Lessons on estimating the distributional impacts of a jobs shock

The economic impact of COVID-19 was massive, with a staggering decline in labor markets across low- and middle-income countries. Estimates from the International Labor Organization (ILO) indicate an 18.5% decline in total hours worked between Q4 2019 and Q2 2020, equivalent to 520 million full-time jobs lost. Phone surveys in 39 developing countries painted an even bleaker picture, with approximately one-third of respondents reporting stopping work during the Spring or Summer of 2020. Despite some recovery between 2020 and 2021, labor markets were still struggling.

A Blind Start

At the outset of the crisis, policymakers faced uncertainty regarding the impacts and distributional effects of the unfolding economic shock. Key questions included how the crisis affected the poor compared to the wealthy and the extent to which poverty increased.  However, collecting and publishing survey data on welfare usually takes many months, if not years, and most data collection was temporarily halted in 2020, making it challenging to answer these crucial questions.

Microsimulations to the Rescue

To address this data gap, microsimulations emerged as a valuable technique for estimating the distributional impacts of the crisis. These simulations involved a five-step process:

1. Projection of Job Losses: Projected job losses based on changes in per capita GDP.
2. Identification of Vulnerable Workers: Estimation of which workers were least likely to be employed based on socio-economic characteristics, sectors of work, and formality status.
3. Simulation of Job Losses: Simulation of job loss and complete loss of labor income among the most vulnerable workers in non-protected sectors, to match the scale of projected job loss.
4. Income Change Simulation: For workers who retained their jobs but were not in protected sectors, simulation of a labor income change equivalent to the percentage change in national accounts private consumption.
5. Remittance Income Simulation: Simulation of a percentage change in remittance income equal to the overall percentage change in remittances for households receiving them.

Assumptions and Realities

Many of these steps relied on certain assumptions. For example:

• Step 1 assumed that employment elasticities from pre-crisis data could accurately project job losses during a crisis.
• Step 2 assumed that the probability of working remained similar pre- and post-crisis.
• Step 3 assumed that workers least likely to be employed were most likely to lose their job.
• Step 4 assumed that workers in protected sectors did not suffer income declines, while those in non-protected sectors experienced uniform percentage earnings declines.
• Step 5 assumed uniform remittance income declines for all households receiving them.

Despite these assumptions, these steps provided valuable insights during the unfolding crisis when time was scarce. 

Insights from Post-Impact Data

Our new paper, “From Middle Class to Poverty,” leverages data collected after the initial impact to probe these issues, and generally finds mixed results.

Employment Elasticities and GDP Predictions: Elasticity-based estimates slightly underestimated employment losses in 2020 but were surprisingly accurate, falling within 5% of survey-based job loss in 11 out of 15 countries.  The estimates were least accurate where the pandemic was particularly disruptive, as measured by the stringency of stay-at-home orders and Google mobility reports. Using elasticities from a previous global crisis period may improve projections.

Microsimulations and Distributional Impacts: Microsimulations in five countries showed a decline in the middle class and an increase in poverty, but the magnitudes of these declines varied widely.  This highlights the importance of considering country-specific factors beyond simply GDP or consumption per capita changes and the pre-crisis welfare distribution.

Accuracy of Microsimulations: In Brazil, microsimulation projections were compared to actual survey data collected for 2020 and 2021. The microsimulations underestimated (pre-transfer) income declines, especially for the wealthiest quintiles.  This discrepancy likely arose because the microsimulations did not account for the distributional differences in the decline of labor earnings, as not all workers experienced the same percentage earnings decline.

The Road Ahead

To strengthen modeling tools, more evaluations and rigorous evidence are needed from various settings. Current distribution-neutral poverty estimates are off by an average of 3.7 percentage points, and microsimulations offer promise for improvement. Additionally, considering labor informality can enhance the accuracy of model projections.

The results from Brazil underscore the importance of strengthening microsimulation models and frequent survey data collection to better understand the distributional impacts of crises.  As we continue to navigate the aftermath of COVID-19, it is crucial to refine our tools for real-time analysis and gather more evidence to inform effective policy responses in the face of future crises.