## wpiea2022089-print-pdf

## Source details

**Canonical URL:** [wpiea2022089-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022089-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022089-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022089-print-pdf.pdf.json)

---

### Introduction — pandemic impact, methods, and main findings
- Pandemic labor-market impacts in Japan:
  - decrease in the number of employed persons by about 1 million (about a 1 percent drop in the ratio of employment to population) during the initial pandemic shock.
  - average earnings declined by 1.4 percent in 2020 due to lower overtime and bonus payments.
- Heterogeneous adverse impacts concentrated on:
  - women, younger age groups, non-regular workers, self-employed, low-income workers.
- Primary driver of disproportionate impacts:
  - greater representation of these groups in the most affected industries, especially contact-intensive services.
- Study design and methods:
  - Combines macro and large micro panel data (employment status, earnings, working hours).
  - Uses machine learning techniques (Double Machine Learning; appendices reference DML and lasso selection).
- Key qualitative findings:
  - part-time workers disproportionately affected; women, young, and low-income workers account for a large share of part-timers.
  - worse earnings outcomes associated with small firms, low education, working while studying, and part-time work.
  - telework ability and training/upskilling generally beneficial.
- Pre-pandemic labor-market context and reform efforts:
  - working-age (15-64) population declined by slightly more than 12 million since 2000, from 87 million to 75 million.
  - female employment rate stood at 71.5 percent as of end 2019 (OECD average 61.5 percent).
  - unemployment rate of 2.4 percent in 2019; job offer to applicant ratio of 1.6 in 2019.
  - non-regular employment rose from 16% in 1995 to 32% in 2019; women account for about 70 percent of non-regular workers.
  - Work Style Reform (WSR): discussions began in 2016; legislation passed in 2018.

### COVID-19 developments, policy responses, and program usage
- Pandemic timeline and major responses:
  - first confirmed case on January 15, 2020.
  - state of emergency declared on April 7, 2020 (initially 7 prefectures), extended nationwide on April 16, 2020; lifted entirely on May 25, 2020.
  - tourism promotion program started July 2020 (excluding Tokyo), extended to Tokyo in October 2020, suspended in December 2020.
  - additional stimulus on December 8, 2020: 73.6 trillion yen (about 13% of GDP).
  - additional measures on November 19, 2021: 78.9 trillion yen (about 14% of GDP).
- Economic support packages:
  - April 20, 2020 package: 117 trillion yen (about 21% of GDP).
- Employment Adjustment Subsidy (EAS) pandemic-era changes (starting April 2020):
  - coverage expanded to all employees (Emergency Employment Safety Subsidy created for those not in employment insurance).
  - grant rate: 1/2 to 2/3 for large companies and 2/3 to 4/5 for SMEs; if companies make no dismissal, rate became 3/4 for large companies and 9/10 for SMEs.
  - upper limit increased from 8,330 yen to 13,500 or 15,000 yen per day.
  - VET (vocational education and training) subsidy: one-day amount increased from 1,200 yen to 2,400 yen for SMEs and 1,800 yen for large companies; VET coverage and flexibility expanded.
  - Ministry of Health, Labor, and Welfare (MHLW) simplified application procedures.
- EAS payments and patterns:
  - total EAS payments during the pandemic far exceeded those during the GFC, peaking close to 600 bn yen in August 2020.
  - payments decreased to roughly 200 billion yen per month in December 2020 and remained at a similar level until late 2021.
  - FY2020 cumulative EAS amount: 3.1 trillion yen (about seven times higher than EAS payments during the entire GFC period).
  - manufacturing, accommodation and restaurants, and wholesale and retail received more than half of EAS payments.
  - government estimates: EAS lowered the unemployment rate by about 2.6 percentage points during April to October 2020.
  - data on number of employees benefiting from EAS not available; EAS payments used as proxy for usage.
- Employment Insurance (EI) adjustments:
  - pre-pandemic EI basic benefits: 50-80% of previous income with a cap for 90-330 days depending on age, contributions, and reason for unemployment.
  - EI eligibility: employees working at least 20 hours per week (including non-regular workers).
  - EI payments temporarily extended by an additional 60 days (with scope variations); in specified circumstances a person receives additional 30 days instead of 60.
  - FY2020 total payment of EI basic benefits increased by about 27 percent compared with the previous fiscal year.
  - total payment for FY2020 is about 42 percent lower than FY2009, reflecting EI’s smaller role in this crisis relative to GFC.

### Macro evidence — employment, earnings, composition, and household aggregates
- Employment and labor-force changes:
  - number of employed persons decreased by about 1 million from March to April 2020.
  - many left labor force instead of filing as unemployed; unemployment rate initially almost unchanged.
  - number of employed persons recovered gradually after April 2020 but remained substantially below pre-crisis level as of late 2021.
  - number registered as unemployed rose gradually until late 2020 and remains about 1.3 times higher than before the pandemic.
  - number of people not in the labor force (NILF) declined to pre-pandemic level in November 2020, but increased again in 2021 during infection waves.
- Earnings and composition:
  - average annual earnings declined by 1.4 percent in 2020.
  - full-time workers: -1.5 percent.
  - part-time workers: -0.6 percent.
  - temporary workers experienced decline in base pay, partially offset by bonuses.
- Industry and worker-type heterogeneity:
  - most affected: contact-intensive services (accommodation and restaurants; living-related and amusement services).
  - manufacturing and transport services hit, especially via lower earnings.
  - real estate and information services: rising employment but falling earnings.
  - medical care, finance and insurance: rising earnings; medical care employment rose.
  - regular vs non-regular: in 2020, number of regular workers grew by 360 thousand persons while non-regular workers fell by 750 thousand.
  - contact-intensive services saw large declines in non-regular workers; women particularly affected due to employment shares.
- Gender, age, and composition:
  - for full year 2020, number of employed persons dropped by 240 thousand for each male and female, but composition differs:
    - male reduction mostly among full-time employees (“mainly at work”).
    - female losses predominantly from part-time women (“work while attending school or housekeeping”).
  - employment losses concentrated in younger age groups.
  - increase in unemployed persons close to double for men compared to women; more women dropped out of workforce rather than registering as unemployed.
  - as of 4Q 2020: 1.2 million males and 0.8 million females remained unemployed.
  - NILF increases for women (esp. ages 15–24 and females 65+).
- Job mobility:
  - frequency of job changes declined for all age groups through end-2021.
  - no evidence of a U.S.-style “great resignation”; increase in desire to change jobs among men on regular contracts.
- Household income, consumption, savings:
  - aggregate compensation of employees dropped by around 8 trillion yen (annualized) in 2Q 2020.
  - disposable income spiked in 2Q 2020 due to unconditional cash transfer of 100,000 yen per person and other supports.
  - household consumption fell; saving ratio increased from 6% to 22% in 2Q 2020.
  - macro-level increase in disposable income does not imply government aid offset adverse effects for every household.

### Micro panel evidence — JPSED panel, sample construction, and groups
- Data source: Japanese Panel Study of Employment Dynamics (JPSED); internet monitoring panel by Recruit Works Institute; annual January surveys since 2016.
- Sample and period:
  - sample size approx. 50,000–60,000 observations per year.
  - 2021 survey: 56,064 individual observations (45,192 continuous; 5,809 new; 5,065 previous samples not continuous).
  - analysis period t = 2018 to 2020.
- Targeted sample restriction:
  - persons employed (at work) from November of year t-1 to January of year t.
  - classification based on Feb–Dec (t) monthly status:
    - Group 1: kept working during the year (employed at work).
      - Group 1-1: job stayers (no job change).
      - Group 1-2: job switchers (had job change).
    - Group 2: became unemployed or left the labor market at least one month.
      - Group 2-1: find a job again.
      - Group 2-2: still unemployed or NILF.
    - Group 3: employed but not at work at least one month during Feb–Dec (t).
- Sample counts (total 2018–2020 = 84,107):
  - Group 1-1 (no job change): 70,341 (2018: 20,369; 2019: 26,082; 2020: 23,890)
  - Group 1-2 (had job change): 7,006 (2018: 2,180; 2019: 2,668; 2020: 2,158)
  - Group 2-1 (find a job): 2,063 (2018: 549; 2019: 721; 2020: 793)
  - Group 2-2 (unemployed or NILF): 2,059 (2018: 553; 2019: 695; 2020: 811)
  - Group 3 (employed not at work): 2,638 (2018: 576; 2019: 742; 2020: 1,320)

### Earnings for job stayers (Group 1-1) — DML findings and heterogeneity
- Dependent variable: ∆y_it = LN(w_it) − LN(w_it−1). Outliers excluded if |∆y_it| > 60%.
- Sample sizes for earnings analysis:
  - full sample: 63,305 observations
  - full-time workers: 50,643 observations
  - part-time workers: 12,662 observations
- Method: Double Machine Learning (cross-fit partialing-out) with lasso selection; variables of interest interacted with 2020 dummy.
- Selected significant DML coefficients and patterns (preserve reported values and significance):
  - industries with significant earnings declines: manufacturing; wholesale and retail; accommodation and restaurants; living-related services.
  - occupations with larger declines: transport/communication; security.
  - firm size:
    - ～9 persons: coefficient -3.5*** (full sample) and -9.4*** (full-time).
  - age:
    - 10～99: coefficient -2.3* (full sample) and -8.4*** (full-time).
  - education:
    - junior high school d20 coefficient -3.5** (full); -1.6*** (full-time); -11.2*** (part-time).
  - working hours (t-1):
    - ～19h d20 coefficient -3.0* (all); 20～29h d20 coefficient -4.1** (all).
  - telework:
    - d20*can telework (t) coefficient 0.8*** (all); 1.0*** (full-time).
  - annual earnings (t-1) bands show large negative d20 coefficients:
    - less than 1M yen d20 coefficient -4.7*** (all)
    - 3-4M yen -7.2*** (all)
    - over 8M yen -8.8*** (all)
  - job level changes:
    - job level down d20 coefficient -4.3** (all); -4.0** (full-time)
    - job level up d20 coefficient -3.2* (all)
  - employment status:
    - regular worker d20 coefficient -3.0**; non-regular worker -2.6*; self-employed -2.8*
- Interpretations emphasized by DML results:
  - sectoral and job-structure variables explain much of the observed heterogeneity; gender, age, non-regular status, self-employment, and low income are not always robust predictors of larger earnings declines once industry and other covariates are controlled for.
  - telework eligibility positively associated with earnings (can telework (t) coefficient 0.82, std.err. 0.29, *** in Appendix Table).
  - telework share rose from less than 5% in 2019 to 14.5% in 2020 in the sample.
  - job level downgrades and fewer promotions in 2020 associated with lower earnings growth.
  - earnings growth was lower across all prior income levels with larger reductions for higher pre-pandemic earnings bands.

### Job switching and labor reallocation (Group 1-2)
- Reallocation patterns:
  - total number of job switchers decreased in 2020.
  - share moving to non-affected industries: 52.7% (2019/18 average) → 57.2% (2020).
  - share moving to affected industries: 8.5% → 7.5%.
  - share moving from non-affected industries: 13.9% → 14.0%.
  - share moving from affected industries: 24.9% → 21.3%.
  - sample sizes: 2019/18 average 2,458; 2020 2,150.
- Estimated earnings changes associated with transitions (controlling for covariates):
  - from non-affected to non-affected: 4.5 (2019/18) → 1.7 (2020)
  - from non-affected to affected: 6.9 → 2.7
  - from affected to non-affected: 0.0 → -2.6
  - from affected to affected: -3.5 → -6.7
- Composition by employment type:
  - for transitions to/from affected industries, non-regular workers accounted for close to twice as many cases as regular workers.
- Training and transitions:
  - higher off-the-job training in 2019 associated with greater likelihood of switching to non-affected industries or from affected to non-affected industries.
- Conclusion:
  - limited reallocation into growth or non-affected sectors in 2020; job changes to non-affected industries associated with higher earnings increases.

### Unemployment, dropouts, and re-employment (Group 2)
- DML estimation for risk of job loss (binary dependent variable; number of obs 72,399):
  - higher risk of job loss in 2020 for:
    - manufacturing (manufacturing(t-1)*d20 odds ratio 1.3**)
    - accommodation and restaurants (accommodation/restaurants(t-1)*d20 odds ratio 1.6**)
  - working less than 20 hours per week associated with higher odds ratio 1.3*.
  - telework variables not significantly associated with lower risk of job loss in 2020.
  - work-related self-learning associated with lower general risk of unemployment/dropout (odds ratio 0.9**), though not stronger in 2020.
- Time to re-employment:
  - about 50% re-employed by year end in 2018, 2019, and 2020.
  - average months to find a new job: 3.0 months in 2020; 2.6 months in 2018 and 2019.
  - top reason for not finding a job: mismatches of job type or content; such mismatches cited more frequently in 2020.
- Job-to-applicant imbalances (Hello Work data, monthly averages):
  - national jobs-to-applicants ratio decreased from 1.6 to 1.2 in 2020.
  - occupation-level imbalances (gaps in millions, 2020):
    - clerical: 3.6 million higher applicants than openings.
    - service: 3.2 million higher openings than applicants.
    - professional/engineering: 2.1 million higher openings than applicants.
  - Tokyo: jobs-to-applicants ratio dipped below one in July 2020.
- NILF motivations and demographics:
  - 2018/2019 top reason: ‘can live without working’.
  - 2020 top reason: ‘no particular reason’; increases in mentions of health, old age, no suitable job, and attending school.
  - pregnancy/childbirth and childcare mentioned less frequently in 2020 (survey timing in December noted).
  - macro evidence: women with young children disproportionately dropped out in 2Q 2020 (school closures effect) but no clear difference in 3Q/4Q 2020.
- Workers reaching mandatory retirement age:
  - re-employment rates in 2020: 17% found a job; 11% unemployed; 72% NILF.
  - avg. 2018/2019: 25% found a job; 7% unemployed; 67% NILF.
  - fewer retirees reemployed in 2020; retired people have significantly higher probability of being unemployed or NILF in 2020 after controls.
- Self-development:
  - in 2020, 56% of those conducting self-development found a new job versus 48% for those not conducting self-development.
  - logistic regressions confirm positive association between prior self-development and finding a new job.

### Temporary leave (Group 3), EAS effects, well-being, and earnings
- Role of EAS:
  - EAS instrumental in allowing firms to place employees on temporary leave while maintaining employment.
  - aggregate “employed persons not at work” spiked in 2Q 2020; declined after 3Q 2020.
  - EAS usage shifted toward shorter leave periods, which may not register as “employed not at work”.
- Leave duration (conditional on at least 1 month on leave):
  - average months on leave:
    - 2020 due to COVID: 2.9 months (conditional on at least 1 month).
    - avg. of 2018/2019: 3.0 months.
  - in 2020: more employees on leave for between 2–4 months; fewer on 1 month or 7+ months.
  - data do not capture leaves shorter than one month; inclusion would raise 2020 average duration.
- Well-being:
  - for leave ≤ 3 months: no difference in reported less life satisfaction/happiness between COVID-related leave and other reasons in 2020.
  - for leave > 3 months: greater dissatisfaction among those on COVID-related leave compared with others.
  - implication: longer COVID-related leaves associated with declines in life satisfaction/happiness; potential jeopardy to motivation and mental health.
- Earnings changes for leave recipients (Table 9 averages preserved):
  - average annual earnings changes (%):
    - due to COVID: -7.5
    - other reasons: -11.8
    - 2020: -4.4
    - 2019: -4.0
    - 2018: (value not shown in excerpt)
  - number of obs: 71,053
  - interpretation: EAS appears to have cushioned earnings reductions; earnings declines for pandemic leave workers less than for those on leave for other reasons.

### DML identification of determinants of being on leave (Group 3 vs Group 1-1)
- Method: DML with logistic lasso; binary dependent variable = 1 if in group 3, 0 if group 1-1.
- Selected DML findings (Table 10):
  - higher tendency to be on leave during 2020 for:
    - non-regular workers
    - self-employed
    - accommodation/restaurants workers
    - production job workers
  - companies arranged leave more frequently for non-regular employees than regular ones.
  - accommodation/restaurants dummy significant in all specifications.
  - telework ability:
    - employees who were able to telework had significantly lower probability of being placed on leave in 2020.
    - during normal times, no relationship between telework ability and being on leave.
  - age, education, gender, and income do not significantly affect probability of being placed on leave after controlling for industry and regular vs non-regular status.
- Significance notation in reported results:
  - *** p<0.001, ** p<0.05, * p<0.1.

### Policy implications and recommendations
- Short-run:
  - target support measures to the most affected industries and occupations, small firms, and part-time workers.
  - EAS furlough scheme successful in maintaining employment and potentially limiting scarring; continued high usage suggests some workers’ skills may be underutilized for extended periods.
- Medium- to long-term:
  - recalibrate support to encourage reallocation of labor from unviable firms to new growth sectors.
  - strengthen active labor market policies and unemployment/transition support to protect the vulnerable while enabling more frequent job transitions.
  - consider strengthening the EI scheme by expanding eligibility.
  - support skill updating and training; inclusion of training into eligible EAS activities during the pandemic is welcome.
  - promote flexible work styles and address obstacles to telework (e.g., use of physical seals and paper documents); government digitalization efforts should aid telework feasibility.
- Structural reforms:
  - address structural inequities for female, young, and non-regular workers irrespective of the pandemic (training, education, elimination of disincentives to full-time and regular work, better availability of childcare and nursing facilities).

### Appendix highlights — DML earnings regression details and notable coefficients
- Estimation details:
  - dependent variable: annual earnings change (in percent); DML with Lasso-selected controls; 10 cross-fit folds.
  - median number of Lasso-selected controls across folds: 285 - 302.
  - number of obs: All sample 63,305; Full time worker 50,643; Part time worker 12,662.
- Selected coefficient magnitudes and notable reported values (preserved exactly as presented):
  - can telework (t): 0.82 (std.err. 0.29) *** (controls 69).
  - d20*can telework (t): coefficients reported include 0.00 (std.err. 0.43) and -0.48 (std.err. 0.44) in some columns.
  - firm size ～9 persons: -3.49 (std.err. 1.35) *** and -9.35 (std.err. 3.28) *** in part-time column.
  - age bands and regional coefficients show large negative estimates in several columns (examples preserved verbatim in appendix).
  - annual earnings pre-pandemic bands display systematically larger negative d20 coefficients for higher earnings:
    - less than 1M yen: -4.66 (std.err. 1.39) ***
    - 1-2M yen: -4.78 (std.err. 1.25) ***
    - 2-3M yen: -6.15 (std.err. 1.17) ***
    - 3-4M yen: -7.21 (std.err. 1.15) ***
    - 4-5M yen: -7.19 (std.err. 1.15) ***
    - 5-6M yen: -7.91 (std.err. 1.16) ***
    - 6-8M yen: -8.46 (std.err. 1.17) ***
    - over 8M yen: -8.81 (std.err. 1.22) ***
- Interpretation:
  - telework ability positively associated with earnings changes for job stayers.
  - industries and occupations exposed to COVID disruptions have significant negative earnings coefficients.
  - part-time workers show larger-magnitude negative coefficients for working-hour and firm-size categories.

_Italic: Source: IMF Working Paper — wpiea2022089-print-pdf (IMF Working Paper material as provided in the source content)._

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

### 1. Introduction

### Pandemic impact summary
- The COVID-19 pandemic and related containment measures hit labor markets worldwide, with highly unequal effects across workers.
- Japan experienced:
  - a decrease in the number of employed persons by about 1 million (equal to about a 1 percent drop in the ratio of employment to population) during the initial pandemic shock.
  - average earnings declined by 1.4 percent in 2020 due to lower overtime and bonus payments.
- Employment impacts varied greatly across industries and worker attributes including skill level, gender, and employment type.

### Study contribution and methods
- The paper draws on macro and a large micro panel data set (including employment status, earnings, and working hours) to analyze labor market dynamics in Japan during the COVID-19 shock.
- The study:
  - covers more aspects of the labor market impact and goes beyond the initial period of the COVID-19 shock relative to existing literature.
  - examines heterogeneity across the labor force, telework and human capital effects, and policy implications.
  - relies on machine learning techniques that are more agnostic of variable selection than traditional econometric approaches (Double Machine Learning is referenced in appendices).

### Main empirical findings (summary)
- Sectoral differences are a crucial driver of labor market outcomes during the pandemic.
- Disproportionately affected groups (greater declines in earnings and higher risk of losing employment) include:
  - women,
  - younger age groups,
  - non-regular workers,
  - self-employed,
  - low-income workers.
- These disproportionate impacts are primarily driven by greater relative representation of these groups in the most affected industries, especially contact-intensive services.
- Employment outcomes: part-time workers were disproportionately affected; women, young, and low-income workers account for a large share of part-timers.
- Earnings outcomes: worse outcomes are associated with:
  - small firms,
  - low education,
  - working while studying,
  - part-time work.
- Additional empirical evidence:
  - need to improve childcare and related support offerings,
  - reduced labor market mobility and resource allocation due to skill mismatches,
  - general beneficial effects of more training and upskilling,
  - importance of being able to telework.

### Relation to existing literature
- Prior studies cited and their findings:
  - Hoshi et al. (2021): women, age groups 31-45 (work absences) and 60-65 (unemployment), non-regular, self-employed, and low-educated workers were more affected (micro data Feb–Jun 2020).
  - Kikuchi et al. (2021): simulation suggests most severely affected are female, non-regular, low-skilled, engaged in social jobs, and unable to work remotely.
  - Kikuchi et al. (2020): crisis would hit low-income groups disproportionately and exacerbate inequality.
  - Fukai et al. (2020): machine learning through June 2020 shows unemployed or part-time workers in hotel and restaurant industry and services occupations, younger and female workers more affected.
  - Yamamoto et al. (2021) (Keio survey May 2020): employment status (regular or non-regular) and firm size were main determinants of employment outcome differences.
  - Sumita (2021): no significant increases in unemployment and job changes due to pandemic though household income decreased, especially for women.
  - Telework studies:
    - Okubo (2020): teleworking increased from 6% in January to 17% in June 2020.
    - Kotera (2020): estimate that potentially about 30% of workers could work from home in Japan.
    - Morikawa (2021a): reported productivity of work from home improved from 2020 to 2021 but on average is lower than office work.
    - Kitagawa et al. (2021): poor home office and communication technology setups main reasons for productivity losses with work from home.
    - Ishii et al. (2021): telework mitigated reduction of income and working hours during April and May 2020, addressing reverse causality concerns.

### Organization of the paper
- Section 2: background on pandemic developments and labor market policies in Japan.
- Section 3: overview of the Japanese labor market during the pandemic using macro data.
- Section 4: analysis of labor market developments using micro panel data.
- Section 5: conclusion and policy implications.

### Background on the Japanese labor market (selected pre-pandemic facts)
- Demographics and participation:
  - The working-age (15-64) population declined by slightly more than 12 million since 2000, from 87 million to 75 million.
  - Employment expanded despite the decline in the working-age population due to fast increases in employment rate of working-age women.
  - Female employment rate stood at 71.5 percent as of end 2019, above the OECD average of 61.5 percent.
  - Employment rates of the elderly (65+) rebounded to levels in the 1990s.
- Labor market tightness pre-pandemic:
  - unemployment rate of 2.4 percent in 2019.
  - job offer to applicant ratio of 1.6 in 2019.
- Employment systems and composition:
  - Lifetime employment remains important; share of male workers with job tenure of 10 or more years for ages 35-45 is more than 60 percent.
  - Non-regular employment rose from 16% in 1995 to 32% in 2019; women account for about 70 percent of non-regular workers.
  - Non-regular employment associated with lower wages and lower job security and less firm investment in training.
- Gender and working hours:
  - Lifetime employment primarily applies to regular male workers and is associated with long working hours, contributing to gender inequality.
  - Factors keeping women out of lifetime employment include insufficient childcare support and need for flexibility and shorter working hours.

### Policy context (pre-pandemic reforms)
- Work Style Reform (WSR):
  - Began discussions in 2016; legislation passed in 2018.
  - Aims to increase employment opportunities and productivity through measures such as caps on excessive overtime, promotion of flexible working styles, equal pay for equal work, and shifting focus from working hours to output.
- Other reform efforts:
  - expansion of childcare and nursing facilities and skills training to increase female workforce participation and employment.

*Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022089-print-pdf.pdf*

### 2.2 COVID-19 Developments

### 2.2 COVID-19 Developments

### Pandemic timeline and major government responses
- First confirmed COVID-19 case in Japan on January 15, 2020.
- Government requests: cancellation/postponement of major events and temporary school closures at end of February 2020.
- State of emergency declared on April 7, 2020, initially in 7 prefectures including the Tokyo metropolitan area; extended to entire country on April 16, 2020.
- Government requested prevention measures during the state of emergency, including refraining from going out, temporary business closure, and reducing the number of commuters.
- Economic support measures announced on April 20, 2020, totaling 117 trillion yen (about 21% of GDP).
- State of emergency gradually lifted from mid-May 2020 and entirely lifted on May 25, 2020.
- Tourism promotion program started in July 2020 (excluding Tokyo), extended to Tokyo in October 2020, suspended entirely in December 2020 as infections rose.
- Additional economic stimulus package adopted on December 8, 2020, worth 73.6 trillion yen (about 13% of GDP), focused on promoting digitalization and green technologies and extending ongoing support measures.
- Government adopted additional economic measures on November 19, 2021, totaling 78.9 trillion yen (about 14% of GDP), including special benefits to households with children.

### Infection waves and containment in 2021
- Japan experienced three major infection waves in 2021 requiring repeated behavioral and economic restrictions.
- Second state of emergency declared in January 2021 in Tokyo and neighboring regions and seven other prefectures; lifted in all areas on March 21, 2021.
- Third state of emergency declared on April 25, 2021 in four prefectures including Tokyo; additional six prefectures added shortly after.
- Requested prevention measures in 2021 included shorter operating hours for restaurants, closure of large commercial facilities such as malls, and banning spectators from events.
- State of emergency lifted between June 21 to July 11, 2021 except for Okinawa.
- New infections soared from late July 2021; 21 prefectures placed under a state of emergency in August 2021; new cases peaked around late August 2021; emergency declaration lifted in all prefectures on September 30, 2021.
- Infections remained low until the emergence of the Omicron variant at the end of 2021.

### Employment Adjustment Subsidy (EAS): usage, generosity, and timing
- EAS provides financial assistance to businesses that maintain employment by placing employees on temporary leave; firms are required to pay a leave allowance, and EAS reimburses a percentage (grant rate) of this leave allowance with a cap.
- Leave eligible under the scheme can be as short as 1 hour.
- EAS origins: established during the 1974 oil crisis; extensively used during the global financial crisis (GFC) when requirements were relaxed.
- Pandemic-era changes (starting April 2020):
  - Expanded coverage to all employees (Emergency Employment Safety Subsidy created to cover those not enrolled in employment insurance).
  - Raised the grant rate and the reimbursement cap.
  - Grant rate became 1/2 to 2/3 for large companies and 2/3 to 4/5 for SMEs. If companies make no dismissal, the rate became 3/4 for large companies and 9/10 for SMEs.
  - The upper limit of subsidy increased from 8,330 yen to 13,500 or 15,000 yen per day.
  - Increased flexibility by allowing firms to place employees in vocational education and training (VET) programs instead of leave; VET subsidized amount and requirements were relaxed.
  - Initially, an additional 1,200 yen was disbursed per worker for one day VET; during the pandemic, the amount increased to 2,400 yen for SMEs and 1,800 yen for large companies. VET coverage was expanded, and days can be split between VET and leave.
  - Ministry of Health, Labor, and Welfare (MHLW) simplified application procedures to accelerate disbursements.
- EAS payments and patterns:
  - Total EAS payments during the pandemic far exceeded those during the GFC, reaching a peak of close to 600 bn yen in August 2020.
  - Payments decreased to roughly 200 billion yen per month in December 2020 and remained at a similar level until late 2021.
  - During FY2020, the total cumulative amount of 3.1 trillion yen was provided, about seven times higher than the EAS payments during the entire GFC period.
  - Among industries, manufacturing, accommodation and restaurants, and wholesale and retail received more than half of EAS payments.
  - Japanese government estimates: EAS lowered the unemployment rate by about 2.6 percentage points during April to October 2020.
  - Data on the number of employees benefitting from EAS is not available; EAS payments used as a proxy for EAS usage.
- Interpretation of EAS usage versus employment status:
  - Large increase in “employed persons not at work” (temporarily on leave) across all age groups primarily in April and May 2020.
  - Workers on leave jumped by over 4 million persons relative to the previous year in April 2020, declining fast to a level only slightly higher than pre-crisis from around June 2020.
  - Discrepancy noted: EAS payments remained substantially above pre-pandemic levels through 2020 and 2021 while employed persons not at work normalized—suggests firms continued using EAS for short-term leave and that more generous terms and additional VET disbursements raised EAS payments.

### Employment Insurance (EI) and complementary labor supports
- Pre-pandemic EI basic benefits: equivalent to 50-80% of previous income with a cap for 90-330 days depending on age, duration of contributions, and reason for becoming unemployed.
- Eligibility: Employees working at least 20 hours per week, including non-regular workers, are eligible for EI coverage.
- COVID-19 adjustments:
  - EI benefit payments were temporarily extended by an additional 60 days (with scope of eligible workers varying depending on whether the unemployment date is before, during, or after a state of emergency; a person receives additional 30 days instead of 60 in specified circumstances).
  - In FY2020, the total payment of EI basic benefits increased by about 27 percent compared with the previous fiscal year, due to more recipients and a longer average benefit period.
  - The total payment for FY2020 is about 42 percent lower than FY2009, implying EI was less important in this crisis compared with the GFC because generous EAS support helped prevent a large increase in unemployment.
- Other labor-market supports:
  - Individuals who cannot receive the leave allowance for some reason can directly apply for grants and allowances.
  - Assistance measures for people forced to take leave due to childcare obligations.
  - Career counseling support through Hello Work (public employment security service) was strengthened for non-regular workers, and eligibility for job-seeker support training was expanded.

*IMF Working Paper — 2.2 COVID-19 Developments (excerpt)*

### 3. Labor Market Developments During the

### 3. Labor Market Developments During the Pandemic – Evidence from Macro Data

### Initial shock, employment, unemployment, and NILF
- The number of employed persons decreased by about 1 million from March to April 2020.
- Many people left the labor market instead of filing as unemployed, leaving the unemployment rate initially almost unchanged.
- The number of people not in the labor force (NILF) gradually declined reaching the pre-pandemic level in November 2020, but increased again in 2021 during COVID-19 infection waves in May/June and in the Fall.
- The number of employed persons recovered gradually after the sharp initial drop in April 2020, but as of late 2021 remains substantially below the pre-crisis level with major damage resulting from renewed COVID-19 waves.
- The number of people registered as unemployed rose gradually until leveling off in late 2020 and remains about 1.3 times higher than before the pandemic.

### Earnings and composition effects
- Average annual earnings declined by 1.4 percent in 2020 due to reduced overtime and bonus payments.
- For full-time workers the average annual earnings change was -1.5 percent.
- Part-time workers experienced a smaller reduction of -0.6 percent; this may reflect compositional effects as lower-wage part-time workers lost jobs and special bonuses for part-time workers in some sectors in 2020 including healthcare and education.
- Temporary workers experienced a decline in base pay, partially offset by bonus payments, reflecting the more flexible nature of temporary contracts.

### Industry- and worker-type heterogeneity
- Contact-intensive services (accommodation and restaurants; living-related and amusement services) were most affected.
- Manufacturing and transport services were also hit, especially through lower earnings.
- Real estate and information services experienced rising employment but falling earnings.
- Medical care and finance and insurance saw rising earnings; in medical care employment also rose.
- Non-regular workers bore the brunt of the shock: in 2020, the number of regular workers grew by 360 thousand persons while the number of non-regular workers fell by 750 thousand.
- Contact-intensive services saw large declines in non-regular workers, with especially women affected due to their greater employment share in these industries.
- New jobs in industries that added employment (health care, information/communication services, real estate, public sector, education) were mostly regular positions, although health care and the public sector also added a significant number of non-regular positions filled by women.

### Gender, age, and compositional changes
- For the full year 2020, the number of employed persons dropped by 240 thousand persons for each male and female, with compositional differences by gender:
  - Male employment reduction was mostly among full-time employees (“mainly at work”).
  - Female employment losses stemmed predominantly from women working part-time (“work while attending school or housekeeping”).
- Employment losses for both genders were concentrated in younger age groups.
- The support from EAS is visible in the large increase in “employed persons not at work” (temporarily on leave) across all age and gender groups.
- The increase in unemployed persons was close to double for men compared to women, suggesting more women dropped out of the workforce rather than registering as unemployed.
- As of 4Q 2020, 1.2 million males and 0.8 million females remained unemployed.
- For those not in the labor force, there was a big increase for women but not for men; especially young workers ages 15 to 24 dropped out of the workforce, although the reason changed from attending school to “other”.
- Females aged 65 and over stand out for dropping out of the labor force in large numbers.

### Job mobility and job-change preferences
- The frequency of job changes declined for all age groups during the pandemic and remained depressed through the end of 2021, reflecting fewer open positions.
- There is no evidence of a “great resignation” in Japan as seen in the U.S., but there has been an increase in the number of employees expressing the wish to change jobs, especially among men on regular work contracts.
- The number of job switchers may increase with the labor market recovery, but given desired-job-change data it is unlikely to reach U.S. proportions.

### NILF motivations and gender differences
- During the pandemic, people not in the labor force became less interested in finding employment; the number of people not wishing to work increased, particularly for ages 15 to 24 and females aged 65 and over.
- Reasons for leaving previous jobs and becoming NILF vary with pronounced gender differences: in Q4 2020 more women indicated marriage, childbirth, or childcare as the reason, suggesting that the burden of school closures and other pandemic restrictions fell especially on women.

### Household income, consumption, and savings
- Aggregate compensation of employees dropped by around 8 trillion yen (annualized) in 2Q 2020.
- Disposable income spiked in 2Q 2020 due to the government's unconditional cash transfer of 100,000 yen per person and other support programs.
- Household consumption fell, and the saving ratio increased from 6% to 22% in 2Q 2020.
- The macro-level increase in disposable income does not imply government aid offset the pandemic's adverse effects for every household; impacts differed significantly depending on households' work situations.

### Summary conclusion
- Macro data show the pandemic's impact varies substantially across gender, age, regular vs non-regular status, and industry, underscoring the importance of accounting for workers’ heterogeneous attributes when analyzing the pandemic impact.
- There are limits to insights from macro data on heterogeneity; a rich micro data set is used in subsequent analysis to address those limits.

*Source: IMF Working Paper — chapter "3. Labor Market Developments During the Pandemic – Evidence from Macro Data"*

### 4. Labor Market Developments During the

### 4. Labor Market Developments During the Pandemic – Evidence from Micro Panel Data

### 4.1 Panel Data Set
- Data source: Japanese Panel Study of Employment Dynamics (JPSED) provided by the Recruit Works Institute; internet monitoring panel survey conducted every January since 2016. Sample includes men and women aged 15 years and over. Sampling designed to be representative by gender, age, type of employment, district block, and educational background.
- Survey content: roughly 100 questions about basic attributes and the previous year’s labor indicators, including employment status and dynamics, working hours, and annual earned income.
- Sample size: fluctuates with approximately 50,000 to 60,000 individual observations each year.
  - 2021 survey: 56,064 individual observations; 45,192 continuous from previous years, 5,809 new individuals, 5,065 in previous samples but not continuously.
- Analysis period: t = 2018 to 2020.
- Sample construction (targeted sample):
  - Restriction: persons employed (at work) from November of year t-1 to January of year t.
  - Classification based on Feb–Dec (t) monthly status:
    - Group 1: kept working during the year (employed at work). Subdivided:
      - Group 1-1: job stayers (no job change).
      - Group 1-2: job switchers (had job change).
    - Group 2: became unemployed or left the labor market for at least one month during the year. Subdivided:
      - Group 2-1: find a job again.
      - Group 2-2: still unemployed or NILF.
    - Group 3: employed but not at work at least one month during Feb–Dec (t).
- Sample counts by year and group (Table 2 / Figure 13 aggregates):
  - Total (2018–2020): 84,107
  - Group 1-1 (no job change): 70,341 (2018: 20,369; 2019: 26,082; 2020: 23,890)
  - Group 1-2 (had job change): 7,006 (2018: 2,180; 2019: 2,668; 2020: 2,158)
  - Group 2-1 (find a job): 2,063 (2018: 549; 2019: 721; 2020: 793)
  - Group 2-2 (unemployed or NILF): 2,059 (2018: 553; 2019: 695; 2020: 811)
  - Group 3 (employed not at work): 2,638 (2018: 576; 2019: 742; 2020: 1,320)

### 4.2 Employed Persons at Work (group 1)
- Definition: workers continuously working during the year without spells of unemployment, NILF, or absence.
- Two analytical focuses:
  - Group 1-1 (no job change): earnings changes.
  - Group 1-2 (job switchers): labor reallocation and earnings implications.

#### 4.2.1 Earnings Developments for Employees with no Job Change (group 1-1)
- Dependent variable: ∆y_it = LN(w_it) − LN(w_it−1) (log annual earnings change). Outliers excluded if |∆y_it| > 60% (threshold based on 2.5 percentile).
- Sample categories analyzed:
  - Full sample: 63,305 observations
  - Full-time workers (35 hours or more per week): 50,643 observations
  - Part-time workers (less than 35 hours per week): 12,662 observations
- Methodology: Double Machine Learning (DML) (cross-fit partialing-out) with lasso selection of controls; variables of interest are feature variables multiplied by a 2020-year dummy to capture pandemic-year movements.
- Key regression findings (selected, Table 3):
  - Significant earnings declines in 2020 concentrated in specific industries, occupations, small firms, and among those with junior high school education only, working while studying, and part-time workers.
  - Affected industries with significant earnings declines: manufacturing; wholesale and retail; accommodation and restaurants; living-related services.
  - Occupations with larger declines: transport/communication; security.
  - Firm size: firms with 9 persons had coefficient -3.5*** (full sample) and -9.4*** (full-time).
  - Age 10～99 had coefficient -2.3* (full sample) and -8.4*** (full-time).
  - Education: junior high school d20 coefficient -3.5** (full), -1.6*** (full-time), -11.2*** (part-time).
  - Working hours (t-1): ～19h d20 coefficient -3.0* (all); 20～29h d20 coefficient -4.1** (all).
  - Telework: d20*can telework (t) coefficient 0.8*** (all), 1.0*** (full-time).
  - Annual earnings (t-1): less than 1M yen d20 coefficient -4.7*** (all); 3-4M yen -7.2*** (all); over 8M yen -8.8*** (all).
  - Job level change: job level down d20 coefficient -4.3** (all); -4.0** (full-time). Job level up d20 coefficient -3.2* (all).
  - Employment status: regular worker d20 coefficient -3.0**; non-regular worker -2.6*; self-employed -2.8*.
- Interpretations:
  - After controlling for covariates, gender, younger age groups, non-regular vs regular status, self-employed, and low-income earners are not robust predictors of larger earnings declines in themselves; rather their overrepresentation in affected industries explains disproportionate impacts.
  - Telework eligibility is associated with higher earnings increases in 2018, 2019, and 2020; telework share in sample rose from less than 5% in 2019 to 14.5% in 2020.
  - Job level downgrades in 2020 increased relative to previous years and are associated with lower earnings growth; fewer promotions in 2020 also limited earnings increases.
  - Earnings growth was lower for all income levels with larger reductions for high income earners.

#### 4.2.2 Job Changes for Employees Continuously at Work Through 2020 (group 1-2)
- Reallocation was limited during the pandemic:
  - Total number of employees switching jobs decreased in pandemic year.
  - Workers tended to move less to affected industries in 2020, and also moved less out of affected industries.
- Table 6 highlights:
  - Share moving to non-affected industries: 52.7% (2019/18 average) → 57.2% (2020).
  - Share moving to affected industries: 8.5% → 7.5%.
  - Share moving from non-affected industries: 13.9% → 14.0%.
  - Share moving from affected industries: 24.9% → 21.3%.
  - Sample sizes: 2019/18 average 2,458; 2020 2,150.
  - Estimated earnings changes associated with transitions (controlling for gender, age, employment status, income level, job level change):
    - From non-affected to non-affected: 4.5 (2019/18) → 1.7 (2020)
    - From non-affected to affected: 6.9 → 2.7
    - From affected to non-affected: 0.0 → -2.6
    - From affected to affected: -3.5 → -6.7
- Regular vs non-regular composition:
  - For job changes between non-affected industries, regular and non-regular workers accounted for similar percentages.
  - For job changes to/from affected industries, non-regular workers accounted for close to twice as many cases as regular workers.
- Off-the-job training is associated with job transition opportunities; those with higher off-the-job training in 2019 more likely to switch between non-affected industries or from affected to non-affected industries.
- Overall: job changes to non-affected industries associated with higher earnings increases; limited labor reallocation into and out of affected industries in 2020.

### 4.3 Unemployed or Dropping out of the Workforce (group 2)
- Objective: identify factors causing job loss/dropout and traits associated with finding a new job.
- Methodology: DML with binary dependent variable (1 if in group 2, 0 if in group 1-1); logit lasso used for selection. Sample size noted: Number of obs 72,399 (Table 7).
- Key findings (selected, Table 7):
  - Statistically significant higher risk of job loss in 2020 for workers in:
    - Manufacturing (manufacturing(t-1)*d20 odds ratio 1.3**)
    - Accommodation and restaurants (accommodation/restaurants(t-1)*d20 odds ratio 1.6**)
  - Working less than 20 hours per week (working hour (～19h) (t-1)*d20) associated with higher odds ratio 1.3*.
  - Telework variables not significantly associated with lower risk of job loss in 2020.
  - Work-related self-learning (self-development) associated with lower risk of unemployment or dropping out in general (odds ratio 0.9**), although effect not stronger in 2020.
- Time to re-employment:
  - About 50% re-employed by year end in 2018, 2019, and 2020.
  - Average months to find a new job: 3.0 months in 2020; 2.6 months in both 2018 and 2019.
  - Top reason for not finding a job: mismatches of job type or content; skill/job-type/content mismatches cited more frequently in 2020 than prior years.
- Job-to-applicant patterns (Hello Work data, monthly averages):
  - National jobs-to-applicants ratio decreased from 1.6 to 1.2 in 2020.
  - Occupation-level imbalances in 2020 (gaps in millions):
    - Clerical: 3.6 million higher applicants than openings.
    - Service: 3.2 million higher openings than applicants.
    - Professional/engineering: 2.1 million higher openings than applicants.
  - Tokyo: jobs-to-applicants ratio dipped below one in July 2020, indicating regional mismatch.
- Reasons for leaving the labor force (NILF) changed in 2020:
  - 2018/2019 top reason: ‘can live without working’.
  - 2020: ‘no particular reason’ became most common; increases in mentions of health, old age, no suitable job, and attending school; pregnancy/childbirth and childcare mentioned less frequently in 2020 (survey timing in December noted).
  - Macro evidence: women with young children dropped out of labor force disproportionately more in 2Q 2020 but no clear difference in 3Q or 4Q 2020—suggests school closures in 2Q forced temporary exits.
- Workers reaching mandatory retirement age:
  - Re-employment rates:
    - 2020: 17% found a job; 11% unemployed; 72% NILF.
    - Avg. of 2018/2019: 25% found a job; 7% unemployed; 67% NILF.
  - Fewer retired people reemployed in 2020 relative to previous years; logistic regression indicates retired people have significantly higher probability of being unemployed or NILF in 2020 even after controls.
- Self-development and re-employment:
  - People who engaged in self-development activities had higher re-employment chances during the pandemic.
  - 2020: 56% of those conducting self-development activities found a new job versus 48% for those not conducting self-development (Figure 17 comparisons: 2020 vs avg. of 18/19).
  - Logistic regression confirms positive association between prior self-development and finding a new job (Appendix 6).

### 4.4 Employees Placed on Temporary Leave in 2020 (group 3)
- Context: EAS program provided employers with subsidy for placing employees on temporary leave while maintaining employment; instrumental in protecting employment.
- Aggregate pattern: number of employed persons not at work spiked in 2Q 2020, declined to slightly above normal after 3Q 2020. Note: EAS usage shifted to shorter leave periods which may not register as “Employed not at work”.
- Duration of leave among those placed on leave for at least one month:
  - Average months on leave:
    - 2020 due to COVID: 2.9 months (conditional on at least 1 month).
    - Avg. of 2018/2019: 3.0 months.
  - Distributional change in 2020: more employees placed on leave for between 2–4 months; fewer on 1 month or 7+ months.
  - Data limitation: leaves shorter than one month not captured; inclusion of those would raise average duration for 2020.
- Well-being associations:
  - For leave duration three months or less: no difference in percentage reporting less life satisfaction or happiness between COVID-related leave and other reasons in 2020.
  - For leave duration over three months: more dissatisfaction among those on leave due to COVID-19 related reasons compared with others.
- Implication: longer COVID-related leaves associated with greater declines in life satisfaction/happiness, which could affect future motivation or productivity.

_Italic: Source: IMF Working Paper — chapter "4. Labor Market Developments During the Pandemic – Evidence from Micro Panel Data" (JPSED-based analysis)._

### 19. This suggest that long pandemic-related leave jeopardizes motivation and potentially mental health.

### wpiea2022089-print-pdf - 19. This suggest that long pandemic-related leave jeopardizes motivation and potentially mental health.

### Findings on pandemic-related leave and earnings
- Long periods of forced leave could cause skills to atrophy and surveys suggests that they reduce motivation and happiness.
- For those placed on leave, EAS seems to have cushioned earnings reductions to some extent in 2020.
- Although the average reduction rate of annual earnings is larger in 2020, the earnings changes for pandemic leave workers are less than for those on leave for other reasons (Table 9).
- Table 9. Average annual earnings changes (%):
  - due to COVID: -7.5
  - other reasons: -11.8
  - 2020: -4.4
  - 2019: -4.0
  - 2018: (value not shown in excerpt)
- Number of obs: 71,053

### DML analysis: identification strategy and selected determinants of being on leave
- Method:
  - Employed Double Machine Learning (DML) with logistic lasso regressions.
  - The binary dependent variable equals one if an observation is in group 3 (employed but on leave) and zero if it is in group 1-1 (job stayers).
  - Control variables are listed in Appendix 2.
- Selected DML results (Table 10):
  - Controlling for various factors, the results suggest higher tendency to be on leave during 2020 for:
    - non-regular workers
    - self-employed
    - restaurant and accommodation workers (accommodation/restaurants)
    - jobs in production (production job)
  - Companies arranged leave more frequently for non-regular employees than for regular ones.
  - The dummy for accommodation and restaurants is significant in all cases, underscoring the considerable adverse pandemic impact on this industry.
  - The higher odds ratio of production workers is consistent with the substantial industrial production decline during 2Q 2020.
  - Age, education, gender, and income do not appear to have affected the probability of being placed on leave after controlling for industry and regular vs non-regular work arrangement.
  - Employees who were able to telework had a significantly lower probability of being placed on leave in 2020.
    - During normal times, there is no relationship between ability to telework and being on leave, but in the pandemic year, telework was of critical importance to continue working.
- Significance notation in selected results:
  - *** p<0.001, ** p<0.05, * p<0.1
  - The hyphen indicates results are not significant at the 10% level. The plus indicates all items are significant.

### Appendix descriptive statistics and key numerical points
- LN(earnings(t)/earnings(t-1)), average:
  - avg. 18/19: 6.4%
  - 2020avg. 18/19 2020avg. 18/19 2020 (values in excerpt): 4.0% -25.0% -30.9% -4.2% -9.6% (preserved verbatim from table excerpt)
- Industry shares (selected examples, avg. 18/19 vs 2020):
  - manufacturing: 16.9% / 17.2% (avg. 18/19) vs 12.1% / 14.2% / 11.5% / 10.1% (2020 entries in excerpt)
  - restaurants/accommodation: 4.6% / 4.0% (avg. 18/19) vs 7.5% / 9.3% / 7.0% / 16.0% (2020 entries in excerpt)
- Sample sizes and group classification (from Appendix):
  - group1 group2 group3 sample size: 25,650      26,048      1,259        1,604        659           1,320
  - Classification: Group 1: Employed at work, Group 2: Unemployed or Not in the Labor Force, Group 3: Employed not at work

### Policy implications and recommendations (conclusions)
- Short-term support measures should be well targeted to the most affected industries and occupations, small firms, and part-time workers.
- Structural inequities for female, young, and non-regular workers should be addressed irrespective of the pandemic (e.g., training, education, elimination of disincentives to full-time and regular work, better availability of childcare and nursing facilities).
- The emphasis on policy support to maintain employment (EAS furlough scheme) was successful in maintaining employment and potentially helped limit scarring, but:
  - Continued high usage rates of EAS support suggest that some workers’ skills and abilities have been underutilized for extended periods.
  - Over time a recalibration of support seems warranted to encourage reallocation of labor from unviable firms to new growth sectors.
- Suggested measures:
  - Active labor market policies and strong unemployment and transition support to protect the vulnerable while enabling more frequent job transitions.
  - Consider strengthening the EI scheme by expanding its eligibility.
  - Support skill updating and training; inclusion of training into eligible activities for EAS support during the pandemic is a welcome step.
  - Continue to promote flexible work styles and address obstacles to telework (e.g., the use of physical seals and paper documents); government digitalization efforts should contribute to greater feasibility of telework.

*IMF WORKING PAPERS The Japanese Labor Market During the COVID-19 Pandemic*

### Appendix 4: DML result of the Earnings Analysis of Job stayers (Group 1-1)

### Appendix 4: DML result of the Earnings Analysis of Job stayers (Group 1-1)

### Dependent variable and estimation details
- Dependent variable: annual earnings change (in percent).
- Estimation method: Double/debiased machine learning (DML) with Lasso-selected controls (median number of controls reported across 10 folds).
- Significance notation: *** p<0.001, ** p<0.05, * p<0.1.
- Number of folds in cross-fit: 10.
- Number of controls (median selected by Lasso): 285 - 302.
- Number of obs:
  - All sample: 63,305
  - Full time worker: 50,643
  - Part time worker: 12,662

### Key statistics and notable coefficient estimates (selected)
- Gender and telework:
  - d20*female(t): Coefficients reported (example entries in table) include -0.58 (std.err. 0.38) for one column and -1.54 (std.err. 1.07) in another column.
  - can telework (t): Coefficient 0.82 (std.err. 0.29) ***; controls 69.
  - d20*can telework (t): Coefficients reported include 0.00 (std.err. 0.43) in one column and -0.48 (std.err. 0.44) in another.
- Age group effects (d20*age(t-1), selected):
  - 25～29: coefficients include 0.07 (std.err. 1.11) and -3.72 (std.err. 1.19) ***.
  - 30～34: coefficients include 0.34 (std.err. 1.04) and -3.42 (std.err. 1.12) ***.
  - 65～: coefficient -0.42 (std.err. 1.15) and -6.14 (std.err. 1.31) *** in another column.
- Education (d20*education (t-1), selected):
  - junior high school: -3.52 (std.err. 1.54) ** ; -1.62 (std.err. 1.43) *** ; -11.21 (std.err. 3.93) *** (across columns).
  - at school: -3.54 (std.err. 1.87) * ; -5.18 (std.err. 1.97) *** ; -7.09 (std.err. 3.81) *.
- Living area (d20*living area (t), selected regional coefficients):
  - Hokkaido: -8.56 (std.err. 1.61) *** ; -4.89 (std.err. 1.61) ***.
  - Kinki: -7.90 (std.err. 1.58) *** ; -4.26 (std.err. 1.57) ***.
  - Kyushu: -7.27 (std.err. 1.58) *** ; -3.61 (std.err. 1.57) **.
- Employment status (d20*employment status (t-1), selected):
  - regular worker: -3.02 (std.err. 1.31) ** (controls 32).
  - non-regular worker: -2.58 (std.err. 1.36) * (controls 46).
  - self-employed: -2.78 (std.err. 1.46) * (controls 35).
- Industry (d20*iIndustry (t-1), selected):
  - manufacturing: -2.35 (std.err. 1.00) ** (controls 34).
  - wholesale/retail: -2.15 (std.err. 1.01) ** (controls 31).
  - accommodation/restaurants: -3.94 (std.err. 1.30) *** (controls 30).
  - real estate: -1.08 (std.err. 1.30) and -2.51 (std.err. 1.30) * in another column.
- Occupation (d20*occupation (t-1), selected from subsequent table):
  - security: -3.11 (std.err. 1.06) *** (controls 32).
  - transportation/communications: -4.46 (std.err. 1.09) *** (controls 32).
  - production: -1.82 (std.err. 0.85) ** (controls 43).
  - professional job: -1.60 (std.err. 0.80) ** (controls 41).
- Working hours (d20*working hour ～19h per week (t-1), etc., selected):
  - ～19h: -2.99 (std.err. 1.75) * (controls 19).
  - 20～29h: -4.07 (std.err. 1.77) ** (controls 24).
  - 30～40h: -2.28 (std.err. 1.65) (controls 23); for full-time worker column, -6.19 (std.err. 1.60) ***.
  - For part-time worker columns, very large negative coefficients reported for several hour bands (e.g., -13.72 (std.err. 4.19) *** for 30～40h).
- Firm size (d20*firm size (t-1), selected):
  - ～9 persons: -3.49 (std.err. 1.35) *** (controls 28) and -9.35 (std.err. 3.28) *** in part-time column.
  - 10～99: -2.29 (std.err. 1.32) * and -8.39 (std.err. 3.23) *** in part-time column.
  - 100～999: -1.88 (std.err. 1.33) and -7.02 (std.err. 3.21) ** in part-time column.
  - 1000～: -1.49 (std.err. 1.34) and -7.55 (std.err. 3.17) ** in part-time column.
- Job level changes (d20*job level change (t), selected):
  - job level up: -3.18 (std.err. 1.70) * (controls 24).
  - job level down: -4.30 (std.err. 1.73) ** (controls 7).
  - job same level: -2.92 (std.err. 1.68) * (controls 12).
- Annual earnings (d20* annual earnings (t-1), selected; large negative estimates concentrated at higher earnings):
  - less than 1M yen: -4.66 (std.err. 1.39) *** (controls 40).
  - 1-2M yen: -4.78 (std.err. 1.25) *** (controls 26).
  - 2-3M yen: -6.15 (std.err. 1.17) *** (controls 28).
  - 3-4M yen: -7.21 (std.err. 1.15) *** (controls 28).
  - 4-5M yen: -7.19 (std.err. 1.15) *** (controls 17).
  - 5-6M yen: -7.91 (std.err. 1.16) *** (controls 24).
  - 6-8M yen: -8.46 (std.err. 1.17) *** (controls 33).
  - over 8M yen: -8.81 (std.err. 1.22) *** (controls 40).
- Spouse status (d20*employment status of spouse (t-1), selected):
  - executives: -3.00 (std.err. 1.17) *** (controls 18).
- Work-related self-learning:
  - work related self-learning (t-1): 0.00 (std.err. 0.17) (controls 42).
  - d20*work related self-learning (t-1): 0.36 (std.err. 0.29) (controls 33).
- Additional notes on model configuration:
  - Tables report coefficients, standard errors, and median number of Lasso-selected controls across 10 folds for three population columns: All sample, Full time worker, Part time worker.
  - Clustering: some columns indicate presence/absence of clustering; specific entries in the table flag "Yes" or "No" for clustering in particular regressions.

### Interpretation highlights from reported estimates
- Telework ability (can telework (t)) is positively associated with annual earnings changes (0.82, std.err. 0.29, ***), indicating a robust positive correlation in the DML estimates for job stayers.
- Industry- and occupation-specific coefficients show significant negative earnings changes for industries and occupations more exposed to COVID-related disruptions (e.g., accommodation/restaurants; transportation/communications; security).
- Annual earnings categories show systematically larger negative coefficients for higher pre-pandemic earnings bands (e.g., 3-4M yen through over 8M yen, with coefficients ranging from -7.21 to -8.81 and highly significant), indicating substantial heterogeneous earnings impacts across prior earnings levels.
- Part-time worker columns often display larger-magnitude negative coefficients for working-hour and firm-size categories, suggesting differential impacts by contract type.

*Appendix 4: DML result of the Earnings Analysis of Job stayers (Group 1-1) — IMF Working Paper material as provided in the source content.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022089-print-pdf.pdf_