COVID-19 lockdown led to an unprecedented increase in inequality

The COVID-19 pandemic has aﬀected households across the globe due to the health impacts but also through indirect socioeconomic eﬀects as a result of the additional stress on the health systems, implications of the lockdowns and other policy measures undertaken by governments. Moreover, there is evidence that these impacts are associated with socioeconomic characteristics of households and could lead to an increase in inequality and poverty. In this paper, we conduct a ﬁrst assessment based on household surveys in a large set of countries and analyze the determinants of income shocks at the household-level and macroeconomic inequality implications. While the average income losses of on average 4% but up to 27% are already high (similar to GDP losses in 2020 of on average 6% in our sample), we ﬁnd an even more striking increase in inequality, of up to several points (1.2 points on average) of the typically very ﬂat Gini index. Across countries, we ﬁnd that on average an additional one percentage point loss in GDP was associated with an increase in the Gini by one percentage point. Analyzing the determinants of the income shock, we ﬁnd strong evidence of heterogeneity with higher likelihood of income shocks for poorer, female-led, and less-educated households. The results indicate that we are experiencing an unprecedented crisis also in terms of economic inequality. The policy response to limit the macroeconomic repercussions therefore should explicitly include reducing inequality. Otherwise, a rebuilt macro-economy, will exhibit a much higher degree of social and economic inequality that are likely to persist.

holds across the globe has also suffered indirect socioeconomic effects due to the additional stress on the health systems, implications of the lockdowns, and other policy measures undertaken by governments.There is evidence that these impacts are associated with socioeconomic characteristics of households and could lead to an increase in inequality and poverty 1 .Most of the existing evidence is based on macroeconomic studies, while some studies use micro-simulation and now-casting exercises have been applied to the COVID-19 crisis in 2020 2,3,4 , empirical analysis using household-level micro surveys are still missing, especially.Furthermore, cross-country comparisons are also unavailable thus far.We use household surveys from 35 countries to investigate the socioeconomic impacts of the pandemic, specifically, inequality and poverty implications and the differentiated impact of household and individual characteristics.
We conduct a first assessment using these household surveys to analyse the socioeconomic determinants of income shocks at the household-level and wider macroeconomic inequality implications.We find strong evidence of heterogeneity with higher likelihood of income shocks for poorer, female-led, and lower-educated households.The results indicate that we are experiencing an unprecedented crisis also in terms of economic inequality, whose policy response will be pivotal to mitigate long-term repercussions of this rise in inequality around the world.While the average income losses of on average 4% but up to 27% are already high (similar to GDP losses in 2020 of on average 6% in our sample), we find an even more striking increase in inequality, of several points (1.2 points on average) of the normally very flat Gini index.The policy response to limit the macroeconomic repercussions therefore should explicitly be targeted at reducing inequality.Otherwise, a rebuilt macro-economy, will exhibit a much higher degree of social and economic inequality that are likely to persist.
The impact of the pandemic on households has been multidimensional, directly due to the health crisis, but importantly also indirectly to economic implications and restrictions due to the lockdown and even through indirect impacts on employers, firms, and the public sector.The World Bank estimates that global per capita GDP will decline by 6.2% in 2020 with Sub-Saharan Africa's per capita GDP expected to decline by 5.3% while that of South Asia is projected to decline by 3% 5 .These impacts threatens to offset development efforts and the decrease in global income inequality achieved in the last forty years 6 .
In terms of empirical evidence based on past pandemics, it has been indeed found that they have led to persistent impacts on economic growth 7 , inequality 8 , and affected absolute poverty 9 .However, most of this evidence is based on crosscountry macroeconomic studies.And while in particular micro-simulation studies and now-casting exercises have been applied to the COVID-19 crisis in 2020 2,3,4 , actual household data based analysis is still missing.Yes their indications show a strong distributional impact of the crisis in particular on market incomes, while in some cases in particular lower-income households in some cases such as the UK even appear to have gained in 2020.Based on household micro-surveys, 10 found impacts of COVID-19 on income and food security for selected regions in India.Yet overall, robust micro-econometric evidence is scarce due to the availability of data that allows the capture of idiosyncratic heterogeneity.
Since the beginning of the pandemic, several countries, in particular in Africa and Asia, have conducted High Frequency Phone Surveys (HFPS) of households linked to the ongoing panel micro studies, including Burkina Faso, Ethiopia, Kenya, Malawi, Mali, Nigeria, South Africa, Uganda, and also in India.Most of these surveys have been conducted in collaboration with the World Bank, however, individual countries have also added COVID-19 modules to their existing household surveys (e.g., South Africa).The main aim of these surveys is to monitor the socioeconomic impacts of the pandemic with a focus on employment, income (wages and business revenue), health, education, food security, and coping strategies, including safety nets.Along with surveys from these developing countries, we also use the Survey of Health, Ageing and Retirement in Europe (SHARE) -a pan-European dataset on public health and socioeconomic conditions in Europe 11 .Descriptions of the datasets used in this paper are available in the c section.
Table A1 in the c provides the descriptive statistics of the main variables used in this paper.India had the highest share of respondents who reported a loss in income.
In the case of pre-COVID income, surveys in South Africa (monthly -February 2020), India (daily wage -March 2020), Kenya (two-weeks period in February 2020), and the SHARE dataset included a question on income before the pandemic.For Nigeria and Uganda, we matched monthly income reported in the 2019 waves of the regular household surveys.Mali reported poverty status of individual households instead of income.The descriptive statistics for the individual countries in the SHARE dataset are provided in Table A2.A number of countries reported that more than 70% of the households suffered a loss in income due to the pandemic; with 78.8% of the respondents in India reporting that their income declined below the pre-lockdown levels.Even the relatively developed countries in the SHARE dataset (EU-27 plus Israel), more than six percent of the respondents had their income reduced.In this paper, we contribute to this research gap by exploiting household surveys in 35 countries that have been conducted to address two main research questions; (1) what are the socioeconomic impacts of the pandemic, in particular in income; hence, inequality and poverty and (2) the differentiated impact of household and individual characteristics.
To investigate the socioeconomic the determinants of income loss due to the pandemic, we use the following Probit specification estimated for all countries with available data and covariates to detect common patterns.

P rob(y
where y i is the income of a household before and after the pandemic-induced lockdown, and Φ represents the cumulative density of the standard normal distribution.
We control for location fixed-effects, socioeconomic and demographic characteristics such as the education, gender, and age of household head, log of pre-lockdown income, and poverty status of households in the case of Mali.

Empirical findings
The regression results for all the countries are shown in Table 1.They suggest that the probability of income loss is generally lower among households with higher educated heads, this is critical as education continues to be beneficial even during a pandemic.In Nigeria, India, Mali, and the SHARE countries -households with female heads have a higher probability of suffering from income loss, however, we find the opposite effect in South Africa.The regression estimates also suggest that households with higher income are less likely to suffer from income loss, further emphasising that the distributional consequences of the pandemic falls disproportionately on the lower income earners.In the case of Mali, which report poverty status of households instead of income, our findings suggest that poor households have a higher probability of suffering from income loss due to the pandemic.

Impact on inequality
The societal impact on income loss and other individual impacts is important in particular given the evidence on incidence among vulnerable groups found in the previous section.Looking at the overall income distribution in all countries under consideration, we now evaluate the macro-economic picture of these income losses at the household-level and underlying representative survey for the countries.
Notably, the panel dimension allows us to produce accurate before and after income distributions.Based on country-level statistics and focusing on past pandemics, it has been found that inequality increases significantly and persistently after pandemic episodes, leading to an increase in the net Gini coefficient by approximately 1% as found in 8 even five years after the pandemic.At a global average Gini index of around 40 points, this implies an increase by around 0.4 points -significant but comparably small.
Here we combine the micro-economic evidence across countries for a short-term and micro-econometric and data-driven estimate of the inequality effect at least in the year of the pandemic for COVID-19.We compute Lorenz curves for net income and consumption -depending on data availability -before and after the pandemic shock and lockdown start.For the countries where only a categorical variable is available and not the monetary value, we approximate the categorical variable reporting perceived impacts on income using the average shock loss reported in previous waves of the survey.
Figure 1 shows the resulting Lorenz curves prior and after the lockdown in various countries, computed on household reported income and consumption values using sampling weights and household size as weights.In most cases, we find a substantial shift of the curve outwards indicating an increase inequality across the full distribution.In case of income and for India, this case is most striking since due to the lockdown, workers' income virtually dropped to zero for a large majority of (informal) workers, at least temporarily.For the SHARE dataset, Figure 2 shows the change in the Lorenz curves across the full distribution (similar as in 13 ) and change in Gini in points pre-and post the March 2020 lockdown.First note the substantial increase in inequality in Southern Europe of up to 5 points, while some countries show even reductions.Looking along the full distribution (left panel), in particular at very low incomes (bottom 10% in all countries and bottom 20% in most countries), the Lorenz curves shifts downwards in almost all countries for the bottom 20% of the population.For higher incomes the differences across countries vary much more including with significant gains in some countries.(shifting inwards of the Lorenz curve).In terms of aggregate inequality statistics, we report the Gini index for all countries and variables, see table A3 in the c.Moreover, we show the latest available historical Gini data based on the UNU-WIDER World Income Inequality Database (WIID), and the GDP growth rate estimated for 2020 as of the October 2020 IMF World Economic Outlook Update.
The income loss (in India) is the highest, with Gini skyrocketing to almost 92 points, as observable from the Lorenz curve.This is due to almost 70% of the respondents reporting their income dropping to zero between February and April 2020.This finding is certainly temporary to a large part, but shows the immediate dramatic impact not only in terms of average income loss but also its high regressivity.
In terms of consumption in the same sample, as expected, the results are much reduced due to consumption smoothing and other measures.However, the findings still shows a substantial increase in the Gini coefficient by almost 7 points, while on average consumption dropped by 3 percent.This again suggests the strong regressive impact of COVID-19 in India.
In South Africa, the income Gini index increased by 3.5 points with an average drop of income of 8%.For Nigeria, we do not see a clear pattern, note, however, that the size of the impact we assume to be constant across the distribution, while the impacted households seem evenly distributed.This also shows that some countries notably in Sub-Saharan Africa seemed to have managed to keep the socioeconomic impact relatively low.In all cases, the Lorenz curve do not seem to intersect showing an unambiguous deterioration of the distribution between February and March/April.
While these impacts certainly show a substantial increase in inequality in most countries, its persistence is yet to be seen.The findings, jointly with the determinants of income shocks found in the previous section, show that the impact could indeed be rather persistent and unequally distributed along various dimensions.
Compared to the long-run estimated impacts, which have been found to be persistent, see 8 , it indicates a strong short-term impact followed by smaller but significant longer term effects.
Based on the large set of countries, we estimated the Gini growth rate across countries and how it relates to the policy stringency of the measures enacted and related GDP impact.The results are shown in table 2. The strongest impact is from GDP growth; a one percent higher GDP loss in 2020 leads to approximately a one per cent increase in the Gini index.The (additional) effect of policy stringency in addition leads to a further increase in inequality.This also has important consequences on increases in poverty, such as the ones projected by 9 , 14 , or 15 , but imply that at least in the short-term, poverty increases could be even higher.

Conclusion
The COVID-19 pandemic and the lockdowns implemented to contain it has had multifaceted impacts on countries and household across the globe.Along with devastating health impacts, the indirect socioeconomic effects have also been substantial in most countries.Our goal in this paper is to provide a first assessment of some of these impacts, notably on the distributional impact and changes in inequality.Using unique micro-surveys conducted to evaluate the impact of pandemic in both developing and developed countries (35 in total), we investigate the effects on household income and inequality.Our econometric analysis suggests that certain socioeconomic characteristics of households could lead to an increase of inequality and poverty in developing countries.Importantly, we find that higher levels of education lowers the probability of income loss even during a pandemic.
To investigate the impact on inequality, we estimate Lorenz curves using household reported values and sampling weights from the surveys.In all cases, we find a substantial shift of the curve outwards indicating an increase inequality across the full distribution.Our findings show an immediate and substantial increase in income and consumption inequality in most countries, of on average 1.2 but up to 5 points in Europe and 7 points in India.While our inequality estimates are short-term in nature, in conjunction with the determinants of income shocks, these impacts could indeed become rather persistent and unequally distributed.
The main determinant of this spike in inequality can be traced back to the general economic crisis; on average, one additional percentage point of GDP in 2020 translated to about one additional percent increase of the Gini coefficient, highlighting the overall macroeconomic importance of a swift recovery also in terms of inequality reduction.
These results highlight the role of two important policy measures: international aid as many of these impacts are within the developing word, and in particular hint to large increases in poverty, and national re-distributive and transfer/safety-net policies to reduce the regressivity of impacts on households.

Econometric methodology
To investigate the socioeconomic the determinants of income loss due to the pandemic, we use the following Probit specification estimated for all countries with available data and covariates to detect common patterns.
P rob(y i,post_lockdown < y i,pre_lockdown where y i is the income of a household before and after the pandemic-induced lockdown, and Φ represents the cumulative density of the standard normal distribution.
We control for location fixed-effects, socioeconomic and demographic characteristics such as the education, gender, and age of household head, log of pre-lockdown income, and poverty status of households in the case of Mali.

Dataset descriptions
Kenya reported 84,000 COVID-19 cases and nearly 1,500 deaths until the end of November.Nairobi so far has reported 45% of all the cases in the country.Lorenz curves before and after the lockdown, developing countries.
Lorenz curves: difference before and after the lockdown (left) and Gini change (right).
Governments across the world implemented various measures to contain the spread of the COVID-19 pandemic.The Oxford COVID-19 Government Response Tracker (OxCGRT) tracks governments' policies and interventions including school closings, travel restrictions, bans on public gatherings, emergency investments in healthcare facilities, new forms of social welfare provision, contact tracing and other interventions to contain the spread of the virus, and augment health systems 12 .We report the stringency of the containment measures in Figure B.3 in the c with the results suggesting that the stringency of the government responses reached peak during March and April.

Figure 1 :
Figure 1: Lorenz curves before and after the lockdown, developing countries.

Figure 2 :
Figure 2: Lorenz curves: difference before and after the lockdown (left) and Gini change (right).

3 .
In the case of India, the number of people who reported that they did not work for income increased form 18.4% in March to 70.4% in the beginning of the post-lockdown period.Furthermore, the average income from March 2020 to April 2020 declined by more than 70% among the households sampled.As a the consequences of these losses in employment and income, the average monthly food expenditure also declined by more 5%.The data for India comes from a collaborative effort by the World Bank, IDinsight, Development Data Lab, and John Hopkins University across six states in India; Jharkhand, Rajasthan, Uttar Pradesh, Andhra Pradesh, Bihar, and Madhya Pradesh.This dataset provides information on agriculture, migration, rural labour markets, consumption patterns, access to relief/safety nets, and healthcare using a sample size on 4,550.Uganda so far reported 21,409 confirmed COVID-19 cases and around twohundred deaths.The High-Frequency Phone Survey was launched in June 2020 by the Uganda Bureau of Statistics (UBOS) with support from the World Bank.The survey tracks the impacts of the pandemic.2,259 households were interviewed from the Uganda National Panel Survey (UNPS) 2019-20.SHARE is the largest longitudinal micro data providing information on public health and socioeconomic living conditions of European individuals 11 .To investigate the health-related and socioeconomic impact of COVID-19 on the risk group of the older individuals, a sub-sample of SHARE's panel respondents were interviewed using a Computer Assisted Telephone Interview (CATI).The survey has been carried out in 27 European countries and Israel from between June and August 2020 and consists of more than 70,000 respondents.The COVID-19 module gathers information on health and health behaviour, mental health, infections and healthcare, changes in work and economic situation, and social networks.

Figures Figure 1
Figures

Table 1 :
Determinants of income loss (Probit analysis)

Table 2 :
Regression of the Gini growth rate on GDP growth and policy stringency This survey is being conducted by the World Bank in collaboration with the Kenya National Bureau of Statistics and the University of California Berkeley.The first sample is a randomly drawn subset of all households that were part of the 2015-16 May 2020 drawn from wave 4 of the General Household Survey-Panel (GHS-Panel) in Nigeria.South Africa reported more than 792,000 cases of COVID-19 and nearly 22,000 deaths as of November 2020.Gauteng, the most populous province in South Africa (population of approximately 15 million) has reported almost 30% of the total number of cases 2 .Data for South Africa suggests an 18% decline in employment between February and April 2020, of which two-thirds were women.In the NIDS-CRAM sur-India has reported more than 9.5 million cases of COVID-19 and more than 138,000 deaths at the end of November.Maharashtra, the second most populous state (population of 112 million) has so far reported the highest number of COVID-19 cases at 1.82 million and more than 47,000 deaths.Among the six states in which the survey was conducted, Andhra Pradesh (the tenth-most populous state in India with a population of more than 49 million) has reported 868,000 (the third highest in India) and almost 7,000 deaths loss of income.Commerce (14%), services (9%), and agriculture (9%) sectors have reported the highest number of job losses.We use the Nigeria COVID-19 National Longitudinal Phone Survey (COVID-19 NLPS) implemented by the National Bureau of Statistics in collaboration with the World Bank.The survey was conducted on a nationally representative sample of 1,950 households between 20 April and 11