Policy Research Working Paper 9859 Inequality under COVID-19 Taking Stock of High-Frequency Data for East Asia and the Pacific Lydia Y. Kim Maria Ana Lugo Andrew D. Mason Ikuko Uochi Poverty and Equity Global Practice & East Asia and the Pacific Region November 2021 Policy Research Working Paper 9859 Abstract While the distributional impacts of the COVID-19 larger among poorer workers who have found it more dif- pandemic have been well-documented in high-income ficult to return to employment. Data on the loss of labor countries, studies in low- and middle-income countries have income also suggests that the pandemic has exacerbated been relatively rare due to data limitations. This paper uses existing inequalities. In addition to being more susceptible pre-pandemic household welfare data and high-frequency to employment and income shocks, poorer households in household phone survey data from seven middle-income East Asia and the Pacific are at higher risk of experiencing countries in East Asia and the Pacific, spanning May 2020 long-term scarring from the pandemic – due to rising food to May 2021, to analyze the distributional impacts of insecurity, increased debt, distress sale of assets, and fewer the pandemic and their implications for equitable recov- distance/interactive learning opportunities for their chil- ery. The results indicate that employment impacts at the dren. Taken together, the findings indicate that inequality extensive margin have been large and widespread across has worsened during the pandemic, raising concerns about the welfare distribution during times of stringent mobil- the prospects for an inclusive recovery in the absence of ity restrictions (low mobility). When mobility restrictions appropriate policy measures. have been relaxed, however, employment impacts have been This paper is a product of the Poverty and Equity Global Practice and the Office of the Chief Economist, East Asia and the Pacific Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at at ykim14@worldbank.org, mlugo1@worldbank.org, amason@worldbank.org, and iuochi@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Inequality under COVID-19: Taking Stock of High-Frequency Data for East Asia and the Pacific Lydia Y. Kim, Maria Ana Lugo, Andrew D. Mason, Ikuko Uochi 1 World Bank Keywords: COVID-19 pandemic, inequality, poverty, East Asia and the Pacific JEL classification: D30, I31, O15, J6. 1 Lydia Kim, Maria Ana Lugo, and Ikuko Uochi are from Poverty and Equity Global Practice, for East Asia and the Pacific; Andrew Mason is from the Chief Economist Office for East Asia and the Pacific. Corresponding authors: mlugo1@worldbank.org and ykim14@worldbank.org. Rinku Murgai (Practice Manager, POV EAP) and Hassan Zaman (EFI Director for EAP) provided overall guidance on this project. We received invaluable comments from Rabia Ali, Tanida Arayavechkit, Kristen Himelein, Aaditya Mattoo, Ambar Narayan, Emilie Perge, Ririn Salwa Purnamasari, Matthew Wai-Poi, and Judy Yang. We thank the country survey teams for providing us access to the datasets and providing feedback on the analysis. Poor and vulnerable workers are more likely to work in less secure, informal employment often characterized by low skills, making them more susceptible to employment and income shocks in the face of an economic crisis such as the COVID-19 pandemic. Various studies in several developed countries have shown that the pandemic has disproportionately affected low-wage, low-skill, informal, and minority workers (Chetty et al., 2020; Gould and Kandra, 2021; OECD, 2020; Gould and Wilson, 2020). At the other extreme, many of those at the very top of the world’s welfare distribution have thrived during the pandemic, further fueling the sense that the pandemic has exacerbated inequalities. In 2021, Forbes added a record 660 new members to its billionaires list and estimated that 86 percent of the world’s billionaires saw increases in their net worth compared to a year ago. However, whether these patterns seen in high-income countries also hold in low and middle-income countries, with vastly different economic structures, access to technology, and government capacity, remains a question. In low and middle-income countries, the distributional impacts of the pandemic may differ from what has been observed globally among high-income countries for several reasons. The poorest households in low and middle-income countries tend to live in remote, rural areas and are more likely to be reliant on agriculture, which has generally been less affected by lockdowns and can act as a fallback when other sectors contract. Self-employment is also more common in low and middle-income economies, particularly among the bottom 40 percent of population, and thus economic impacts may not be reflected in job losses to the same extent as in high-income economies, as some may continue to work – albeit with lower earnings – until forced to fold. Additionally, while the top 20 are more likely to work in service sectors that may be more conducive to remote working arrangements such as finance, information, administration, and other professional activities, they are still significantly engaged in economic sectors for which remote work arrangements is more difficult such as retail, tourism, and manufacturing. Relatively limited scope for remote work means that the pandemic and continuing containment measures in low and middle-income countries may affect workers across the distribution similarly. At the same time, households may differ in their ability to cope and adapt to similar shocks depending on their welfare status at the onset of the crisis. Some coping mechanisms that poorer households in low and middle-income countries may resort to – namely, selling productive assets, increasing indebtedness levels, reducing food intake at critical ages, removing children from school – may have deleterious impacts on human capital and long-term individual and household prospects, which could, in turn, have serious implications for future inequality. Therefore, the question of whether the pandemic has had an equalizing effect or has exacerbated inequalities in low- and middle-income economies merits empirical exploration. Recent studies have employed household survey data from low and middle-income countries around the world and have found that the economic impacts of the pandemic and the recovery have been uneven, with female, younger, less educated, and urban workers being more likely to stop working (Agrawal, et al., 2021; Kugler, et al., 2021). This paper adds to this literature by using high-frequency household survey data from East Asia and Pacific (EAP) countries to examine the socioeconomic and distributional impacts of the pandemic. Specifically, in this paper, we look at the pandemic’s distributional impacts in the context of three questions. First, have poorer workers been more vulnerable to employment and earning losses than wealthier workers, or has the likelihood of experiencing losses been widespread irrespective of welfare status? Second, how have strategies employed by households to cope with these shocks differed across welfare groups? Third, what are the potential longer-term effects that the pandemic has given rise to? The East Asia and Pacific region provides a unique context in which to explore these questions for several reasons. The onset of the pandemic began in the EAP, thereby affecting travel and trade channels within 2 and in and out of the region early on. Border closures and restricted trade with China, in addition to precautionary containment measures that were put in place at the onset of the pandemic had a profound economic impact on many EAP countries in the early months of 2020, before many other countries had felt its toll. Furthermore, despite being geographically close, many developing EAP countries had differing experiences during the pandemic in terms of the severity of local COVID-19 transmission, mobility restrictions, and economic impacts. For example, countries like Vietnam and Lao PDR had few COVID-19 cases throughout 2020 and after a brief period of mobility restrictions and an economic contraction in the first half of 2020, their economies rebounded relatively quickly until experiencing an uptick in cases in mid- 2021. On the other hand, countries such as Indonesia and the Philippines experienced a pandemic-induced health crisis early on which led to economic contractions that crippled the economy for much of 2020. Thus, the prolonged and varied impact of the pandemic across middle-income countries in the EAP region provides a befitting context to explore the distributional impacts of the economic crisis that have unfolded from the pandemic across different economies. Our main source of data is the High Frequency Phone Surveys (HFPS), collected by the World Bank from May 2020 in collaboration with National Statistics Offices of numerous developing countries. 2 While these surveys are not without their caveats due to constraints of the mobile phone survey methodology, 3 they provide timely and useful indicative evidence on how COVID-19 has affected households, in the absence of standard labor force and expenditure surveys. Between May 2020 and June 2021, 11 countries in the EAP region administered the HFPS, of which nine implemented at least two surveys. The HFPS data analyzed in this paper 4 comes from seven of these countries and spans one year of the pandemic, from May 2020 to May 2021. Countries included in the analysis are Indonesia, Lao PDR, Mongolia, Myanmar, Philippines, Papua New Guinea (PNG), and Vietnam. 5 While the design of the HFPS differs slightly across countries, all surveys share a core set of indicators that provide insight into the socioeconomic impacts of the pandemic, including information on employment, income losses, coping mechanisms, food insecurity, and access to services. The paper contributes to the burgeoning literature on the distributional impacts of the COVID-19 pandemic around the world. Numerous studies have documented the labor market and income impacts of the pandemic in high-income countries such as the United States, Japan, and the European Union, among many others. In the United States, Chetty et al. (2020) find that the pandemic led to widespread layoffs, which were greater and more long-lasting for low-wage workers. Other studies such as Gould and Wilson (2020) have documented the disproportionate impact the pandemic has had on black, Hispanic, and female workers in the United States. Several recent studies have used the World Bank’s global harmonized data 2 While most countries administered the HFPS in consultation with their local NSO, Mongolia was the only country in which the NSO was the implementing agency of the HFPS. Country-specific briefs and reports, produced by the country teams, can be found in the website links available here: https://www.worldbank.org/en/topic/poverty/brief/high-frequency-monitoring-surveys 3 All HFPS used here are nationally representative of households that have phones; hence their national representativeness depends on the country’s mobile phone penetration. In addition, by design, the questionnaires are much shorter (ideally, they should take less than 15 minutes to complete, with relatively simple questions) than standard face-to-face surveys, to ensure quality of responses. This means that the resulting indicators collected may differ from the official definitions (for instance, on employment) and that the analysis may be less comprehensive than what can be done with standard labor force surveys or the equivalent. 4 Note that the data used in this paper does not come from the World Bank’s global harmonized data set. While from the same sources of raw data, the data used here has been harmonized specifically for the purposes of this paper. 5 Four countries – Cambodia, Malaysia, Solomon Islands, and Thailand – also administered the HFPS but are not included in this analysis. At the time of this paper, Malaysia and Thailand had only implemented one round, for which data were not yet available. Solomon Islands had low representation of bottom 40 households. Cambodia’s small sample size, particularly in later rounds, made quintile-level disaggregation difficult. 3 from the HFPS to look at the pandemic’s socioeconomic and distributional impacts in the developing world. Kugler et al. (2021) and Khamis et al. (2021) have shown that these surveys are a rapid, efficient, and reliable source of data despite potential shortcomings. Kugler et al. (2021) as well as Bundervoet et al. (2021) both show that women, youth, and less-educated workers were significantly more likely to lose their job and experience reductions in income. Agrawal et al. (2021) use HFPS data from 17 developing countries to look at the distributional impacts of the pandemic, particularly how inequality has played a role in their economic recovery. They find that the lower-skilled, youth, and women were more likely to be impacted initially and that the recovery in employment was not sufficient to reduce the initial disparities in losses. Our main contribution is threefold. First, we examine the differential impact of the pandemic on households across the welfare distribution using recent pre-existing or imputed data of welfare status before the onset of the pandemic, thus building on previous studies which mainly focus on impacts by gender, location, and education. We are able to obtain pre-COVID-19, household-level welfare data for all the high-frequency surveys administered in the EAP, allowing us to analyze the impacts the pandemic has had across welfare quintiles on household-level outcomes such as income, coping mechanisms, food insecurity, access to services, in addition to individual employment. Second, we take advantage of multiple rounds of data in each country, examining the evolution of impacts and coping methods over time: from periods of stringent mobility restrictions and lockdown to reopening and (limited) recovery. Third, unlike the global studies mentioned above, we aim to capture the heterogeneity of experiences across these seven EAP economies while highlighting emerging common patterns. Our study suggests the following. First, at the extensive margin (work stoppages), employment impacts during the pandemic have been substantial among middle-income EAP countries, but particularly during periods of tighter mobility restrictions, either due to sudden or continuing lockdown measures. Second, the pandemic may have exacerbated existing inequalities, although in ways different from those typically observed in high-income countries. When mobility restrictions were high, workers experienced work stoppages across a wide range of sectors, and employment impacts were distributed relatively evenly across the welfare distribution. As economies in the EAP reopened, however, work stoppages became more selective, disproportionally affecting poorer workers. Our analysis also shows that less-educated, female, and urban workers were more likely to stop working. Taken together, these findings suggest unequal impacts of the pandemic that may jeopardize an inclusive recovery and exacerbate inequality in the EAP. Third, despite significant expansions of safety nets in several middle-income EAP countries, the depletion of physical and human capital has been more dire among the poor: When faced with income losses, poorer household are more likely to reduce their food consumption, accumulate debt and sell productive assets, undermining their ability to recover from the crisis. Finally, with prolonged disruptions in education in many EAP countries, differences in distance learning opportunities across welfare groups may be leaving poorer children behind, potentially having negative long-term consequences in inequality. The paper is structured as follows. The first section describes the data and provides summary statistics of the key indicators used in this paper. Section 2 analyzes work stoppages by country-specific quintiles of the welfare distribution, across countries and over time. While impacts have been heterogeneous across countries, notable patterns emerge. Section 3 looks beyond work stoppages at changes in household labor income across the distribution. Section 4 briefly presents differences across quintiles in coping strategies, food insecurity, and access to education, which may have potential long-term implications for inequality within countries. Section 5 concludes. 4 1. Data: High Frequency Phone Surveys (HFPS) in the EAP region Starting from May 2020, the High Frequency Phone Surveys (HFPS) were conducted in 11 middle-income countries in the EAP in order to monitor the socioeconomic impacts of the pandemic. The surveys covered a wide range of topics, including employment, income, food insecurity, access to health and education services, and coping mechanisms, among others. World Bank teams working in each country led the design and collection of the data. The first rounds of surveys were administered in May-June 2020 and subsequent rounds continued into 2021. This paper utilizes data from seven of the 11 countries in which these data were collected (Indonesia, Lao PDR, Mongolia, Myanmar, Philippines, PNG, and Vietnam) and spans the period from May 2020 to May 2021. We include at least two survey rounds for each country to allow examination of the temporal aspects of the pandemic’s impacts. In the EAP, the HFPS relied on one of two sampling methods: Four countries drew from a sample frame of a recent representative household survey, while three used random digit dialing from a roster of phone lines (Table 1). Sampling weights were constructed for all surveys to ensure unbiased estimates from the sample. As the high-frequency surveys were phone-based, the samples are representative of households who have access to a phone. This may limit representativeness in areas and subpopulations for which phone penetration is low. One major contribution of this paper relative to other studies that have used the HFPS is the use of household-level welfare data to estimate the distributional impacts of the pandemic. For countries that drew their sample from a previous pre-pandemic household survey (Mongolia and Vietnam), consumption data from these existing surveys could be used to identify each household’s welfare status before the pandemic. For the other five countries that employed random digit dialing methods or did not have pre- pandemic consumption information, welfare was imputed using a basic set of welfare predictors collected in the HFPS or from recent surveys. As shown in Table 1, various methods employing survey-to-survey multiple imputation such as proxy means testing or Survey of Well-being via Instant and Frequent Tracking (Yoshida, et al., 2015) were used, relying on recent household surveys with household expenditure data. Table 1. Overview of HFPS data used 2020 Construction of Household Population 2020 GDP welfare sample size size per capita Country Survey timing Sampling method indicator in Round 1 (million) (2010 USD) Indonesia R1: May 2020 Subsample of the 2018 Proxy means 4,338 273.5 4,312 R2: May-Jun 2020 Urban Perception, 2018 test R3: Jul 2020 Rural poverty, and the R4: Nov 2020 2020 Digital economy R5: Mar 2021 household surveys Lao PDR R1: Jun-Jul 2020 Random digit dialing Proxy means 2,500 7.3 1,836 R2: Feb-Mar 2021 test Mongolia R1: May 2020 Subsample of 2018 Consumption 1,334 3.3 4,054 R2: Sep 2020 Household Socio- aggregate from R3: Dec 2020 Economic Survey previous survey R4: Apr 2021 Myanmar R1: May 2020 Random digit dialing Survey of Well- 1,500 54.4 1,478 R2: Jun 2020 from an existing list being via Instant R3: Aug 2020 owned by contracted and Frequent R4: Oct 2020 firm for the survey Tracking 5 Philippines R1: Aug 2020 Phone survey based on Survey of Well- 9,448 109.6 2,980 R2: Dec 2020 survey firm sample being via Instant R3: May 2021 frame; Self- and Frequent administered web Tracking surveys PNG R1: Jun-Jul 2020 Random digit dialing Wealth index 3,114 8.9 2,347 R2: Dec 2020 Vietnam R1: Jun 2020 Subsample from Consumption 6,213 97.3 2,123 R2: Jul 2020 representative 2018 aggregate from R3: Sep 2020 survey previous survey Two primary data sets from the HFPS are used in the analysis. First, an individual-level data set that includes information on pre-pandemic employment6 allows examination of how workers were initially impacted, how they have coped and recovered, and who is still left behind. Due to differing sampling methods, characteristics among the sample of respondents vary across countries (Table 2). 7 For example, the Indonesia HFPS only asked about the employment status of the household head or breadwinner. Thus, males are overrepresented in the sample, and the average age is on the higher end among the sample of countries. In Vietnam, information from a previous household survey was used to target the household head as the respondent, while in Lao PDR and Mongolia, the targeted respondent in the first survey round was an adult member of the household who answered the phone or was available to be surveyed. In Myanmar, the Philippines, and PNG, employment data was collected for both the respondent and the household head when the respondent was not the head, but only in PNG was the pooled sample used to measure employment impacts. 8 Thus, in some countries, household heads are overrepresented, though at varying degrees (Table 2). 9 To the extent that household heads differ in terms of age, gender, type and formality of work, and other job characteristics from the average worker, the sample of respondents may not fully capture the pandemic’s impact on the labor force. To control for these differences across countries in respondent characteristics, we conduct the analysis controlling for respondent’s characteristics including relation to the household head, gender, and age. Results are generally not qualitatively different whether these controls are included, so unconditional means are reported throughout the paper unless otherwise noted. Regression results including controls are available in the appendix. The second data set is a cross-sectional household data set that is used to analyze household-level indicators of income, coping mechanisms, food insecurity, and access to services. While it is possible to build a panel data set of households, cross-sectional data are used in order to maximize the number of 6 For Indonesia, Mongolia, Philippines, and Vietnam, we use an individual-level panel data set. However, building a true panel was difficult or purposely not done for Lao PDR, Myanmar, and PNG. Lao PDR and Myanmar added new respondents in each round to account for attrition and maintain representativeness, and thus restricting the sample to just panel respondents across rounds would have limited the sample size substantially. Instead, new respondents were included in the sample, which was possible due to pre-pandemic employment information that was asked to new respondents in each round. For PNG, an entirely new sample was surveyed in round 2, which had greater representation of households in the bottom 40 percent of the welfare distribution. We thus utilize retrospective information from the round 2 survey, which included questions about employment at the time of round 1 as well as pre-pandemic. 7 The respondents of mobile phone surveys in low- and middle-income countries typically underrepresented women, older people, the less educated, and people in rural areas (Lau et al 2019). 8 In an effort to maximize comparability across countries, the pooled sample of individuals were not used for Myanmar and the Philippines, which also did not have representative weights for the pooled sample. In PNG, individual-level weights were only available for the pooled sample of heads and respondents, and hence this sample was used to estimate employment impacts. 9 Indonesia – which targeted the household head or primary breadwinner – is at one extreme, while Philippines and Mongolia are at the other and are more similar to the actual percentage (47 percent in the original Mongolia sample survey, 34 percent in 2015 Philippines Census). 6 observations across countries and rounds and preserve representativeness. Table 2 presents data on unweighted sample sizes from the round 1 survey of each country. Table 2. Respondent characteristics, round 1 Indonesia Lao PDR Mongolia Myanmar Philippines PNG Vietnam Relation (%) Head 93 55 52 52 36 72 63 Spouse 3 21 40 23 35 0 25 Other 4 24 8 24 30 28 12 Average age (years) 46 38 47 41 41 45 51 Male (%) 85 60 33 58 35 51 53 Sample size 4,216 2,500 1,026 1,500 2,924 3,368 6,213 Note: Respondent refers to the subject of the employment module. In all countries, respondents were restricted to adults, defined as 15 years old or older. In Indonesia, the respondent was asked to answer questions in the employment module about the household head or breadwinner. In Mongolia Round 3-4, the respondent (if different from the previous respondent) was asked to report on the previous respondent. In PNG, respondents and household heads (if not the same) are pooled. In the remainder of the paper, we focus on six key indicators to examine the distributional impacts of the pandemic in the context of the aforementioned three questions. Table 3 provides summary statistics by country and round for each of these six outcomes. Table 3. Summary statistics of outcomes (percentage) Sold assets Reduction Used Interactive Stopped or increased Food in labor savings to distance working debt to insecurity income cope learning cope Survey Survey (% of Country round period (% of those (% of households working households (% of (% of (% of with children pre- with labor households) households) households) attending pandemic) income) school pre- pandemic) 1 May-Jun20 24 76 2 May-Jun20 59 32 27 43 Indonesia 3 Jul-Sep20 10 44 25 4 Oct-Dec20 54 25 25 70 5 Jan-Mar21 10 36 24 1 Jul-Sep20 13 56 4 19 7 Lao PDR 2 Jan-Mar21 10 41 1 May-Jun20 19 59 28 5 10 33 2 Jul-Sep20 37 9 Mongolia 3 Oct-Dec20 52 48 17 6 7 38 4 Apr-Jun21 32 51 27 7 4 1 May-Jun20 57 82 48 34 10 2 May-Jun20 33 62 59 39 13 Myanmar 3 Jul-Sep20 23 53 59 33 16 4 Oct-Dec20 34 75 56 40 24 5 7 1 Jul-Sep20 31 75 83 53 50 13 Philippines 2 Oct-Dec20 24 57 76 58 37 33 3 Apr-Jun21 21 63 90 14 35 1 May-Jun20 44 59 52 30 44 PNG 2 Oct-Dec20 33 1 May-Jun20 3 32 16 5 58 Vietnam 2 Jul-Sep20 9 31 8 2 3 Jul-Sep20 7 27 8 1 First, we look at work stoppages and reductions in labor income in order to assess how the pandemic has affected employment and income for workers and households across the welfare distribution. Work stoppages are defined as the share of respondents (i.e., subjects of the employment module) working pre- pandemic who were not working in the seven days preceding the survey. Reductions in labor income are defined as the share of households with wage or farm/non-farm business income that reported experiencing a reduction in income from these sources since the previous round. 10 We also assume that work stoppages translate into losses in labor income. 11 To examine the coping strategies households have employed to deal with pandemic-induced shocks, we examine the use of savings in addition to coping mechanisms that involve selling productive assets or are potentially debt-increasing – namely, crediting purchases, delaying payments, or borrowing from friends, family, moneylenders, or other sources. Both outcomes look at use of coping mechanisms among all households since the previous survey round. Finally, we explore potential longer-term effects on human capital accumulation by looking at food insecurity and engagement in interactive distance learning among school-age children. In this paper, we define food insecurity as the share of households that reported that they had either run out of food, went hungry, or was hungry but did not eat, due to a lack of money or resources in the 30 days preceding the survey. Engagement in interactive distance learning is the share of households with children attending school before the pandemic whose children either used mobile apps, online or in-person meetings/sessions with a teacher or tutor, or other online learning platforms in the week prior to the survey. Mobility and timing of the surveys Across and within the seven EAP countries included in this analysis, mobility restrictions varied substantially between May 2020 and May 2021, 12 reflecting the variation in the severity of the pandemic and related 10 Looking at changes in income between rounds does not take into consideration net changes in income since pre-pandemic. For example, a household may have faced an income loss between pre-pandemic and round 1 but may have experienced an income gain between round 1 and round 2. Without information on the magnitude of these changes, it is difficult to say whether the household experienced a net loss, net gain, or income stayed the same relative to before the pandemic. Information on net losses are only available for a few country-rounds, so we look at changes in income between rounds instead. 11 Data from Mongolia and Myanmar, for instance, indicate that the majority of those who stop working do not have a job to return to or have a job to return to but have experienced a reduction in wages. Thus, the assumption of lower labor income appears to be reasonable. 12 Mobility is measured using data from the Google Mobility Reports, and the mobility index is here defined as the average percentage change in mobility to retail and recreation, transit stations, parks and workplace, relative to a pre-pandemic baseline level. We also analyzed the stringency index - as reported by the Oxford Coronavirus Government Response Tracker (OxCGRT) 8 containment measures over time and across countries. While most countries experienced an initial shock that restricted mobility between March and April of 2020, some countries, like the Philippines and Indonesia, experienced more prolonged reductions in mobility with little recovery over time (Figure 1a). Other countries such as Myanmar and Mongolia showed signs of recovery in mobility in May 2020, soon after the initial lockdown, but later in the year faced sharp declines in mobility due to a renewed surge in COVID-19 cases. Finally, countries like Lao PDR, PNG, and Vietnam experienced relative normalcy in mobility during 2020 after the initial shock, reflecting comparatively low local transmission of COVID-19 until 2021. Multiple rounds of the HFPS have helped capture the socioeconomic and distributional impacts of the pandemic at varying periods of each country’s experience during the pandemic. For example, in Indonesia and the Philippines, the first survey round occurred at a time when mobility was on average more than 40 percent lower than baseline levels, while the subsequent round with employment and income data occurred at a time of relative improvement (Figure 1b). On the other hand, the first round of the HFPS in Mongolia and Myanmar was implemented at a period of comparative recovery, while later rounds (rounds 3 and 4 for Mongolia, round 4 for Myanmar) took place during a time of lockdown in each country. 13 While mobility varied relatively less across rounds in Lao PDR, PNG, and Vietnam, it was still below pre-pandemic levels in Lao PDR and Vietnam during the included survey rounds. Figure 1a. Some countries experienced more severe containment measures that restricted mobility Percent change in mobility compared to pre-pandemic baseline Figure 1b. Survey timing and frequency differed across countries HFPS field dates and mobility index project - but found that in some countries, there was a significant disconnect between restrictions imposed and actual movement of people as indicated by the mobility data (in some countries, measures were stringent but mobility during the same period varied significantly), particularly as the pandemic progressed. The variation in work stoppages and earnings was more closely related to mobility than stringency. 13 In Mongolia’s round 4 survey, the reference period for employment was not the week preceding the survey but April 5-11, 2021, before the lockdown occurred. 9 Note: The mobility index in an average of the percentage change in mobility to retail and recreation locations, transit stations, parks, and workplaces. The 7-day rolling average is shown. The change is calculated relative to a pre-pandemic baseline. Grey areas indicate when each round of the HFPS was fielded in the specified country. Indonesia had five rounds, but only the first, third, and fifth included information on employment and income. Source: Google Mobility Reports; HFPS 2. Work stoppages: Employment impacts of the pandemic in developing EAP countries The HFPS provides insight into the employment impacts of the pandemic at varying points during 2020- 2021 in each country. In particular, we look at employment impacts at the extensive margin – namely, work stoppages 14 – rather than at the intensive margin due to limitations in the data. The HFPS suggests that the initial impact of the pandemic on work stoppages was significant in most countries 15 but particularly in 14 As described in the previous section, work stoppages are defined as the share of respondents working pre-pandemic who are not working in the week preceding the survey. A worker may be not working even though she is still employed. 15 The initial impact of the pandemic on work stoppages may even be underestimated, given that most round one surveys occurred after the steep drop in mobility between March and May 2020 (see Figure 1). 10 Myanmar, Papua New Guinea (PNG), and the Philippines, where a third or more of respondents working pre-pandemic were not working at the time of the survey (Figure 2). Most countries experienced periods of recovery in the latter half of 2020, when the rate of work stoppages declined. However, some countries like Mongolia and Myanmar faced renewed shocks due to a surge in COVID-19 cases late in 2020. Figure 2. Work stoppages have varied across countries and periods Share of respondents working pre-pandemic who are not working Note: Work stopppages are defined as the share of respondents working pre-pandemic who were not working in the week preceding the survey. Source: HFPS; Round 2 is used for Myanmar May-Jun 2020; Round 3 is used for Vietnam Jul-Sep 2020 Differences across countries in mobility restrictions have had implications for the distributional impact of the pandemic on employment in the EAP. 16 Across and within countries, periods of lower mobility have generally been associated with higher average work stoppages (Figure 3A, Appendix Table 1). 17 Moreover, lower mobility, either due to continuing or sudden lockdown measures, has been associated with widespread employment impacts across the welfare distribution (Figure 3B, Appendix Table 1). In other words, during these periods, workers were similarly likely to stop working, irrespective of their position in the welfare distribution. Figure 3B shows that times of tighter (de facto) mobility restrictions were associated with a more uniform distribution in work stoppages across welfare groups: The top 20 to bottom 20 ratio in work stoppages was closer to one (denoted by the horizontal dotted line) during times of low mobility, indicating a similar likelihood to stop working. 18 As mobility restrictions were relaxed, businesses and workplaces reopened, and economies recovered, poorer workers – often employed in less secure, informal jobs – were less likely to return to work. In figure 16 Results shown in this note are unconditional averages unless otherwise noted. Controlling for survey period or respondent characteristics such as being the household head, age, and gender, do not qualitatively change results. Greater susceptibility to seasonal labor among poorer workers may also be a concern when looking at employment patterns throughout the course of the year. Data show that while seasonality is indeed a commonly cited reason for being out of work, it does not explain differences that can be observed across the welfare distribution. 17 Among the countries considered, PNG appears as an exception. Despite having relatively normal internal mobility, work stoppages in the country are unusually high, likely associated with agricultural seasonality (World Bank and UNICEF 2021). 18 Vietnam represents an outlier with work stoppages being generally low, but among those that stopped working, wealthier workers were more likely to be affected. This can be explained by two factors which are discussed in detail below: 1) higher participation among poorer quintiles in agriculture; and 2) higher likelihood of poorer households to live in rural areas. 11 3B, we see that with increasing mobility, differences in work stoppages between the top 20 and bottom 20 became larger, with the bottom 20 being more likely to stop working. Similar results can be seen in the regressions in Appendix Table 1. The coefficient on quintile 5 (relative to quintile 1, the omitted category) is negative and significant in periods of higher mobility (omitted), while this effect disappears when mobility is lower. Figure 3. Work stoppages across country-rounds are negatively correlated with mobility. Poorer workers were more likely to stop working during times of high mobility. A. Mobility and work stoppages B. Mobility and Top 20 to Bottom 20 ratio in work stoppages Note: Each point represents a country-round. The mobility index in an average of the percentage change in mobility to retail and recreation locations, transit stations, parks, and workplaces. The average of the mobility index is calculated for each country- round. The regression lines exclude outliers (i.e., PNG in Panel A and Vietnam in Panel B). See footnotes 17 and 18 for further details. Source: HFPS, Google mobility reports Below, countries are categorized into groups to shed light on different experiences across countries in the region. Mongolia and Myanmar – Between April and May 2020, Mongolia and Myanmar experienced significant economic shocks and a drop in mobility due to precautionary containment measures that were put in place, in addition to the decline in global demand, particularly from China (World Bank 2021a). Round 1 of both the Mongolia and Myanmar high-frequency surveys took place after this shock during a period when mobility was relatively high (Figure 1b). At this time, poorer workers in both countries were less likely to resume work, leading to higher work stoppages (Figure 4). In Mongolia, poorer workers were significantly more likely to be out of work in May-June 2020, with workers in the bottom quintile being twice as likely to have stopped working as those at the top. A similar downward gradient in work stoppages across quintiles was also visible in Myanmar between May and September 2020. However, as stringent mobility restrictions were reinstated in both countries late in 2020 due to a sudden increase in locally transmitted cases, differences across the distribution dissipated. As sectors with high participation among workers in the top 20 such as retail and previously less affected sectors such as transportation, finance, real estate, professional activities suffered contractions, the pandemic’s impact on employment became more widespread across the distribution. 12 Figure 4. Work stoppages in Mongolia and Myanmar Mongolia Myanmar Note: 95% confidence intervals for difference from Q1 are shown (rather than from zero). Including controls for respondent characteristics do not qualitatively change results. See Appendix Table 2 for regression tables with controls. Source: HFPS Lao PDR, PNG, and Vietnam – After a short decline in mobility in March to April 2020, Lao PDR, PNG, and Vietnam experienced relatively normal mobility throughout the rest of the year and into 2021. Yet, in Lao PDR and PNG, poorer workers found it harder to resume work in 2020 and thus were more likely to be out of work compared to their wealthier counterparts: In 2020, the poorest quintile in Lao PDR was twice as likely as the wealthiest quintile to stop working, while in PNG, the poorest quintile was 2.3 to 2.9 times more likely to be out of work that the top quintile (Figure 5). While the prevalence of work stoppages was generally very low in Vietnam, a different gradient is evident, with wealthier workers marginally more likely to have stopped working. This result, however, is explained by two factors: 1) differences in sectoral participation across welfare quintiles; and 2) differences in residential area areas across quintiles. As with many other middle-income EAP countries, agriculture was one of the least affected sectors in Vietnam in terms of work stoppages. However, among the sample of countries, participation among bottom 20 workers in agriculture is relatively high 19 in Vietnam (74 percent), and differences across quintiles in agricultural employment are among the largest. Nevertheless, data on work stoppages is also available over a longer reference period in Vietnam (not only the week prior to the survey), which shows a clear regressive gradient, with households in the poorer quintiles being significantly more likely to have been affected by job losses relative to those in wealthier ones (World Bank 2021b). 19 Vietnam is the second highest after PNG, in which 88 percent of bottom 20 workers were employed in agriculture pre- pandemic according to the HFPS. In PNG, however, agriculture was one of the most severely affected sectors. 13 Figure 5. Work stoppages in Lao PDR, PNG, and Vietnam Lao PDR PNG Vietnam Note: 95% confidence intervals for difference from Q1 are shown (rather than from zero). Including controls for respondent characteristics do not qualitatively change results. See Appendix Table 2 for regression tables with controls. Source: HFPS; Round 3 is used for Vietnam Jul-Sep 2020 Indonesia and the Philippines – Like other countries discussed above, Indonesia and the Philippines experienced a steep decline in mobility early in 2020, but unlike the others, they faced a pandemic-induced health crisis relatively early in 2020 and a prolonged period of depressed mobility (lower than -20 percent). Far-reaching mobility restrictions translated into employment shocks that were similarly widespread across the welfare distribution (Figure 6). A wide range of sectors experienced work stoppages during early stages of the pandemic, including sectors such as retail, transportation, and finance in which significant shares of wealthier workers were employed. For example, in Indonesia, HFPS data indicate that more than half of workers in the top 20 percent of the welfare distribution were employed in retail, transportation, or 14 manufacturing before the pandemic (compared to 32 percent among the bottom 20), i.e., sectors that have traditionally required physical presence and thus are not very amenable to home-based work. These sectors were also among the hardest hit early on, with 28 percent of workers employed in them pre- pandemic being out of work by May 2020. Even within sectors, those at the bottom and at the top of the distribution were on average similarly likely to be out of work in Indonesia and the Philippines, despite official data showing higher rates of job security and formality among the wealthier. Towards the end of 2020 and into 2021, the economy in Indonesia and the Philippines started to recover, yet mobility remained well below pre-pandemic levels and below the levels of most other countries in the EAP (Figure 1). At this time, signs of potentially unequal effects emerged, but differences across quintiles in work stoppages were largely not statistically significant across periods (Figure 6). However, evidence suggests that recovery may have been slower for poorer workers in the Philippines and Indonesia. In the Philippines, workers in the top 60 percent of the distribution were more likely to return to work than those in the poorest 20 percent between August and December 2020. 20 In Indonesia, while differences across welfare quintiles were not statistically significant, the top 40 experienced a faster recovery in employment after May 2020, and results from March 2021 also suggest that more educated workers had a faster rebound, particularly relative to those in the bottom quintile. 21 Moreover, as will be shown below, wealthier workers in both countries were less likely to face reductions in labor income compared to poorer ones after the initial impact of the pandemic. Figure 6. Work stoppages in Indonesia and the Philippines Indonesia Philippines 20 While work stoppages between August 2020 and May 2021 paint a slightly different picture, data shows that compared to top 60 workers, bottom 40 workers were more likely to stop working due to COVID-related reasons in May 2021, such as being ill/quarantined or having to take care of kids or ill family members. 21 See Round 5 report for Indonesia: https://thedocs.worldbank.org/en/doc/0485c24dfb867c75fcf15e98fed9fb13- 0070012021/original/Indonesia-HiFy-COVID-19-Round-5.pdf 15 Note: 95% confidence intervals for difference from Q1 are shown (rather than from zero). Including controls for respondent characteristics do not qualitatively change results. See Appendix Table 2 for regression tables with controls. Source: HFPS What may be driving unequal impacts on employment? In short, evidence from these seven EAP countries suggests that the pandemic may have exacerbated inequalities in employment, particularly when the economy was reopening, and mobility restrictions were less stringent. But in periods when restrictions were more severe (and overall mobility was low), work stoppages were not significantly different across income groups (Appendix Table 1). What might be driving these results? Information on economic structures in middle-income EAP countries and characteristics of jobs and workers may hint to the mediating factors behind the observed welfare gradient at times of recovery or relatively normalcy in mobility. First, differences observed between top and bottom workers at times of high or low mobility may reflect welfare differences in economic sector of employment. In general, workers in the top 20 were more likely to be engaged in sectors that were generally less likely to experience work stoppages – for example, public administration, professional, financial sectors (Figure 7, Panel A). Indeed, when restrictions were relatively relaxed and the economy started to reopen, work stoppages were selective, with sectors with high top 20 participation such as professional and financial activities, experiencing low rates of work stoppages (Figure 7, Panel B). This is consistent with the unequal impacts we find during the periods of low mobility. However, during periods of de facto high mobility restrictions, work stoppages were widespread across most economic sectors, even including typically less vulnerable sectors in which the wealthiest are more likely to be employed. This, combined with the fact that top 20 participation was still significant in sectors that were relatively hard hit, 22 may attenuate the (potential) regressive gradient in work stoppages, especially in times of stringent mobility restrictions. - Nonetheless, sector of employment alone does not appear to explain differences between the top 20 and the bottom 20 in the likelihood to stop working during times of higher mobility. Table 1 in the Appendix shows that the coefficient on quintile 5 does not change significantly once we control for the most recent sector of employment. One possible explanation for this result is that the industry classification available in the HFPS is not granular enough to pick up differences across quintiles in employment sector. These results suggest that differences across the welfare distribution within sectors (e.g., occupation, ability to work from home, job formality, job security) likely have contributed more to observed differences among the top 20 and bottom 20. 22 Across countries in the sample, an average of 46 percent of workers in quintile 5 worked in manufacturing, transportation, retail, or other services pre-pandemic. 16 Figure 7. Sectors in which top 20 workers are more likely to be employed have faced lower work stoppages A. Work stoppages vs. top 20 sectoral C. Work stoppages by education of respondent participation B. Work stoppages by sector and mobility Note: In figure A, each dot represents a country-round-sector. Highly affected service sectors include accommodations and food and other service activities; Not as affected service sectors include retail, public administration, professional, financial, education, and health. In figures B and C, work stoppages are estimated over a pooled sample of countries and survey rounds. In figure B, higher mobility includes country-rounds with an average of the mobility index above -20 percent; lower mobility includes country-rounds with an average mobility index below -20 percent. Conditional means are taken from a regression of work stoppages on sector fixed effects and controls including country fixed effects, respondent gender, age, and head status. For figures A and B, the most recently engaged sector is used for all rounds. Source: HFPS; Google mobility reports While we are not able to observe many job characteristics that can differentiate workers within sectors, we can analyze differences according to workers’ characteristics. Indeed, within and across countries, we find that workers with tertiary education have consistently been less likely than those with below secondary or 17 secondary education to stop working. Workers with tertiary education were on average 6.6 percentage points less likely to be out of work than those with less than high school education (Figure 7, Panel C and Appendix Table 3). Interestingly, differences between workers with tertiary education and those without decrease once we control for sector of employment: The relative advantage of tertiary educated workers to those with less than high school education declines to 3.7 percentage points (Appendix Table 3). Yet, the difference remains significant. suggesting that differences within sector associated with type of occupation, job formality/security, and the ability to work from home may be driving results. In addition, when looking at differences across quintiles, controlling for education, the coefficient on quintile 5 is reduced only slightly (from -0.131 to -0.126) and remains significant (Appendix Table 1, column 5). A third hypothesis, though one that we are not able to test here, is that even within certain economic sectors, poorer workers are generally more likely to work in micro or small firms, which have been disproportionately affected to the pandemic (Freund and Pesme, 2021). Adopting digital technology such as digital platforms and digital solutions has been a common response among businesses to dealing with decreased sales during the pandemic. Findings from the Business Pulse Surveys suggest that digital adoption is associated with a smaller decline in sales. However, these surveys show that smaller businesses have been significantly less likely to adopt digital technology than larger ones (Freund and Pesme, 2021). To note, in addition to differences in work stoppages across quintiles and by education levels, female workers were more likely to stop working once we control for education and sector of employment, according to pooled regression analyses (Appendix Table 2, 3). However, there is some variation across countries (Figure 8, Appendix Table 2). In places like Indonesia and the Philippines, women were less likely than men to return to work at times of economic reopening, even conditional on sector of employment (Figure 8). 23 In other places such as Mongolia, differences between men and women in the likelihood of stopping work were only marginally statistically significant initially, but women were significantly less likely than men to return to work, thus enlarging the gender gap in work stoppages across time. As lockdown measures led to school closures, increased caregiving activities largely fell on women, presenting barriers to employment. In addition, in some countries, women were more likely to work in sectors that were harder hit by the pandemic such as retail, manufacturing, and tourism. Thus, in the longer-term, it is possible that women may face greater difficulty recovering from the pandemic. 23 Uneven recovery from work stoppages may be just one part of the gender story: Analysis based on the Indonesia Labor Force Surveys, however, shows that six months into the pandemic, women in Indonesia were more likely to join the labor force for the first time, largely in unskilled, informal, and agricultural employment (Halim et al., 2021). 18 Figure 8. Male-Female difference in work stoppages (percentage points) Note: Each bar represents the male coefficient in a regression by country and survey round of work stoppages on a male dummy and (most recent) employment sector fixed effects. Asterisks denote levels of statistical significance. Source: HFPS In sum, while periods of stringent government containment measures may have had similarly far-reaching impacts on workers across the welfare distribution, in areas or times of recovery or relative normalcy in terms of mobility, those at the top may have been more shielded from the economic impacts of the pandemic. The partially mixed picture of work stoppages (with unequal impacts only at times of lower mobility restrictions) contrasts what has typically been observed in high-income countries. While beyond the scope of the present note, as mentioned in section 1, this result may be due to the tendency of the poor and vulnerable in middle-income EAP countries to reside in rural areas and be engaged in agriculture, which was relatively less affected during the pandemic. Moreover, limited scope for home-based work, as well as higher engagement in self-employment activities in middle-income EAP may have influenced our results. The next section looks at household labor income losses, complementing the analysis on work stoppages and providing a more comprehensive picture of the economic impacts of the pandemic. 3. Household labor incomes: Looking beyond work stoppages HFPS data on labor income losses may paint a more comprehensive picture than work stoppages of the extent of the pandemic’s impact on livelihoods for two main reasons. First, the HFPS typically only asks about the employment status of one or two members of the household, and overwhelmingly the household head in most countries (Table 2). Thus, to the extent that employment patterns and job characteristics differ between the head and other members of the household or that additional household members begin to work as the main income earner loses employment (added worker effect), these surveys may not fully capture the pandemic’s impact on employment. Second, employment status is sensitive to type of employment, for example whether one is a wage employee or a self-employed worker. Self-employed workers such as farmers and business owners have greater discretion over whether or not they stop working or lose their job compared to employees. For example, a shop owner may not have any customers but may continue to keep shop until they are forced to close their business. They may lay off employees as demand continues to decline, but only when the business folds will the shop owner be considered 19 unemployed. Indeed, self-employment in the EAP is quite extensive, ranging from 36 percent in the Philippines to 75 percent in PNG as a share of the labor force. 24 In middle-income countries such as those in the EAP, poorer workers are generally more likely to be self- employed, while wealthier workers are more likely to work in salaried jobs (Fields 2019). Self-employment is relatively high in the agriculture sector and certain service sectors such as retail trade, which also had lower work stoppages (Figure 9, Panel A). Thus, looking solely at employment changes may understate the impact of the pandemic on inequality if poorer workers are less likely to report being out of work due to the nature of their employment. Additionally, work stoppages only consider the extensive margin of employment impacts. Data from the World Bank’s Business Pulse Surveys suggest that firms have been more likely to reduce work hours, reduce wages, and/or grant paid or unpaid leave: Two-thirds of businesses in the sample made adjustments to their payroll, but only about one in five actually fired workers (Freund and Pesme, 2021). As a result, those who do not face job losses may still face reduced working hours and wages. 25 Income changes can thus provide a more accurate picture of the impact of the pandemic on inequality. Indeed, in all of the countries analyzed, the share of households that experienced labor income losses was larger than the share of respondents that experienced work stoppages. Moreover, in most countries, households were significantly more likely to experience reductions in business income than in wages (Figure 9, Panel B). Figure 9. While the self-employed may be less likely to stop working, income losses from self- employment were significantly larger than those from wage labor A. Self-employment and work stoppages B. Wage vs. Non-farm business income losses Note: Data from six countries are included: Indonesia, Lao PDR, Mongolia, Myanmar, the Philippines, and Vietnam. The share of workers who are self-employed come from the most recent available pre-pandemic LFS for each country. Share of households with incomes losses are the average across rounds by country. Source: ILOSTAT, HFPS Wealthier households engaged in wage or farm/non-farm business were less likely to report experiencing a labor income decline relative to the previous period. Across countries, households in the top 20 were on average about 5 percentage points less likely than those in the bottom 20 to experience losses in labor 24 Based on 2019 data from the World Development Indicators. 25 It is also plausible that losses in employment do not necessarily mean losses in income, particularly if work stoppages are temporary. However, as further elaborated in the Appendix, evidence suggests that while some work stoppages may not have resulted in reductions in income, the large majority has been associated with declines in wages or profits. 20 income (Table 4). When controlling for survey period, in all countries besides Vietnam, households in the wealthiest quintile were less likely to report that they had experienced a decline in their household labor income relative to the previous period. However, only in Indonesia and the Philippines were differences statistically significant at the 5 percent level. As previously mentioned, in Vietnam, the upwards gradient in income losses across quintiles is explained by relatively high participation among households in the bottom 20 in agriculture, which was more resilient during the pandemic compared to other sectors. The difference between the poorest and the wealthiest households is particularly stark in urban areas (7-8 percentage points), while differences across quintiles are small in magnitude and statistically insignificant in rural areas (Appendix Table 4). This result in part reflects the resilience of agriculture during the pandemic in many EAP countries. Moreover, when disaggregated by income source, analysis shows that differences across quintiles in income losses from farm and non-farm business are small and statistically insignificant at the 5 percent level (Appendix Table 5). On the other hand, welfare differences – particularly between the wealthiest and poorest quintiles – in reductions in wages are large (11 percentage points) and highly statistically significant, suggesting that much of the increased inequality in labor income losses is attributable to wage income shocks. There are two important caveats worth highlighting. First, the information collected does not capture the cumulative effect of labor income declines. In each round, the respondents are asked about changes in labor income relative to the previous round. Without information on the magnitude of the loss, it is difficult to assess whether the same households experiencing income losses in one period are able to recover it in the next. While information on net income changes relative to pre-pandemic are available for a few countries in later rounds, this data is not available for the majority of country-rounds. Second, and relatedly, even if the prevalence of labor income losses is similar across the distribution, we are not able to speak about the magnitude of the loss (in absolute or relative terms) which may differ across the distribution. However, information collected in later rounds in Mongolia and Indonesia on percentage changes in income compared to pre-pandemic may offer some insight. The percentage decline in income relative to pre-pandemic is not statistically different between poorer and wealthier households for either country for the months inquired. In any case, income losses are more likely to have serious and long-term consequences for poor and vulnerable households, for whom even small reductions in income can be crippling. 21 Table 4. Wealthier households were less likely to experience reduction in wage or business income relative to previous period, though this is not significant everywhere 60 Share of households with wage/business income 50 40 30 20 10 0 Q1 Q2 Q3 Q4 Q5 Note: The reference period for looking at changes in income are since the previous round. The exception is the first round, for which the reference period is pre-pandemic. Results from an OLS regression are shown. A probit regression produces qualitatively similar results. Income is not available for PNG. Source: HFPS 4. Potential long-term effects of the pandemic: Coping mechanisms, food insecurity and access to education Consistently, poorer households have been more likely than wealthier households to rely on coping mechanisms that may inhibit recovery such as selling scarce assets or livestock, crediting purchases, delaying payments, and borrowing (Figure 10). Across countries, households in the bottom 20 were on average 16 percentage points more likely than those in the top 20 to rely on such methods of coping. 22 Instead, in most countries, the wealthiest quintile was 9 percentage points more likely than the poorest to use their savings to cope with the economic impacts of the pandemic. For instance, in Indonesia, 42 percent of households in the top 20 used their savings to cope with lower labor income losses, compared to 26 percent among the poorest quintile. Thus, while the pandemic may have had sweeping impacts across the welfare distribution, among those that have experienced a shock, poorer households have been forced to rely on coping mechanisms that may have more severe consequences for their future prospects. Figure 10. Poorer households that experienced labor income losses were more likely to sell assets or increase debt than the wealthier counterparts Sold assets or increased debt to cope 100 experienced labor income losses 80 Percent of households who 60 40 20 0 Pooled Indonesia Lao PDR Mongolia Myanmar Philippines Vietnam Q1 Q2 Q3 Q4 Q5 Relied on savings to cope 80 experienced labor income losses Percent of households who 60 40 20 0 Pooled Indonesia Lao PDR Mongolia Myanmar Philippines Vietnam Q1 Q2 Q3 Q4 Q5 Reduced food consumption to cope 100 experienced labor income losses 80 Percent of households who 60 40 20 0 Pooled Indonesia Lao PDR Mongolia Myanmar Philippines Vietnam Q1 Q2 Q3 Q4 Q5 Note: Coping mechanisms that involve selling assets or increasing debt include: selling farm or non-farm assets, crediting purchases, delaying payments, and borrowing from friends, family, moneylenders, or other sources. 95% confidence intervals for difference from Q1 are shown (rather than from zero). The pooled regression is unconditional; results including country and period fixed effects are similar. Source: HFPS 23 One type of extreme coping mechanism with potential short and long-term consequences is reduction of food intake. While in general, poorer households were more likely to reduce food consumption to cope with the effects of the pandemic, differences across quintiles were not particularly stark in most countries, and in some, differences were negligible (Figure 10). However, reduced food consumption is likely to be more detrimental for the poorest households, who are more prone to food insecurity. Poor and vulnerable households tend to spend a large share of their income on food, and thus even a slight income shock may put them at risk of food insecurity or greater food insecurity, which could have serious long-term implications on human capital, particularly for children (McGovern et al., 2017). Indeed, analysis shows that across countries, households in the bottom quintile are on average 17-19 percentage points more likely to be food insecure than those in the wealthiest quintile (Table 5, columns 1-2). 26 This result is robust to controlling for labor income losses in Table 5, column 3, suggesting that other factors may be driving differences across the distribution in food insecurity. However, in the absence of information on whether or not the household was already food insecure even before the pandemic hit, it is difficult to know how much of the differences across quintiles can be attributable to the pandemic. 27 While the prevalence of labor income reductions was similar across the welfare distribution, it is possible that differences in the type of mechanisms employed to cope with these negative impacts has contributed to greater food insecurity among poorer households. Table 5, column 4 shows that households that sold assets or increased their level of indebtedness were more likely to be food insecure. Yet, controlling for these coping mechanisms reduces, but does not alter substantively, the regressive gradient observed across quintiles in food insecurity. This finding suggests that poorer households may have been more likely to experience food insecurity partially due to higher reliance on coping mechanisms that increased indebtedness or were otherwise harmful to household welfare. Table 5. Regression of food insecurity on welfare quintiles, labor income loss, and coping mechanisms (pooled) Food insecurity 40 Percent of households that experienced a reduction in 30 labor income 20 10 0 Q1 Q2 Q3 Q4 Q5 26 This result is consistent within each country as well. 27 Correlational evidence does indicate, however, that the pandemic may have put households, particularly those in the bottom 20, at greater risk of food insecurity. 24 Note: The sample includes households from Indonesia, Mongolia, Myanmar, Philippines, and PNG that are not missing coping and income data. The reference period for food insecurity is the past 30 days for all countries except Indonesia, for which it is the past 7 days. The final dimension that we analyze that may have long term implications relates to children’s access to education. School closures may have a real and detrimental effect on learning outcomes for children, particularly if prolonged over an extended period of time. In Indonesia, an estimated 62.5 million students from pre-primary to higher education have been affected by school closures that started in March 2020 and continued late into the year. 28 In the absence of online learning platforms or other tools that enable effective home learning, school closures have the potential to make a lasting impact on human capital accumulation for school-age children. Estimates suggest that students in the EAP stand to lose an average of about two-thirds of a year of learning-adjusted years of schooling, with loses twice as high in countries such as Cambodia, Lao PDR and Myanmar (World Bank 2021c). These losses may have the longer-term effect of negatively impacting the future earning potential of individuals, reducing educational and income mobility (Narayan et al. 2018, World Bank 2018). Across most countries analyzed, children in poorer households were at greater risk of being left behind: Among households with children attending school pre-pandemic, those in the bottom quintile were on average 16 percentage points less likely than those in the top quintile to engage in face-to-face learning or utilize online, mobile, or video platforms that allow more interactive learning (Figure 11). In the Philippines, for instance, students from the richest quintiles were twice as likely to engage in these more interactive educational activities than those at in the bottom quintile (38 vis-à-vis 16 percent, respectively). Most of these mediums require access to technology, particularly the internet or a computer or mobile phone, presenting a greater challenge for children at the bottom of the welfare distribution to engage in more interactive methods of learning at home. 28 https://blogs.worldbank.org/eastasiapacific/covid-19-and-learning-inequities-indonesia-four-ways-bridge-gap 25 Figure 11. Children in wealthier households have been more likely to be engaged in more interactive educational activities than poorer ones Interactive distance learning by country and quintile Q1 Q2 Q3 Q4 Q5 100 Percent of households with children attending school 80 before the pandemic 60 40 20 0 Pooled Indonesia Lao PDR Mongolia Myanmar Philippines Vietnam Note: Modes of interactive distance learning include mobile apps, online or in-person meetings/sessions with a teacher or tutor, or other online learning platforms. 95% confidence intervals for difference from Q1 are shown (rather than from zero). The pooled regression includes country and period fixed effects. Education data are only available for Round 4 in Myanmar and Round 2 and 4 in Indonesia. Source: HFPS 5. Prospects for inclusive recovery This paper presents evidence from the High Frequency Phone Surveys (HFPS) indicative of the risk of rising inequality both in the short-and long-terms across a selected set of EAP countries. Unlike in developed countries, work stoppages and labor income have been relatively widespread at times of economic closure and downturn, although there is some evidence that the top 20 have been able shield themselves more than workers at the bottom of the distribution when economic activity resumed. The data on potentially harmful coping mechanisms, food insecurity, and distance learning suggest that the impacts could be long lasting, suggesting that the recovery may be uneven. These results from high-frequency survey data, should be taken as indicative of trends, in the absence of official household and labor force surveys. The HFPS, while timely and informative for looking at how households have weathered during the pandemic on a number of dimensions, is not without limitations. As mentioned in the paper, limited numbers of survey rounds and a respondent sample can only provide a partial (and imperfect) picture of the impacts. In addition, the rapid nature of the phone surveys means that the questionnaires may lack the level of detail necessary to look at precise mechanisms of the pandemic’s economic impact on workers and households. As nationally representative household or labor surveys become available, a more comprehensive assessment of the extent to which the pandemic and its recovery have increased inequality in the region will be important. Not discussed in this note is the important role that the expansion of safety nets played in many countries in the region in mitigating to some extent these negative impacts. Several countries in EAP have made unprecedented efforts to scale up their social protection systems in a relatively short amount of time. In Mongolia, according to the December 2020 survey, recipients of the Child Money Program have largely used the benefit to purchase food, while a third to save for future usage, potentially increasing their buffer. Among recipients, 8 percent declared that the transfers have completely mitigated the negative impact of 26 the pandemic, and 81 percent only partially. 29 In Indonesia, the increased reliance on government assistance among the bottom 40 (reaching 80 percent in November) may have helped families avoid resorting to the more destructive coping mechanisms. The share of households cutting down on food consumption declined from nearly 70 percent in June to less than 50 percent in November. 30 Yet, other losses -such as those of human capital- may be harder to rebuild. As the economies in the region recover, unprecedented levels of fiscal stimulus will need to unwind. An informed assessment of which groups have been most affected and the extent to which government actions have mitigated these effects will be critical. Social protection responses originally designed to be of limited duration were often extended as employment and income shocks continued. Going forward, countries will need to consider how to support an inclusive recovery in the face of mounting fiscal pressure. Finally, the roll-out of the vaccines will play a critical role in achieving a sustained recovery from the pandemic. Countries that have vaccinated their population more rapidly tend to be those with stronger recovery capacity (World Bank 2021c). 31 While vaccination levels are ramping up quickly in many EAP countries such as China, Cambodia, and Malaysia, they remain low in the poorest countries, including Myanmar and PNG, as well as some of the largest, such as Indonesia, Philippines, and Vietnam. Recent data from the region suggests that acceptance is relatively high in most countries, but that timely access to sizeable stocks is proving difficult. In addition, even when countries do manage to procure the needed vaccines, reaching poor and underserved populations may be challenging due to inadequacy in the distribution infrastructure. In addition, there is a risk that poorer households could opt out disproportionately if they are required to pay for vaccinations. 29 Results of Mongolia COVID-19 Household Response Phone Survey (Round 3). January 2021. 30 Indonesia High-frequency monitoring of COVID-19 impact (brief for round 4). January 2021. 31 According to estimates in that report, 10-percentage point higher vaccination coverage was associated with a one-half of a percentage point higher quarterly gross domestic product growth. 27 References Agrawal, Sarthak; Cojocaru, Alexandru; Montalva, Veronica; Narayan, Ambar; Bundervoet, Tom; Ten, Andrey. (2021). “COVID-19 and Inequality: How Unequal Was the Recovery from the Initial Shock?”. World Bank, Washington, DC. Bundervoet, Tom; Davalos, Maria E.; Garcia, Natalia. (2021). “The Short-Term Impacts of COVID-19 on Households in Developing Countries: An Overview Based on a Harmonized Data Set of High-Frequency Surveys.” Policy Research Working Paper No. 9582. World Bank, Washington, DC Chetty, Raj, Friedman, John, Hendren, Nathaniel, and Stepner, Michael and the Opportunity Insights Team (2020). “How did covid-19 and stabilization policies affect spending and employment? a new real-time economic tracker based on private sector data”. National Bureau of Economic Research No. 27431, Cambridge, MA. Gould, Elise, and Jori Kandra. (2021). “Wages Grew in 2020 Because the Bottom Fell Out of the Low-Wage Labor Market: The State of Working America 2020 Wages Report”. Economic Policy Institute, February 2021. Gould, Elise, V. Wilson (2020). “Black workers face two of the most lethal preexisting conditions for coronavirus—racism and economic inequality” (Rep.). Washington, D.C: Economic Policy Institute. Fields, Gary. (2019) “Self-employment and poverty in developing countries”. IZA World of Labor 2019: 60 doi: 10.15185/izawol.60.v2 Freund, Caroline, and Jean Denis Pesme. (2021). “Supporting Firms in Restructuring and Recovery” (English). Washington, D.C. : World Bank Group. Halim, Daniel, Sean Hambali, and Ririn Purnamasari (2021). “The gendered impacts of the COVID-19 pandemic on Indonesia’s labor market: how does it shape the future of women’s work?”. Jakarta, Indonesia: World Bank Group. Khamis, Melanie; Prinz, Daniel; Newhouse, David; Palacios-Lopez, Amparo; Pape, Utz; Weber, Michael. (2021). “The Early Labor Market Impacts of COVID-19 in Developing Countries: Evidence from High-Frequency Phone Surveys”. Jobs Working Paper No. 58. World Bank, Washington, DC Kugler, Maurice; Viollaz, Mariana; Duque, Daniel; Gaddis, Isis; Newhouse, David; Palacios-Lopez, Amparo; Weber, Michael. (2021). “How Did the COVID-19 Crisis Affect Different Types of Workers in the Developing World?”. World Bank, Washington, DC. Lau, C. Q., Cronberg, A., Marks, L., & Amaya, A. (2019). “In Search of the Optimal Mode for Mobile Phone Surveys in Developing Countries. A Comparison of IVR, SMS, and CATI in Nigeria.” Survey Research Methods, 13(3), 305-318. McGovern, Mark E., Aditi Krishna, victor M Aguayo, SV Subramanian (2017). “A review of the evidence linking child stunting to economic outcomes”. International Journal of Epidemiology, vol.46, no. 4, pp. 1171-1191. Narayan, Ambar; Van der Weide, Roy; Cojocaru, Alexandru; Lakner, Christoph; Redaelli, Silvia; Mahler, Daniel Gerszon; Ramasubbaiah, Rakesh Gupta N.; Thewissen, Stefan. (2018). “Fair Progress? Economic Mobility Across Generations Around the World”. Equity and Development. Washington, DC: World Bank World Bank. (2018). “World Bank East Asia and Pacific Economic Update, October 2018: Navigating Uncertainty”. Washington, DC: World Bank. World Bank. (2021a). Uneven Recovery: East Asia and Pacific Economic Update, April 2021. Washington DC: World Bank. World Bank. (2021b). A Year Deferred: Early Experiences and Lessons from COVID-19 in Vietnam. Washington DC: World Bank. World Bank. (2021c). Long COVID: East Asia and Pacific Economic Update, October 2021. Washington DC: World Bank. World Bank; UNICEF. (2021). Papua New Guinea High Frequency Phone Survey on COVID-19, December 2020 to January 2021. World Bank, Washington, DC. © World Bank. 28 Statistical appendix Appendix Table 1. Regression of work stoppages on mobility and respondent characteristics Note: The sample is restricted to those with non-missing education data. Relatively normal (higher) mobility indicates an average mobility index above -20 percent. More stringent (lower) mobility indicates an average mobility index equal to or below -20 percent. The omitted mobility category is “Higher mobility”. Vietnam is excluded from the sample for reasons stated in section 2. 29 Appendix Table 2. Regression of work stoppages on welfare quintiles controlling for respondent characteristics 30 Note: In all tables, “respondent” refers to the subject of the employment module. 31 Appendix Table 3. Regression of work stoppages on respondent’s* education. Note: The omitted category for education is less than high school. The omitted category for sector is agriculture. Results from an OLS regression are shown. A probit regression produces qualitatively similar results. “Respondent” refers to the subject of the employment module. The sample is restricted to those working pre-pandemic. The sector refers to the most recent sector in which the respondent worked. Appendix Table 4. Regression of labor income losses by residential area 32 Note: The reference period for looking at changes in income are since the previous round. The exception is the first round, for which the reference period is pre-pandemic. Results from an OLS regression are shown. The pooled sample excludes PNG, for which income data is not available. Appendix Table 5. Regression of labor income losses by source 33 Note: The reference period for looking at changes in income are since the previous round. The exception is the first round, for which the reference period is pre-pandemic. Results from an OLS regression are shown. The pooled sample excludes PNG, for which income data is not available. 34