## Will Working from Home Stick in Developing Economies?

## Source details

**Canonical URL:** [Will Working from Home Stick in Developing Economies?](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023112-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023112-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023112-print-pdf.pdf.json)

---

### Data gap and new survey dataset
- A data gap on the shift to working from home in developing economies is addressed with a new online survey dataset compiled by the IMF Research Department covering 10 developing countries: Argentina, Brazil, Ecuador, India, Indonesia, Pakistan, Peru, South Africa, Türkiye, and Vietnam.
- Survey scope and timing:
  - About 500 respondents from each country; total respondents = 5,058 (even distribution across the 10 countries).
  - Fieldwork for the baseline (second) round conducted between October and November 2021.
  - Two rounds of online surveys; return rate from first- to second-round respondents = 13 percent (635 respondents).
- Retrospective periods tied to WHO pandemic declaration on March 11, 2020:
  - Before end-March 2020 (before the onset of the pandemic)
  - Between end-March 2020 and end-March 2021 (during the first year of the pandemic)
  - After end-March 2021 (after one year of the pandemic)

### Measures and construction of telework indicators
- Telework measured two ways:
  - Share of telework days per week: computed as one minus the ratio of commute workdays over total workdays per week, using mid-points of range responses (e.g., 1.5 days for “1-2 days”).
  - Categorical indicator: “Commute only”, “Work from home only”, and “Work from home some days” (the latter two often combined as “some telework”).
- Survey response ranges for usual workdays and commute days: Zero or had never worked; 1-2 days; 3-4 days; 5 days or more.
- Analysis focuses on main income earners; parallel questions asked when respondent was not the main income earner.

### Cross-country variation in telework adoption (key findings)
- Aggregate telework measures (shares of telework days per week, percent, with population weights):
  - Before March 2020: 9.7
  - April 2020–March 2021: 16.9
  - After 2021 (April 2021–October/November 2021): 16.8
  - Change: April 2020–March 2021 = 7.2; After 2021 = -0.1
- Selected country-level telework shares (shares of telework days per week, percent):
  - India: Before March 2020 = 23.4; April 2020–March 2021 = 28.6; After 2021 = 29.4
  - Argentina: Before March 2020 = 9.8; April 2020–March 2021 = 27.4; After 2021 = 19.6
  - Total sample pre-pandemic telework share (for main income earner) = 9.7
- Cross-country heterogeneity:
  - Large increases (>10 percentage points) in Argentina, Ecuador, and Peru (ranging 11.7-17.6 percentage points).
  - Very limited adjustment in Indonesia and Türkiye (ranging 0.7 to 1.2 percentage points).
  - Overall telework levels in these countries were much below the U.S. level in most cases.
- Regression evidence:
  - A multivariate regression on internet access, years of schooling, and log GDP per capita shows about 70 percent (or 60 percent excluding India) of cross-country variation in telework share upon the pandemic is unexplained.
  - Correlations: almost zero correlation (0.08) between pre-pandemic telework levels and the increases upon the pandemic; telework days per week generally positively correlated with population share of internet users as of 2019 and average year of total schooling for people age 15-64 as of 2015. Excluding India, correlation coefficients are 0.46 (internet users) and 0.70 (educational attainment).

### Transitions and worker characteristics
- Transitions across arrangements:
  - Telework jobs increased at the onset of the pandemic; parttime jobs also rose notably.
  - Almost 20 percent of respondents who did not work from home before the pandemic shifted to an arrangement with some telework.
  - Almost half of pre-pandemic teleworkers shifted to fully on-site jobs at the onset of the pandemic.
  - More workers adjusted from fulltime to parttime than adjusted to telework to preserve fulltime employment.
- Groups more likely to increase telework days:
  - Workers with higher educational attainments.
  - Older workers.
  - Those living with elderly people.
  - Workers in jobs easy to conduct from home.
- Other patterns:
  - Workers in industries deemed essential tended to telework less.
  - Male workers telework less than female workers (not statistically significant).
  - Low-income households worked more in arrangements without commuting before the pandemic (particularly Peru, Vietnam); high-income households increased telework days more since the onset.

### Income differentials and severe incidents
- Estimated income differentials associated with telework (percent deviations from 100; significance preserved exactly as presented):
  - Full sample, “Telework: all workdays”: Before = -33.33***; April 2020–March 2021 = -3.22; After = -17.04**.
  - Education: low, “Telework: all workdays”: Before = -56.06***; April 2020–March 2021 = -17.30; After = -33.69*.
  - Education: middle, “Telework: all workdays”: Before = -36.58*; April 2020–March 2021 = -2.78; After = -4.62.
  - Education: high, “Telework: all workdays”: Before = -18.02*; April 2020–March 2021 = 0.12; After = -17.04*.
- Parttime income differentials (full sample, percent deviation from 100):
  - 4 days per week: Before = -4.16; April 2020–March 2021 = 16.19**; After = 4.74.
  - 3 days per week: Before = -19.76***; April 2020–March 2021 = -9.42*; After = -13.40**.
  - 2 days per week: Before = -32.83***; April 2020–March 2021 = -29.39***; After = -28.39***.
- Regression summaries (telework days in the week as dependent variable; selected coefficients, percentage-point effects):
  - Education (main earner) = High: Before = 0.06; April 2020–March 2021 = 5.01***; After = 0.83.
  - Age (main earner) = Older (>50): Before = 1.03; April 2020–March 2021 = 6.07**; After = 7.04***.
  - Initial Income Bin = 1 (lowest): Before = 5.70***; April 2020–March 2021 = 5.57***; After = 4.58**.
  - Full time worker (main earner): Before = -6.96; April 2020–March 2021 = -7.50**; After = -11.35***.
  - Live with household member(s) above age 66: Before = -9.09***; April 2020–March 2021 = 23.06**; After = 22.82**.
  - Hard to work from home: Before = -0.44; April 2020–March 2021 = -0.74**; After = -0.92***.
- Severe incidents (selected total effects, preserved exactly):
  - Telework: all workdays — Lost job: April 2020–March 2021 = -4.73*; After = -2.20.
  - Parttime: 2 days per week — Lost job: April 2020–March 2021 = 1.63; After = 5.59***.
  - Income bin effects (examples):
    - Initial Income Bin = 1 (lowest) — Lost job: April 2020–March 2021 = 11.10***; After = 0.48.
    - Initial Income Bin = 5 (highest) — Lost job: April 2020–March 2021 = -6.15**; After = 3.22.

### Model to disentangle preferences versus profitability
- Model environment and primitives:
  - Static, discrete-time, infinite horizon; closed economy with continuum of heterogeneous households measure one and representative firm with continuum of jobs.
  - Two production technologies k ∈ {F, R}: F requires on-site worker; R requires remote worker.
  - Production with coordination externalities: per-unit output y_k ≝ a_k h_k, where h_k ≝ (L_k / (L_F + L_R))^ρ and ρ ∈ [0,1).
  - Matching frictions: matching function μ(n, N) ≝ (q_0 n)^b N^{1−b}; job-seeker hiring probability q ≝ (q_0 n / N)^b.
  - Nash bargaining (asymmetric) with firm bargaining power b and worker power 1−b; with u = 0 and θ = 1 − b yields closed-form wages w_k = (1 − b) y_k.
  - Job choice probability p_k = β_k y_k^{θ/(1−b)} / ∑_j β_j y_j^{θ/(1−b)}; under θ = 1 − b simplifies to p_k = β_k y_k / ∑_j β_j y_j.
- Key normalizations and calibrated values used in analysis:
  - u = 0; b = 0.3; q_0 = 1; η = b; θ = 1 − b; β_F = 1.
  - Remaining parameters to calibrate: a_F, a_R, β_R, ρ. Illustrative ρ values used: 0 or 0.9.
- Calibration targets mapped to model objects:
  - q from average number of days worked in a week.
  - p_F and p_R from average commuting and telework days among workdays.
  - w_R / w_F from estimated relative income of remote vs on-site workers.
- Closed-form equilibrium (under parameter restrictions):
  - p_F = 1 / { (β_F a_F / β_R a_R)^{1/(ρ−1)} + 1 }, p_R = 1 − p_F.
  - q = (a_F p_F^{1+ρ} + a_R p_R^{1+ρ})^{b/(1−b)}.
  - Endogenous variables: L_k = q p_k; h_k = p_k^ρ; y_k = a_k p_k^ρ; w_k = (1 − b) a_k p_k^ρ.
- Identification and interpretation:
  - β_R can be calibrated from earnings-adjusted odds ratio: β_R = (w_R / w_F)^−1 p_R / p_F (β_R calibration does not depend on ρ).
  - If ρ = 0, relative profitability a_R / a_F equals relative earnings w_R / w_F.
  - Both β_R (preference) and a_R / a_F (profitability) affect p_R; their product determines p_k, but they have different implications for wages and outputs.

### Empirical implications and externalities
- Observed patterns imply an upward shift in workers’ willingness to work remotely at the pandemic onset: telework share increased more than proportional to the reduction in estimated relative income for remote workers (preference shift).
- Education heterogeneity:
  - Low-education workers: calibration suggests a decline in implied telework preferences — low-education groups did not increase telework shares as much as improvements in relative income would imply, pointing to constraints (limited access to technologies, skill mismatch).
  - Evidence: 80 percent of low-education respondents who changed jobs reported new jobs had “less risks” of COVID-19 infection (72-74 percent for other education groups); low-education respondents who reported “less risks” took parttime jobs more frequently than telework jobs.
- Externality amplification:
  - Economy-wide externalities (coordination) can magnify small fundamental differences (internet access, education) into large cross-country differences in telework shares.
  - Example comparison: Brazil vs Vietnam — similar fundamentals but different telework shares; introducing externality parameter ρ = 0.9 brings implied profitability changes closer across the two countries, indicating amplification by externalities.

### Policy implications and obstacles to sustained telework
- Barriers identified for sustaining remote work in developing economies:
  - Skill mismatch.
  - Limited access to telework technologies.
  - Internet access essential to telework.
- Policy-relevant points:
  - If externalities matter, public policy intervention may improve social outcomes by addressing coordination/externality issues and social preferences regarding work arrangements.
  - Mitigating access and skill obstacles could determine whether remote work sustains in developing economies.
  - More research is needed to explore the role of externalities and optimal public interventions.
- Dataset utility:
  - Survey contains variables for basic demographics, employment characteristics, incidents (illness, lack of food, rent difficulties), receipt of public support, and political sentiments — useful for analyzing distributional implications of the pandemic.

*Source: IMF Working Paper excerpt "Will Working from Home Stick in Developing Economies?" (wpiea2023112-print-pdf).*

### 1. Introduction ........................................................................................................

### 1. Introduction

### Data gap and new survey dataset
- A data gap exists on how significant the shift to working from home was in developing economies.
- A new online survey dataset compiled by the IMF Research Department covers 10 developing countries: Argentina, Brazil, Ecuador, India, Indonesia, Pakistan, Peru, South Africa, Türkiye, and Vietnam.
- The survey includes about 500 respondents from each country and a total of 5,058 respondents (even distribution across the 10 countries), conducted between October and November 2021.
- The questionnaire gathered retrospective information on workdays and commuting for three periods tied to the WHO’s declaration of a global pandemic on March 11, 2020:
  - Before end-March 2020 (before the onset of the pandemic)
  - Between end-March 2020 and end-March 2021 (during the first year of the pandemic)
  - After end-March 2021 (after one year of the pandemic)

### Measures and construction of telework indicators
- Telework measured two ways:
  - Share of telework days per week: computed as one minus the ratio of commute workdays over total workdays per week, taking the mid-point of range responses (e.g., 1.5 days for “1-2 days”).
  - Categorical indicator: “Commute only”, “Work from home only”, and “Work from home some days” (the latter two often combined as “some telework”).
- Survey questions on usual workdays and usual commute days used multiple-choice ranges: Zero or had never worked; 1-2 days; 3-4 days; 5 days or more.
- If respondent was not the main income earner, parallel questions were asked for the main income earner; analysis focuses on main income earners.

### Cross-country variation in telework adoption (key findings)
- Substantial cross-country differences in the degree of the shift to working from home were observed.
- Telework increases:
  - Large increases of more than 10 percentage points occurred in Argentina, Ecuador, and Peru (ranging 11.7-17.6 percentage points).
  - Very limited adjustment in Indonesia and Türkiye, ranging from 0.7 to 1.2 percentage points.
- Pre-pandemic versus pandemic-level observations:
  - In most cases, overall telework levels in these countries were much below the U.S. level.
  - Parttime jobs increased in many cases, likely as an alternative adjustment to the pandemic situation.
  - Telework days increased more for workers with higher educational attainments.
- Correlations:
  - Almost zero correlation (0.08) between pre-pandemic telework levels and the increases upon the pandemic.
  - Telework days per week (before and after onset) were generally positively correlated with population share of internet users as of 2019 and average year of total schooling for people age 15-64 as of 2015. When excluding India, the correlation coefficients are 0.46 and 0.70 for internet users and educational attainment, respectively.

### Income differentials and dynamics
- Remote work was prevalent for low-income households before the pandemic in many surveyed countries (particularly Peru, Vietnam).
- Estimated income differentials:
  - An estimated income discount for home-based work was sizable as of 2019, at 33 percent.
  - This discount almost disappeared at the onset of the pandemic and then partially reversed to -17 percent in late 2021.
- These estimated associations align with a strong increase in the telework share by high-income households.

### Model to disentangle preferences versus profitability
- A simple equilibrium model is developed to disentangle telework choice into:
  - Preference shifts toward telework (“telework adoption”).
  - Choices based on profitability differences between telework and on-site jobs (“telework substitution”).
- Key model features and assumptions:
  - Built on the idea of Bick, Blandin, and Mertens (2022) but assumes a search-matching friction in the labor market, tying job profitability to workers’ earnings.
  - Telework choice emerges from a comparison between job profitability and workers’ willingness to work remotely.
  - Under simplifying parametric assumptions, the model can be calibrated using observed telework share, total employment share, and an estimated relative income from telework versus on-site work.
- Externalities and coordination:
  - The model incorporates externalities in profitability depending on the economy-wide share of the same work modality:
    - Local agglomeration spillovers may increase profitability of in-office jobs when more jobs are on-site.
    - Strategic complementarities may make telework jobs more viable if counterpart firms or clients adopt remote arrangements.
  - Such coordination mechanisms can induce diverse telework responses to small differences in fundamentals (e.g., internet access, educational attainment), implying that sustainability of telework could hinge on relatively small observable differences.

### Contributions and implications
- The paper narrows the data gap on telework in developing economies by providing direct survey evidence, complementing studies that impute telework ability from U.S. O*NET data.
- Policy-relevant implications noted:
  - Whether working from home will persist depends on coordination mechanisms and small differences in fundamentals such as internet access and educational attainment.
  - Working from home has implications for productivity, inequality, resilience, and inflation pressures.

*IMF WORKING PAPERS — Will Working from Home Stick in Developing Economies? — 1. Introduction. Source: IMF Research Department, survey and model summary; canonical URL: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023112-print-pdf.pdf*

### 0.41 and 0.69, respectively). But a simple multivariate regression on these two features and the logarithm of

### Will Working from Home Stick in Developing Economies?

### Cross-country variation in telework uptake and persistence
- A simple multivariate regression on internet access, years of schooling, and the logarithm of GDP per capita shows that about 70 percent (or 60 percent excluding India) of the cross-country variation in the telework share upon the pandemic is unexplained.
- After one year from the onset of the pandemic:
  - Telework levels were mostly kept unchanged in many countries.
  - Argentina reversed its largest increase among the 10 countries by 8 percentage points, while maintaining the share of telework days at 20 percent.
  - India’s telework level did not change much despite widespread Delta infections in April-June 2021.
  - Vietnam increased the share of telework days to 23 percent.
- Variations unexplained by fundamentals remain at about 70 percent (reducing to about 40 percent if India is excluded).

### Transitions across work arrangements during the pandemic
- Telework jobs increased at the onset of the pandemic, and parttime jobs also rose notably.
- Almost 20 percent of respondents who did not work from home before the pandemic shifted to an arrangement with some telework.
- Almost half of pre-pandemic teleworkers shifted to fully on-site jobs at the onset of the pandemic.
- Many workers shifted to parttime jobs; more than two thirds of parttime jobs were on-site.
- There were more workers who adjusted from fulltime to parttime than those who adjusted to telework to preserve fulltime employment.
- Possible explanation: telework prior to the pandemic in developing economies may have been different in essence, requiring adjustments under forced social distancing.

### Worker characteristics associated with telework increases
- Groups more likely to increase telework days: workers with higher educational attainments, older workers, those living with elderly people, and workers in jobs easy to conduct from home.
- Workers in industries deemed essential tended to telework less.
- Male workers telework less than female workers (not statistically significant).
- Low-income people worked more in arrangements without commuting before the pandemic (particularly Peru, Vietnam), but high-income households increased telework days more since the onset.

### Income differentials by work arrangement
- Estimated income discount associated with working remotely:
  - Approximately 33 percent before the pandemic in the 10 developing economies in the sample.
  - Comparable historical U.S. figure: 25-27 percent in 1980.
  - At the onset of the pandemic, the discount became almost zero (or just 3 percent, without statistical significance).
  - Later in 2021 the discount widened again to 17 percent.
- Income discount is larger and more persistent for less educated workers.
- Parttime work income differentials:
  - Estimated income discount for two workdays per week was largely negative with statistical significance.
  - For more than three workdays per week the discount was not as severe.
  - For four days per week, it turns positive since the pandemic (possible compositional effect: forced distancing reduced fulltime workers’ days, mechanically increasing income associated with those working less than 5 days).
- Telework was associated with a lower chance of job loss at the onset, but not significantly associated with incidents such as suffering from any kind of illness, lack of food, or rent delinquencies.
- Low income levels were associated with a higher chance of severe incidents; high income levels with a lower chance.
- Working parttime tends to be positively associated with occurrence of severe incidents, implying either income effects outweighed reduced exposure benefits or reverse causality.

### Model separating preferences and profitability of telework
- Model framework:
  - Search-matching friction between representative firms and heterogeneous workers over work arrangements.
  - Two production technologies: F (on-site) and R (remote).
  - Nash bargaining selects work arrangement and wage after matching.
  - Profitability parameters: a_F and a_R. Externalities can make on-site (or remote) work more profitable as their shares increase.
  - Worker preference distribution parameters: β_F (normalized to 1) and β_R (willingness to work remotely relative to on-site).
- Calibration strategy:
  - β_R can be calibrated to an income-adjusted odds ratio of choosing remote relative to on-site work.
  - Excess telework shares not proportional to income differentials are attributed to preferences (β_R), i.e., residual factors such as regulatory requirements, skill mismatches, and access to telework technologies.
  - COVID-19 containment measures can affect both β_R and relative profitability a_R/a_F.

### Empirical implications from the model
- Observed telework experiences imply an upward shift in workers’ willingness to work remotely at the onset of the pandemic: telework share increased more than proportional to the reduction in estimated relative income for remote workers.
- Education heterogeneity:
  - Low education category shows an unintuitive downward shift in implied telework preferences: they did not increase telework shares as much as implied by improvements in relative income—suggesting constraints (limited access to technologies, skill mismatch) rather than preferences.
  - Evidence: 80 percent of low-education respondents who changed jobs reported their new jobs had “less risks” of COVID-19 infection, compared to 72-74 percent for other education groups; low-education respondents who reported “less risks” took parttime jobs more frequently than telework jobs.
- Externality amplification:
  - Cross-country differences may be amplified by economy-wide externalities.
  - Example: Brazil and Vietnam have similar observed fundamentals (internet access, schooling) but different telework shares (Vietnam higher than Brazil).
  - Baseline calibration without externality shows large differences in implied relative profitability changes (Brazil nearly 4x increase at pandemic onset vs. 20 percent for Vietnam).
  - Introducing externality (ρ=0.9) brings the implied changes closer across the two countries, indicating externality can magnify small fundamental differences into large outcomes.

*Source: IMF Working Paper excerpt from "Will Working from Home Stick in Developing Economies?"*

### 4. Conclusion

### wpiea2023112-print-pdf - 4. Conclusion

### Survey evidence on telework adoption in developing economies
- The IMF–Nielsen online survey fills an information gap on telework shifts in developing economies during the pandemic by covering 10 developing countries.
- Sample and fieldwork details:
  - Two rounds of online surveys; the second round conducted in October and November 2021 is used as the baseline.
  - The return rate from first- to second-round respondents was 13 percent (635 respondents).
  - For each round, the number of respondents per country is about 500. Total sample count (2nd round) = 5,058.
  - The 10 countries: Argentina, Brazil, Ecuador, India, Indonesia, Pakistan, Peru, South Africa, Türkiye, and Vietnam.
- Aggregate telework measures (Appendix Table A3, shares of telework days per week, percent, with population weights):
  - Before March 2020: 9.7
  - April 2020–March 2021: 16.9
  - After 2021 (April 2021–October/November 2021): 16.8
  - Change: April 2020–March 2021 = 7.2; After 2021 = -0.1
- Key country-level telework statistics (selected from Appendix Table A3, shares of telework days per week, percent):
  - India: Before March 2020 = 23.4; April 2020–March 2021 = 28.6; After 2021 = 29.4
  - Argentina: Before March 2020 = 9.8; April 2020–March 2021 = 27.4; After 2021 = 19.6
  - Total sample pre-pandemic telework share (for main income earner) = 9.7

### Model-based analysis and interpretation
- A simple parsimonious model was developed to separate two drivers of telework choices:
  - Workers’ preferences for remote work.
  - Relative profitability across work arrangements.
- Main modeling result:
  - Clear increase in preferences toward working remotely for highly educated workers, corroborating findings for advanced and emerging market economies (Aksoy and others 2022).
- For workers with low education, results are mixed:
  - Model calibration indicates a decline in telework preferences for low education workers.
  - A different portion of the survey suggests the preference shift toward telework would have been slightly higher for the low education category.
  - The mixed result likely reflects obstacles not captured by the model (e.g., skill mismatch, limited access to telework technologies, internet access).

### Barriers, externalities, and policy implications
- Identified obstacles to sustaining remote work in developing economies:
  - Skill mismatch.
  - Limited access to telework technologies.
  - Internet access is essential to telework (cited examples in the text).
- Economy-wide externalities among the same work arrangements may explain cross-country heterogeneity beyond observed fundamentals.
- Policy implications and possibilities:
  - If externalities are important, there may be scope for public policy intervention to improve social outcomes based on social preferences regarding work arrangements.
  - Mitigating access and skill obstacles could be key to whether remote work will sustain in developing economies.
  - More research is needed to explore the role of externalities and optimal public interventions.

### Dataset utility and additional variables for future research
- The online survey dataset contains variables useful for analyzing distributional implications of the pandemic in developing economies:
  - Basic demographics and employment characteristics.
  - Responses on whether households suffered from illness, lack of food, or difficulty in paying rent.
  - Variables on receipt of public support and political sentiments.
- The dataset can help fill information gaps on economic behaviors and responses during the pandemic in developing economies.

*Source: wpiea2023112-print-pdf - 4. Conclusion*

### Appendix Figure A5. Not only telework but also parttime jobs increased during the pandemic

### Appendix Figure A5. Not only telework but also parttime jobs increased during the pandemic

### Figure and notes
- Chart type: Sankey plot (produced by the Stata command sankey_plot (Rios-Avila 2022)).
- Data source: IMF COVID-19 Custom Tracker (2nd round).
- Year labels for exposition:
  - “2019” corresponds to before March 2020
  - “2020” corresponds to April 2020–March 2021
  - “2021” corresponds to April 2021–October/November 2021
- Numbers in parentheses in charts indicate shares to the total in percent (with population weights).

### Telework and part-time job trends (summary of Appendix Figures A5–A7)
- Both telework and part-time work shares increased during the pandemic periods covered (“2020” and “2021”) relative to “2019”.
- Appendix Figure A6 examines the relationship between telework share in “2021” and changes in country-level weighted average of workdays in a week from “2020” to “2021”:
  - Panel A: unconditional on working (full sample, captures extensive and intensive margins).
  - Panel B: conditional on working (restricted to those who worked at least one day, focuses on intensive margin).
- Appendix Figure A7 examines how COVID-19 infection risk considerations when changing jobs relate to telework and part-time shares, disaggregated by education; charts use main income earner work-arrangement questions (Q25a–Q26c or Q8a–Q9c) and job-change/infection-risk questions (Q24/Q7, Q21/Q4).

### Characteristics associated with telework (Appendix B — regression overview)
- Estimation method: linear regressions with region-level fixed effects (93 regions across 10 countries), standard errors clustered at region level; alternative setups tested for robustness (country fixed effects, clustering at city level, telework dummy, exclusion of top income bin).
- Key control/constructed variables:
  - Urban dummy: regions with habitants over 300,000 (United Nations 2018).
  - Indexes of “essential sector”, “physical proximity”, and “hard to work from home” from Mongey, Pilossoph, and Weinberg (2021) via industry-occupation matches.

### Telework determinants (selected coefficients from Appendix Table B1)
- Telework days in the week (main earner) — sample averages and R-squared:
  - Number of Respondents (observations): 3,889 (Before March 2020), 3,771 (April 2020 to March 2021), 3,534 (After April 2021).
  - Average Dep. Var.: 0.10 (Before), 0.16 (April 2020–March 2021), 0.16 (After April 2021).
  - Within R squared (overall R-squared): 0.06 (0.16) (Before), 0.05 (0.16) (April 2020–March 2021), 0.07 (0.19) (After).
- Selected coefficient estimates (telework days in the week as dependent variable; values in percentage points):
  - Education (main earner) = High: 0.06 (Before), 5.01*** (April 2020–March 2021), 0.83 (After).
  - Age (main earner) = Older (>50): 1.03 (Before), 6.07** (April 2020–March 2021), 7.04*** (After).
  - Initial Income Bin = 1 (lowest): 5.70*** (Before), 5.57*** (April 2020–March 2021), 4.58** (After).
  - Full time worker (main earner): -6.96 (Before), -7.50** (April 2020–March 2021), -11.35*** (After).
  - Live with household member(s) above age 66: -9.09*** (Before), 23.06** (April 2020–March 2021), 22.82** (After).
  - Hard to work from home: -0.44 (Before), -0.74** (April 2020–March 2021), -0.92*** (After).

### Income levels and telework (Appendix Table B2 — selected coefficients)
- Dependent variable: natural log of annual household income (observations and averages):
  - Number of Respondents (observations): 3,888 (Before), 3,771 (April 2020–March 2021), 3,533 (After).
  - Average Dep. Var.: 12.74 (Before), 12.75 (April 2020–March 2021), 12.91 (After).
  - R-squared: 0.915 (Before), 0.903 (April 2020–March 2021), 0.910 (After).
- Selected coefficient estimates (telework and other controls):
  - Telework days in the week (From 0 to 1 main earner): -0.41*** (Before), -0.03 (April 2020–March 2021), -0.19** (After).
  - Education (main earner) = Low: -0.46*** (Before), -0.42*** (April 2020–March 2021), -0.45*** (After).
  - Education (main earner) = High: 0.27*** (Before), 0.31*** (April 2020–March 2021), 0.24*** (After).
  - Workdays in the week (From 0 to 1 main earner): 0.89*** (Before), 1.25*** (April 2020–March 2021), 0.95*** (After).
  - Essential sector: 0.22*** (Before), 0.31*** (April 2020–March 2021), 0.21** (After).
  - Hard to work from home: -0.06*** (Before), -0.05*** (April 2020–March 2021), -0.04*** (After).

### Estimated income differentials by work arrangement and education (Appendix Table B3)
- Panel A. Full sample (percent deviation from 100; significance preserved):
  - Telework: all workdays: -33.33*** (Before), -3.22 (April 2020–March 2021), -17.04** (After).
  - Parttime: 4 days per week: -4.16 (Before), 16.19** (April 2020–March 2021), 4.74 (After).
  - Parttime: 3 days per week: -19.76*** (Before), -9.42* (April 2020–March 2021), -13.40** (After).
  - Parttime: 2 days per week: -32.83*** (Before), -29.39*** (April 2020–March 2021), -28.39*** (After).
- Panel B. Education: low:
  - Telework: all workdays: -56.06*** (Before), -17.30 (April 2020–March 2021), -33.69* (After).
  - Parttime: 2 days per week: -33.90** (Before), -20.56 (April 2020–March 2021), -43.26*** (After).
- Panel C. Education: middle:
  - Telework: all workdays: -36.58* (Before), -2.78 (April 2020–March 2021), -4.62 (After).
  - Parttime: 4 days per week: -20.30* (Before), 37.37** (April 2020–March 2021), 9.03 (After).
- Panel D. Education: high:
  - Telework: all workdays: -18.02* (Before), 0.12 (April 2020–March 2021), -17.04* (After).
  - Parttime: 2 days per week: -32.63*** (Before), -30.12*** (April 2020–March 2021), -26.25*** (After).
- Notes on interpretation:
  - “Telework: all workdays” = worker does not commute at all during workdays.
  - “Parttime” = worked less than 5 days per week.
  - Estimates converted to percent changes using exponential function (Stata nlcom); presented as deviation from 100 (e.g., -33.3 percent corresponds to exp(coef) ≃ 0.667).

### Severe incidents related to the pandemic (Appendix Table B4 — selected estimates)
- Dependent variables: dummies for incidents during respective periods (Lost job, Felt ill, Lack of food, Rent payment delinquencies).
- Sample sizes and averages:
  - No. Obs.: 3,771 (April 2020 to March 2021) and 3,534 (After April 2021) for many outcomes; averages vary by outcome.
  - Average Dep. Var.: Lost job: 0.22 (April 2020 to March 2021), 0.15 (After); Felt ill: 0.28, 0.26; Lack of food: 0.24, 0.14; Rent payment delinquencies: 0.27, 0.22.
- Estimated total effects (selected coefficients, preserved exactly):
  - Telework: all workdays — Lost job: -4.73* (April 2020 to March 2021), -2.20 (After); Felt ill: 0.05, 3.29; Lack of food: 0.04, 0.89; Rent payment delinquencies: 0.72, -0.25.
  - Parttime: 2 days per week — Lost job: 1.63, 5.59***; Felt ill: 3.24, 2.93; Lack of food: 2.35, 8.53***; Rent payment delinquencies: 0.72, 9.29***.
- Income bin effects (examples):
  - Initial Income Bin = 1 (lowest) — Lost job: 11.10*** (April 2020 to March 2021), 0.48 (After); Lack of food: 9.73***, 3.19; Rent payment delinquencies: 7.52**, -3.43.
  - Initial Income Bin = 5 (highest) — Lost job: -6.15** (April 2020 to March 2021), 3.22 (After); Lack of food: -6.74***, 1.63; Rent payment delinquencies: -3.17, -5.66**.

### Key takeaways and interpretation
- Telework adoption rose during the pandemic, alongside increases in part-time work.
- Telework was associated with substantial negative income differentials prior to the pandemic (e.g., -33.33*** percent for full telework in the full sample before March 2020) but the income discount narrowed sharply at the onset of the pandemic (e.g., -3.22 percent April 2020–March 2021) and remained partially persistent after April 2021 (e.g., -17.04** percent).
- Income discounts from telework were largest for workers with low education (e.g., -56.06*** percent before March 2020 for full telework) and diminished with higher education.
- Part-time work is also associated with income penalties, especially for fewer workdays per week (e.g., -32.83*** percent for 2 days per week in the full sample before March 2020).
- Telework reduced the probability of job loss during the first year of the pandemic (telework: all workdays estimate -4.73* for April 2020–March 2021), but had limited association with other severe incidents.
- Lower initial income bins experienced higher frequencies of severe incidents; higher income bins were better able to cope with incidents except for falling ill.

*Source: IMF COVID-19 Custom Tracker (2nd round) and Appendix material for “Will Working from Home Stick in Developing Economies?” (IMF Working Paper).*

### Appendix Table B1 are used, except for the inclusion of the telework share. “Initial Income Bin” is an approximated hous

### Appendix C. Model details

### Model environment and structure
- Time is discrete, the time horizon is infinite, and decision making is static.
- The economy is closed with a continuum of heterogeneous households of measure one and a representative firm operating a continuum of jobs.
- Two production technologies: k ∈ {F, R} where F requires an on-site worker and R requires a remote worker.
- Externality in production efficiency depends on economy-wide shares of on-site or telework arrangements (strategic complementarities).
- Coordination cost elasticity parameter: ρ ∈ [0,1). No externality if ρ = 0; ρ = 1 is excluded to avoid continuum/no equilibrium.

### Production technology (key expressions)
- Per-unit output for technology k:
  - y_k ≝ a_k h_k, where h_k ≝ (L_k / (L_F + L_R))^ρ, k ∈ {F, R}, ρ ∈ [0,1)
- Aggregate output:
  - Y ≝ y_F L_F + y_R L_R = a_F h_F L_F + a_R h_R L_R
  - Potential output without coordination costs: Y_0 ≝ a_F L_F + a_R L_R
  - Coordination reduces output because (L_k / (L_F + L_R))^ρ ∈ [0,1] for k ∈ {F, R}

### Preferences
- Households obtain linear utility from consumption c and idiosyncratic utility x_k from working in job type k.
- Joint distribution of (x_F, x_R): G(x_F, x_R) ≝ exp(−β_F x_F^−θ − β_R x_R^−θ) with density g(x_F, x_R).
- Each x_k independently follows a Fréchet distribution with parameter θ.
- All jobs last one period; saving is abstracted; u is the unemployment value taken as u = 0 (simplifying assumption so every household applies for a job).
- Value of applying for a job: W(x_F, x_R) = q w_{k*} x_{k*} > 0 = u, where k* is the job type chosen in Nash bargaining.

### Labor market with frictions
- Matching function (Cobb–Douglas with elasticity b):
  - μ(n, N) ≝ (q_0 n)^b N^{1−b}
- Job-seeker hiring probability:
  - q ≝ μ(n, N) / N = (q_0 n / N)^b
- Vacancy fill probability:
  - μ(n, N) / n = μ(n, N) / N × (N / n) = q (N / n)
- Nash bargaining (asymmetric) with firm bargaining power b and worker power 1−b; Hosios condition applied.
- Conditional Nash product maximized yields wage (with u = 0):
  - w_k = (1 − b) y_k
- Job type chosen if it maximizes y_k^{1/(1−b)} x_k (ties have probability zero; if tie then random ½ but probability zero in practice).
- Probability a matched job is type k:
  - p_k ≝ P[type k chosen] = β_k y_k^{θ/(1−b)} / ∑_{j∈{F,R}} β_j y_j^{θ/(1−b)}
- Job posting cost: ex-ante cost η in units of final consumption goods; free entry condition:
  - q_N n ∑_{k∈{F,R}} p_k {y_k − w_k} = η

### Equilibrium definition and conditions
An equilibrium is wages (w_F, w_R) > 0 and allocations satisfying:
1. Household optimization: N = 1 under u = 0 and W(x_F, x_R) = q w_{k*} x_{k*} > 0 = u.
2. Free entry: q_N n ∑_{k} p_k {y_k − w_k} = η, with normalization η = b.
3. Job-match rate consistency: q = (q_0 n / N)^b ≤ 1.
4. Nash bargaining (with u = 0 and θ = 1 − b):
   - w_k = (1 − b) y_k
   - p_k = β_k y_k^{θ/(1−b)} / ∑_{j} β_j y_j^{θ/(1−b)} = β_k y_k / ∑_{j} β_j y_j (given θ = 1 − b)
   - k* determined by comparison of y_R^{1/(1−b)} x_R and y_F^{1/(1−b)} x_F
5. Market clearing:
   - L_k = q N p_k for each k ∈ {F, R}
   - ∑_k w_k L_k + (1 − q) N u + η ∑_k L_k = Y, with u = 0 and Y ≝ ∑_k y_k L_k = a_F h_F L_F + a_R h_R L_R
   - Coordination functions: h_k ≝ (L_k / (L_F + L_R))^ρ

- With normalizations q_0 = 1, η = b, N = 1, and θ = 1−b, equilibrium simplifies to:
  - p_k = β_k a_k h_k / (β_F a_F h_F + β_R a_R h_R) = β_k a_k p_k^ρ / (β_F a_F p_F^ρ + β_R a_R p_R^ρ)
  - q = (a_F h_F p_F + a_R h_R p_R)^{b/(1−b)} = (a_F p_F^{1+ρ} + a_R p_R^{1+ρ})^{b/(1−b)}

- Solved closed-form equilibrium (unique under parameter restrictions):
  - p_F = 1 / { (β_F a_F / β_R a_R)^{1/(ρ−1)} + 1 }, p_R = 1 − p_F
  - q = (a_F p_F^{1+ρ} + a_R p_R^{1+ρ})^{b/(1−b)}
  - n = Y = q ∑_k a_k p_k^{1+ρ}
  - Endogenous variables:
    - L_k = q p_k
    - h_k = p_k^ρ
    - y_k = a_k p_k^ρ
    - w_k = (1 − b) a_k p_k^ρ
    - N = 1 (u = 0)

### Parameter normalization and calibrations (assumptions used)
- Fixed/calibrated normalizations:
  - u = 0
  - b = 0.3 (i.e., labor share is 0.7)
  - q_0 = 1
  - η = b
  - θ = 1 − b
  - β_F = 1 (normalization)
- Remaining parameters to calibrate: a_F, a_R, β_R, ρ. For illustrative purposes, ρ is set either to 0 or 0.9 in the analysis.
- Data points used for calibration correspond to model objects (q, p_F, p_R, w_R / w_F):
  - (1) average number of days when people work in a week → q
  - (2) average number of commuting days among days people work in a week → p_F
  - (3) average number of telework days among days people work in a week → p_R
  - (4) estimated relative income of remote workers relative to on-site workers → w_R / w_F
- Calibrations derived:
  - w_R / w_F = y_R / y_F = (a_R p_R^ρ) / (a_F p_F^ρ) = p_R β_F p_F / (p_F β_R) ⇒ β_R = (w_R / w_F)^−1 p_R / p_F
  - a_R / a_F = (w_R / w_F) (p_R / p_F)^−ρ
  - a_F determined by:
    - q = (a_F p_F^{1+ρ} + a_R p_R^{1+ρ})^{b/(1−b)} ⇒ a_F = [p_F^{1+ρ} + (w_R / w_F) (p_R / p_F)^{−ρ} p_R^{1+ρ}]^{−1} q^{(1−b)/b}
    - Expressed compactly in the text as:
      - a_F = [ p_F^{1+ρ} + (w_R / w_F) (p_R / p_F)^{−ρ} p_R^{1+ρ} ]^{−1} q^{(1−b)/b}

- Interpretation notes:
  - β_R can be interpreted as an earnings-adjusted odds ratio for remote work relative to on-site work.
  - β_R calibration does not depend on ρ.
  - If ρ = 0, relative profitability a_R / a_F equals relative earnings w_R / w_F.
  - Increasing either β_R or a_R / a_F increases p_R; only their product β_R (a_R / a_F) matters for p_k, but they affect wages and outputs differently.

### Derivations (key steps and expressions)
- Nash bargaining first-order condition (with u = 0 leads to closed form):
  - w_k = (1 − b) y_k
- Conditional Nash product Φ_k (u = 0 and after simplification):
  - Φ_k = b^b (1 − b)^{1−b} y_k (x_k)^{1−b}
  - Equivalent condition for choosing k: compare y_k^{1/(1−b)} x_k across k
- With x_k Fréchet distributed: distribution of y_k^{1/(1−b)} x_k leads to closed-form p_k:
  - p_k = β_k y_k^{θ/(1−b)} / ∑_j β_j y_j^{θ/(1−b)}
  - Under normalization θ = 1 − b this simplifies to:
    - p_k = β_k y_k / ∑_j β_j y_j
- Substituting y_k = a_k p_k^ρ yields fixed-point condition:
  - β_F a_F p_F^ρ / (β_F a_F p_F^ρ + β_R a_R p_R^ρ) = p_F
  - Solving yields interior solution:
    - p_F = 1 / { (β_F a_F / β_R a_R)^{1/(ρ−1)} + 1 }

### Equilibrium employment rate and output (free entry normalization)
- Free entry and normalizations imply n = Y (since η = b and q_0 = 1 and N = 1).
- Matching probability:
  - q = Y^b = (∑_k a_k h_k L_k)^b = (q ∑_k a_k p_k^{1+ρ})^b ⇒ q = (∑_k a_k p_k^{1+ρ})^{b/(1−b)}
- Final equilibrium relation used for calibration:
  - q = (a_F p_F^{1+ρ} + a_R p_R^{1+ρ})^{b/(1−b)}

### Parameter restriction for equilibrium existence
- Existence requires q ≤ 1. Under normalization q = Y^b this reduces to Y ≤ 1 for b > 0.
- Equilibrium condition in (a_F h_F, a_R h_R)-plane forms an ellipse:
  - β_F [a_F h_F − Y^2]^2 + β_R [a_R h_R − Y^2]^2 = Y^2 (β_F + β_R)^4
- For Y = 1, inequality for existence:
  - β_F [a_F h_F − 1/2]^2 + β_R [a_R h_R − 1/2]^2 ≤ (β_F + β_R) / 4
- Points inside the ellipse correspond to equilibria with Y ≤ 1. A unique mapping exists between (a_F h_F, a_R h_R) and equilibrium outcomes (Y, p_F, p_R) for β_F, β_R > 0 and ρ < 1.
- Identification: given (a_F h_F, a_R h_R) and (β_F, β_R, ρ), a_F and a_R are uniquely determined:
  - a_F = [1 + (β_R a_R h_R) / (β_F a_F h_F)]^ρ a_F h_F (expression shows mapping between variables to identify a_F)

*IMF WORKING PAPERS — Appendix C, "Model details", as provided.*

### References

### References (Will Working from Home Stick in Developing Economies?)

### Major thematic coverage in the references
- COVID-19 and labor markets, including lockdown impacts, labor market dynamics, and contact-intensive industries.
- Working from home / remote work: prevalence, persistence, productivity, wage effects, and city-level implications.
- Digitalization and internet adoption: trends during COVID-19, digital infrastructure investment needs, and determinants of digital adoption.
- Inequality, poverty, and socioeconomic impacts of COVID-19 across advanced and developing economies.
- Methods and data sources: occupation and industry classifications, high-frequency phone surveys, time-use surveys, and global datasets (e.g., International Financial Statistics, World Development Indicators, povcalnet).
- Theoretical and empirical foundations: matching and search models, trade and geography, bargaining problems, probability theory, and stochastic methods.

### Key cited works and identifiers (selected exact items as given)
- Aksoy, Cevat Giray, Jose Maria Barrero, Nicholas Bloom, Steven J. Davis, Mathias Dolls, and Pablo Zarate. 2022. “Working from home around the world.” BFI Working Paper No. 2022-124, September.
- Alon, Titan, Matthias Doepke, Jane Olmstead-Rumsey, and Michèle Tertilt. 2020. “The Impact of COVID-19 on Gender Equality.” National Bureau of Economic Research Working Paper 26947.
- Alonso, Cristian, Tanuj Bhojwani, Emine Hanedar, Dinar Prihardini, Gerardo Una and Kateryna Zhabska. 2023. “Stacking up the Benefits: Lessons from India’s Digital Journey.” IMF Working Paper 23/78.
- Amaglobeli, David, Mengfei Gu, Mariano Moszoro, Yue Zhou, and Patricia Escalante. 2023. “Internet Adoption Trends during COVID-19” IMF Note 2023/001.
- Barrero, Jose Maria, Nicholas Bloom, and Steven J. Davis, 2021a. “Why working from home will stick.” National Bureau of Economic Research Working Paper 28731.
- Barrero, Jose Maria, Nicholas Bloom, and Steven J. Davis, 2021b. “Internet Access and its Implications for Productivity, Inequality, and Resilience.” In Rebuilding the Post-Pandemic Economy.
- Barrero, Jose Maria, Nicholas Bloom, Steven J. Davis, Brent Meyer, and Emil Mihaylov. 2022. “The Shift to Remote Work Lessens Wage-Growth Pressures.” Retrieved at: https://wfhresearch.com/wp-content/uploads/2022/06/The-Shift-to-Remote-Work-Lessens-Wage-Growth-Pressures-20-June-2022.pdf, accessed on July 22, 2022.
- Barro, Robert, and Jong-Wha Lee. 2013. “A New Data Set of Educational Attainment in the World, 1950-2010.” Journal of Development Economics, 104, 184-198. Dataset v.3.0, updated in September 2021.
- Bick, Alexander, Adam Blandin, and Karel Mertens. 2022. “Work from Home Before and After the COVID-19 Outbreak.” September, Working Paper 2017, Federal Reserve Bank of Dallas.
- Brussevich, Mariya, Era Dabla-Norris, and Salma Khalid. 2020. “Who will Bear the Brunt of Lockdown Policies? Evidence from Tele-workability Measures Across Countries.” IMF Working Paper 20/88.
- Dingel, Jonathan I., and Brent Neiman. 2020. “How many jobs can be done at home?” BFI White Paper, June, Becker Friedman Institute.
- Egger, D., et al. 2021. “Falling living standards during the covid-19 crisis: Quantitative evidence from nine developing countries.” Science Advances 7(6), eabe0997.
- Jaumotte, Florence, et al. 2023. “Digitalization during the Covid-19 Crisis: Implications for Productivity and Labor Markets in Advanced Economies.” Staff Discussion Note SDN2023/003.
- Kumar, Utkarsh, David Amaglobeli, and Mariano Moszoro. 2023. “Determinants and Social Dividends of Digital Adoption.” IMF Working Paper 23/65.
- Oughton, Edward, J., David Amaglobeli, and Mariano Moszoro. 2023. “Estimating Digital Infrastructure Investment Needs to Achieve Universal Broadband.” IMF Working Paper 23/27.
- World Bank. 2020. Purchasing Power Parities and the Size of World Economies: Results from the 2017 International Comparison Program. Doi: 10.1596/978-1-4648-1530-0.
- World Bank. 2022a. World Development Indicators (database). https://datacatalog.worldbank.org/dataset/world-development-indicators, updated on July 25, 2022, accessed on July 29, 2022.
- International Labour Organization (ILO). 2012. International Standard Classification of Occupations, 2008: ISCO-08.
- United Nations. 2008. International Standard Industrial Classification of All Economic Activities, Revision 4. Statistical papers, Series M No. 4/Rev.4.
- Rios-Avila, Fernando. 2022. "SANKEY_PLOT: Stata module to produce Sankey plots," Statistical Software Components S459067, Boston College Department of Economics, revised 24 Jun 2022.

### Data sources and repositories cited
- International Financial Statistics (database), accessed in July 2022.
- povcalnet (Global Poverty Monitoring Technical Note, No. 9) — povcalnet command for estimating global poverty in Stata.
- World Development Indicators (database), accessed July 29, 2022.
- Poverty and Inequality Platform (database), updated June 22, 2022, accessed in July 2022.
- High-frequency/household monitoring surveys and rapid surveys referenced for country-specific COVID-19 evidence.

*References section of "Will Working from Home Stick in Developing Economies? Working Paper No. WP/2023/112"*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023112-print-pdf.pdf_