Policy Research Working Paper 10166 Protect Incomes or Protect Jobs? The Role of Social Policies in Post-Pandemic Recovery Asli Demirgüç-Kunt Michael Lokshin Iván Torre Europe and Central Asia Region Office of the Chief Economist September 2022 Policy Research Working Paper 10166 Abstract This paper examines the effectiveness of income protection had a significant economic impact only in countries with and job protection policies for the post-pandemic economic weaker pre-pandemic social insurance systems. In countries recovery of the second half of 2020 through 2021. The with broader coverage of the social insurance system, the paper is based on a new data set of the budgets of social income and job protection programs appear to have had protection programs implemented as a part of the pan- a limited impact on post-pandemic recovery. Because the demic stimulus package in 154 countries. The empirical structural economic changes induced by the pandemic are analysis shows that, in the short run, higher expenditure expected to materialize fully in several years, more research on job protection measures is associated with more robust is needed to understand the longer-term effects of job pro- gross domestic product growth, increased employment, tection and income protection policies on labor markets and decreased inactivity and poverty rates compared to the and economic recovery. expansion of income protection programs. Both policies This paper is a product of the Office of the Chief Economist, Europe and Central Asia 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 ademirguckunt@gmail.com, mlokshin@worldbank.org, and itorre@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 Protect Incomes or Protect Jobs? The Role of Social Policies in Post-Pandemic Recovery Asli Demirgüç-Kunt, Michael Lokshin and Iván Torre* Declaration of Interest: none Keywords: pandemic, labor market policies, social protection, cash transfers, unemployment insurance, job retention JEL: O4, J65, I3 *Asli Demirgüç-Kunt is non-resident fellow at the Center for Global Development. This work was carried out while she was at the World Bank. Michael M. Lokshin and Iván Torre are at the World Bank. This paper’s findings, interpretations, and conclusions are entirely those of the authors and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. We thank Aylén Rodriguez Ferrari for her excellent support and Martin Ravallion for his constructive comments. 1. Introduction The recent global events have forced policy makers to reevaluate the trade-offs between income protection and job protection policies in terms of their impact on the post-pandemic economic recovery. During 2020, unemployment skyrocketed from 3.5 to 8.3 percent in the US and from 5.4. to 9.5 percent in Canada, but it increased only moderately from 5.2 percent to 6.3 percent in, for example, Scandinavian countries. Many European economies tried to preserve employment, keeping workers attached to their employers. In contrast, countries in North America focused on supporting the incomes of the affected populations. Policy choices adopted by governments to protect their population played the leading role in defining the trajectories of early economic recovery (e.g., Demirguc-Kunt et al. 2021). Job retention schemes have been used by many countries to mitigate the economic shock of COVID-19. These schemes consist of short-time work arrangements, furloughs, and/or wage subsidies funded by the government (Drahokoupil and Müller, 2021). During the first peak of the pandemic in mid-2020, governments of OECD countries supported about 50 million jobs. On average, high-income countries spent an equivalent to 1.5% of their GDP on wage subsidies. By reducing labor costs, job retention schemes allowed firms to adjust work hours, thereby reducing the number of jobs to be terminated and lowering the risk of unemployment. These schemes tend to provide stronger support to workers who are temporarily not working than do the unemployment benefits (OECD 2021a). The UK government subsidized the wages of 9.6 million employees – about a third of the UK workforce. The labor market regulation programs were present in 178 countries (Gentilini et al. 2022). The massive income-support measures deployed during the pandemic through stimulus packages incorporated cash transfer programs, expanded existing and introduced new social assistance programs. Cash transfers reached unprecedented levels of coverage, but unevenly so across countries and largely concentrated in the early phase of the pandemic (Gentilini, 2022). Cash- based policies implemented in 203 countries, and social pension programs established by 48 countries represented 44 percent of the world’s total social assistance measures (Gentilini et al. 2022). In most countries, the labor market policies introduced in response to the COVID-19 pandemic pursued two main goals: to limit social hardship caused by the pandemic and ensure a rapid and 2 sustainable post-pandemic economic recovery. The weights of the job protection and income protection measures in the policy response, in theory, should depend on the government’s perception of the nature of the crisis. The optimal response to transitory and exogenous shocks, such as natural disasters, should focus on job preservation by subsidizing businesses to maintain jobs and reduce workers’ welfare losses. Such policies support workers and ensure that firms can jumpstart their activities once the crisis is over. Among these policies are short-time work and temporary layoff schemes and administrative measures to limit workers’ dismissal (i.e., Giupponi and Landais 2018). But the impact of the current crisis might be more structural and permanent. The pandemic-induced changes forced businesses to develop new value chains that rely on digitization and automation processes, and many companies learned to operate with fewer workers (McKensey 2020). In developed countries, automation and digitalization are expected to increase demand for high-skill occupations. At the same time, remote work and reduced travel will likely suppress demand and lower wages in the less-skilled service sectors such as hospitality, food, and janitorial services (Ding and Molina 2020). The cost of capital, high degree of informality, and barriers to technology diffusion might depress wages and increase unemployment in developing countries and emerging economies. These structural changes will require significant reallocation of resources. Policies focusing on job protection may` hinder the movement of labor from unviable jobs to better- performing industries and slow the recovery (Barrero et al., 2020). Unemployment insurance systems complemented by comprehensive social assistance services and cash transfers could effectively support workers during the transition between jobs, ensuring rapid recovery. But these policies come at a cost. Weak employment protection and reliance on unemployment insurance schemes could lead to excessive dismissals. To reduce costs, businesses may lay off workers, putting extra pressure on unemployment insurance (Cahuc and Zylberberg 2008). In countries with low levels of wage replacement or low insurance eligibility and coverage, inadequate job protection could increase the risks of extreme poverty and adverse health outcomes (O’Campo et al. 2015). This paper attempts to assess which combinations of job protection and income protection policies lead to better economic outcomes. The analysis focuses on the short-term effects of different policy mixes on the post-pandemic recovery during 2020 – 2021. 3 We assembled a new database with information on budget allocations for every social protection program implemented in response to the pandemic in 155 countries. The database derives the universe of social protection programs from the Global Database on Social Protection Responses to COVID-19 (Gentilini et al. 2022), with a cut-off date for policies announced before June 1, 2021. Using these data, we estimate the relative contributions of income protection and job protection policies to the overall social protection response. Our empirical model estimates the relationship between four economic outcomes (GDP growth, unemployment, rates of labor inactivity, and poverty) and variables representing expenditures on the two types of social protection policies and country-level controls. Our analysis shows that, in the short run, higher expenditure on job protection is associated with a more robust economic recovery, increased employment, and decreased inactivity and poverty rates; expansion of income protection measures appears to have no significant economic effect. We conduct a series of robustness checks with empirical specifications that use alternative definitions of our dependent variables. These estimations qualitatively confirm our main results on the greater effectiveness of job protection policies in promoting economic recovery during the second half of 2020 and the first half of 2021. We then take our analysis further to account for the effects of pre-pandemic characteristics of social protection systems. We find that social protection response had a significant economic impact in countries with weaker pre-pandemic social insurance systems. In countries with broader social insurance coverage, income and job protection programs appear to have a limited impact on post-pandemic recovery. We speculate that some programs implemented during the pandemic - particularly those protecting jobs - may have been redundant in these countries as automatic stabilizers were already in place. This paper contributes to the emerging literature on the effectiveness of various types of social protection programs in facilitating post-pandemic recovery. To our knowledge, it offers one of the first empirical cross-country analyses of the short-term impact of job protection and income protection programs on economic growth, unemployment, labor inactivity, and poverty. The findings of this paper could be used to design more effective policies to address the main challenges in the post-pandemic world and find the right balance between job preservation and 4 reallocation of resources in the economy. We hope that our paper will motivate further research on the relationship between social protection systems and economic outcomes in the longer run. The paper is organized as follows. The next section describes the latest literature on the effect of employment protection and unemployment insurance policies on various labor market outcomes. Section 3 discusses the sources of data used in the analysis. Section 4 presents descriptive statistics on the nature of the social protection response to the pandemic and describes the short-term evolution of economic outcomes in 2020-21. Section 5 outlines our empirical strategy, followed by a presentation of the main results in Section 6. Section 7 summarizes the paper’s main findings. 2. Trade-offs between job protection and income protection: Recent literature Many studies have examined the separate effects of employment protection and unemployment insurance, but relatively few analyze the combined effect of these policies on welfare and labor market outcomes. Policies that protect incomes and policies that protect jobs are often contrasted in a debate about poverty reduction in low-income developing countries. Seen as rights-based, these policies guarantee employment (“the right to work”), meaning that anyone who wants a job at a predetermined wage rate can get it, or guarantee income (“the right to income”), meaning that everyone with an income below a certain threshold receives a cash transfer that brings their income to the threshold level. An extensive body of literature on the pros and cons of these two approaches concludes that “differentiated Universal Basic Income” might dominate both policies as a more transparent and less bureaucratically costly instrument of poverty reduction (i.e. Ravallion 2019). The early theoretical works on unemployment insurance included research by Baily (1978), Shanell and Weiss (1979), Hansen and Imrohoroglu (1992), and Hopenhayn and Nicolini (1997). More recent studies – for example, Acemoglu and Shimer (1999), Chetty (2006), Hassler and Rodriguez Mora (2008), Landais et al. (2010), and Boeri and Macis (2010) – further developed the theoretical frameworks and presented empirical evidence on the significant effects of unemployment benefits on gross job turnover, coming primarily from higher rates of job destruction, as well as on inter-industry reallocation. Studies on the effect of employment protection on sectoral reallocation and job destruction include Lazear (1990), Bertola and Rogerson (1997), Ljungqvist (2002), Bertola (2004), Rogerson et al. (2005), Piccirilli (2010), Bartelsman et al. (2016), and Karabay and McLaren (2011). This 5 literature concludes that strong employment protection negatively affects labor market performance and hinders technological innovation by limiting the size of high-risk, innovative sectors. The European policy debate attracted significant attention to comparing the performances of the European and US labor markets in the last decades. The rigidity of the European labor market, which is heavily reliant on employment protection policies, was often blamed for slowing down technological change and maintaining high unemployment rates. In response to these criticisms, Pissarides (2001) developed a theoretical model demonstrating that employment protection does not reduce job creation if chosen optimally. He shows that employment protection can insure against income risk when moral hazard prevents unemployment insurance from providing sufficient cover. Workers pay for employment protection during the productive phase of the job and continue receiving wages when the job is no longer productive. If court fees are high and enforcement of job contracts is costly, government legislation can provide a cheaper alternative to enforcing the contract between the employer and a worker than private contracts. In countries with inadequate employment insurance provision and strict employment protection measures, increasing the generosity of unemployment benefits could lead to faster destruction of unproductive jobs. Blanchard and Tirole (2008) emphasize the importance of analyzing the joint optimal design of unemployment insurance and employment protection programs. They show that under different assumptions, employment protection could be a substitute for unemployment insurance, but that such substitution comes at the cost of production efficiency. Making unemployment benefits conditional on search and acceptance of jobs could improve insurance, lower employment protection, and decrease production inefficiencies. Lommerud et al. (2018) extend the Blanchard and Tirole (2008) model to investigate the impact of “flexicurity” on investment in technology and job creation. Flexicurity refers to a situation in which unemployment benefits are combined with low employment protection, and the payroll tax is used to balance public budgets (i.e., Andersen 2012). Lommerud et al. (2018) show that low firing costs are bad for technology investment but good for job creation. Weaker job protection policies increase the number of dismissals and reduce investment in production-improving 6 technology. At the same time, lower dismissal costs decrease firms’ costs of retaining workers by diminishing their bargaining power. Anesi and De Donder (2013) study the political economy of combining employment protection and unemployment insurance policies. They show that the election process and the formation of political parties could influence the relative weights of the policies. In the direct citizen-candidate model, low-skilled workers are decisive and push for maximum employment protection and unemployment benefits. Allowing for the party formation results in policy equilibria in which high-skilled workers form a coalition with the unemployed and support a combination of policies with high unemployment benefits and lower job protection levels. Several recent papers analyze the impact of the COVID-19 pandemic on the labor market and its relationship to social policies. Eymeoud et al. (2021) contrast US and European labor market trends during the pandemic. They demonstrate that the EU countries avoided a sharp increase in unemployment in the early stages of the pandemic by funding short-time work programs. However, these programs may slow the reallocation of workers from shrinking to growing sectors of the economy during the recovery. Aidukaite et al. (2021) and Beland et al. (2021) study the labor market response to the COVID-19 crisis in Hungary, Lithuania, Poland, the Slovak Republic, and the OECD countries. Both papers find that labor market policies in these countries depended on pre-pandemic social policy choices and focused mostly on job protection measures. Finamor and Scott (2021) document a negative association between the generosity of unemployment insurance and employment in the US during the pandemic. Notably, the employment gap remained constant even after the expansion of benefits expired, indicating that fear of infection and childcare options could be important factors explaining the persistent rates of labor inactivity in the US. 3. Data The analysis in this paper relies on a new country-level data set of expenditure on social protection measures implemented in response to the COVID-19 pandemic put together by the authors. Our database uses information from the Global Database on Social Protection Responses to COVID- 19 (GDSPRC) (Gentilini et al. 2022). The GDSPRC contains information on the duration, target population, and other technical aspects of these measures. We collected detailed budget 7 information for each social protection program listed in the GDSPRC that was announced before June 1, 2021 from official documents (including IMF Article IV revisions and related documents from other international organizations), government websites, and news sources. We quantified the expenditures of social protection programs in 154 countries. This set of countries forms the core sample for the empirical analysis in this paper (see Appendix Table A.1 for the list of countries). We also used several auxiliary data sources. Information on countries’ pre-pandemic social protection landscape comes from the ASPIRE database (World Bank 2021a) and OECD’s Social Spending database (OECD 2021b). Data on employment, inactivity, and growth outcomes comes from ILO (2021), the IMF (2021), and the World Bank (World Bank 2021b, 2022b). Poverty projections are from Lakner et al. (2020). Information on the size of the informal sector is from the Informal Economy Database produced by the World Bank (Elgin et al. 2021). An essential feature of our analysis is the distinction between social programs focused on income protection and programs focused on job protection. Table 1 shows a correspondence between social protection areas, categories of social protection, as defined in GDSPRC, and the policy focus (on either income protection or job protection). All categories included in the social assistance policy areas focus on income protection; the categories of social insurance and labor market policy areas are split between those focusing on income protection and those focusing on job protection. 4. The nature of the social protection response and the impact of the pandemic Countries across the world implemented substantial economic measures in response to the economic shock of the pandemic. The size of the economic mitigation measures (also called the “stimulus budget”) averaged about 5.6 percent of GDP (Table 2). In high-income countries, the average stimulus package reached 10 percent of GDP; in lower-middle and low-income countries, the average size of the stimulus package was about 3 percent of GDP. Pandemic stimulus packages also included infrastructure spending and general business support measures. The share of these non-social protection policies in the total stimulus budget was highest at the extremes of the world income distribution (Figure 1). Low-income countries allocated about 35 percent of their total mitigation budgets to health spending. Health spending represented only 13 percent of the total size of the mitigation budget in high-income countries. 8 Social protection policies were an essential part of stimulus packages. Globally, the average size of the social protection response budget was 2.0 percent of countries’ GDP. Like the overall stimulus budget, the size of the social protection budget varied widely across countries at every income level (Figure 2). On average, high-income countries allocated almost 3.5 percent of their GDP to social protection response, with several countries exceeding 5 percent of GDP. Upper- middle-income countries’ social protection response budget averaged 1.9 percent of GDP, almost half of their overall stimulus package. Lower-middle-income countries’ budget was 1.0 percent of GDP. Among low-income countries, the social protection response budget was just 0.8 percent of GDP – less than a fourth of these countries’ overall stimulus budget. Countries’ stimulus packages differed not only by size but also by composition. Low-income countries allocated almost all of their social protection response budget to income protection measures. On average, these countries devoted 0.4 percent of GDP (about half of their social protection response budget) to unconditional cash transfers, followed by in-kind and food transfers for about 0.3 percent of GDP and 0.1 percent of GDP on waivers of utility bills. Job protection measures, such as wage subsidies and waivers of social insurance contributions, were practically absent in low-income countries. Lower-middle income countries had a policy mix similar to that of low-income countries, with 0.9 percent of GDP going to income protection measures out of a total social protection response budget of almost 1 percent of GDP. Upper-middle-income countries spent slightly less than a third of their total social protection response budget (or about 0.6 percent of GDP) on job protection measures. The higher a country’s income, the more it spent on job protection measures (Figure 3). In high-income countries, job protection measures were most prevalent, amounting to about 1.9 percent of GDP, or more than half of their social protection response budget. About 1.4 percent of GDP (or 40 percent of the SP response budget) was allocated to wage subsidies, with some countries exceeding 2 percent of GDP. Waivers of social insurance contributions amounted to about 0.6 percent of GDP in high-income countries. With respect to income protection measures, which amounted to about 1.6 percent of GDP in this country income group, the largest were unconditional cash transfers (0.7 percent of GDP), unemployment support (0.4 percent of GDP), and utility and financial waivers (0.3 percent of GDP). 9 A clear pattern emerges from this analysis: low-income and lower-middle income countries devoted most of their (small) social protection budgets to measures preserving the income of their citizens, through either direct cash transfers or the provision of in-kind and food transfers. In contrast, higher-income countries aimed their (larger) social protection budgets at preserving jobs by directly subsidizing wages and reducing the social insurance costs to employers. Upper-middle- income countries combined both policy approaches -preserving income and preserving jobs- with unconditional cash transfers taking the forefront. The economic impact of the pandemic was substantial: In every country in the world, except for Burundi and Mongolia, including the minority of countries that recorded positive growth in GDP in that year, employment declined in 2020 (Figure 4). The correlation between the drop in employment and the decline in GDP is relatively weak, however, countries that faced similar declines in GDP reported very different changes in employment. Among countries in which GDP dropped by -4.5 – -3.5 percent (about the median change of -3.8 percent), the decline in employment rate ranged from -6.8 to -0.4 percentage points. Differences in structural characteristics of the economy –such as labor intensity across and within sectors affected by the pandemic shock– may explain part of this variance. But differences in social and labor market policy responses to the crisis may also be driving the relationship between change in employment and change in GDP (Eymeoud et al. 2021). Simple correlation analysis shows that the policy focus of the social protection response -leaning towards either job protection or income protection measures- affects the socio-economic outcomes during the first year of the pandemic (Figure 5). Countries that spent a larger share of their social protection response budget on job protection measures (as opposed to income protection) observed, at least in the short term, smaller decreases in employment (panel b) and smaller increases in inactivity and poverty (panels c and d), while there was no clear pattern for changes in GDP (panel a). 5. Theoretical priors and empirical specification In this paper, we analyze how the allocation of the pandemic stimulus budget between income and job protection measures affects socio-economic outcomes during the pandemic and early post- pandemic period. The effects of these two types of policies differ both in the short and the long run. 10 By allowing firms to jumpstart their operations immediately after pandemic restrictions are lifted, the job protection measures could promote faster short-term recovery. However, this might come at the cost of indirectly subsidizing inefficient businesses, reducing the rate of creative destruction, and ultimately hindering long-term recovery (Barrero et al. 2020). 1 The moral hazard of generous unemployment benefits and cash transfer programs could slow the recovery in the short run but facilitate the reallocation of workers to the most productive sectors of the economy, securing sustainable long-term recovery. Income and job protection measures affect the behavior of workers and firms and induce fiscal externalities. The combination of these effects influences the equilibrium in the labor market and the pace of economic recovery. Social protection systems could generate inefficiencies in the labor market because of the heterogeneity of economic shocks across sectors and industries. Inadequate job protection policies and/or overly generous unemployment insurance may result in excessive firing and lower rates of productive job matches. By reducing the incentives to search for more productive matches, job protection programs might delay the efficient reallocation of workers (Jager et al. 2019). This effect could be especially pronounced if the shock is permanent. When only a few jobs are available, incentivizing workers to search less might be welfare-enhancing (Michaillat and Saez 2019). At the same time, both types of programs have direct welfare-improving effects for beneficiaries by smoothing their consumption and thus reducing poverty. Unemployment insurance, cash, and in-kind transfers help households mitigate the impact of negative income shocks. Such measures could influence economic growth through the fiscal multiplier effect of increased consumption (McKay and Reis 2016). Job protection programs also have a direct consumption-smoothing effect on the well-being of workers who would have been laid off without such programs. In that sense, they may play a dual role of preserving productive job matches and protecting workers’ incomes. However, while unemployment insurance and other transfers primarily protect vulnerable groups (such as youth and individuals with low levels of education), employment protection programs tend to protect mostly insiders and better-educated workers (Cahuc and Carcillo 2011). 1 Better job security might moderate workers’ resistance to introducing new technologies and work practices (Akerlof 1984). 11 In countries with large informal sectors, income protection programs could be the dominant mode of social protection as they can reach a broader share of the population, particularly vulnerable people (Bottan et al. 2021). In contrast, job protection measures may be effective in countries with primarily formal economies. The generosity of unemployment insurance, direct cash transfers, and job protection policies also affect labor market tightness, unemployment, and job participation rates. However, the effects of increases in the generosity of unemployment insurance on the labor market are much stronger than the effects of job protection policies (Giupponi and Landais 2020). Based on these theoretical considerations and empirical evidence, we expect that by mid-2021, the countries that gave higher priority to job protection policies in their pandemic mitigation portfolio would have had lower unemployment and job inactivity levels than countries that expanded and increased the generosity of their unemployment insurance policies and/or implemented large cash or in-kind transfers (e.g., Schwellnus et al. 2020). We also expect that, on average, job protection measures would be more effective in promoting short-term economic recovery, as firms can restart their activities as soon as nonpharmaceutical interventions are lifted. The poverty rate should respond more to generous unemployment insurance and cash transfers, as these measures directly support household consumption. The job protection measures should also reduce poverty, albeit to a lesser degree. We analyze the impact of different types of social policies on four economic outcomes. To assess the pace of economic recovery, we use the difference in GDP levels between the 2021 World Bank estimate 2 and the pre-pandemic forecast for the same year. 3 This indicator quantifies how close economic activity is to its pre-pandemic level. It can be interpreted as a measure of the strength of the economic recovery. We use the difference in the pre-pandemic (January 2020 forecast) and post-pandemic (2020, actual) poverty headcount rates in a country to evaluate the impact of social policies on poverty. 4 We use differences in the employment and inactivity rates between 2019 and 2020 to assess the impact of the various allocations of social policies on labor market outcomes. 2 Values for the 2021 GDP are based on the estimates by the World Bank Global Economic Prospects of January 2022 (World Bank 2022). For countries missing in that publication, the estimates of the IMF World Economic Outlook of October 2021 (IMF 2021) are used. 3 The pre-pandemic forecast corresponds to the values published by the World Bank Global Economic Prospects of January 2020 (World Bank 2020a). For countries missing in that publication, the values of IMF World Economic Outlook of October 2019 (IMF 2019) are used. 4 We use the poverty headcount rates corresponding to the $5.5 (2017 PPP) per day international poverty line as projected by Lakner et al. (2020). 12 Unlike GDP and poverty, the measures of labor market outcomes are expressed in relation to their pre-pandemic values. In this sense, the measures used for labor market outcomes may include some variation caused by pre-pandemic trends. Estimation results based on these measures may therefore have to be interpreted with caution. We use several controls in estimating our empirical model. As we show in the previous section, the pre-pandemic country’s per capita GDP affects the size of the stimulus budget and the relative allocation of funds to job protection and income protection policy measures. Pre-pandemic GDP can also affect the speed of recovery and other outcomes of interest. We control for the share of services in pre-pandemic GDP, and the size of the informal sector as these sectors suffered disproportionately from the pandemic (OECD 2020, World Bank 2020b). In addition, the degree of informality may affect the effectiveness of job protection and income protection policies in mitigating the impact of the pandemic. The descriptive statistics of the variables used in our empirical analysis are shown in Appendix Table A.2. In our main estimations, we regress the four economic outcomes on variables representing expenditures on the two types of social protection policies and country-level controls: + 3 = + 1 + 2 (1) + 1 + 2 + 3 + where DYc is one of the four indicators described above for country c. IncomeProtectionGDPc is the expenditure on income protection measures, JobProtectionGDPc is expenditure on job protection policies, and NonSPResponseGDPc is the stimulus budget spent on measures other than social protection in country c, all expressed in percentage of 2019 GDP. LogGDPpcc is the log of 2019 per capita GDP in PPP dollars. ServicesGDPc represents the percentage of the services sector in country c’s GDP in 2019. Informalityc is the share of informal output in GDP in the last available pre-pandemic year, expressed in percentage points. 6. Main results Table 3 summarizes the main results of estimating equation (1). Expenditure on income protection measures appears to have no significant correlation with any of the four economic outcomes. Expenditure on job protection measures seems to be associated with a more robust short-term recovery, an increase in employment, a decrease in the inactivity rate, and a decrease in the poverty 13 headcount rate. These correlations are in line with our theoretical priors, although the association with the poverty rate is stronger than expected, given that no statistically significant correlation is found between poverty and income protection measures. 5 The magnitude of the elasticities between expenditure on job protection and the four economic outcomes is substantial. An increase in job protection expenditures of 1 percentage point of GDP is associated with an increase in GDP of about the same magnitude in 2021 (Table 3, column 2); an increase in employment of about 0.4 percent of the working-age population during 2020 (column 4) and an equivalent decrease in the inactivity rate in the same year (column 6). These results suggest that preserving employment during 2020 by subsidizing firms’ wage bills brought levels of economic activity during 2021 closer to the pre-pandemic trend. The longer-term effects are unclear, as evidence from business enterprise surveys suggests that job protection measures were associated with less labor reallocation from low productivity firms to high productivity firms. This finding would imply a loss of efficiency in the economy in the longer run (Demirgüç-Kunt et al. 2022). The share of the population with incomes under the $5.50 a day international poverty line is about 0.4 percentage points lower for every percentage point of GDP spent on job protection measures (Table 3, column 8). From a cross-country perspective, this result suggests that preserving labor earnings may have a stronger effect on short-term poverty reduction than direct income protection measures. However, income protection programs appear to be more effective in preventing extreme poverty in low-income settings during the pandemic (Bottan et al., 2021). To assess the robustness of these results, we estimate a series of alternative specifications of equation (1), shown in Table 4. In specification I, all dependent variables are expressed as the difference between the values observed in 2020 and 2019. In specification II, the GDP and poverty headcount rate (at the $5.50 a day poverty line) are expressed as the difference between the value observed in 2020 and the pre-pandemic projection for the same year. In specification III, the GDP and poverty headcount rate ($5.50 a day) are expressed as the difference between the value observed in 2021 and the pre-pandemic projection for the same year. In specification IV, the 5 We replicated all the results presented in this paper by including the total number of COVID-19-related deaths per million of population in specification (1). This variable is insignificant is all estimations and the inclusion of it produces no meaningful changes in cooficinets compared to our baseline specification. 14 poverty headcount rate is expressed as the difference between the value observed in 2020 and the value observed in 2019, using the international poverty line corresponding to a country’s income group ($1.90 a day for low-income countries, $3.20 a day for lower-middle income countries and $5.50 a day for upper-middle income countries) and the national poverty line for high-income countries. In all four specifications, the main correlations hold in qualitative terms. Specifications V-VIII include, as additional regressors, the coverage of the social protection system before the pandemic – distinguishing between coverage of the social insurance subsystem (specification V, expressing all dependent variables as a difference between 2020 and 2019, and specification VI, where GDP and poverty headcount rate are expressed as the difference between the value observed in 2020 and the pre-pandemic projection for the same year) and coverage of the social assistance subsystem (specification VII, expressing all dependent variables as the difference between 2020 and 2019, and specification VIII, where GDP and poverty headcount rate are expressed as the difference between the value observed in 2020 and the pre-pandemic projection for the same year). These regressors control for a direct effect of the pre-existing social protection systems on top of which the social protection responses to the pandemic were implemented. The results of estimating specifications V-VIII show that the main correlations between the four economic outcomes and income and job protection expenditures during the pandemic are qualitatively similar to the baseline estimates after accounting for the direct effect of characteristics of the pre-existing social protection system. However, the magnitudes of the effects on employment and inactivity rates are slightly smaller and less significant. These results lead us to perform further exploratory analysis of the relationship between the pre-pandemic characteristics of the social protection systems and the performance of the social protection response to the pandemic, which we detail in the next section. 6.1 The role of preexisting social protection systems The pandemic social protection response relied partly on the pre-existing system of social protection: about 30 percent of the social protection programs implemented during the pandemic were an adaptation or a new benefit of a pre-existing program (Gentilini et al. 2022). New programs may also have benefited from the existing social protection infrastructure. In this sense, 15 the economic effects of the social protection measures could be a function of the pre-existing social protection systems (e.g., Aidukaite et al. (2021), Beland et al. (2021)). To explore this possibility, we focus on a characteristic of social protection systems before the pandemic for which there is comparable cross-country information: the coverage of social insurance and social assistance programs, defined as the share of a country’s population covered by either type of program. Other relevant characteristics of the pre-existing social protection systems are coverage within the eligible population, or the average value of the benefits received. Our data are insufficient for conclusive cross-country comparisons of these characteristics and their impact on the economic effects of the pandemic response. 6 In Table 5, we present the results of estimations in which expenditure on income and job protection is interacted with a dummy variable indicating whether a country’s pre-pandemic coverage of social insurance or social assistance is above the sample median. Columns 1 and 2 show that the effect of income protection expenditure is not statistically significant for either the countries with high or low social protection (social insurance or social assistance) coverage. 7 In the case of job protection expenditure, the effect is not statistically significant for countries with low social protection coverage, and it is only significant at the 10% level for countries with high social protection coverage. 8 This finding suggests that the baseline result of a positive and statistically significant association of job protection measures with economic recovery could be attributed to the effects of pre-pandemic characteristics of social protection systems. The effects of income and job protection expenditure on employment and inactivity during the pandemic depend on the pre-pandemic coverage of social insurance programs. The results in columns 3 and 5 of Table 5 show that in countries with low coverage of social insurance programs, expenditure on income protection measures is correlated with a decrease in employment and an increase in inactivity. Expenditure on job protection measures is positively correlated with increased employment and shows no correlation with inactivity rates. 6 The sample size of our cross-country regressions will drop to about 85 observations if we use countries for which data on all relevant characteristics of the social protection systems are available. 7 The main coefficient is not statistically significant in itself and it is also not statistically significant jointly with the interaction coefficients. 8 The coefficients on job protection expenditure and its interaction with either high social insurance coverage or high social assistance coverage are jointly significant only at the 10% level. The sample size of this analysis is reduced compared to the baseline analysis, and therefore the results should be interpreted with caution. 16 In countries with higher coverage of social insurance programs, the effects of both types of social protection expenditure on employment and inactivity are much smaller, as shown by the statistically significant interaction effect. 9 These results suggest that the social protection response during the pandemic may have had stronger effects on labor markets in countries with low pre- pandemic social insurance coverage. This finding is not surprising given the objective of social insurance systems – to insure individuals and households against shocks and allow them to smooth consumption over time. To the extent that the programs implemented during the pandemic play a similar role as social insurance systems in terms of protecting people against economic shocks, it is expected that the effects of these programs are weaker in countries in which an advanced social insurance system was already in place. In columns 4 and 6 of Table 5, income and job protection expenditures are interacted with the pre- pandemic coverage of social assistance systems. Unlike in the case of social insurance, there is no statistically significant effect of income protection expenditure on employment and inactivity once social assistance coverage is included as an interaction factor. In the case of job protection expenditure, the joint significance test results show that the positive effect on employment and the negative effect on inactivity is statistically significant at the 5% level only in countries with high social assistance coverage. This suggests that, differently from social insurance, social assistance coverage may actually enhance or complement the effects of job protection policies – although more evidence is needed to assert this. Columns 7 and 8 of Table 5 show the effects of the social protection response interacted with pre- pandemic characteristics of social protection systems on poverty rates. In line with the results on employment and inactivity, pre-pandemic coverage of social insurance programs appears to be a significant factor determining the economic effectiveness of social protection responses (column 7). A puzzling finding is that income protection expenditure appears to be associated with a higher poverty headcount rate in countries with low social insurance coverage. This finding could indicate reverse causality, as countries with low social insurance coverage may have been more vulnerable to an increase in poverty, to begin with – precisely because they lacked automatic stabilizer 9 The coefficients of job protection expenditure and its interaction with high social insurance coverage are not different from zero at the 10% significance level, suggesting that the effect of job protection on employment and inactivity in countries with high social insurance coverage could actually be zero. 17 mechanisms – and may, therefore, have implemented larger income protection programs in response. 10 Job protection expenditures have a stronger, positive impact on poverty rates in countries with low social insurance coverage. These programs have a much smaller effect in countries with more extensive social insurance systems before the pandemic. The results in column 8, where the two types of social protection expenditure are interacted with pre-pandemic social assistance coverage, suggest that, just like in the case of the effects on employment and inactivity, social assistance may actually enhance or complement the effect of job protection programs. 6.2 Addressing the potential bias in cross-country correlations The interpretations of our results might raise concerns about the potential influence of omitted variables. Three variables could be endogenous in specification (1) – the levels of expenditure on income protection programs (IncomeProtectionGDP), on job protection programs (JobsProtectionGDP), and on non-social protection programs (NonSPResponseGDP). To address this potential omitted variable bias, the standard instrumental variable (IV) approach requires using three instruments for these endogenous variables. However, such macro-level expenditure variables are highly inter-correlated, and finding exogenous instruments for each of them appears to be virtually impossible. We revert to an alternative approach and formulate a different specification (2) where we regress our variables of interest on the share of the social protection response budget allocated to job protection measures (JobsProtectionShare): = + 1 ℎ + 1 + 2 (2) + 3 + The сcoefficient β1 then captures the effect of the focus of the social protection response on different socio-economic outcomes. This partial correlation coefficient is more appropriately interpreted in terms of its sign and statistical significance rather than of its magnitude. To instrument the share of the social protection response budget allocated to job protection policies in equation (2), we use the number of months elapsed in March 2020 (at the onset of the COVID- 10 This result does not hold when a different poverty line is used according to each country’s income group (see footnote 4). 18 19 pandemic) since the last national elections. Our exclusion restriction relies on the assumption that the share of social protection response spending allocated to job protection is correlated with a country’s electoral cycle, but the economic effects of the pandemic (our dependent variables) are not correlated with the electoral cycle. There is substantial evidence that social spending follows an electoral cycle (Nordhaus, 1975): in particular, governments change the composition of spending during electoral years (Brender and Drazen, 2013; Drazen and Eslava, 2010; Han, 2021). Governments in countries that had elections just before the start of the pandemic may have implemented social protection policies differently from the governments that were up for reelection in the second part of 2020 or 2021. 11 Because the pandemic altered the election schedules in many countries, the variable we use to capture the electoral cycle is the number of months elapsed since the last election (before the pandemic). Table 6, column 1, presents the result of the first stage of instrumenting the share of the social protection response budget allocated to job protection measures by the number of months since the last election. We exclude countries with suppressed political competition or with interim non- democratic governments. 12 The number of months since the last election is highly significant and negatively correlated with the share of social protection spending allocated to job protection, indicating that the electoral cycle influences the composition of social protection response packages. The results of the second stage of the IV estimation (presented in columns 3, 5, 7, and 9 of table 6) are qualitatively equivalent to our main results of estimation of equation (1) – a strong country’s focus on job protection policies is associated with a higher GDP in 2021, a higher employment rate, a lower inactivity rate, and a lower poverty headcount rate. While the sign of the effects is consistent with our main results, the coefficients on the share allocated to job protection are not statistically significant. The associated F-statistic has a value corresponding to a size distortion of 11 While we are a priori agnostic on the direction of the effect, evidence from Latin America suggests that governments expand cash transfers in the run-up to elections (Bueno, 2021), therefore implying in our case that income protection policies would be a priori more present in countries where elections were scheduled after March 2020 than in countries that had a recent election before that date. 12 We exclude countries where political competition is suppressed according to the Polity5 project (a value of 1 in the scale 1-10 of the POLCOMP variable) and also countries with interim non-democratic governments in March 2020 (Bolivia, Sudan) or that have little effective control of their territories (Central African Republic). Singapore and Tajikistan are also excluded as political competition in these countries is virtually suppressed as well despite having a value higher than 1 in POLCOMP. 19 above 20% - implying that our instrument barely passes the weak instrument test (Stock and Yogo 2005). Therefore, the second stage results of the IV analysis need to be treated with caution. Indeed, there are limits to how far cross-country comparisons, with limited degrees of freedom, can be considered conclusive about the effects of social protection policies on economic outcomes in a causal inference setting. 7. Conclusions In response to the COVID-19 pandemic, countries across the world implemented an array of social protection measures, some of them aiming at protecting the income and livelihoods of families, others aiming at protecting employment through job retention schemes. To assess the short-term impacts of these policies on post-COVID-19 economic recovery, we put together a database on the social protection response expenditure during the pandemic. It reveals that, on average, countries spent about 2 percent of GDP on social protection measures, although high-income countries spent almost five times more than low-income countries. Expenditure on income protection measures was about 1.2 percent of GDP, with less variation across country income groups than expenditure on job protection. That spending averaged 0.8 percent of GDP but ranged from almost zero in low-income countries to 1.9 percent of GDP in high-income countries. Our analysis shows that countries that allocated larger shares of their stimulus budget to job protection measures seem to experience more robust GDP growth, higher employment, and lower inactivity and poverty rates than countries that allocated a larger share of the stimulus package to income protection measures. These results could be partly explained by the pre-pandemic coverage of social insurance programs. In countries with broader coverage, the income and job protection programs implemented during the pandemic had a smaller economic effect. The social protection response had a significant effect in countries with limited pre-pandemic social insurance coverage. In countries with advanced social protection systems, some social protection programs implemented during the pandemic may have been redundant as automatic stabilizers were already in place. In addition, job protection programs may have limited labor reallocation from low productivity to high productivity firms during the post-pandemic recovery. While the analysis in this paper focuses mainly on the “instrumental” value of social protection, that is, on the role of social protection for economic recovery and growth, we recognize the 20 principal “intrinsic” value of social protection in direct transfers to the welfare of recipients. In that, the income protection measures play the most crucial role. These findings underscore the importance of investing in the development of social protection systems during periods of economic growth to be able to protect households and firms during economic downturns without implementing additional measures. The effects of structural economic changes induced by the pandemic (the new value chains, production processes, and labor institutions) are expected to fully materialize in five to eight years. More research is therefore needed to understand the longer-term effects of various types of social policies on labor markets and economic recovery. 21 8. References Acemoglu, D. and R. Shimer (1999). “Efficient Unemployment Insurance,” Journal of Political Economy 107: 893–928. Aidukaite, J, Saxonberg, S, Szelewa, D, and D. Szikra (2021). “Social policy in the face of a global pandemic: Policy responses to the COVID-19 crisis in Central and Eastern Europe.” Social Policy Administration 55: 358– 373. Akerlof, G. (1984). An economic theorist’s book of tales: Essays that entertain the consequences of new assumptions. New York, NY, Cambridge University Press. Andersen, T. (2012). “A Flexicurity Labour Market in the Great Recession: the Case of Denmark.” De Economist 160: 117–140. Anesi, V. and P. De Donder (2013). “A Coalitional Theory of Unemployment Insurance and Employment Protection.” Economic Theory 52: 941–977. Baily, M. (1978). “Some Aspects of Optimal Unemployment Insurance,” Journal of Public Economics 10: 379–402. Ball, L., Leigh, D. and P. Loungani (2013). “Okun’s Law: Fit at Fifty?” Working paper No. w18668. National Bureau of Economic Research, Cambridge, MA. Bartelsman, E., Gautier, P. and J. De Wind (2016). “Employment Protection, Technology Choice, and Worker Allocation.” International Economic Review 57: 787-826. Barrero, J., Bloom N. and S. Davis (2020). “COVID-19 Is Also a Reallocation Shock,” National Bureau of Economic Research, Cambridge, MA, http://dx.doi.org/10.3386/w27137. Béland, D., Cantillon, B., Hick, R. and A. Moreira (2021). “Social policy in the face of a global pandemic: Policy responses to the COVID-19 crisis.” Social Policy Administration 55: 249– 260. Bertola, G. and R. Rogerson (1997). “Institutions and labor reallocation,” European Economic Review 41(6): 1147-1171. Bertola, G. (2004). “A Pure Theory of Job Security and Labour Income Risk,” Review of Economic Studies 71(1): 43-61. Blanchard, O. and J. Tirole, (2008). “The Joint Design of Unemployment Insurance and Employment Protection: a First Pass.” Journal of the European Economic Association 6:45–77. Boeri, T. and M. Macis (2010). “Do Unemployment Benefits Promote or Hinder Job Reallocation?” Journal of Development Economics 93: 109–125. Bottan, N., Hoffmann, B. and D. Vera-Cossio (2021). “Stepping up during a crisis: The unintended effects of a noncontributory pension program during the Covid-19 pandemic.” Journal of Development Economics 150 Cahuc, P. and S. Carcillo (2011). “Is Short-Time Work a Good Method to Keep Unemployment Down?” Nordic Economic Policy Review, 1 (1): 133–165. Cahuc P. and A. Zylberberg (2008). “Optimum income taxation and layoff taxes.” Journal of Public Economics 92 (10-11): 2003-2019. 22 Cazes, S., Verick, S., and F. Al-Hussami (2012). “Diverging trends in unemployment in the United States and Europe: Evidence from Okun’s law and the global financial crisis.” Employment Working Papers, ILO. Chetty, R. (2006). “A General Formula for the Optimal Level of Social Insurance,” Journal of Public Economics 90: 1879–1901. Demirgüç-Kunt, A., Lokshin, M. and I. Torre (2021). “Opening-up Trajectories and Economic Recovery: Lessons after the First Wave of the COVID-19 Pandemic.” CESifo Economic Studies 67(3), 332-369. Ding, L. and J. Molina (2020). “Forced Automation’ by COVID-19? Early Trends from Current Population Survey Data.” Federal Reserve Bank of Philadelphia. https://philadelphiafed.org/-/media/community-development/publications/discussion- papers/discussion-paper_automation.pdf Drahokoupil, J. and T. Müller (2021). “Job retention schemes in Europe: A lifeline during the Covid-19 pandemic.” ETUI Working Paper 2021.07. Available at https://www.etui.org/publications/job-retention-schemes-europe Elgin, C., M. A. Kose, F. Ohnsorge, and S. Yu (2021). “Understanding Informality.” CERP Discussion Paper 16497, Centre for Economic Policy Research, London. Eyméoud, J., Petrosky-Nadeau, N., Santaeulàlia-Llopis, R. and E. Wasmer (2021). “Contrasting US and European Job Markets during COVID-19.” FRBSF Economic Letter, 2021-05 Finamor, L., and D. Scott (2021). “Labor market trends and unemployment insurance generosity during the pandemic.” Economics Letters, 199. Gentilini, U. (2022). Cash Transfers in Pandemic Times : Evidence, Practices, and Implications from the Largest Scale Up in History. Washington, DC : World Bank Gentilini, U., Almenfi, M., TMM Iyengar, H., Okamura, Y., Downes, J. A., Dale, P., Weber, M., Newhouse, D., Rodriguez Alas, C., Kamran, M., Mujica, I., Fontenez, M., Ezzat, M., Asieduah, S., Martinez, V., Hartley, G., Demarco, G., Abels, M., Zafar, U., Urteaga, E., Valleriani, G., Muhindo, J., and S. Aziz (2022). “Social Protection and Jobs Responses to COVID-19: A Real-Time Review of Country Measures”. World Bank, Version 16 (February 2). Washington DC Giupponi, G. and C. Landais (2018). “Subsidizing Labor Hoarding in Recessions: The Employment & Welfare Effects of Short Time Work.” CEPR Discussion Paper No. DP13310, Available at SSRN: https://ssrn.com/abstract=3287057 Giupponi, G. and C. Landais (2020). “Subsidizing Labor Hoarding in Recessions: The Employment and Welfare Effects of Short TimeWork,” Discussion Paper 13310, CEPR. Hansen, G. and A. Imrohoroglu (1992). “The Role of Unemployment Insurance in an Economy with Liquidity Constraints and Moral Hazard,” Journal of Political Economy, 100(1): pages 118-142. Hassler, J. and J. Rodriguez Mora (2008). “Unemployment Insurance Design: Inducing Moving and Retraining.” European Economic Review 52: 757–779. Hopenhayn, H. and J. Nicolini (1997). “Optimal Unemployment Insurance.” Journal of Political Economy, 105(2): 412–438. 23 International Labour Organization. (2021). ILOSTAT database [database]. Available from https://ilostat.ilo.org/data/. IMF (2019). World Economic Outlook October 2019: Global Manufacturing Downturn, Rising Trade Barriers. Washington, DC: IMF IMF (2021). World Economic Outlook October 2021: Recovery During a Pandemic. Washington, DC: IMF Jäger, S., Schoefer, B. and J. Zweimüller (2019). “Marginal Jobs and Job Surplus: A Test of the Efficiency of Separations,” NBER Working Paper 25492, National Bureau of Economic Research. Karabay, B. and J. McLaren (2011). “Pareto-Improving Firing Costs?” European Economic Review 55: 1083–1093. Knotek, E. (2007). How useful is Okun’s law? Federal Reserve Bank of Kansas City, Kansas. Lakner, C., Mahler, D. G., Negre, M., and E. Beer Prydz (2020) “How Much Does Reducing Inequality Matter for Global Poverty?” Global Poverty Monitoring Technical Note 13, World Bank. Landais, C., Michaillat, P., and E. Saez (2010). “Optimal Unemployment Insurance over the Business Cycle.” NBER Working Paper No. 16526. Lazear, E. (1990). “Job Security Provisions and Employment.” Quarterly Journal of Economics 105: 699–726. Ljungqvist, L. (2002). “How Do Layoff Costs Affect Employment?” The Economic Journal 112(482): 829 – 853. Lommerud, K., Straume, O. and S. Vagstad (2018). “Employment Protection and Unemployment Benefits: On Technology Adoption and Job Creation in a Matching Model.” Scandinavian Journal of Economics 120(3): 763–793. McKay, A. and R. Reis (2016). “The Role of Automatic Stabilizers in the US Business Cycle,” Econometrica, 84 (1): 141–194. McKinsey (2020). The Next Normal: The recovery will be digital. McKinsey Global Publishing. Michaillat, P. and E. Saez (2019). “Optimal Public Expenditure with Inefficient Unemployment,” The Review of Economic Studies, 86 (3): 1301–1331. O’Campo, P., Molnar, A., Ng, E., Renahy, E., Mitchell, C., Shankardass, K., St. John, A., Bambra, C. and C. Muntaner (2015). “Social welfare matters: A realist review of when, how, and why unemployment insurance impacts poverty and health.” Social Science & Medicine, 132: 88-94 OECD (2020). “Corporate sector vulnerabilities during the Covid-19 outbreak: assessment and policy responses.” OECD Policy Responses to Coronavirus (COVID-19). Available at http://www.oecd.org/coronavirus/policy-responses/corporate-sector-vulnerabilities- during-the-covid-19-outbreak-a6e670ea/. OECD (2021a). “Job retention schemes during the COVID-19 lockdown and beyond” OECD Policy Responses to Coronavirus (COVID-19). OECD (2021b), Social spending (database). doi: 10.1787/7497563b-e 24 Okun, A. (1962). “Potential GNP: Its Measurement and Significance.” Reprinted as Cowles Foundation Paper 190. Pissarides, C. (2001). “Employment Protection,” Labour Economics, 8: 131–159. Ravallion, M. (2019). “Guaranteed employment or guaranteed income?” World Development, 115: 209-221. Rogerson, R., Shimer, R. and R. Wright (2005). “Search-Theoretic Models of the Labor Market: A Survey.” Journal of Economic Literature, 43 (4): 959-988. Shavell, S. and L. Weiss (1979). “The Optimal Payment of Unemployment Insurance Benefits over Time.” Journal of Political Economy 87: 1347–1362. Schwellnus, C., Koelle, M. and B. Stadler (2020). “Flattening the unemployment curve? Policies to support workers’ income and promote a speedy labour market recovery.” Policy note, OECD Economics Department. Stock, J. and M. Yogo (2005). “Testing for Weak Instruments in Linear IV Regression,” Chapter 5 in Andrews DWK Identification and Inference for Econometric Models. New York: Cambridge University Press: 80-108. World Bank (2020a). Global Economic Prospects, January 2020. Washington, DC: World Bank. doi: 10.1596/978-1-4648-1468-6 World Bank (2020b). Global Economic Prospects, June 2020. Washington, DC: World Bank. doi: 10.1596/978-1-4648-1553-9 World Bank (2021a). ASPIRE: The Atlas of Social Protection Indicators of Resilience and Equity. Washington, DC: World Bank. Data and methodology available at https://www.worldbank.org/en/data/datatopics/aspire World Bank (2021b). Global Economic Prospects, June 2021. Washington, DC: World Bank. doi:10.1596/978-1-4648-1665-9. World Bank (2022). Global Economic Prospects, January 2022. Washington, DC: World Bank. doi: 10.1596/978-1-4648-1758-8 World Bank (2022b). World Development Indicators. Washington, DC: World Bank. 25 Figure 1 – Stimulus budget, expressed in % of 2019 GDP Note: this graph plots the size of economic mitigation measures (distinguishing between social protection and other measures, expressed in percent of 2019 GDP) by log GDP per capita (2019 data, expressed in USD at PPP). The share of each type of spending is plotted by a LOWESS (Locally Weighted Scatterplot Smoothing) function. Source: Social protection budget data are from the authors’ dataset. Data on the budget for economic mitigation measures are from the IMF (2021). GDP data are from World Bank (2022b). 26 Figure 2 – Size of social protection response budget and GDP per capita Note: this graph plots, for every country in the world, the pandemic social protection response budget, as a percent of the 2019 GDP (vertical axis) and the log 2019 GDP per capita, in USD PPP prices (horizontal axis). Source: Social protection budget data are from the authors’ dataset. GDP data are from World Bank (2022b). 27 Figure 3 – Share of social protection response budget by program type of measure Note: this graph plots the social protection response budget (by type of program, expressed in percent of 2019 GDP) by log GDP per capita (2019 data, expressed in USD at PPP). The share of each type of program is plotted by a LOWESS function. Detailed program codes can be found in Table 1. “Other job protection measures” includes program codes 2.3, 2.4, 2.5, 3.2, 35, 3.6, and 3.7. “Other income protection measures” includes program codes 1.2, 1.3, 1.5, 1.6, 2.1, and 3.4. Source: Social protection budget data are from the authors’ dataset, expanded from Gentilini et al. (2022). GDP data are from World Bank (2022b). 28 Figure 4 – GDP growth and change in employment rate, 2020 Note: this graph plots, for every country in the world, the change in employment rate between 2019 and 2020 (vertical axis) and the change in GDP between 2019 and 2020 (horizontal axis). The employment rate is defined as the share of employed over the working-age population. The change between 2019 and 2020 is expressed in percentage points. The GDP change is expressed in percentage variation with respect to the 2019 level. The solid black line plots the linear fit between the two variables. Source: employment data are from ILO (2021), and GDP data are from IMF (2021) and World Bank (2021). 29 Figure 5 – Change in socio-economic outcomes (2020) and job protection measures Note: this graph plots the change in different socio-economic outcomes during 2020 (vertical axis) and the job protection measures expressed as a share of the social protection pandemic response budget mix (horizontal axis) for 133 countries. The solid black line plots a linear fit between the two variables in each panel. Panel (a) plots the change in GDP during 2020 in percentage points. Panel (b) plots the change in the employment rate during 2020 in percentage points, where the employment rate is defined as the share of employed over the working-age population. Panel c plots the change in the inactivity rate during 2020 in percentage points, where the inactivity rate is defined as the share of inactive over the working-age population. Panel d plots the change in the poverty headcount rate ($5.5 a day international poverty line) Source: Social protection budget data are from the authors’ dataset, expanded from Gentilini et al. (2022). Employment data are from ILO (2021) and GDP data from World Bank (2021). 30 Table 1 – Classification of social protection programs by policy focus Social protection area Social protection category Policy focus 1.1 Unconditional cash transfers 1.2 Conditional cash transfers 1.3 Social pensions (non-contributory) Social assistance 1.4 Unconditional food and in-kind transfers Income protection 1.5 Conditional in-kind transfers (school feeding) 1.6 Public works 1.7 Utility and financial obligations waivers/reductions 2.1 Pensions 2.2 Social insurance contributionsa 2.3 Paid leave Social insurance Job protection 2.4 Workers’ compensation 2.5 Health insurance 2.6 Unemployment/out of work income support Income protection 3.1 Activation measures 3.2 Redistribution of labor Job protection 3.3 Wage subsidies Labor market policies 3.4 Other active labor market policiesb Income protection 3.5 Labor income support 3.6 Labor regulatory adjustment and enforcement Job protection 3.7 Firm liquidity support Notes: a. Measures that involve withdrawals from individual retirement accounts were excluded from this analysis because their fiscal impact could not be unambiguously identified. Waivers of social insurance contribution for firms were included when budget cost estimates were available. Thus, all measures under this category were classified as focusing on job protection. b. This category includes entrepreneurship support, startup incentives, and employment measures for people with disabilities. Given their focus on bringing people into employment (similar to activation measures), measures in this category were classified as focusing on income protection as they are not tied to an individual having had a job. 31 Table 2 – Stimulus and social protection response budget Low- Lower-middle Upper-middle High- All income income income income countries Variable countries countries countries countries All values are country averages expressed in % of 2019 GDP Stimulus budget 3.14 3.00 4.26 10.03 5.58 of which Health response budget 1.06 0.69 0.72 1.34 0.95 Social protection response budget 0.75 0.97 1.91 3.48 1.99 Income protection measures 0.75 0.90 1.31 1.57 1.20 of which Unconditional cash transfers 0.38 0.60 0.74 0.68 0.63 In-kind and food transfers 0.28 0.13 0.03 0.04 0.10 Utility and financial waivers 0.07 0.07 0.19 0.27 0.16 Unemployment/out-of-work 0.00 0.01 0.11 0.38 0.15 income support Job protection measures 0.00 0.07 0.59 1.92 0.79 of which Social insurance contributions 0.00 0.01 0.21 0.21 0.13 waivers Wage subsidies 0.00 0.05 0.35 1.55 0.60 Number of countries 19 40 38 45 142 Note: see table 1 for the classification of social protection into policy focus areas (income protection and job protection). UCT = Unconditional Cash Transfers. Source: Social protection budget data is from the authors’ dataset. Stimulus budget and health response data from IMF (2021). 32 Table 3: Cross-country regressions of the effect of social protection response on socio-economic outcomes Note: this table reports estimates from the following regression model: = + 1 + 2 + 3 + 1 + 2 + 3 + Where Sc is a socio-economic outcome for country c. IncomeProtectionGDPc indicates the expenditure in social protection measures aimed at income protection of country c, expressed in percentage points of the 2019 GDP, JobProtectionGDPc is the expenditure in social protection measures aimed at job protection of country c, expressed in percentage points of the 2019 GDP and NonSPResponseGDPc is the expenditure of pandemic response measures in policy areas other than social protection in country c, expressed in percentage points of the 2019 GDP. LogGDPpcc is the log of GDP per capita of country c in 2019, expressed in dollars at PPP. ServicesGDPc represents the share of the services sector in the GDP of country c in 2019, expressed in percentage points. Informalityc represents the share of informal output in the GDP of country c in the last available pre-pandemic year, expressed in percentage points. All estimations are by OLS with robust (HC3) standard errors. *** indicates that the coefficient is significant at 1% level, ** - at 5% level, * - at 10% level. Change in GDP Change in employment rate Change in inactivity rate Change in poverty rate (1) (2) (3) (4) (5) (6) (7) (8) Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Social protection response Expenditure on income protection, % -0.092 0.267 0.145 0.289 -0.162 0.125 -0.175 0.201 0.104 0.098 0.127 0.153 0.127 0.116 0.195 0.151 of GDP Expenditure on job protection, % of 0.583 0.391 1.132** 0.502 0.446*** 0.165 0.324 0.196 -0.405*** 0.155 -0.382** 0.156 -0.444*** 0.136 -0.488*** 0.178 GDP Non-SP response expenditure, % of 0.282** 0.125 0.220* 0.122 0.049 0.065 0.055 0.087 -0.018 0.053 -0.015 0.073 -0.129** 0.052 -0.141** 0.055 GDP Country characteristics Log GDP per capita, 2019 0.097 0.402 0.874 0.520 -0.183 0.143 -0.054 0.251 -0.095 0.122 -0.203 0.206 -0.014 0.157 -0.230 0.209 Share of services sector, % of GDP -0.157* 0.080 -0.043 0.027 0.038* 0.021 0.049* 0.027 Share of informal output, % of GDP 0.002 0.047 -0.044 0.033 0.035 0.028 0.015 0.018 (DGE) Number of countries 147 134 145 134 145 134 148 134 R2 0.09 0.15 0.07 0.12 0.10 0.16 0.14 0.19 33 Table 4 Sensitivity analysis Dependent variables Specification Budget on each Δ GDP Δ Employment Δ Inactivity rate Δ Poverty rate type of program (% rate of GDP) Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Income protection 0.193 0.271 -0.175 0.201 0.127 0.153 0.092 0.116 I Job protection 1.108** 0.501 0.324 0.198 -0.382** 0.156 -0.438*** 0.152 # of Countries 134 134 134 132 Income protection 0.151 0.288 0.195 0.151 II Job protection 0.974* 0.525 -0.488*** 0.178 # of Countries 134 134 Income protection 0.145 0.289 0.175 0.142 III Job protection 1.132** 0.502 -0.435** 0.171 # of Countries 134 134 Income protection 0.219 0.139 IV Job protection -0.697** 0.301 # of Countries 113 Income protection 0.166 0.320 -0.321* 0.192 0.217 0.155 0.128 0.140 V Job protection 1.279** 0.607 0.184 0.212 -0.317* 0.169 -0.514** 0.201 # of Countries 115 115 115 113 Income protection 0.153 0.318 0.161 0.178 VI Job protection 1.356** 0.598 -0.616*** 0.227 # of Countries 115 115 Income protection 0.421 0.340 -0.022 0.222 0.046 0.200 0.021 0.149 VII Job protection 1.547** 0.677 0.352 0.251 -0.386* 0.215 -0.569*** 0.208 # of Countries 109 109 109 107 Income protection 0.378 0.359 0.067 0.195 VIII Job protection 1.680** 0.669 -0.650*** 0.238 # of Countries 109 109 Description of scenarios I – In this specification, all the dependent variables are expressed as a difference between the value observed in 2020 and the value observed in 2019. II – In this specification, the GDP and poverty headcount rate ($5.5 a day line) are expressed as a difference between the value observed in 2020 and the pre-pandemic projection for the same year. III - In this specification, the GDP and poverty headcount rate ($5.5 a day line) are expressed as a difference between the value observed in 2021 and the pre-pandemic projection for the same year. IV - In this specification, the poverty headcount rate is expressed as a difference between the value observed in 2020 and the value observed in 2019, using the international poverty line corresponding to a country’s income group ($1.9 a day for low- income countries, $3.2 a day for lower-middle income countries and $5.5 a day for upper-middle income countries) and the national poverty line for high-income countries. V - In this specification, all the dependent variables are expressed as a difference between the value observed in 2020 and the value observed in 2019, and the level of pre-pandemic social insurance coverage is included as an additional control variable. VI - In this specification, the GDP and poverty headcount rate ($5.5 a day line) are expressed as a difference between the value observed in 2021 and the pre-pandemic projection for the same year, and the level of pre-pandemic social insurance coverage is included as an additional control variable. VII - In this specification, all the dependent variables are expressed as a difference between the value observed in 2020 and the value observed in 2019. The level of pre-pandemic social assistance coverage is included as an additional control variable. VIII - In this specification, the GDP and poverty headcount rate ($5.5 a day line) are expressed as a difference between the value observed in 2021 and the pre-pandemic projection for the same year and the level of pre-pandemic social assistance coverage is included as an additional control variable. 34 Table 5: Cross-country regressions of the effect of social protection response on socio-economic outcomes Note: this table reports estimates from the following regression model: = + 1 + 2 × _ + 3 + 4 × _ + 35 + 1 + 2 + 3 + 4 _ + Where Sc is a socio-economic outcome for country c. IncomeProtectionGDPc is the expenditure in social protection measures aimed at income protection of country c, expressed in percentage points of the 2019 GDP, JobProtectionGDPc is the expenditure in social protection measures aimed at job protection of country c, expressed in percentage points of the 2019 GDP and NonSPResponseGDPc is the expenditure of pandemic response measures in policy areas other than social protection in country c, expressed in percentage points of the 2019 GDP. LogGDPpcc is the log of GDP per capita of country c in 2019, expressed in dollars at PPP. ServicesGDPc represents the share of the services sector in the GDP of country c in 2019, expressed in percentage points. Informalityc represents the share of informal output in the GDP of country c in the last available pre-pandemic year, expressed in percentage points. HI_SPCoveragec is a dummy value that indicates whether the share of the population of country c covered by the social protection system is above the sample median. Two subsystems are distinguished – social assistance and social insurance. All estimations are by OLS with robust (HC3) standard errors. *** indicates that the coefficient is significant at 1% level, ** - at 5% level, * - at 10% level. Change in GDP Change in employment rate Change in inactivity rate Change in poverty rate (1) (2) (3) (4) (5) (6) (7) (8) Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Expenditure on income protection, % of -1.674 1.250 1.041 0.711 -1.308** 0.506 0.172 0.198 1.129** 0.438 -0.132 0.175 1.222** 0.603 -0.096 0.159 GDP Expenditure on income protection × 2.060 1.262 1.142* 0.580 -1.052** 0.490 -1.239** 0.621 High social insurance coverage Expenditure on income protection × -1.546 0.946 -0.500 0.524 0.452 0.445 0.276 0.339 High social assistance coverage Expenditure on job protection, % of 2.628 2.175 1.477* 0.853 1.790** 0.749 0.132 0.312 -1.166 0.802 -0.176 0.249 -2.137*** 0.790 -0.305 0.218 GDP Expenditure on job protection × -1.221 1.961 -1.539** 0.692 0.779 0.755 1.630** 0.759 High social insurance coverage Expenditure on job protection × 0.403 1.034 0.480 0.343 -0.452* 0.268 -0.567* 0.323 High social assistance coverage High social insurance coverage 0.484 2.332 0.442 0.753 -0.230 0.590 -0.471 0.922 (above the median) High social assistance coverage -0.280 1.592 -0.829 0.820 0.605 0.704 0.897 0.624 (above the median) Test of joint significance – main variable F-stat P-value F-stat P-value F-stat P-value F-stat P-value F-stat P-value F-stat P-value F-stat P-value F-stat P-value and interaction with SP coverage Income protection 2.04 0.135 1.43 0.245 3.79 0.026 0.60 0.552 3.58 0.031 0.58 0.561 2.06 0.133 0.36 0.699 Job protection 3.00 0.054 2.54 0.084 2.87 0.061 3.24 0.043 2.64 0.076 5.35 0.006 5.58 0.005 3.94 0.023 Number of countries 115 109 115 109 115 109 113 107 R2 0.18 0.20 0.22 0.21 0.25 0.24 0.32 0.25 35 Table 6: Cross-country regressions of the effect of social protection response on socio-economic outcomes Note: this table reports estimates from the following regression model: = + 1 ℎ + 1 + 2 + 3 + Where Sc is a socio-economic outcome for country c. JobProtectionSharec is the share of the social protection pandemic response expenditure that was allocated to measures aimed at job protection in country c. LogGDPpcc is the log of GDP per capita of country c in 2019, expressed in dollars at PPP. ServicesGDPc represents the share of the services sector in the GDP of country c in 2019, expressed in percentage points. Informalityc represents the share of informal output in the GDP of country c in the last available pre-pandemic year, expressed in percentage points. All estimations in columns 2, 4, 6, and 8 are by OLS with robust (HC3) standard errors. Columns 3, 5, 7, and 9 are estimated by 2SLS with robust standard errors. The first stage results are presented in column 1. *** indicates that the coefficient is significant at 1% level, ** - at 5% level, * - at 10% level. The Stock-Yogo weak identification test critical values are: 10% maximal IV size=16.38 15%=8.96 20%=6.66 25%=5.9363. Share of SP Change in GDP Change in employment rate Change in inactivity rate Change in poverty rate response allocated to job protection (1) (2) (3) (4) (5) (6) (7) (8) (9) First stage IV OLS IV OLS IV OLS IV OLS IV Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Coef. S. Err. Social protection response Share of SP response budget allocated 5.894** 2.267 14.02 10.36 2.527*** 0.632 5.803 4.178 -2.078*** 0.487 -4.765 3.635 -2.707*** 0.803 -2.622 3.822 to job protection measures Country characteristics Log GDP per capita, 2019 0.119*** 0.025 1.138 0.747 0.159 1.268 -0.091 0.279 -0.486 0.605 -0.113 0.236 0.210 0.540 -0.034 0.253 -0.044 0.442 Share of services sector, % of GDP 0.000 0.003 -0.186* 0.108 -0.197** 0.096 -0.050* 0.026 -0.055* 0.028 0.036* 0.020 0.040* 0.022 0.029 0.026 0.029 0.026 Share of informal output, % of GDP -0.007*** 0.002 -0.011 0.061 0.045 0.10 -0.039 0.033 -0.016 0.037 0.030 0.028 0.012 0.029 0.025 0.022 0.025 0.033 (DGE) Instrumental variable Months elapsed since last general -0.003*** 0.001 election (March 2020) Cragg-Donald Wald F statistic 6.028 6.028 6.028 6.028 6.028 Number of countries 126 126 126 126 126 126 126 126 126 R2 0.47 0.14 0.04 0.14 0.03 0.17 0.07 0.15 0.15 36 APPENDIX TABLES Table A.1 Countries included in the core sample Albania Georgia Niger United Arab Emirates Ghana Nigeria Argentina Gambia, The Netherlands Armenia Guinea-Bissau Norway Australia Equatorial Guinea New Zealand Austria Greece Pakistan Azerbaijan Guatemala Panama Belgium Honduras Peru Benin Croatia Philippines Burkina Faso Hungary Poland Bangladesh Indonesia Portugal Bulgaria India Paraguay Bahrain Ireland Qatar Bahamas Iran, Islamic Rep. Romania Bosnia and Herzegovina Iceland Russian Federation Bolivia Israel Rwanda Brazil Italy Saudi Arabia Botswana Jamaica Sudan Central African Rep. Jordan Senegal Switzerland Japan Singapore Chile Kazakhstan Sierra Leone China Kenya El Salvador Côte d’Ivoire Kyrgyz Rep. Slovak Republic Cameroon Cambodia Slovenia Congo, Dem. Rep. Korea, Rep. Sweden Congo, Rep. Lao People’s Dem. Rep. Eswatini Colombia Liberia Chad Comoros Sri Lanka Togo Cabo Verde Lesotho, Kingdom of Thailand Costa Rica Lithuania Tajikistan Cyprus Luxembourg Trinidad and Tobago Czech Rep. Latvia Tunisia Germany Morocco Türkiye Denmark Madagascar Uganda Dominican Rep. Maldives Ukraine Algeria Mexico Uruguay Ecuador North Macedonia United States Egypt, Arab Rep. Mali Vietnam Spain Malta South Africa Estonia Myanmar Zambia Ethiopia Mongolia Zimbabwe Finland Mozambique Fiji Mauritius France Malawi Gabon Malaysia United Kingdom Namibia 37 Table A.2 Descriptive statistics Variable Mean Std. Dev. Min Max Obs Change in GDP, 2020-2019 (p.p.) -4.347 5.378 -31.98 6.06 152 Change in GDP, 2021 actual to pre- -6.645 5.432 -27.74 12.09 150 pandemic forecast (p.p.) Change in employment rate, 2019- -2.500 2.160 -14.40 0.62 150 2020 (p.p.) Change in inactivity rate, 2019-2020 1.932 1.826 -0.41 12.69 151 (p.p.) Change in poverty rate ($5.5), 2019- 1.267 1.998 -1.29 12.22 152 2020 (p.p.) Change in poverty rate ($5.5), 2020 1.870 2.319 -0.60 12.80 154 pre-Covid projection to 2020 (p.p.) Change in poverty rate ($5.5), 2021 1.998 2.591 -0.80 15.60 154 pre-Covid projection to 2021 (p.p.) Change in poverty rate (line by 1.382 2.065 -5.20 12.22 131 income group), 2019-2020 (p.p.) Pandemic expenditure in income 1.310 1.782 0 13.46 154 protection measures (p.p. of GDP) Pandemic expenditure in job 0.740 1.253 0 6.48 154 protection measures (p.p. of GDP) Pandemic expenditure in non-SP 3.573 3.448 0 16.89 148 measures (p.p. of GDP) Log GDP per capita, 2019 9.458 1.1708 6.81 11.70 154 Share of services in GDP, last pre- 55.49 10.936 31.12 89.98 152 pandemic year available (p.p.) Share of informal output (DGE), last 28.267 11.237 7.97 63.40 139 pre-pandemic year available (p.p.) Social Insurance Coverage, last pre- 21.381 18.314 0.37 61.24 129 pandemic year available (p.p. of population) Social Assistance Coverage, last pre- 27.754 24.995 0 93.25 123 pandemic year available (p.p. of population) 38