on the gender dummy is negative for all 8 SAR countries (and the coefficient for 5 SAR countries is also statistically significant at the conventional level). In addition, the results in Annex B, Table B4 (using a regression with a gender dummy variable interacted with sector of employment and education variables) suggest that the gender gaps in job losses were particularly large in the service sector and among uneducated individuals. But given a small number of observations, the latter results on differential gender gaps by sector and education should be interpreted with extreme caution. Annex B, Tables B5 and B6 report similar gender analysis for the job recovery rates. 13 Note: The sample includes individuals who were employed in January 2020. The age is measured at the time of the survey. The job loss category includes individuals who lost or changed the job, which they were employed in January 2020, between March 2020 and the time of the survey. Moreover, a relatively higher rate of job losses is observed among the youngest age cohort in all SAR countries. In most SAR countries, individuals aged 15 to 25 years old, who belong to the youngest age cohort, experienced one of the highest job loss rates relative to all other age cohorts within the same country (Figure 6). Moreover, in every SAR country, the difference in job loss rate between this youngest and the oldest (46 years or older) age cohort is also very striking. For instance, in Sri Lanka, the job loss rates in the youngest and the oldest age cohorts are 20 and 9 percent, respectively. Similarly, in Nepal and Bangladesh, the job loss rate in the youngest age cohort is 8 and 20 percentage points higher, respectively, compared to their counterparts in the oldest age cohort. This differential job loss rate by age cohort is perhaps not surprising. A loss of older, more experienced workers typically leads to a relatively higher loss of employer-specific human capital, which can only be acquired from on-the-job learning. In addition, firms also face a lower sunk cost of laying off younger workers relative to their older counterparts due to previous investments that firms would have made on worker training, particularly during the early stages of their employment tenure. But a prolonged absence from the labor market, especially in the early stages of the labor life cycle, has shown in the past to have severe and long-term negative impact on later life income trajectories (Oreopoulos et al., 2012). Therefore, examining the recovery rate and the bottlenecks in the recovery process could be important to mitigate some of the long-term effects, especially among younger workers who were negatively affected. At the same time, high job losses among older cohorts in some SAR countries raise the risk of permanently losing human capital, crucial for the recovery process. In countries like Nepal, Sri Lanka, Bangladesh, the Maldives, and Pakistan, the job loss rate among the oldest age cohort 46 years or older is not trivial (23, 9, 6, 17, and 9 percent, respectively). Moreover, our estimates in Figure 6 also show very similar levels of job loss rates among individuals aged 26 to 35 years old, and those between 36 and 45 years old. In this context, the slow recovery process could lead to discouragements and early retirements among older workers, and this loss of human capital due to a decline in the labor supply could have long-term impacts on future economic growth prospects. Figure 7: Firm closure rates 31% 21% 13% 13% 7% 7% 5% 4% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes firms in January 2020 that shut down by the time of the survey defined based on individuals who worked as an own-account worker, and employed at least one additional worker in January 2020, but were unemployed at the time of the survey. 14 High closure rates among business enterprises and firms in most SAR countries that suggest a significant loss of entrepreneurial capital. This includes Nepal, Sri Lanka, Bhutan, the Maldives, Pakistan, and Bangladesh, with firm closure rates of 31, 21, 13, 12, 7, and 7 percent, respectively, after March 2020 (Figure 7).12 Such a high incidence of closures is particularly problematic if new firms face large entry costs due to regulatory barriers and limited access to credit. While the survey did not measure the capital stock of business enterprises owned by survey respondents, it collected information on firm size that is likely to be correlated with its capital stock. In 5 out of 8 SAR countries, most firms that shut down have more than one employee in addition to the owner (Figure 8).13 Twenty-two percent of firms that reported a shut down in India have five or more employees. Such relatively larger firms (with five or more employees) make up 43, 31, 23, and 12 percent of the firms that shut down in countries like the Maldives, Nepal, Bangladesh, and Pakistan, respectively. This indicates a potential loss of entrepreneurial capital that is significant, that could further hamper recovery. Figure 8: Size distribution of firms that closed 1 worker 2 worker 3 workers 4 wokrers more than 5 LKA 60% 18% 4%1% 17% PAK 67% 10% 7% 4% 12% NPL 36% 17% 9% 6% 31% MDV 30% 13% 13% 1% 43% IND 36% 29% 8% 5% 22% BTN 41% 33% 8% 7% 11% BGD 46% 22% 5% 4% 23% AFG 62% 15% 15% 7% 1% Note: The sample includes firms in January 2020 that shut down by the time of the survey defined based on individuals who worked as an own-account worker, and employed at least one additional worker in January 2020, but were unemployed at the time of the survey. Firm size is based on the total number of employees in January 2020. 12 Business enterprises and firms are defined based on the recorded work type of survey respondents: own-account workers who employed at least one additional worker in January 2020. A firm is considered to have closed if an own-account worker with at least one other employee in January 2020 reported being unemployed at the time of the survey. Due to data limitations, our definition of firm closure will exclude firms that closed but the owner was employed as a wage worker at the time of the survey. Any owner who closed business and was employed as wage worker at the time of the survey was not asked for details about the nature of his previous business enterprise, such as the sector or size of organization, which are at the core of discussion in this section. In this regard, the firm closure rate is an underestimate. 13 Firm size is based on the total number of employees in January 2020. The survey only asks how many employees work in the enterprise and does not distinguish between unpaid family workers or wage employees outside the household. 15 Differences in firm size between those that survived and those that closed suggest that small household-based firms and microenterprises were disproportionately affected relative to larger firms. Figure 9 depicts the difference in the size distributions of firms between those that remained operational till the survey date and those that reported shutting down. 14 Across almost all SAR countries, there is a considerably higher proportion of smaller firms (with only one employee) among those that shut down compared to firms that survived. Moreover, this difference in survival rates between small and large firms is likely to be even higher, given our data could underestimate the size of the operational firms if such firms had downsized between January 2020 and the time of the survey to cope with declining market demand.15 These results suggest that policies that could facilitate the entry of new firms and the expansion of existing firms through rehiring, following the likely rise in market demand in the future could also play an important role in recovery. Figure 9: Difference in size distributions of firms that survived and firms that closed 1 worker 2 worker 3 workers 4 wokrers more than 5 35% 13% 12% 10% 10% 9% 8% 8% 6% 6% 6% 5% 5% 4% 3% 3% 3% 1% 1% 1% 0% 0% 0% 0% -1% -3% -3% -3% -4% -4% -4% -6% -6% -7% -8% -13% -15% -22% -24% -29% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes: (i) firms in January 2020 that were operational at the time of the survey defined based on individuals who worked as an own-account worker with at least one additional worker and in the same business in January 2020 and at the time of the survey; and (ii) firms in January 2020 that shut down by the time of the survey defined based on individuals who worked as an own-account worker, and employed at least one additional worker in January 2020, but were unemployed at the time of the survey. 14 Annex B, Figure B3 plots the difference in the size distributions of firms between those that remained operational and those that reported shutting down in the pooled sample of all 8 SAR countries. 15 For firms that shut down, their size is measured based on the total number of employees in January 2020. 16 Figure 10: Difference in sectoral distributions of new firms and firms that closed 42% Agriculture Manufacturing Service 15% 13% 12% 12% 10% 9% 7% 5% 5% 1% 0% 0% -2% -3% -6% -7% -7% -11% -12% -16% -17% -18% -30% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes: (i) firms in January 2020 that shut down by the time of the survey defined based on individuals who worked as an own-account worker, and employed at least one additional worker in January 2020, but were unemployed at the time of the survey; and (ii) new firms that opened after January 2020 based on individuals who report working as an own-account worker with at least one additional employee at the time of the survey, but were either unemployed or employed in a different job as a wage worker (not as an own-account worker) or employed in a difference business (as an own-account worker) in January 2020. Changes to the sectoral composition of employment may pose a considerable risk to recovery, depending on whether negatively affected workers can easily access new employment opportunities in different sectors and industries. Among new firms16 that opened after the first lockdown, the agriculture sector accounted for 40 and 35 percent of all new firm openings in Nepal and Sri Lanka, respectively, whereas agricultural enterprises accounted for only 27 and 23 percent of old firm closures during the same time period in the two respective countries (Figure 10).17 These differences by sector indicate key differences in the sectoral skills of negatively affected workers relative to the new employment opportunities being generated via new firms. This could have important implications for whether workers who lost their jobs can be easily reabsorbed into the labor market by new firms. In this context, worker re-training programs can better facilitate such job transitions and ensure positive income trajectories during the recovery process. Moreover, new investments in relevant worker skills that reflect the new market demand will also be crucial to sustaining any 16 New firms are measured based on individuals who report working as an own-account worker with at least one additional employee at the time of the survey but were either unemployed or employed in a different job as a wage worker (not as an own-account worker) or employed in a different business (as an own-account worker) in January 2020. 17 Annex B, Figure B4 plots the difference in the sectoral distributions between the new firms and those that closed in the pooled sample of all 8 SAR countries. 17 future economic growth by ensuring that new firms are not constrained in rehiring due to the lack of appropriate worker skills. Figure 11: Difference in spatial distributions of new firms and firms that closed Low region Medium region High region 27% 26% 17% 11% 11% 11% 7% 4% 2% 1% 1% 0% -1% -3% -5% -7% -8% -8% -9% -11% -13% -16% -18% -19% AFG BGD BTN IND MDV* NPL PAK LKA Note: The Y-axis plots spatial differences in net firm openings, where new firms and firm closures are defined as in Figure 10. For consistency, administrative regions within each country are ranked (and then categorized into high, medium, and low, on the x- axis) based on the job loss rate (of workers employed in January 2020) in the region. A district constitutes the smallest administrative region in all SAR countries except the Maldives, where a province is the smallest administrative unit collected in the survey. At the same time, spatial differences in firm openings and closures within each country underscore the key role of domestic migration and mobility in the recovery process. In Figure 11, we divide local administrative regions within each country into three categories based on the rate of job loss in the region, with “high� indicating the most affected regions in a country and “low� signifying the least affected areas. Figure 11 depicts considerable spatial differences in the concentration of new firm openings relative to firm closures within each country.18 These differences underscore the potential geographic realignment of new employment opportunities within these countries, which could incentivize workers to move across provincial and district borders for better economic opportunities. Such internal movement of labor could, however, be constrained for at least some subgroups due to high financial as well as non-financial costs associated with relocation. In addition, new firms are likely to face difficulties in hiring individuals from farther away due to their inability to perfectly observe and reward worker skills. Both these factors may generate significant mismatches of workers across geographic regions as well as over skills that could seriously undermine labor income trajectories as well as the overall rate of recovery within each country. 18 Annex B, Figure B5 plots the difference in the spatial distributions between the new firms and those that closed in the pooled sample of all 8 SAR countries. 18 Given these findings on the immediate labor market impacts of COVID-19 induced restrictions, it will be critical from a long-term welfare perspective to be able to continue to monitor different stages of a potentially long recovery process, along with identifying, in a timely manner, possible bottlenecks that could undermine any part of this recovery. This, first and foremost, requires appropriate measures of labor market outcomes on the recovery of jobs, and measures that describe the nature of new employment opportunities and benefits available to workers, including changes in labor earnings trajectories. The current round of data used in this paper also provides some key indicators related to the early recovery seen in SAR countries following the first lockdowns. Furthermore, future rounds of the survey can complement these results on the early stages of the recovery process to provide a more in-depth examination of factors that could pose serious risks to labor welfare in the long-term. Such longitudinal surveys, which can track the same individuals over time and across their multiple employment stints, provide a more accurate and detailed description of the recovery process, and help better analyze the roles of external factors and policies like gender mobility and safety net programs in influencing workers’ as well as firms’ inter-temporal decisions in the labor market during different stages of the recovery process. This will be crucial to formulating appropriate policies that ensure that the recovery of jobs and labor earnings is robust, and more importantly, that it is inclusive, such that historically disadvantaged subgroups, including women and low-skilled workers in SAR countries, are not disproportionately left out of the recovery process. 19 Section 2: The early stage of the recovery process About six months after the first set of COVID-19 induced lockdowns, the labor market recovery from first wave had already begun, but the rate of recovery was slow, and it varied within SAR. In Nepal, which experienced the most severe job losses in the region, the job recovery rate is only five percent, even six months after the first lockdown was introduced (Figure 12).19 In comparison, Bangladesh and Sri Lanka have experienced a slightly more robust, albeit still weak, recovery of 14 and 11 percent, respectively, during the same period. Figure 12: Job recovery rates 28% 25% 23% 18% 14% 11% 11% 5% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who lost or changed the job that they were employed in January 2020. Job recovery rate is defined based on individuals who were employed in a different job either in March 2020 or at the time of the survey. The recovery rate is positively correlated with the duration of time that has elapsed since the first lockdown and the revival of physical and economic activities. In Pakistan and India—where the survey was conducted about 11 months after the first lockdown—the job recovery rates are the highest in the region (25 and 28 percent, respectively).20 It is , 19 An individual who lost or changed the job that they were employed in January 2020, and was employed in a different job after March 2020 or at the time of the survey is defined to have recovered a job. 20 To estimate country-specific versus month-specific differences in the job recovery rates, we run a pooled regression with country and month of interview dummy variables, and restrict the sample to October-December 2020 (the time period that overlapped with surveys in all countries). The results in Annex B, Table B7 show that the job recovery rate in Nepal was significantly lower relative to other SAR countries even after controlling for the month of the interview (in line with the results in Figure 12). In addition, in Annex B, Table B8, we estimate within-country job recovery rates by the number of months since the introduction of the lockdown. These within-country time trends in job recovery do not provide evidence of a strong positive correlation between the recovery rate and the duration of time elapsed since the first lockdown within each country. These results however should be interpreted with caution due to small sample size and potential sample selection into survey months that could be correlated with job recovery outcomes. 20 therefore, likely that a modest labor market recovery also accompanies the resumption of physical and economic activities. Figure 13 shows the cross-country comparison between the recovery rate and the resumption of economic activities and physical mobility based on Google mobility data for six (out of eight) SAR countries for which the latter data was available.21 We find a positive correlation between job recovery rate and the revival of economic activities in a country (correlation coefficient = 0.123). For instance, Nepal, with its Google Mobility index at the time of the survey still almost 40 percent below that before the introduction of a lockdown (worst in the region), also has the lowest recovery rate out of all SAR countries. One outlier to this trend is India: it has one of the most robust job recovery rates in SAR, despite its Google Mobility index continuing to lag significantly below the pre-COVID level (second worst in the region). In countries like Sri Lanka, Afghanistan, and Bangladesh, mobility bounced back to, and in some cases even exceeded, pre- COVID levels by the time of the survey. However, in these countries, only 11 to 18 percent of the lost jobs had been recovered during the same period, suggesting that other factors apart from the resumption of physical mobility could potentially influence the rate of recovery. Figure 13: Correlation between job recovery rates and resumption of activities 20 AVG. % CHANGE IN GOOGLE 10 LKA AFG MOBILITY INDEX 0 BGD -10 PAK -20 -30 IND -40 NPL -50 0% 5% 10% 15% 20% 25% 30% RECOVERY RATE Note: Average % change in Google mobility index is calculated by taking an equally weighted mean over the average % changes in each of the four Google mobility outcomes: (i) retail and recreation, (ii) grocery and pharmacy, (iii) transit stations, and (iv) workplaces. For these four measures, we calculate % change (i) during the survey period (weighted average of all survey days) with respect to (ii) the last day prior to the introduction of the first lockdown. Job recovery rate is defined based on individuals who were employed in January 2020, but they were employed in a different job either in March 2020 or at the time of the survey. Source: Google COVID-19 Community Mobility Report (https://www.google.com/covid19/ mobility/), last updated July, 2021. 21 We calculate the average percentage change in Google mobility index by taking an equally weighted mean over the average percentage changes in each of the four Google mobility outcomes: (i) retail and recreation, (ii) grocery and pharmacy, (iii) transit stations, and (iv) workplaces. For these four measures, we calculate percentage change (i) during the survey period (weighted average of all survey days) with respect to (ii) the last day prior to the introduction of the first lockdown. 21 Observed within-country differences in the recovery rate by key demographic characteristics could also worsen pre- existing inequalities in labor market outcomes. Across all SAR countries, we find a relatively higher job recovery rate among male compared to female workers who lost their pre-COVID job (Figure 14). This result is particularly worrisome since job loss rates were also relatively higher among female than male in most SAR countries (Figure 5). Hence, these results indicate that female workers in SAR countries were double-cursed: they are more likely to have lost their employment immediately following the first lockdown, and they face greater difficulty in recovering a job in the early stages of the recovery process relative to their male counterparts. More importantly, reversing this trend in recovery is likely to play a significant role in ensuring that the long-term effects from the COVID-19 induced restrictions on gender inequalities in the labor market are limited. Figure 14: Job recovery rates, by gender Male Female 32% 28% 26% 23% 20% 18% 15% 14% 12% 12% 9% 7% 7% 5% 3% 1% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who lost or changed the job that they were employed in January 2020. Job recovery rate is defined based on individuals who were employed in a different job either in March 2020 or at the time of the survey. Furthermore, within-sector and within-education level analyses suggest that the gender bias in the recovery rate may not be fully explained by the difference in the educational attainment and the sector-specific comparative advantage between male and female workers. In all SAR countries, the female recovery rate is strictly less than the male recovery rate for both types of workers—those with and without education (Annex B, Figure B6, Panel A). Moreover, across the three sectors in each country, female workers who lost their pre-COVID job in a particular sector are less likely to recover a job (in any sector) compared to their male counterparts who also lost their job in the same sector (Annex B, Figure B6, Panel B).22 Uneducated and low-skilled workers were disproportionately left out of the recovery process in SAR countries. We measure worker productivity using two alternate measures: (i) formal educational attainment and (ii) baseline 22 The only exception is the manufacturing sector in Nepal and Maldives, where the job recovery rates are higher among female workers than male workers. 22 occupation characteristics based on the ISCO code. Figure 15, Panel A shows that the recovery rate is considerably lower among workers without any formal education than those with some education (in 7 out of 8 SAR countries). Likewise, the recovery rate is also lower among workers who were previously employed in an elementary occupation (based on the ISCO code)23 than those employed in a relatively higher skilled occupation (Figure 15, Panel B). Taken together, these results provide strong evidence that workers with the least education and skills, who are often among the most disadvantaged and vulnerable in the population, may have faced difficulty in finding new employment opportunities during this early period of recovery. Figure 15: Job recovery rates, by education and by occupation quality Panel A (education) Panel B (occupation quality) No education Some education Elemetary occupation Others 32% 31% 29% 28% 27% 25% 23% 23% 22% 22% 21% 17% 16% 16% 15% 14% 12% 12% 11% 11% 10% 10% 11% 9% 9% 9% 8% 5% 5% 4% 3% 0% AFG BGD BTN IND MDV NPL PAK LKA AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who lost or changed the job that they were employed in January 2020. Job recovery rate is defined based on individuals who were employed in a different job either in March 2020 or at the time of the survey. 23 The ISCO classification states “Elementary occupations consist of simple and routine tasks which mainly require the use of hand-held tools and often some physical effort.� These include selling goods in streets and public places, or from door to door; providing various street services; cleaning, washing, pressing; taking care of apartment houses, hotels, offices and other buildings; washing windows and other glass surfaces of buildings; delivering messages or goods; carrying luggage; doorkeeping and property watching; stocking vending machines or reading and emptying meters; collecting garbage; sweeping streets and similar places; performing various simple farming, fishing, hunting or trapping tasks performing simple tasks connected with mining, construction and manufacturing including product-sorting and simple hand-assembling of components; packing by hand; freight handling; pedalling or hand-guiding vehicles to transport passengers and goods; driving animal-drawn vehicles or machinery. 23 Figure 16: Job recovery rates, by age cohort 15-25 years 26-35 years 36-45 years 46 or older 30% 30% 30% 30% 28% 28% 26% 25% 24% 23% 21% 19% 17% 16% 17% 16% 15% 15% 15% 14% 13% 11% 11% 11% 9% 8% 6% 5% 4% 2% 1% AFG BGD BTN 1% IND MDV NPL PAK LKA Note: The sample includes individuals who lost or changed the job that they were employed in January 2020. Job recovery rate is defined based on individuals who were employed in a different job either in March 2020 or at the time of the survey. The recovery rate is also the highest among the youngest age cohort, and it decreases with age raising the risks of early retirements due to discouragement among older experienced workers. In Nepal, the job recovery rate among 15-25 years old is six percent compared to a two percent recovery rate among the oldest age cohort (46 years and older). We also find similar patterns of differential recovery rates that are correlated with age in all SAR countries (Figure 16). In so far as older cohorts have a relatively higher reservation wage due to their likely higher accumulated savings and a stronger preference for leisure, the above results suggest that possibly low market wages in this early stage of the recovery process may have discouraged older workers from re-entering the labor market. At the same time, if market wages remain low, a prolonged absence from the labor market could further hurt their labor market outcomes and raise the risk of early retirements. The recovery of jobs, however, has not been accompanied by the recovery in labor earnings suggesting that the current recovery process could, in large part, be driven by a decline in market wages while market demand continues to languish. Among individuals who recovered a job after losing their pre-COVID job, a significant proportion of them still earn less than their actual earnings before March 2020. This figure ranges from 18 to 69 percent across different SAR countries, suggesting that job recovery has not led to a recovery in earnings for a large share of affected workers (Table 2, Column 1). This could be due to various reasons including a potential skill mismatch during rehiring, which can lower worker productivity and earnings. On the other hand, the results could also underlie a continued low market demand for labor, with declining or stagnant market wages. We also find a similar rate of earning losses even among workers who kept the same employment since January 2020.24 For instance, in Nepal, 51 percent of workers who never lost their pre-COVID job earned less at the time of the survey 24 The SAR COVID-19 survey did not ask the respondents if they experienced reduced hours at work. However, findings from other regions (for instance, LAC) suggest shorter hours at work and job rotations accompanied reduced labor earnings (World Bank, 2021c) 24 relative to their earnings before March 2020 (Table 2, Column 2). In comparison, 56 percent of workers who lost and recovered their job earn less than their pre-COVID jobs. This suggests that high unemployment rate during the early recovery period could continue to suppress market wages, including for workers who are currently employed, and the recovery in market wages and worker earnings would crucially depend on a significant rebound in market demand for labor in the near future. Table 2: Share of recovered and unaffected workers with lower earnings Those who lost and Those who never recovered job lost job (1) (2) Afghanistan 42% 38% Bangladesh 69% 53% Bhutan 18% 26% Maldives 40% 42% Nepal 56% 51% Pakistan 45% 37% Sri Lanka 55% 48% India 46% 40% Note: The earnings loss category includes individuals who were employed in both January 2020 and at the time of the survey (or in March 2020), and they reported earning less than that prior to March 2020. The sample in Column (1) includes individuals who lost or changed the job that they were employed in January 2020 and were employed in a different job either in March 2020 or at the time of the survey; and in Column (2) includes individuals who never lost or changed the job that they were employed in January 2020, between March 2020 and the time of the survey. The slow pace of recovery of both jobs as well as market wages raises the risks of older cohorts permanently dropping out of the labor force, thereby taking with them valuable human capital that could be important for recovery. Among those who lost their pre-COVID job and were unemployed at the time of the survey, 78, 70, and 79 percent report not looking for a job in Nepal, Bangladesh, and Sri Lanka, respectively (Table 3, Column 1). We find similar rates of discouragement among unemployed workers in other SAR countries. Crucially, the discouragement rates are also usually higher among the older age cohorts than their younger counterparts. In Nepal, for example, we find a discouragement rate of 88 percent among unemployed workers 46 years or older, compared to only 71 percent among workers 15 to 25 years old (Table 3, Columns 2 and 5). Table 3: Discouragement, by age cohort Age cohorts All 15-25 years 26-35 years 36-45 years 46 or older (1) (2) (3) (4) (5) Afghanistan 45% 34% 47% 42% 60% Bangladesh 70% 64% 71% 81% 83% Bhutan 80% 75% 68% 85% 91% 25 India 79% 66% 70% 76% 83% Maldives 75% 77% 66% 86% 92% Nepal 81% 71% 77% 84% 88% Pakistan 78% 67% 54% 55% 58% Sri Lanka 57% 78% 83% 75% 79% Note: The sample includes individuals who were employed in January 2020 but were unemployed at the time of the survey. Discouragement is defined based on whether the individual reports actively looking for a job during the week of the survey. The ongoing recovery process is also likely to be non-linear, characterized by employment volatility and multiple job transitions. Figure 17 illustrates three different types of employment trajectories experienced by workers following a loss of their pre-COVID job. The first category comprises workers who never recovered a job, all of whom, at the time of the survey, have experienced a single but a possibly long unemployment spell. The low recovery rates in SAR countries explain the large share of workers who fall under this category of affected workers, which ranges from 38 percent in Bangladesh to 85 percent in Nepal, thereby raising the risk of discouragements among a large proportion of their workforce. The remaining two categories include workers who found a new job after losing their pre-COVID job, including those who may have subsequently lost their post-COVID job. The difference between the two recovered groups underscores the risk and volatility associated with the early stages of recovery that will continue to characterize the recovery process in the future. For a large share of workers, their early recovery path already includes multiple employment stints that are temporary and punctuated by short unemployment spells. In Bangladesh, for instance, 71 percent of those who had recovered a job (row 1 + row 2), had already lost it again by the time of the survey (Figure 17). While our data does not include the total number of different employment stints, the results show that this group had already experienced at least two separate unemployment spells between January 2020 and the time of the survey (an average period of 10 months). This pattern of recovery is prevalent in all SAR countries. Figure 17: Job transitions during recovery Spells AFG BGD BTN MDV IND NPL PAK LKA Pre- 1st COVID unemp. Post-COVID job 18% 14% 11% 23%% 25% 5% 28% 11% job spell Pre- 1st Post- 2nd COVID unemp. COVID unemp. 6% 46% 8% 17% 25% 8% 5% 8% job spell job spell Pre- COVID 1st unemployment spell 76% 40% 81% 60% 50% 87% 67% 81% job 100% 100% 100% 100% 100% 100% 100% 100% Note: The sample includes individuals who lost or changed the job that they were employed in January 2020. 26 Such labor market trajectories observed during this early stage of the recovery process could foreshadow the nature of the recovery process going forward, which would involve multiple employment stints punctuated with short, independent unemployment spells. While the restrictions on physical and economic mobility are less stringent and vary substantially across the SAR, each subsequent wave of COVID-1925 has defined more localized lockdowns. High- frequency data on labor market outcomes will therefore be required to monitor individuals’ recovery paths over time, including their income progression and transfer of skills across their different employment spells. A repeated panel will play an important role in better understanding the evolving labor market conditions as well as informing policies to facilitate the recovery process. In addition, any such future data also needs to capture workers’ key labor market decisions, such as their job search strategies and labor market choices during both their employment and unemployment spells, which can directly affect their long-term income trajectories and other benefits in the labor market. In addition, these economic impacts associated with the COVID-19 crisis could result in worsening inequality, by inducing reliance on risky coping strategies and posing risks to human development outcomes particularly for poor and vulnerable households (World Bank, forthcoming, World Bank 2021a). Accordingly, the second Round of the SAR CPMS is designed to track labor market trajectories during the recovery process, as well as coping strategies and human capital investments. 25 As per the Oxford COVID-19 Government Response Tracker, the stringency index during the first set of lockdowns was above 78 for the seven SAR countries (Figure B7). The adopted lockdown measures have been less stringent with each subsequent wave and vary substantially between countries. Apart from Afghanistan where the closures due to COVID-19 have been minimal, the stringency index ranged between 94 (Nepal) and 70 (Bhutan), and 72 (India) and 24.5 (Pakistan) during the second and the third wave of infections. 27 Section 3: Worker skills and the risks to recovery Employment transitions among workers who recovered a job are characterized by movements within and across different sectors. In all three sectors, we observe both inflows and outflows of workers (Figure 18). Across countries, 15 to 40 percent of workers recovered jobs in a different sector than the sector that they were originally employed in in January 2020 (Table 4, Column 1). In Sri Lanka, Nepal, Afghanistan, Pakistan, and India, more than one-third of individuals who recovered a job after losing their pre-COVID job switched their sector of employment. Figure 18: Sectoral transitions during recovery Ag-Ag Ag-Man Ag-Ser Man-Man Man-Ag Man-Ser Ser-Ser Ser-Man Ser-Ag LKA 11% 2% 8% 8% 2% 18% 34% 13% 4% P A K 2% 4% 4% 11% 4% 9% 49% 16% 3% NPL 3% 5% 8% 15% 3% 15% 37% 9% 6% M D V 2%3% 7% 1% 5% 0% 76% 2%3% IND 13% 3% 7% 10% 3% 10% 43% 8% 3% B T N 0% 7% 12% 3% 10% 63% 2%3% BGD 18% 5% 3% 14% 0% 8% 44% 6% 3% A F G 2% 8% 4% 7% 5% 12% 47% 6% 9% Note: The sample includes individuals who changed the job that they were employed in January 2020. Each transition category describes a unique sectoral transition (or the absence of a sectoral transition), where the first half of the legend name refers to the pre-COVID sector of employment in January 2020, and the second part refers to the new sector of employment in a different job in March 2020 or at the time of the survey. 28 Table 4: Change in sector, industry, and job type during recovery Moved sector Moved industry Moved to a lower (1) (2) skilled job (3) Afghanistan 43% 80% 56% Bangladesh 24% 42% 22% Bhutan 25% 65% 47% India 34% 59% 33% Maldives 15% 60% 27% Nepal 45% 69% 43% Pakistan 39% 79% 40% Sri Lanka 46% 69% 44% Note: The sample includes individuals who changed the job that they were employed in January 2020. The transition to a lower skilled job is defined based on the 1-digit ISCO code (ISCO-08 categories), where a higher value of the ISCO code signifies lower skilled job, and the change in industry of employment is defined based on the 1-digit ISIC code. These sectoral transitions may represent a potential misallocation of worker skills during the recovery process. A more detailed measurement of industry of employment based on the 1-digit ISIC code reveals the risk of misallocation of worker skills, even within sectors. In Sri Lanka, 69 percent of workers who recovered a job were employed in a different industry compared to that in January 2020 (Table 4, Column 2). We also find high rates of movement across industries during recovery in countries like Afghanistan, Pakistan, Nepal, and India (80, 79, 69, and 59 percent, respectively). Such high rates of industry transition during rehiring is likely to lead to losses in industry-specific human capital stock that could hamper economic recovery, along with a potential decline in individual earnings even among those who recovered a job if such transitions are associated with the misallocation of worker skills. In Column 3 of Table 4, we find that the high rate of industry transitions observed in SAR countries is also accompanied by a decline in the quality of jobs (measured based on the 1-digit ISCO code). Among workers who recovered a job in Afghanistan and Bhutan, 56 and 47 percent of them, respectively, were employed in a new occupation with a higher ISCO code compared to that of their pre-COVID job (signifying a decline in occupation quality). This share is also high in Nepal (43 percent), Sri Lanka (44 percent), and Pakistan (40 percent). Moreover, a change in the industry of employment is positively correlated with a decline in the worker’s job quality and earnings. For instance, in Bangladesh, among those who recovered a job within the same industry, 89 percent of them recovered the job with the same occupation quality or better. In contrast, among workers who changed industries during recovery, more than 64 percent of them were employed in a lower quality job compared to the occupation characteristics of their original job in January 2020. This trend is consistent across all countries in SAR, as seen in Table 5, Columns 1 and 2. 29 Table 5: Correlation between industry transition and job quality Lower skilled jobs Lower earnings Did not change Changed Did not change Changed industry industry industry industry (1) (2) (3) (4) Afghanistan 12% 64% 53% 38% Bangladesh 5% 36% 69% 66% Bhutan 4% 69% 31% 11% India 8% 51% 44% 47% Maldives 17% 41% 54% 31% Nepal 11% 60% 65% 51% Pakistan 21% 45% 46% 45% Sri Lanka 13% 60% 50% 58% Note: The samples in columns 1 and 3 include individuals who changed the job that they were employed in January 2020, and changed their industry of employment (defined based on the 1-digit ISIC code). The samples in columns 2 and 4 include individuals who changed the job that they were employed in January 2020, and did not change their industry of employment (defined based on the 1-digit ISIC code). The transition to a lower skilled job is defined based on the 1-digit ISCO code (ISCO-08 categories), where a higher value of the ISCO code signifies lower skilled job. Lower earnings category is defined based on whether the individual’s earnings in the month (or week) preceding the survey date is lower than the earnings in a typical month before March 2020. While the above pattern of results is consistent across all SAR countries, a similar analysis on the correlation between industry transition and a change in worker salary shows cross-country differences within the region (Table 5, Columns 3 and 4). In countries like India and Sri Lanka, we find that a change in the industry of employment increased the likelihood of a decline in earnings (relative to those that did not change industry during their employment transition). In contrast, workers in Nepal and Bhutan, who changed industry while recovering their job, were relatively less likely to have also experienced a decline in their earnings compared to their counterparts who recovered a job within the same industry. Nevertheless, industry transitions, including in the latter set of countries, could still negatively affect future economic growth prospects and undermine the recovery process in the long term. Overall, we observe high rates of sectoral and industry transitions that have characterized the early recovery process in all SAR countries, which could have both immediate and long-term implications for worker earnings as well as the country’s economic growth prospects. It is important to examine various factors and bottlenecks that underlie these increased risk of skill misallocations, which are typically associated with inefficiencies related to firm hiring and worker job search processes to ensure that such job transitions do not slowdown the pace of recovery in the future. In this context, we use information from the current data to examine, to the extent possible, some of these factors—in particular, labor mobility and access to social safety nets—that can directly impact individual’s employment opportunities as well as their decisions related to labor market participation and job search. Future rounds of labor survey data from the SAR CPMS or comparable country level panel surveys that capture more detailed outcomes related to job search strategies employed by workers and rehiring decisions made by firms are likely to play an important role in informing effective policies to address these market inefficiencies and facilitate recovery. 30 31 Section 4: The role of migration in the recovery process Labor mobility has played an important role in the recovery process. We observe a high rate of geographic mobility among individuals who lost their pre-COVID job. Across all SAR countries, the likelihood of moving to a new location of residence between March 2020 and the time of the survey is 3 to 13 percentage points higher among workers who lost their pre-COVID job compared to those who did not lose their pre-COVID job (Figure 19).26 More importantly, the recovery rate is also higher among those who moved (relative to those that did not), implying that spatial mobility of workers could have played a key role in the recovery process (Figure 20). Figure 19: Physical mobility of labor Did not lose a job Lost a job 18% 15% 14% 10% 10% 10% 9% 9% 5% 5% 5% 5% 3% 2% 2% 2% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020. The job loss category includes individuals who lost or changed the job, which they were employed in January 2020, between March 2020 and the time of the survey. Labor mobility is defined based on whether an individual resides in a new location (within the same country) at the time of the survey compared to the location of residence prior to March 2020. 26 We also find that a rural destination accounts for 43% of all migration in SAR ranging from 11% in Afghanistan to over 50% in Sri Lanka and Maldives (Annex B, Table B9). The migration to rural areas is also about 10 percentage points higher among the migrant sample who lost their pre-COVID jobs (relative to those migrants who did not lose a job). Since the survey did not ask the type of residence before the individual migrated, we are however unable to identify urban-rural migration patterns in the data. 32 Figure 20: Job recovery rates, by physical mobility Those who moved Those who did not move 36% 30% 28% 28% 24% 24% 23% 22% 18% 16% 15% 13% 10% 10% 7% 5% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who lost or changed the job that they were employed in January 2020. Job recovery rate is defined based on individuals who lost a job after January 2020, and reported working in a different job in March 2020 or the week of the survey. Labor mobility is defined based on whether an individual resides in a new location (within the same country) at the time of the survey compared to the location of residence prior to March 2020. In all SAR countries (except the Maldives), internal migration significantly improves the likelihood of finding a new job among those who lost their pre-COVID job. In Bangladesh, the recovery rate among those that moved since the lockdown is 28 percent compared to only 13 percent among those that did not move (Figure 19). Likewise, in Nepal, India, and Sri Lanka, the rate of recovery is 5 to 8 percentage points higher among the migrant group (relative to non- migrants). However, the positive correlation between mobility and recovery is likely to be confounded by various worker characteristics, including socio-economic and demographic characteristics, that directly influence both outcomes. Importantly, if labor mobility is, in part, correlated with worker characteristics like gender, which can negatively affect mobility, this could also lead to an unequal recovery, with specific socio-economic and demographic subgroups being left out of the recovery process. Examining some of the key characteristics related to internal migration highlights the potential bottlenecks in labor mobility that could hamper recovery and underscores the risks of differing recovery rates across different demographic groups. Across all SAR countries, the age gradient in migration patterns is evident, where younger individuals were more likely to move geographies following a loss of employment relative to the older cohorts. For instance, 24, 19, and 14 percent of individuals in the youngest cohort aged 25 years or younger moved to a different location within their country following a job loss in Nepal, India, and Sri Lanka, respectively (Table 6). In contrast, the rate of internal migration among their counterparts in the oldest cohort aged 46 or older was significantly lower in all three countries (14, 5, and 6 percent, respectively). 33 Table 6: Physical mobility following a job loss, by age cohort and gender Gender Age cohort Male Female 15-25 years 26-35 years 36-45 years 46 or older (1) (2) (3) (4) (5) (6) Afghanistan 14% 0% 10% 9% 4% 24% Bangladesh 12% 5% 11% 10% 10% 8% Bhutan 13% 19% 26% 16% 0% 17% Maldives 11% 8% 14% 5% 16% 4% Nepal 19% 17% 24% 18% 9% 14% Pakistan 6% 3% 7% 6% 2% 3% Sri Lanka 11% 5% 14% 10% 9% 6% India 17% 9% 19% 15% 16% 5% Note: The sample includes individuals who lost or changed the job that they were employed in January 2020. Labor mobility is defined based on whether an individual resides in a new location (within the same country) at the time of the survey compared to the location of residence prior to March 2020. In addition, we find differential migration rates by gender, which indicates potential constraints to mobility faced by women in SAR countries that could hamper their recovery. In 8 out of 9 SAR countries, the internal migration rate following a job loss for men was higher than for women who lost employment. This gender gap in labor mobility was particularly high in countries like Afghanistan, Bangladesh, Sri Lanka, and India compared to the Maldives and Nepal, where the internal migration rate was still higher among men, but the gender difference was relatively muted (Table 5, Columns 1 and 2). More importantly, these differences in migration rates by key demographic characteristics like gender and age, align with the differential rates of job recovery observed earlier in Figures 13 and 15. Addressing such bottlenecks in physical mobility will, therefore, help ensure that certain subgroups are not disproportionately left out of the recovery process that can further worsen pre-existing inequalities in labor market outcomes in the long-term. Workers are also likely to face monetary and non-monetary constraints to labor mobility that increase with distance. In each of the eight SAR countries, most of the observed migration is limited to within the same district or province (Figure 21). In Nepal, 59 percent of migrants moved to a new location within the same district, while 25 percent moved to a different district within the same province, and only 16 percent moved to a different province. Similarly, in India, this distribution of migration within the same district, across different districts within the same state or province, and to a different province is 66, 15, and 19 percent, respectively. These domestic migration patterns in SAR countries suggest that labor mobility could be costly, especially across administrative boundaries and to regions that are farther away, which restricts workers from seeking better employment opportunities even within their own country. These costs could include both the monetary (such as transportation costs, loss of location-tied social assistance and other services, etc.) and non-monetary costs (such as psychological costs, loss of social network, etc.) of moving. In addition, domestic migrants could also face disadvantages in the labor market due to the non-transferability (due to differences in language, business environment, customer preferences, licensing, etc.) and imperfect observability (due to lack of referrals and accreditation systems) of worker skills across regions. These bottlenecks to internal migration could seriously undermine recovery especially in light of the early evidence on the geographic realignment of new employment opportunities within each country as reported in Figure 11. 34 Figure 21: Types of physical mobility following a job loss Within same district Different district in the same province Different province 75% 66% 61% 60% 59% 55% 45% 45% 43% 42% 40% 39% 28% 27% 25% 25% 19% 16% 15% 15% 0% AFG BGD BTN IND MDV* NPL PAK* LKA* Note: The sample includes individuals who lost or changed the job that they were employed in January 2020 and were residing in a new location (within the same country) at the time of the survey compared to the location of residence prior to March 2020. *In the Maldives (MDV), Pakistan (PAK), and Sri Lanka (LKA), the data only allows us to measure migration (i) within and (ii) across provinces, and it does not allow us to measure migration within a same district. In many SAR countries, migration-induced recovery is correlated with a greater risk of misallocation of worker skills across sectors and industries, along with a decline in job quality and labor earnings. In Bangladesh, Bhutan, Nepal, and Sri Lanka, the likelihood of changing the sector of employment among workers who recovered a job is relatively higher among those who moved their location of residence (migrants) compared to those who did not move (non-migrants). For example, 75 percent of domestic migrants in Nepal who recovered a job changed their sector of employment, while only 41 percent of non-migrants in Nepal underwent a similar sectoral transition during their recovery (Figure 22, Panel A). In line with these results, we also find a strong correlation between labor mobility and the change in the industry of employment during recovery in Bangladesh, Bhutan, the Maldives, Nepal, and India (Figure 22, Panel B). 35 Figure 22: Sectoral and industry transitions in recovery, by physical mobility Panel A (sector) Panel B (industry) Migrants Non-migrants Migrants Non-migrants 75% 93% 88% 86% 85% 79% 75% 75% 70% 69% 54% 61% 58% 57% 44% 55% 54% 41% 41% 40% 39% 36% 35% 34% 34% 39% 29% 23% 22% 20% 15% 8% AFG BGD BTN IND MDV NPL PAK LKA AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who lost or changed the job that they were employed in January 2020 and were employed in a different job either in March 2020 or during the week of the survey. Labor mobility is defined based on whether an individual resides in a new location (within the same country) at the time of the survey compared to the location of residence before March 1st, 2020. The change in industry of employment is defined based on the 1-digit ISIC code. In 7 out of 8 SAR countries, job recovery led through domestic migration was accompanied by a decline in the occupation quality of new jobs. For instance, 86, 65, 52, and 50 percent of domestic migrants in Bhutan, the Maldives, Nepal, and Pakistan, respectively, recovered a job that was of lower quality (based on the 1-digit ISCO code) relative to their pre-COVID job. This share is only 60, 20, 50, and 40 percent, respectively, among those who did not migrate following a job loss (Figure 23, Panel A). Moreover, in Nepal, the Maldives, and Afghanistan, a large share of domestic migrants who recovered a job also experienced a decline in their earnings (75, 73, and 53, respectively) compared to the non-migrants (57, 42, and 41 percent, respectively). These results are in line with the evidence of high sectoral and industry transitions experienced by workers during recovery, with such mismatches in worker skills more concentrated among those who changed location during the recovery process, leading to a decline in the quality of their new jobs along with lowered labor earnings. 36 Figure 23: Change in job quality and earnings in recovery, by physical mobility Panel A (lower skilled jobs) Panel B (lower earnings) Migrants Non-migrants Migrants Non-migrants 75% 73% 71% 86% 57% 56% 53% 65% 64% 50% 60% 47% 46% 52% 42% 41% 41% 50% 50% 45% 43% 40% 40% 40% 30% 28% 32% 23% 27% 20% 18% 11% AFG BGD BTN IND MDV NPL PAK LKA AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who lost or changed the job that they were employed in January 2020 and were employed in a different job either in March 2020 or at the time of the survey. Labor mobility is defined based on whether an individual resides in a new location (within the same country) at the time of the survey compared to the location of residence in January 2020. The transition to a lower skilled job is defined based on the 1-digit ISCO code (ISCO-08 categories), where a higher value of the ISCO code signifies a lower skilled job. These results suggest that while migration facilitates recovery, workers in some SAR countries face difficulties in transferring skills when they relocate to a new location that reduces their economic returns to migration, and thereby also hampers the recovery process. Such transferability issues can arise from administrative as well as information barriers. These can be addressed however by harmonizing labor laws across different regions, and by investing in active labor market policies like skill accreditation programs, which help new employers better evaluate worker skills, and job fairs that reduce the search costs for both workers and employers (McKenzie and Yang, 2014). In addition, when information barriers make up a significant cost of finding a job, policies that enable workers to expand their job search both across regions and over time, can facilitate long-term recovery by allowing individuals to find skill appropriate jobs, albeit at the cost of potentially prolonging their unemployment spells in the short term. 37 Section 5: Unemployment spells, discouragement, and the role of social safety nets Across all SAR countries, most workers who lost the job they had before the lockdown continue to remain unemployed at the time of the survey, with unemployment spells that are long and still ongoing. The average duration of an unemployment spell among those unemployed at the time of the survey ranges between one to more than five months across different SAR countries. Moreover, in India and Bangladesh, 39 and 36 percent of unemployed workers, respectively, have unemployment spells that span more than five months, while 54 percent of unemployed workers in Nepal have been unemployed for more than five months (Figure 24). These results are also in line with the low recovery rates observed in SAR countries, as noted in Section 2. Figure 24: Distribution of the length of unemployment spells Less than 1 mth 1-2 mths 2-3 mths 3-4 mths 4-5 mths More than 5 mths LKA 17% 41% 17% 6% 2% 16% PAK 17% 39% 11% 12% 5% 17% NPL 8% 15% 11% 6% 5% 54% MDV 28% 16% 19% 5% 2% 30% IND 13% 22% 12% 11% 3% 39% BTN 13% 15% 21% 7% 15% 30% BGD 17% 22% 12% 9% 3% 36% AFG 17% 29% 18% 7% 6% 22% Note: The sample includes individuals who were employed in January 2020 but were unemployed at the time of the survey. The length of unemployment spells is measured in months (mths). Longer unemployment spells could raise the risk to worker discouragements leading them to permanently drop out of the labor force, thereby losing valuable human capital that could be important for recovery. Figure 25 shows discouragement rates among individuals who lost a pre-COVID job and were unemployed at the time of the survey by quintiles based on the length of their unemployment spell. In all SAR countries, the share of unemployed individuals who report not looking for a job is extremely high across all quintiles. For instance, in India, the share of discouraged workers in the bottom quintile (those with the shortest unemployment spell) is 78 percent, while this share is 75 percent in the top quintile (those with the longest unemployment spell). The high rate of discouragement, especially among those who were only recently unemployed, indicates to a severe lack of labor demand prevalent in SAR countries. This is also in line with the low recovery rates and the high unemployment rates observed previously in 38 Figures 11 and 17. More importantly, if recovery continues to languish, the long-term effects of a prolonged absence from the labor market due to a high rate of discouragement could be detrimental for the future economic growth prospects of countries in the South Asia region. Figure 25: Discouragement, by length of unemployment spell Bottom quintile Second quintile Third quintile Fourth quintile Top quintile 100% 100% 88% 87% 86% 84% 83% 83% 82% 82% 82% 80% 80% 78% 78% 76% 75% 74% 73% 73% 73% 73% 71% 70% 69% 67% 66% 65% 63% 63% 62% 59% 58% 58% 56% 56% 55% 53% 36% 20% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020 but were unemployed at the time of the survey. The length of unemployment spells is measured in months (mths). The bottom quintile includes individuals with the shortest unemployment spell, while the top quintile includes individuals with the longest unemployment spell at the time of the survey. Discouragement is defined based on whether the individual reports actively looking for a job during the week of the survey. At the same time, worker unemployment spells are also positively correlated with their access to social safety nets, which could have enabled them to prolong their job search and thereby improve their future labor market outcomes. In many SAR countries, a longer duration of unemployment is associated with a higher receipt of household assistance through formal programs (Figure 26). For example, in Bhutan, 70 percent of unemployed individuals in the top quintile (of unemployment spell) received some form of household assistance from the government, non-governmental organizations, or the like, whereas, only 13 percent in the bottom quintile received any such assistance. This trend is also prevalent in other SAR countries like India and the Maldives. However, such household assistance is low among all unemployed individuals (across all quintiles) in Afghanistan, Nepal, Pakistan, and Bangladesh, where access to formal assistance is typically low. In these latter countries, those unemployed, especially in the midst of a long spell, may also have had to rely on costly coping strategies such as a sale of productive assets or high-cost debt, which could significantly hamper their future labor market outcomes (as observed in Nepal, see World Bank 2021a). 39 Figure 26: Access to formal assistance, by length of unemployment spell Bottom quintile Second quintile Third quintile Fourth quintile Top quintile 73% 70% 69% 66% 58% 56% 51% 48% 48% 48% 43% 40% 35% 31% 27% 26% 22% 21% 20% 18% 16% 16% 16% 16% 15% 14% 14% 13% 12% 10% 10% 9% 8% 7% 6% 4% 1% 0% 0% 0% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020 but were unemployed at the time of the survey. The length of unemployment spells is measured in months (mths). The bottom quintile includes individuals with the shortest unemployment spell, while the top quintile includes individuals with the longest unemployment spell at the time of the survey. Access to formal assistance is defined based on whether the household of the individual received any assistance from the government or a non-governmental organization. In countries with high international migration, remittances from absent household members can also play an important role in protecting households from a negative labor market shock. In all SAR countries, we observe a high share of remittance recipients among unemployed individuals (between 38 and 78 percent). Specifically, 79 and 76 percent of unemployed individuals in the top quintile received remittances in Bangladesh and Bhutan, respectively. In comparison 62 and 13 percent of unemployed individuals in the bottom quintile received remittances from abroad in the two countries, respectively (Figure 27). More importantly, in 7 out of 8 SAR countries, the share of remittance recipients far exceeds the share of unemployed individuals with access to formal assistance, highlighting the importance of international migration and remittances during recovery in the entire region. 40 Figure 27: Remittances, by length of unemployment spell Bottom quintile Second quintile Third quintile Fourth quintile Top quintile 100% 100% 100% 100% 87% 87% 82% 80% 79% 78% 76% 75% 71% 70% 69% 69% 68% 62% 57% 56% 56% 55% 55% 54% 54% 53% 51% 48% 47% 44% 38% 37% 36% 29% 13% 11% 0% 0% 0% 0% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020 but were unemployed at the time of the survey. The length of unemployment spells is measured in months (mths). The bottom quintile includes individuals with the shortest unemployment spell, while the top quintile includes individuals with the longest unemployment spell at the time of the survey. Receipt of remittance is defined based on whether the household of the individual usually receives any remittance from a household member who is currently a (domestic or international) migrant worker. In addition, subsistence agriculture has provided a key safety net for many unemployed workers. Across all SAR countries, we find high rates of entry into subsistence agriculture following the lockdowns. For instance, in Sri Lanka, 48 percent of individuals who were not engaged in subsistence agriculture one year before the survey, were engaged in subsistence activity at the time of the survey (Figure 28). Moreover, in all SAR countries, this rate of entry into subsistence agriculture is considerably higher among those who were unemployed at the time of the survey compared to those employed (Figure 28). Specifically, across all SAR countries, we find approximately ten percentage points difference between employed and unemployed individuals in the entry and continued participation in subsistence agriculture at the time of the survey. Recent engagement in subsistence agriculture is also correlated with longer unemployment spells, thereby providing the most affected with a valuable source of non-monetary income, and at the same time, perhaps allowing individuals to prolong their unemployment spell for better future labor market outcomes. In 6 out of 8 SAR countries, the rate of entry into subsistence agriculture is the highest among unemployed individuals in the top quintile (those with the longest unemployment spell). In Afghanistan, Nepal, and Bangladesh, more than 50 percent of unemployed individuals in the top quintile started to engage in subsistence agriculture within the last one year (Figure 29). In contrast, this share is 20, 26, and 18 percent among individuals in the bottom quintile in these three respective countries. This result is particularly striking given we previously found fairly low access to formal assistance in each of these three countries, including among those with the longest unemployment spells. 41 Figure 28: Recent engagement in subsistence agriculture, by employment status All Employed Unemployed 53% 48% 46% 37% 35% 28% 28% 28% 28% 27% 26% 23% 23% 21% 21% 21% 21% 18% 18% 17% 17% 15% 14% 14% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020, while the two subsamples include those individuals who were employed at the time of the survey, and those who were unemployed at the time of the survey. Recent engagement in subsistence agriculture is defined based on individuals who were engaged in subsistence agriculture at the time of the survey, but were not engaged in subsistence activity one year prior to the survey. Figure 29: Recent engagement in subsistence agriculture, by length of unemployment spell Bottom quintile Second quintile Third quintile Fourth quintile Top quintile 76% 66% 59% 57% 53% 53% 52% 52% 50% 46% 46% 45% 44% 42% 40% 37% 36% 32% 30% 26% 25% 22% 22% 21% 21% 20% 20% 19% 18% 18% 18% 17% 17% 16% 15% 3% 0% 0% 0% 0% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020 but were unemployed at the time of the survey. The length of unemployment spells is measured in months (mths). The bottom quintile includes individuals with the shortest unemployment spell, while the top quintile includes individuals with the longest unemployment spell at the time of the survey. Recent engagement in subsistence agriculture is defined based on individuals who were engaged in subsistence agriculture at the time of the survey, but were not engaged in subsistence activity one year prior to the survey. These results highlight the importance of formal and informal social safety nets, including subsistence agriculture and remittances, in protecting worker welfare during unemployment spells. In a context with an overall low coverage of social assistance programs, as in the case of Afghanistan, Nepal, and Bangladesh, access to these informal mechanisms could be even more crucial in protecting the most severely affected in the region. Moreover, high instances of 42 international migration observed in countries like Bangladesh, Nepal, India, and Pakistan, imply that the role of remittance as a safety net mechanism should also not be underestimated. More importantly, in so far as these social safety nets allow workers to better invest in job search by temporarily prolonging their unemployment spell, an improved access to such safety nets can address some of the issues previously identified related to worker skill misallocations and internal migration, which have undermined key labor market outcomes in the medium-term as well as risked long-term recovery. 43 Conclusion The imposition of swift and stringent lockdowns aimed to curb the rising infection rates of COVID-19 in early 2020 led to a drastic, abrupt disruption of physical mobility and economic activities in all eight SAR countries. Labor market exposures to these COVID-induced lockdowns and restrictions have been widespread and deep, with significant loss of pre-COVID employment and earnings with direct implications for immediate worker welfare. Moreover, the structural and geographic realignments in the distribution of jobs across all countries in SAR seen in the first round of the SAR CPMS survey can have long-term impacts on the labor market and compound the recovery process. Given this, future rounds of data, which track the evolving labor market conditions, including wage adjustments or the lack thereof, will be crucial to addressing the critical challenges and risks to a robust and inclusive recovery in the region. In most SAR countries, the continuous promulgation of different forms of local and national restrictions have also meant that the recovery of jobs has been slow, and the early recovery exhibits worrying signs that historically poor and vulnerable groups, including women, low skilled, and older workers, could be disproportionally left out of the recovery process. Reversing this trend will require additional data that can provide key insights into various mechanisms that underlie the pace and the nature of the recovery process, along with identifying, in a timely manner, appropriate policy measures that can support inclusive recovery and ensure that COVID-related disruptions do not contribute to rising inequality over the longer term. Soon after the completion of the first round of data collection, the South Asia region also experienced the second wave of COVID-19 infections in May 2021, with considerably more severe health impacts on the population relative to the first wave in early March 2020. This is likely to have disrupted the nascent recovery process documented in this paper, and further slow and complicate the long-term recovery. Interestingly, in most SAR countries, lockdown measures during this second wave were not as strictly imposed as in early 2020, and they were intermittent and short-lived. Figure B7 highlights this fact: all SAR countries relaxed many of the restrictions throughout 2021, including during the period encompassing the second wave. The ease of restrictions and less stringent impositions were primarily driven by concerns about economic instability. In this context, new rounds of data will help monitor the state of recovery during this critical period of a severe health crisis and examine the impacts of such subsequent waves of COVID-19 infections and the repeated public health crisis on the long-term recovery. Follow-up rounds of the World Bank’s COVID-19 monitoring surveys have been implemented in 6 of the 8 SAR countries27. In Afghanistan, where multiple rounds have been collected, the second round in May-June 2021 covered the second wave during 2021. In Bangladesh, round 2 was implemented between July and October 2021, coinciding with the rollout of follow-up survey in 4 of the 8 SAR countries (Bhutan, Nepal, Pakistan, and Sri Lanka), which is almost a year after the first round. Data collection began in September 2021 and concluded in January 2022. The second round tracks the labor market outcomes for the same respondent at the time of the survey (last quarter of 2021), job transitions, layoffs, and searches since the previous round, and the 27 The Maldives Bureau of Statistics has implemented a sub-sample phone follow up to the 2019 Household Income and Expenditure survey which covers the 2021 wave. 44 reasons behind these transitions. In addition, the survey is collecting information on the health impacts of COVID-19 faced by these households as well as the availability and access to COVID vaccines. 45 References Balcázar, Carlos Felipe, Sonal Desai, Rinku Murgai and Ambar Narayan. 2016. Why did Poverty Decline in India? A Non- Parametric Decomposition Exercise. World Bank Working Paper 7602. World Bank: Washington, DC. McKenzie, David and Dean Yang. 2014. Evidence on Policies to Increase the Development Impacts of International Migration. Policy Research Working Paper 7057. World Bank, Washington, DC. Oreopoulos, Philip, Till Von Wachter, and Andrew Heisz. 2012. The Short- and Long-Term Career Effects of Graduating in a Recession. American Economic Journal: Applied Economics 4(1): 1-29. World Bank. 2002. Poverty Assessment: Poverty in Pakistan - Vulnerabilities, Social Caps, and Rural Dynamics. Washington, DC. © World Bank. World Bank. 2016a. Afghanistan: Trends in Poverty and Inequality 2007-17. © World Bank: Washington, DC. World Bank. 2016b. Moving up the Ladder: Poverty Reduction and Social Mobility in Nepal. World Bank Group: Nepal. World Bank. 2017. Maldives: Poverty and Inequality in Maldives. World Bank: Washington, DC. World Bank. 2019a. Bangladesh Poverty Assessment: Facing Old and New Frontiers in Poverty Reduction. Washington DC. © World Bank. World Bank. 2019b. Poverty, Vulnerability and Welfare in Bhutan: Progress and Challenges. World Bank: Washington, DC. World Bank. 2019c. Poverty and Equity Brief: Pakistan. © World Bank. World Bank. 2021a. Risks to Poverty, Vulnerability, and Inequality from COVID-19: Nepal Light Poverty Assessment. World Bank, Washington, DC. World Bank. 2021b. Agricultural Productivity, Diversification and Gender: Background Report to Sri Lanka Poverty Assessment. World Bank: Washington, DC. World Bank. 2021c. Taking Stock of COVID-19 Labor Policy Responses in Developing Countries. World Bank Group: Washington DC. World Bank. forthcoming. Cox’s Bazar Panel Survey: Rapid Follow-Up Round 3. Coping with the Health and Economic Impacts of COVID-19 in Cox’s Bazar. World Bank: Washington, DC. 46 Annex A: Sampling weights The phone survey sample was drawn using Random Digit Dialing (RDD), where respondents are contacted using random plausible phone numbers who consent to be interviewed. These phone-based surveys may not be representative of these overall populations since they are likely to exclude those who do not own mobile phones, individuals who may be outside network coverage, and those who choose not to pick up phone calls from unknown phone numbers. Weighting techniques adjust the estimates by considering the different likelihood of being included in the sample based on the overall population's observable characteristics. Therefore, weighting allows the sample estimates to be closer to the overall population, thereby attenuating sample selection biases but not necessarily eliminating them due to the confounding effects from unobservable characteristics that will remain. The most recent representative national survey, from each country, was used to compute weights for these surveys, following a 2-step process. In the first step, the sample from this nationally representative survey was divided into cells based on following time-invariant demographic characteristics – location, age, gender, and education level of the respondents.28 Following this, the weighted share of respondents (over age 15) in each cell was estimated using the survey weights in the national sample. For every cell, the weighted share of respondents over age 15, and in households with at least one mobile phone was also computed. Using these two proportions, the original survey weights were then adjusted to be representative of those older than 15 years and from households with at least one mobile phone. Table A1: Most recent national Survey Survey Year Afghanistan Integrated Expenditure and Labor Force Survey 2016 Bangladesh Household Income and Expenditure Survey 2016-17 Bhutan Bhutan Living Standards Survey 2017 India Periodic Labor Force Survey 2018-19 Maldives Household Income and Expenditure Survey 2019 Nepal Nepal Labor Force Survey 2017-18 Pakistan Pakistan Social and Living Standards Measurement Survey 2018-19 Sri Lanka Household Income and Expenditure Survey 2016 28These observable characteristics are chosen based on the common demographic variables in the phone surveys and the nationally representative sample such that cells are non-empty in all surveys. Education is time-invariant for the target sample of adult respondents. 47 Second, the SAR COVID-19 Monitoring Phone Survey sample was also divided into cells based on the same set of variables: location, age, gender, and education level of the respondents; and the share of respondents is computed for each cell. The final step in estimating weights for the phone surveys was to inversely adjust the weights for each cell from the national survey, by the inverse of the share of respondents for each cell in the phone survey. At this stage, the resulting cells have not been further inflated or adjusted to be the same size as in the population, as is standard practice in household surveys. This is a deliberate choice to be transparent about the potential shortcomings of these types of phone surveys. Nevertheless, the RDD surveys do not appear to underrepresent groups who have been particularly impacted by the COVID-induced economic crisis such as less educated workers, and workers in rural areas. Given the focus on tracking labor market impacts, the share of female respondents and labor force participants is also adequate for analytical purposes, without an adjustment of weights to reflect population proportions. 48 Annex B: Additional Tables and Figures Table B1: Timeline of introduction of restrictions in SAR countries Start date of first Formal first national measure lockdown Afghanistan March 22 Sequential lockdowns in provinces starting with Herat on March 22nd, and extending to Logar, Kabul and Kandahar on March 26th, and so on. Nationally, wedding halls, restaurants, and other public areas including mosques were closed for public use. School closure took effect from March 16th. Bangladesh March 23 National 10-day lockdown which was March 23- May 30 extended till May 31st Bhutan March 23 International travel banned and all August 11-August 31 international borders sealed India March 24 March 24-May30 Maldives March 11 Three resorts closed, and schools shut. Male went into a three-week lockdown beginning April 17 Nepal March 24 March 24-July 21 Pakistan March 21 The Government of Sindh announced a April 1-April 30 lockdown in the province for 14 days from 23 March, ordering all public transport, markets, offices, shopping malls, restaurants, and public areas to be shut down. All major cities and provinces put in place restrictions on movements by March 24th. Sri Lanka March 21 March 21 – April 27 49 Table B2: Differential job loss rates, by baseline demographic and employment characteristics (1) (2) (3) (4) (5) 26-35 years -0.0114 (0.00882) 36-45 years -0.0308*** (0.00974) 46 or older -0.0224* (0.0126) Female 0.0111 (0.0112) Uneducated -0.0124 (0.0130) Manufacturing/Industry 0.0250** (0.0124) Services 0.0337*** (0.0104) Elementary occupation 0.000374 (0.0123) Rural 0.00165 0.00266 0.00193 0.0103 0.00277 (0.00764) (0.00797) (0.00761) (0.00798) (0.00806) Constant 0.0862*** 0.0670*** 0.0731*** 0.0508*** 0.0751*** (0.00794) (0.00518) (0.00491) (0.00889) (0.00520) Observations 27657 27644 27657 24069 25408 R-squared 0.002 0.000 0.000 0.003 0.000 Note: The table reports the results from a pooled regression with the dependent variable indicating whether an individual lost or changed the job, which they were employed in January 2020, between March 2020 and the time of the survey . In Column 1 – 5, the independent variables include dummy variables for age cohort, gender, education, sector of employment, and occupation quality of the pre-COVID job, respectively. In all columns, the regression specification also controls for the location of residence (urban/rural) and the country of interview. The sample includes individuals who were employed in January 2020. The sample size changes across regressions in different columns due to missing information (no response by respondent) on the attribute added as an independent variable in the column. Robust standard errors reported in parentheses. Levels of significance reported at * 10% ** 5% *** 1%. 50 Table B3: Differential job loss rates, by gender (Country-level regressions with additional controls) (1) (2) (3) (4) (5) (6) (7) (8) AFG BGD BTN MDV NPL PAK LKA IND Female 0.0201 0.00450 -0.0232 0.0120 0.115*** -0.0216 0.0227 0.0211 (0.0248) (0.0148) (0.0292) (0.0307) (0.0295) (0.0155) (0.0173) (0.0142) Uneducated -0.000286 -0.0301** 0.0340 0.0797 -0.0329 0.000882 0.00262 -0.00767 (0.0104) (0.0125) (0.0289) (0.0663) (0.0390) (0.0104) (0.0774) (0.0145) Manufacturing/Industry 0.0286 0.0611*** -0.000695 -0.0352 0.0893** 0.0381*** 0.0169 0.0149 (0.0200) (0.0182) (0.0406) (0.0634) (0.0367) (0.0125) (0.0230) (0.0134) Services 0.00498 0.0230* 0.0192 -0.0811 0.0454 0.0401*** 0.0300 0.0233** (0.0140) (0.0121) (0.0321) (0.0544) (0.0319) (0.0115) (0.0223) (0.0110) Constant 0.0372** 0.105*** 0.196*** 0.404*** 0.291*** 0.0509*** 0.108*** 0.106*** (0.0167) (0.0186) (0.0516) (0.0924) (0.0454) (0.0152) (0.0397) (0.0195) Observations 2839 4215 929 1014 1801 4758 2846 5655 R-squared 0.012 0.019 0.024 0.071 0.072 0.005 0.024 0.019 Note: The table reports the results from country-level regressions with the dependent variable indicating whether an individual recovered a job. The independent variable includes dummy variable indicating respondent’s gender. The regression specification also includes dummy variables for the month of interview, age cohort, education, sector of employment, and location of residence as additional controls. The sample includes individuals who lost or changed the job that they were employed in January 2020. Robust standard errors reported in parentheses. Levels of significance reported at * 10% ** 5% *** 1%. 51 Table B4: Differential job loss rates, by gender and education/sector of employment (1) (2) (3) (4) (5) (6) (7) (8) AFG BGD BTN MDV NPL PAK LKA IND Female 0.00299 0.0240 0.00498 0.0380 0.272*** 0.00195 -0.0130 0.0608* (0.0923) (0.0251) (0.0546) (0.149) (0.0609) (0.0304) (0.0331) (0.0312) Uneducated -0.00137 -0.00886 0.0424 0.0817 0.00896 0.0104 -0.0669*** 0.00101 (0.0108) (0.0140) (0.0358) (0.0762) (0.0507) (0.0112) (0.0198) (0.0152) Manufacturing/Industry 0.0320 0.0633*** 0.0439 -0.0684 0.149*** 0.0411*** -0.0129 0.0202* (0.0208) (0.0205) (0.0551) (0.0671) (0.0381) (0.0130) (0.0270) (0.0120) Services 0.00341 0.0155 0.0187 -0.0705 0.124*** 0.0392*** 0.0219 0.0402*** (0.0132) (0.0134) (0.0415) (0.0566) (0.0321) (0.0126) (0.0284) (0.0111) Female # uneducated 0.0170 -0.101*** -0.0216 0.00449 -0.118 -0.0463* 0.147 -0.0269 (0.0499) (0.0301) (0.0523) (0.138) (0.0755) (0.0280) (0.174) (0.0307) Female # Manufacturing/Industry -0.00948 0.000195 -0.137* 0.0601 -0.142 -0.0162 0.0773 -0.0138 (0.0891) (0.0366) (0.0742) (0.169) (0.0872) (0.0330) (0.0486) (0.0485) Female # Services 0.0253 0.0483* -0.00161 -0.0425 -0.205*** 0.00803 0.0207 -0.0581 (0.0960) (0.0250) (0.0606) (0.152) (0.0690) (0.0325) (0.0404) (0.0356) Constant 0.0379** 0.100*** 0.188*** 0.401*** 0.227*** 0.0470*** 0.119*** 0.0934*** (0.0160) (0.0183) (0.0545) (0.0945) (0.0439) (0.0158) (0.0429) (0.0210) Observations 2839 4215 929 1014 1801 4758 2846 5655 R-squared 0.013 0.025 0.029 0.073 0.080 0.007 0.028 0.021 Note: The table reports the results from country-level regressions with the dependent variable indicating whether an individual recovered a job. The independent variable includes dummy variable indicating respondent’s gender, and the interaction variables betwee n gender and education and gender and sector of employment. The regression specification also includes dummy variables for the month of interview, age cohort, education, sector of employment, and location of residence as additional controls. The sample includes individuals who lost or changed the job that they were employed in January 2020. Robust standard errors reported in parentheses. Levels of significance reported at * 10% ** 5% *** 1%. 52 Table B5: Differential job recovery rates, by gender (Country-level regressions with additional controls) (1) (2) (3) (4) (5) (6) (7) (8) AFG BGD BTN MDV NPL PAK LKA IND Female -0.230*** -0.103*** -0.101** -0.0409 -0.0302* -0.256*** -0.0443 -0.0647 (0.0479) (0.0303) (0.0485) (0.0724) (0.0176) (0.0526) (0.0285) (0.0513) Uneducated -0.0159 -0.0415 0.0368 -0.271** 0.0603 0.00379 -0.113*** -0.0813 (0.0645) (0.0317) (0.0585) (0.111) (0.0394) (0.0420) (0.0342) (0.0579) Manufacturing/Industry -0.149 0.0883** 0.00900 -0.147 -0.00853 -0.0805 -0.0389 0.00258 (0.0979) (0.0427) (0.0620) (0.129) (0.0283) (0.0747) (0.0539) (0.0596) Services -0.0655 0.0734** 0.0554 -0.0833 0.00347 0.0438 -0.0372 0.0800 (0.0980) (0.0334) (0.0572) (0.121) (0.0239) (0.0767) (0.0573) (0.0555) Constant 0.459*** 0.196*** 0.115 0.379** 0.0573 0.410*** 0.248** 0.277*** (0.107) (0.0498) (0.0811) (0.178) (0.0361) (0.0839) (0.103) (0.0729) Observations 252 964 215 204 875 841 748 679 R-squared 0.157 0.045 0.096 0.104 0.020 0.083 0.020 0.078 Note: The table reports the results from country-level regressions with the dependent variable indicating whether an individual lost or changed the job, which they were employed in January 2020, between March 2020 and the time of the survey. The independent variable includes dummy variable indicating respondent’s gender. The regression specification also includes dummy variables for the month of interview, age cohort, education, sector of employment, and location of residence as additional controls. The sample includes individuals who were employed in January 2020. Robust standard errors reported in parentheses. Levels of significance reported at * 10% ** 5% *** 1%. 53 Table B6: Differential job recovery rates, by gender and education/sector of employment (1) (2) (3) (4) (5) (6) (7) (8) AFG BGD BTN MDV NPL PAK LKA IND Female -0.288** -0.0544 -0.0169 -0.184 -0.0540 -0.349*** 0.0222 0.109 (0.125) (0.0583) (0.0998) (0.225) (0.0401) (0.0822) (0.106) (0.108) Uneducated -0.0141 -0.0340 0.0714 -0.295* 0.0749 -0.0152 -0.142*** -0.0268 (0.0731) (0.0375) (0.0756) (0.154) (0.0527) (0.0444) (0.0436) (0.0765) Manufacturing/Industry -0.168 0.0857* 0.0135 -0.232 -0.0458 -0.0801 0.00177 0.0249 (0.111) (0.0476) (0.0767) (0.162) (0.0344) (0.0768) (0.0687) (0.0616) Services -0.0718 0.0824** 0.0567 -0.128 -0.00303 0.0358 -0.00815 0.162*** (0.107) (0.0366) (0.0802) (0.150) (0.0345) (0.0792) (0.0694) (0.0586) Female # uneducated -0.0173 -0.0274 -0.133 0.0483 -0.0512 0.0790 0.0219 -0.160 (0.0835) (0.0698) (0.0981) (0.158) (0.0655) (0.107) (0.0829) (0.106) Female # Manufacturing/Industry 0.101 -0.0219 -0.0918 0.212 0.104* 0.0449 -0.0934 -0.00584 (0.131) (0.0881) (0.102) (0.257) (0.0617) (0.0946) (0.112) (0.134) Female # Services 0.0571 -0.0621 -0.0433 0.141 0.00146 0.0762 -0.0708 -0.248** (0.129) (0.0596) (0.111) (0.239) (0.0439) (0.109) (0.109) (0.113) Constant 0.464*** 0.195*** 0.0962 0.430** 0.0717* 0.419*** 0.215* 0.213*** (0.112) (0.0506) (0.0957) (0.198) (0.0377) (0.0872) (0.122) (0.0794) Observations 252 964 215 204 875 841 748 679 R-squared 0.159 0.046 0.103 0.107 0.031 0.084 0.022 0.095 Note: The table reports the results from country-level regressions with the dependent variable indicating whether an individual lost or changed the job, which they were employed in January 2020, between March 2020 and the time of the survey. The independent variable includes dummy variable indicating respondent’s gender, and the interaction variables between gender and education and gender and sector of employment. The regression specification also includes dummy variables for the month of interview, age cohort, education, sector of employment, and location of residence as additional controls. The sample includes individuals who were employed in January 2020. Robust standard errors reported in parentheses. Levels of significance reported at * 10% ** 5% *** 1%. 54 Table B7: Pooled regression with month and country of interview dummies (Aug20 – Apr21 sample) (1) Pooled sample September -0.00999 (0.0185) October 0.0263 (0.0309) November -0.0265 (0.0465) December 0.0281 (0.0536) Afghanistan 0.188*** (0.0364) Bangladesh 0.133*** (0.0202) Bhutan 0.101*** (0.0284) Maldives 0.112*** (0.0410) Pakistan 0.216*** (0.0563) Sri Lanka 0.0902** (0.0374) India 0.166*** (0.0440) Constant 0.128*** (0.0434) Observations 4062 R-squared 0.076 Note: The table reports the results from a pooled regression with the dependent variable indicating whether an individual recovered a job on the independent categorical variables for country (Nepal omitted) and survey month (August omitted month). The regression specification also controls for individual’s age cohort, gender, education, sector of employment, and location of residence (rural/urban). The sample includes individuals who lost or changed the job that they were employed in January 2020. The sample is also restricted to individuals who were interviewed between August 2020 and December 2020. Robust standard errors reported in parentheses. Levels of significance reported at * 10% ** 5% *** 1%. 55 Table B8. Country-level regressions with the number of months since lockdown dummies (1) (2) (3) (4) (5) (6) (7) (8) AFG BGD BTN MDV NPL PAK LKA IND Months since lockdown 4 0 0 (.) (.) - 5 0.0503 0 0 0 0.000895 (0.0928) (.) (.) (.) (0.0244) 6 0.101 -0.0383 0.0767 0.226*** -0.0325 0 (0.0906) (0.0306) (0.0488) (0.0707) (0.0250) (.) - 7 -0.0490 -0.0441 0.0541 0.000360 0.0346 -0.0850* 0 (0.0937) (0.0389) (0.0967) (0.0644) (0.0426) (0.0452) (.) 8 -0.00755 -0.0691 0.187** 0 -0.0901** 0.0124 (0.102) (0.0571) (0.0806) (.) (0.0451) (0.0604) 9 0.261*** -0.0445 0.128 0.0209 (0.0823) (0.0461) (0.133) (0.0670) 10 0.568*** -0.0270 -0.00347 (0.173) (0.0532) (0.0574) 11 0.319* -0.160** 0.0702 (0.166) (0.0634) (0.0631) 12 0.291** 0.268*** (0.146) (0.104) 13 -0.156*** (0.0566) Constant 0.248** 0.288*** 0.145** 0.0699 0.0642** 0.442*** 0.242*** 0.263*** (0.0985) (0.0358) (0.0643) (0.0536) (0.0289) (0.0519) (0.0608) (0.0673) Observations 270 1037 218 232 886 900 787 737 R-squared 0.131 0.034 0.059 0.086 0.017 0.074 0.031 0.079 Note: The table reports the results from country-level regressions with the dependent variable indicating whether an individual recovered a job on the independent dummy variables for the number of months since the lockdown was imposed in the country (the omitted category is the number of months with a zero coefficient). The omitted category differs between countries due to the variation in the month in which survey rolled out in those countries. The regression specification also controls for individual’s age cohort, gender, education, sector of employment, and location of residence (rural/urban). The sample includes individuals who lost or changed the job that they were employed in January 2020. The sample spans the entire sample period from August 2020 to April 2021.Robust standard errors reported in parentheses. Levels of significance reported at * 10% ** 5% *** 1%. 56 Table B9: Share of migrants with a rural destination (1) (2) (3) Full Sample Lost a job Did not lose a job Afghanistan 12% 18% 9% Bangladesh 39% 41% 38% Bhutan 46% 50% 38% Maldives 62% 59% 63% Nepal 49% 54% 42% Pakistan 32% 33% 31% Sri Lanka 52% 59% 47% India 48% 58% 44% All SAR countries 44% 49% 40% Note: The table reports the percentage of individuals who migrated to a rural area among all those who migrated after March 2020. Columns 2 and 3 report the share of rural migrants for the following two subgroups: those who migrated and lost their pre-COVID job, and those who migrated but did not lose their pre-COVID job, respectively. 57 Figure B1: Sectoral composition of workforce in January 2020 Agriculture Manufacturing Services 48% 45% 49% 46% 52% 53% 53% 80% 24% 26% 15% 28% 26% 27% 37% 32% 31% 14% 22% 26% 20% 26% 15% 6% AFG BGD BTN IND MDV NPL PAK LKA Note: The sample includes individuals who were employed in January 2020. Source: SAR COVID-19 Monitoring Phone Survey, Round 1. 58 Figure B2: Job loss rates, by gender and sector of employment for males and females a. Education levels b. Sector of employment Male Female Male Female 11% 14% 12% 0% SERVICES 17% NO EDUCATION LKA LKA 8% 8% 10% 11% MANUFACTURING/I… SOME EDUCATION AGRICULTURE 8% 8% 7% SERVICES 4% 5% 6% 7% NO EDUCATION 4% PAK PAK MANUFACTURING/I… 7% SOME EDUCATION 8% AGRICULTURE 2% 26% 35% 24% SERVICES 26% 27% NO EDUCATION 31% NPL NPL 43% 45% MANUFACTURING/I… 41% 15% 14% SOME EDUCATION AGRICULTURE 15%16% 18% SERVICES 27% NO EDUCATION MDV MDV 28% MANUFACTURING/I… 14% 15% 23% SOME EDUCATION 24% AGRICULTURE 9% SERVICES 7% 8% 5% NO EDUCATION 7% IND IND 9% 11% MANUFACTURING/I… 7% SOME EDUCATION 9% AGRICULTURE 4% 13% 16%15% 13% SERVICES 10% NO EDUCATION BTN BTN MANUFACTURING/I… 4% 13% 14% 13%11% 12% SOME EDUCATION 7% 10% AGRICULTURE SERVICES 7% NO EDUCATION 2% BGD BGD 11% MANUFACTURING/I… 9% 16% SOME EDUCATION 5% 5% AGRICULTURE 0% SERVICES 9% 5% 11% NO EDUCATION AFG AFG 8% MANUFACTURING/I… 9% 5% SOME EDUCATION 8% 4% AGRICULTURE 9% Note: The sample includes individuals who were employed in January 2020. Source: SAR COVID-19 Monitoring Phone Survey, Round 1. 59 Figure B3: Difference in size distributions of new firms and firms that closed 10% 2% 1% -3% -9% 1 WORKER 2 WORKER 3 WORKERS 4 WOKRERS MORE THAN 5 Note: The sample includes: (i) firms in January 2020 that were operational at the time of the survey defined based on individuals who worked as an own-account worker with at least one additional worker and in the same business in January 2020 and at the time of the survey; and (ii) firms in January 2020 that shut down by the time of the survey defined based on individuals who worked as an own-account worker, and employed at least one additional worker in January 2020, but were unemployed at the time of the survey. Figure B4: Difference in sectoral distributions of new firms and firms that closed 6% -3% -3% AGRICULTURE MANUFACTURING SERVICE Note: The sample includes: (i) firms in January 2020 that shut down by the time of the survey defined based on individuals who worked as an own-account worker, and employed at least one additional worker in January 2020, but were unemployed at the time of the survey; and (ii) new firms that opened after January 2020 based on individuals who report working as an own-account worker with at least one additional employee at the time of the survey, but were either unemployed or employed in a different job as a wage worker (not as an own-account worker) or employed in a difference business (as an own-account worker) in January 2020. 60 Figure B5: Difference in spatial distributions of new firms and firms that closed High impact region -5% Medium impact region 5% Low impact region 0% -6% -4% -2% 0% 2% 4% 6% Note: The Y-axis plots spatial differences in net firm openings, where new firms and firm closures are defined as in Figure 10. For consistency, administrative regions within each country are ranked (and then categorized into high, medium, and low, on the x-axis) based on the job loss rate (of workers employed in January 2020) in the region. A district constitutes the smallest administrative region in all SAR countries except the Maldives, where a province is the smallest administrative unit collected in the survey. 61 Figure B6: Job recovery rates, by education and sector of employment for males and females a. Education levels b. Sector of employment Male Female Male Female 20% Services 38% No education 18% 10% IND IND Manufacturing 24% 29% 22% Some education 18% 23% Agriculture 20% 0% Services 13% No education 8% 0% LKA LKA Manufacturing 13% 13% 6% Some education 13% 8% Agriculture 12% 28% Services 39% No education 20% 14% PAK PAK Manufacturing 27% 5% Some education 33% 9% Agriculture 35% 0% 11% Services 7% No education 2% 5% NPL NPL Manufacturing 3% 8% Some education 5% 3% Agriculture 7% 2% 8% Services 30% No education 20% 0% MDV MDV Manufacturing 18% 24% Some education 27% 20% Agriculture 37% 26% 15% Services 18% No education 10% 0% BTN BTN Manufacturing 11% 0% Some education 14% 10% Agriculture 7% 0% 11% Services 19% No education 8% 4% BGD BGD Manufacturing 20% 13% Some education 17% 8% 11% Agriculture 0% 11% Services 29% No education 4% 4% AFG 14% AFG Manufacturing 0% Some education 17% 31% 11% Agriculture 0% Note: The sample includes individuals who were employed in January 2020. Source: SAR COVID-19 Monitoring Phone Survey, Round 1. 62 Figure B7: COVID-19 Stringency Index Note: Oxford COVID-19 Government Response Tracker, Blavatnik School of Government, University of Oxford. Last updated on January 18th, 2022. 63