## wpiea2021279-print-pdf

## Source details

**Canonical URL:** [wpiea2021279-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021279-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021279-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021279-print-pdf.pdf.json)

---

### I. Introduction
- As of June 24, 2021, India has more than 30 million confirmed cases.
- The stringent national lockdowns led to a 24 percent decline in GDP in the second quarter of last year.
- Objective: develop a de-facto state-level social distancing measure by combining an ordinal index of containment stringency and observed mobility trends at the state level to quantify impacts on health and economic outcomes and examine the role of state-level factors (health infrastructure, population and economic characteristics, governance).
- Empirical approach: local projections methods (Jordà 2005); comprehensive daily state-level dataset; sample period runs from mid-February 2020 to mid-January 2021, covering the first wave.

### II. Data and stylized facts
- Mobility and containment:
  - Google mobility at the state level; containment measures from a new IMF containment index and Oxford University index used for robustness.
  - Mobility components used: Google retail and recreation mobility and transit stations mobility.
- Health indicators and daily economic proxies:
  - Official confirmed cases at state level from CEIC.
  - Electricity consumption from POSOCO.
  - NO2 emissions from Air Quality Open Data Platform.
  - Flights data from FlightRadar24.
- State characteristics (sources indicated in Annex Table A1): health care spending per capita; doctor density; population density; share of population aged 60 and over; share with comorbidities; share of services; GDP per capita; governance.
- Sample: daily data from February 15, 2020 to January 12, 2021 (34 states for main analyses; some tables report different panel counts).
- State heterogeneity examples:
  - Health: Maharashtra, Kerala, Karnataka, and Andhra Pradesh among worst hit in absolute and per-capita confirmed cases.
  - Economic: NO2 emissions declined most in West Bengal and Gujarat; energy consumption falls largest in Sikkim, Himachal Pradesh, and Punjab.
  - Mobility: mobility decline in Maharashtra much more marked than in Rajasthan after national lockdowns.

### III. Mobility, social distancing, and containment
- Effective state-level social distancing index:
  - Constructed from national-level IMF containment index (covers six economic sectors: international travel, schools, retail, industry, services, public gatherings) and state-level mobility; IMF index captures announcements about future changes.
  - Mobility proxies: retail and recreation mobility (discretionary spending), transit stations mobility (workplace behavior) — highest absolute correlation with Oxford Stringency Index in India.
- Observed relationships:
  - High effective social distancing coincides with sharp reductions in growth rate of confirmed cases (example states: Maharashtra, Rajasthan).
  - Negative relationship between social distancing and energy consumption at state level.

### IV. The impact of effective social distancing (local projections results)
- Methodology:
  - Local projections (Jordà 2005) estimating impulse responses for horizons h=0,..,30.
  - Regression controls: lagged infections and economic variables; state-specific time trends; fixed effects; robust standard errors clustered at state level.
  - Balanced sample of 34 states; estimation period February 15, 2020 to January 12, 2021.
- Aggregate health impacts (selected reported results):
  - Social distancing and containment reduced confirmed COVID-19 cases by around 80 percent relative to a baseline of no social distancing and containment (impulse responses in log-percentage points converted using (e^{θ_h}−1)∗100).
  - Table A3 selected coefficients (el i effect on cumulative confirmed cases):
    - f1.ccase: -0.0535** (SE 0.0216)
    - f5.ccase: -0.159* (SE 0.0865)
    - f10.ccase: -0.475*** (SE 0.150)
    - f11.ccase: -0.570*** (SE 0.165)
    - f20.ccase: -1.291*** (SE 0.230)
    - f21.ccase: -1.318*** (SE 0.234)
    - f30.ccase: -1.387*** (SE 0.285)
  - Autoregressive dynamics: L.ccase coefficients generally positive and significant (examples: 0.172*** at f1; 0.615*** at f10; 0.399** at f30).
- Aggregate economic impacts (selected reported results):
  - Energy consumption growth:
    - Growth on average declined by around 12 percentage points after about 10 days, with some recovery thereafter.
    - Table A3 selected coefficients (el i effect on energy consumption growth):
      - f1d.en_gr: -0.0225*** (SE 0.00407)
      - f5d.en_gr: -0.0906*** (SE 0.0136)
      - f10d.en_gr: -0.128*** (SE 0.0205)
      - f11d.en_gr: -0.127*** (SE 0.0213)
      - f20d.en_gr: -0.0612** (SE 0.0263)
      - f21d.en_gr: -0.0516* (SE 0.0267)
      - f30d.en_gr: 0.0195 (SE 0.0283)
  - NO2 emissions:
    - Cumulative reduction by almost 86 percent after 30 days in states with stringent social distancing, relative to a baseline of no social distancing and containment (impulse estimates reported in Table A4; selected el i coefficients for NO2):
      - f1.n: -0.0384*** (SE 0.0111)
      - f5.n: -0.478*** (SE 0.128)
      - f10.n: -1.427*** (SE 0.403)
      - f20.n: -2.264* (SE 1.210)
      - f30.n: -2.683 (SE 2.428)
  - Flights:
    - Decline of over 75 percent at peak.
    - Table A4 selected el i coefficients for flights:
      - f1d.ff: -0.308*** (SE 0.0501)
      - f5d.ff: -1.000*** (SE 0.158)
      - f10d.ff: -1.410*** (SE 0.216)
      - f20d.ff: -1.208*** (SE 0.219)
      - f30d.ff: -0.455** (SE 0.196)
- Model fit and samples (selected):
  - R-squared for cases: 0.211 (f1) to 0.676 (f30); observations decline across horizons (e.g., f1: 9,008; f10: 7,608; f20: 4,198; f30: 1,110); number of state panels: 31.
  - Energy models: R-squared from 0.925 (f1) to 0.468 (f10); f1 observations 10,558; number of states 32.
  - NO2 models: number of state panels 15; f1 observations 4,957; f30 observations 522.
  - Flights models: number of state panels 29; f1 observations 9,559; f30 observations 719.
- Robustness:
  - Results hold using alternative high-frequency proxies (NO2, flights), using de-facto mobility-only measure, and using principal component analysis of stringency and mobility.
  - Robust to alternative lag structures, longer horizons, alternative standard errors (e.g., Driscoll-Kray), sub-samples, and additional controls (non-linear trends, daily time fixed effects).

### V. What drives heterogeneity at the state level?
- Factors examined (six categories): (i) health infrastructure, (ii) vulnerability of the population to COVID-19, (iii) ease of social distancing, (iv) economic structure, (v) income level, (vi) government effectiveness.
- Estimation approach:
  - Augmented local projections using smooth transition autoregressive model (Granger and Terävistra 1993); impulse responses vary smoothly with normalized state-specific characteristic z; cutoff F(z)=0.5 separates low and high regimes.
- Health outcomes — key heterogeneous effects:
  - Social distancing is more effective in reducing cases where:
    - Health infrastructure is stronger (health expenditure per capita; doctor density).
      - Table A5 — interaction with doctor density (el i coefficients):
        - High doctor density (c.dh#c.eli):
          - f1.ccase: -0.0927** (SE 0.0389)
          - f10.ccase: -0.788** (SE 0.320)
          - f20.ccase: -1.911*** (SE 0.517)
          - f30.ccase: -2.031*** (SE 0.630)
        - Low doctor density (c.dl#c.eli):
          - f1.ccase: -0.0261 (SE 0.0316)
          - f10.ccase: -0.144 (SE 0.270)
          - f20.ccase: -0.803* (SE 0.448)
          - f30.ccase: -0.706 (SE 0.570)
      - Quantitative example: cases decline by close to 90 percent for states with high health expenditure and doctor density, relative to baseline of no social distancing and containment (Annex Figure A2 and Table A5).
    - Share of vulnerable population is higher (share with comorbidities, share aged 60+, life expectancy).
    - Literacy rates are higher.
    - Population density is lower (easier to social distance).
    - Government effectiveness is higher (governance performance index).
- Economic outcomes — key heterogeneous effects:
  - Social distancing imposes larger economic costs in states with:
    - Higher shares of services, particularly contact-intensive services.
      - Table A6 — interaction with share of services (el i coefficients for energy growth, high services):
        - f1d.en_gr: -0.0268*** (SE 0.00640)
        - f5d.en_gr: -0.102*** (SE 0.0274)
        - f10d.en_gr: -0.152*** (SE 0.0445)
        - f11d.en_gr: -0.154*** (SE 0.0463)
        - f20d.en_gr: -0.0835* (SE 0.0448)
        - f30d.en_gr: 0.0290 (SE 0.0510)
      - Low share of services shows smaller and less statistically significant impacts.
    - Higher urbanization.
  - Persistence and recovery patterns:
    - Impact more persistent in states with lower GDP per capita and weaker health care infrastructure (lower doctor density).
    - Richer states: larger initial impact (about a week to a fortnight) but bounce back after about 20 days.
    - States with higher doctor density: smaller and more transitory economic costs, with positive benefits after about 20 days.
    - Low government effectiveness associated with more profound and persistent economic impacts; high government effectiveness associated with lower and more transient impacts.

### VI. Conclusions and policy implications
- Main conclusions:
  - Social distancing and containment were very effective in reducing COVID-19 cases but imposed high economic costs.
  - Health benefits of social distancing are greatest where health infrastructure is stronger, vulnerable population shares and literacy rates are higher, population density is lower, and governance is better.
  - Economic costs are largest where the share of services and urbanization are high; costs are more persistent in lower GDP per capita and weaker health infrastructure states; better governance reduces economic costs.
- Policy implications and recommendations:
  - Adequate spending and better health care infrastructure are crucial for containing health crises and limiting economic costs.
  - Strengthen health infrastructure (health care spending per capita, doctor density) to improve both health outcomes and economic recovery speed.
  - Improve governance performance to enhance effectiveness of containment measures and reduce economic fallout.
  - Target social safety nets to support the most vulnerable populations, including the urban poor and workers in service-intensive sectors.

### Annex figures and tables — key data and methodological notes
- Estimation sample and period common to figures and tables: sample of 34 states using daily data from February 15, 2020 to January 12, 2021; confidence bands at 95 and 75 percent; horizons indicated as "ℎ=0, 1, ...퐻퐻" and lag structure "ℓ=1, 2 ...ℒ".
- Figure A1 robustness variants:
  - Mobility-only measure; PCA of official containment measures and mobility; Oxford stringency measure and mobility.
  - Note: impulse responses shown for cases (cumulative) and energy consumption growth (daily) with confidence bands at 95 and 75 percent; model specification and estimation details provided verbatim in figure notes.
- Figures A2–A9 show interaction analyses for:
  - Health infrastructure (Health Expenditure per Capita; Doctor Density)
  - Comorbidities (Diabetes Mellitus; Chronic Respiratory Diseases; Cardiovascular Diseases; Cancer)
  - Demographics (Share of population over 60; Life expectancy)
  - Implementation ease (Population Density; Literacy Rate)
  - Services share and contact-intensive services
  - Urbanization
  - Income level and doctor density (State GDP per Capita; Doctor Density)
  - Government effectiveness (Health Outcome; Economic Outcome)
  - Common note text for Figures A2–A9 provides the smooth transition local projection specification used.
- Annex Table A1: data definitions and sources (CEIC, POSOCO, Air Quality Open Data Platform, FlightRadar24, Google Mobility Report, IMF Lockdown Stringency Index, Oxford Stringency Index, RBI, CMIE, Census, NIPFP).
- Annex Table A2: cross-correlations among state-level characteristics (selected correlations preserved exactly, e.g., Doctor Density with Comorbidities: 0.79; Income with Urbanisation: 0.87; Governance with Doctor Density: 0.71).
- Annex Tables A3–A6: detailed regression results, selected coefficients and standard errors reported above; significance annotations: *** p<0.01, ** p<0.05, * p<0.1.

*Source: IMF staff analysis from the document "Annex Tables" of wpiea2021279-print-pdf (state-level analysis of COVID-19 impacts in India).*

### References .............................................................................................................

### wpiea2021279-print-pdf - References .............................................................................................................

### Page entries
- References ............................................................................................................................... 21
- Annex Figures ......................................................................................................................... 23

*wpiea2021279-print-pdf - References*

### Annex Tables ...........................................................................................................

### wpiea2021279-print-pdf - Annex Tables ...........................................................................................................

### I. Introduction
- As of June 24, 2021, India has more than 30 million confirmed cases.
- The stringent national lockdowns led to a 24 percent decline in GDP in the second quarter of last year.
- Objective: develop a de-facto state-level social distancing measure by combining an ordinal index of containment stringency and observed mobility trends at the state level to quantify impacts on health and economic outcomes and examine the role of state-level factors (health infrastructure, population and economic characteristics, governance).
- Empirical approach: local projections methods (Jordà 2005); comprehensive daily state-level dataset; sample period runs from mid-February 2020 to mid-January 2021, covering the first wave.

### II. Data and stylized facts
- Mobility: Google mobility at the state level; containment measures from a new IMF containment index and Oxford University index used for robustness.
- Health indicators: official confirmed cases at state level from CEIC.
- Daily economic indicators: electricity consumption from POSOCO, NO2 emissions from Air Quality Open Data Platform, flights data from FlightRadar24.
- State characteristics: health care spending per capita, doctor density, population density, share of population aged 60 and over, share with comorbidities, share of services, GDP per capita, governance (compiled from RBI, NSO, NIPFP, Census of India, Density of India State of Population Census, CMIE, CEIC).
- Sample: daily data from mid-February 2020 to mid-January 2021; state-level characteristics based on latest available information.
- State heterogeneity examples:
  - Health: Maharashtra, Kerala, Karnataka, and Andhra Pradesh among worst hit in absolute and per-capita confirmed cases.
  - Economic: NO2 emissions declined most in West Bengal and Gujarat; energy consumption falls largest in Sikkim, Himachal Pradesh, and Punjab.
  - Mobility: sharp inter-state differences after national lockdowns; example: mobility decline in Maharashtra much more marked than in Rajasthan.

### III. Mobility, social distancing, and containment
- Effective state-level social distancing index constructed from national-level IMF containment index and state-level mobility; national IMF index covers six economic sectors (international travel, schools, retail, industry, services, public gatherings) and captures announcements about future changes.
- Mobility components used: Google retail and recreation mobility (proxy for discretionary spending) and transit stations mobility (proxy for workplace behavior); these have highest absolute correlation with Oxford Stringency Index in India.
- Observations:
  - High effective social distancing coincides with sharp reductions in growth rate of confirmed cases (example states: Maharashtra, Rajasthan).
  - Negative relationship between social distancing and energy consumption at state level across states with different energy patterns.

### IV. The impact of effective social distancing
- Methodology:
  - Local projections (Jordà 2005) estimating impulse responses for horizons h=0,..,30.
  - Regression specification controls for lagged infections and economic variables, state-specific time trends, fixed effects; robust standard errors clustered at state level.
  - Balanced sample of 34 states; estimation period February 15, 2020 to January 12, 2021.
- Aggregate results:
  - Social distancing and containment reduced confirmed COVID-19 cases by around 80 percent relative to a baseline of no social distancing and containment (impulse responses in log-percentage points converted using (e^{θ_h}−1)∗100).
  - Energy consumption growth on average declined by around 12 percentage points after about 10 days, with some recovery thereafter.
  - NO2 emissions: cumulative reduction by almost 86 percent after 30 days in states with stringent social distancing, relative to a baseline of no social distancing and containment.
  - Flights: decline of over 75 percent at peak.
- Robustness:
  - Results hold using alternative high-frequency proxies (NO2, flights), using de-facto mobility-only measure, and using principal component analysis of stringency and mobility.
  - Robust to alternative lag structures, longer horizons, alternative standard errors (e.g., Driscoll-Kray), sub-samples, and additional controls (non-linear trends, daily time fixed effects).
- Reference: detailed regression tables underlying impulse responses reported in Annex Tables A3 and A4.

### V. What drives heterogeneity at the state level?
- Factors examined (six categories): (i) health infrastructure, (ii) vulnerability of the population to COVID-19, (iii) ease of social distancing, (iv) economic structure, (v) income level, (vi) government effectiveness.
- Estimation approach for heterogeneity:
  - Augmented local projections using smooth transition autoregressive model (Granger and Terävistra 1993).
  - Equation allows impulse responses to vary smoothly with a normalized state-specific characteristic z (zero mean, unit variance); weighting function F(z) in (0,1) with cutoff F(z)=0.5 between low and high regimes.
- Health outcomes — key findings:
  - Social distancing and containment measures are more effective in reducing cases where:
    - Health infrastructure is stronger (proxied by health care spending per capita and doctor density). Example: Delhi, Himachal Pradesh, Uttarakhand have high per-capita health expenditure; Andhra Pradesh, Tamil Nadu, Karnataka have high doctor density.
    - Share of vulnerable population is higher (share with comorbidities, share aged 60+, life expectancy); states with larger shares of comorbidities benefit more from social distancing.
    - Literacy rates are higher.
    - Population density is lower (easier to social distance).
    - Government effectiveness is higher (governance performance index from NIPFP).
  - Quantitative example: cases decline by close to 90 percent for states with high health expenditure and doctor density, relative to baseline of no social distancing and containment (see Annex Figure A2; regression tables in Annex Tables A5 and A6).
- Economic outcomes — key findings:
  - Social distancing and containment measures have the most adverse economic impact in states with:
    - Higher shares of services, particularly contact-intensive services (e.g., Delhi, Chandigarh have highest share of services in state GDP).
    - Higher urbanization — sharp decline in energy consumption immediately following social distancing with gradual recovery as adaptation (e.g., remote work).
  - Persistence and recovery:
    - Impact tends to be more persistent in states with lower GDP per capita and weaker health care infrastructure (lower doctor density).
    - Richer states: more affected initially (about a week to a fortnight) but bounce back after about 20 days; poorer states see more persistent decline.
    - States with higher doctor density: smaller and more transitory economic costs, with positive benefits after about 20 days.
    - Low government effectiveness associated with more profound and persistent economic impacts; high government effectiveness associated with lower and more transient impacts.

### VI. Conclusions and policy implications
- Main conclusions:
  - Social distancing and containment were very effective in reducing COVID-19 cases but imposed high economic costs.
  - Health benefits of social distancing are greatest where health infrastructure is stronger, vulnerable population shares and literacy rates are higher, population density is lower, and governance is better.
  - Economic costs are largest where the share of services and urbanization are high; costs are more persistent in lower GDP per capita and weaker health infrastructure states; better governance reduces economic costs.
- Policy implications and recommendations:
  - Adequate spending and better health care infrastructure are crucial for containing health crises and limiting economic costs.
  - Strengthen health infrastructure (health care spending per capita, doctor density) to improve both health outcomes and economic recovery speed.
  - Improve governance performance to enhance effectiveness of containment measures and reduce economic fallout.
  - Target social safety nets to support the most vulnerable populations, including the urban poor and workers in service-intensive sectors, given the large economic impact on services and urban centers.

*Source: IMF staff analysis from the document "Annex Tables" of wpiea2021279-print-pdf (state-level analysis of COVID-19 impacts in India).*

### REFERENCES

### REFERENCES

### Cited literature
- Beyer, R., Franco-Bedoya, S . and Galdo, V. 2021. 'Examining the economic impact of COVID-19 in India through daily electricity consumption and nighttime light intensity'. World Development, Volume 140, 2021.  
- Beyer, R., Jain, T., and Sinha, S. 2020. 'Lights Out ? COVID-19 Containment Policies and Economic Activity '. Policy Research Working Paper Series 9485, The World Bank.  
- Brotherhood L., Kircher P., Santos C., Tertilt M., 2020. An economic model of the COVID-19 epidemic: the importance of testing and age-specific policies. CEPR Discussion Paper DP14695, Centre for Economic Policy Research.  
- Chinazzi et al., 2020. The effect of travel restrictions on the spread of the 2019 novel coronavirus (COVID-19) outbreak. Science, 368, 395-400.  
- Deb, P., Furceri, D., Ostry, J. and Tawk, N. 2020a. 'The Effect of Containment Measures on the COVID-19 Pandemic'. London, Centre for Economic Policy Research.  
- Deb, P., Furceri, D., Ostry, J. and Tawk, N. 2020b. 'The Economic Effects of COVID-19 Containment Measures'. London, Centre for Economic Policy Research.  
- Eichenbaum M., Rebelo S., Trabandt, M., 2020. The Macroeconomics of Epidemics. NBER Working Papers 26882, National Bureau of Economic Research, Inc.  
- Engler, P., Pouokam, N., Guzman, D. R., and I. Yakadina. 2021. “Fiscal Redistribution and Inequality in the Time of a Pandemic”. IMF Working Paper, forthcoming.  
- Forslid R., Herzing M., 2020. Assessing the consequences of quarantines during a pandemic. CEPR Discussion Paper DP14699, Centre for Economic Policy Research.  
- Furceri, D., Kothari, S., and L. Zhang. 2021. ‘The Effects of COVID-19 Containment Measures on the Asia Pacific Region’. IMF Working Paper, forthcoming.  
- Granger, C.W.J. and Teräsvirta, T., 1993. Modelling Nonlinear Economic Relationships Oxford University Press. New York.  
- Hale, T., Angrist, N., Goldszmidt, R., Kira, B., Petherick, A., Phillips, T., Webster, S., Cameron-Blake, E., Hallas, L., Majumdar, S. and Tatlow, H., 2021. A global panel database of pandemic policies (Oxford COVID-19 Government Response Tracker). Nature Human Behaviour, 5(4), pp.529-538.  
- Hsiang, S., Allen, D., Annan-Phan, S., Bell, K., Bolliger, I., Chong, T., Druckenmiller, H., Huang, L.Y., Hultgren, A., Krasovich, E. and Lau, P., 2020. The effect of large-scale anti-contagion policies on the COVID-19 pandemic. Nature, pp.1-9.  
- IMF. 2020a. ' COVID-19 Lockdowns and Exits in Asia: Some Lessons'. Asia and Pacific Regional Economic Outlook, Chapter 2.  
- IMF. 2020b. 'The Great Lockdown: Dissecting the Economic Effects'. World Economic Outlook, Chapter 2.  
- Jordà, Ò., 2005. Estimation and Inference of Impulse Responses by Local Projections. American economic review, 95(1), pp.161-182.  
- Kahalé, N., 2020. On the economic impact of social distancing measures. Available at SSRN 3578415.  
- Kraemer et al., 2020. The effect of human mobility and control measures on the COVID-19 epidemic in China. Science, 368, 493–497.  
- Maloney W. and Taksin T., 2020. Determinants of social distancing and economic activity during COVID-19: A global view. Covid Economics: Vetted and Real-Time Papers. 2020(13), pp.157-177.  
- Mundle, S., Chakraborty, P., Chowdhury, S., and Sikar, S., 2012. 'The Quality of Governance : How Have Indian States Performed?'. National Institute of Public Finance and Policy Working Paper No. 2012-104.  
- Tian, H., et al., 2020. An investigation of transmission control measures during the first 50 days of the COVID-19 epidemic in China. Science10.1126/science.abb6105 (2020).

### Notes on methodology and models referenced in text
- Local projections / impulse response framework: referenced Jordà, Ò., 2005. Estimation and Inference of Impulse Responses by Local Projections.  
- Nonlinear modelling reference: Granger, C.W.J. and Teräsvirta, T., 1993. Modelling Nonlinear Economic Relationships.  
- Epidemiological and policy interaction models: Brotherhood et al. 2020; Eichenbaum et al. 2020; Forslid & Herzing 2020.

---

### ANNEX FIGURES

### Figure A1 — Robustness to alternative measures of social distancing
- Variants shown:
  - Social distancing measured by state level changes in mobility only  
  - Social distancing measured by Principal Component Analysis (PCA) of official containment measures and mobility  
  - Social distancing measured by the Oxford stringency measure and mobility
- Note text (verbatim):
  - "Impulse response functions are estimated using a sample of 34 states using daily data from February 15, 2020 to January 12, 2021. The graphs show the cumulative response for cases and daily response for energy consumption growth with confidence bands at 95 and 75 percent. The horizontal axis shows the response x days after effective social distancing. Estimates based on ∆퐼퐼푖푖,푡푡+ℎ = 푢푢푖푖 +qℎ푐푐푖푖,푡푡 +푋푋′푖푖,푡푡 Γℎ + ∑휓휓ℎ,ℓ ∆퐼퐼푖푖,푡푡−ℓ ℒ ℓ=1 +휀휀푖푖,푡푡+ℎ where ∆퐼퐼푖푖,푡푡+ℎ =퐼퐼푖푖,푡푡+ℎ − 퐼퐼푖푖,푡푡+ℎ−1 and 퐼퐼푖푖,푡푡 is the log of the number of infections or economic indicator in state 푖푖 at date 푡푡. The model is estimated at each horizon ℎ=0, 1, ...퐻퐻, with a lag structure ℓ=1, 2 ...ℒ; 푐푐푖푖,푡푡 is the measure of effective state-level containment and social distancing; 푋푋 is a matrix of time varying control variables and state specific time trend; and 푢푢푖푖 captures state level fixed effects."

### Figures A2–A9 — Interaction analyses (health infrastructure, comorbidities, demographics, implementation ease, services share, urbanization, income, government effectiveness)
- Figure A2: Interaction with Health Infrastructure
  - Panels: Health Expenditure per Capita; Doctor Density, number of doctors per 1000 people
- Figure A3: Interaction with Comorbidities
  - Panels: Prevalence of Diabetes Mellitus, Chronic Respiratory Diseases, Cardiovascular Diseases and Cancer
- Figure A4: Interaction with Share of Population over 60 and Life Expectancy
  - Panels: Share of population over 60 years of Age; Life expectancy
- Figure A5: Interaction with Ease of Implementing Social Distancing
  - Panels: Population Density; Literacy Rate
- Figure A6: Interaction with Share of Services
  - Panels: Services Share; Contact-intensive Services Share
- Figure A7: Interaction with Urbanization
  - Panel: (urbanization interaction depicted)
- Figure A8: Interaction with income level and health infrastructure
  - Panels: State GDP per Capita; Doctor Density, number of doctors per 1000 people
- Figure A9: Role of Government Effectiveness
  - Panels: Health Outcome; Economic Outcome

- Common note text for Figures A2–A9 (verbatim):
  - "Note: Impulse response functions are estimated using a sample of 34 states using daily data from February 15, 2020 to January 12, 2021. The graph shows the cumulative response and confidence bands at 95 and 75 percent. The horizontal axis shows the response x days after effective social distancing. Estimates based on ∆퐼퐼푖푖,푡푡+ℎ = 푢푢푖푖 + 휃휃ℎ퐿퐿퐹퐹(푧푧푖푖,푡푡)푐푐푖푖,푡푡 +휃휃ℎ퐻퐻(1−퐹퐹(푧푧푖푖,푡푡))푐푐푖푖,푡푡 + 푋푋′푖푖,푡푡 훤훤ℎ + ∑퐹퐹(푧푧푖푖,푡푡) 휓휓ℎ,ℓ ∆퐼퐼푖푖,푡푡−ℓ ℒ ℓ=1 + ∑(1−퐹퐹(푧푧푖푖,푡푡))휓휓ℎ,ℓ ∆퐼퐼푖푖,푡푡−ℓ ℒ ℓ=1 +휀휀푖푖,푡푡+ℎ with 퐹퐹(푧푧푖푖푡푡) = 푒푒푒푒푒푒 −훾훾푧푧푖푖푖푖 (1−푒푒푒푒푒푒 −훾훾푧푧푖푖푖푖 ),훾훾>0 where ∆퐼퐼푖푖,푡푡+ℎ =퐼퐼푖푖,푡푡+ℎ −퐼퐼푖푖,푡푡+ℎ−1 and 퐼퐼푖푖,푡푡 is the log of the number of infections in state 푖푖 at date 푡푡. The model is estimated at each horizon ℎ=0, 1, ...퐻퐻, with a lag structure ℓ=1, 2 ...ℒ; 푐푐푖푖,푡푡 is the measure of effective state-level containment and social distancing; 푋푋 is a matrix of time varying control variables and state specific time trend; and 푢푢푖푖 captures state level fixed effects."

- Estimation sample and period common to all figures:
  - "sample of 34 states using daily data from February 15, 2020 to January 12, 2021"
  - Confidence bands shown at "95 and 75 percent"
  - Horizons indicated as "ℎ=0, 1, ...퐻퐻" and lag structure "ℓ=1, 2 ...ℒ"

*Content derived from wpiea2021279-print-pdf - REFERENCES*

### ANNEX TABLES

### ANNEX TABLES

### Table A1: Data Definitions and Sources
- Total confirmed COVID cases: CEIC
- Daily confirmed COVID cases: CEIC
- Electricity consumption: POSOCO
- NO2 emissions: Air Quality Open Data Platform
- Daily indicators—Flight (sum of daily departure and arrivals): FlightRadar24
- Google mobility indicators—retail and recreation, transit stations: Google Mobility Report
- National lockdown stringency: IMF Lockdown Stringency Index (Furceri, Kothari and Zhang (forthcoming)); Oxford Stringency Index (Oxford University)
- Health infrastructure:
  - Health expenditure per capita: RBI: State Finances 2020-2021
  - Doctor density: Doctors registered with State Medical Councils/Medical Council of India in 2016, per thousand (CMIE, CEIC, and IMF staff calculations)
- Vulnerable population: Population over 60 (Share/percent of population in 60+ age cohort) — RBI: State Finances 2020-2021
- Comorbidities (prevalence in 2017): Diabetes Mellitus; Chronic Respiratory Diseases; Cardio Vascular Diseases; Cancer — RBI: State Finances 2020-2021
- Life expectancy in years: RBI: State Finances 2020-2021
- Ease of social distancing:
  - Population density per sq.km in 2011: Density of India State of Population Census 2011
  - Literacy rate in percent among persons of age 7 years and above: NSS Report No.585
- Economic structure:
  - GDP shares by sector (percent of trade, hotels, transport, etc. in state of GDP): CMIE
  - Percent of services in state GDP: CMIE
  - Urbanization: Percentage of urban population to total population in 2001 (Census of India Chapter IV: Trends in Urbanization)
- Income: State GDP per capita for 2018-19 — RBI: State Finances 2020-2021
- Governance: Government effectiveness (Governance ranking, average of average of ranks, actual data) — National Institute of Public Finance and Policy: S. Mundle et. al. (2012)

### Table A2: Cross-Correlations — State-Level Characteristics (selected entries)
- Health Expenditure per capita correlations:
  - with Doctor Density: 0.40
  - with Comorbidities: 0.46
  - with Share of 60+ population: 0.41
  - with Life expectancy: 0.12
- Doctor Density correlations:
  - with Comorbidities: 0.79
  - with Share of 60+ population: 0.83
  - with Life expectancy: 0.64
- Population density correlations:
  - with Health Expenditure per capita: -0.46
  - with Life expectancy: 0.32
- Literacy rate correlations:
  - with Life expectancy: 0.57
  - with Share of services: 0.45
- Share of services correlations:
  - with Urbanisation: 0.26
- Income (GDP per capita) correlations:
  - with Health Expenditure per capita: 0.46
  - with Governance (Government effectiveness): 0.64
  - with Urbanisation: 0.87
- Governance (Government effectiveness) correlations:
  - with Health Expenditure per capita: 0.64
  - with Doctor Density: 0.71

### Table A3: Impact of Effective Social Distancing — Confirmed cases cumulative difference (selected lags and statistics)
- Coefficient on el i (effect of effective social distancing) across f1.ccase to f30.ccase:
  - f1.ccase: -0.0535** (SE 0.0216)
  - f5.ccase: -0.159* (SE 0.0865)
  - f10.ccase: -0.475*** (SE 0.150)
  - f11.ccase: -0.570*** (SE 0.165)
  - f20.ccase: -1.291*** (SE 0.230)
  - f21.ccase: -1.318*** (SE 0.234)
  - f30.ccase: -1.387*** (SE 0.285)
- Autoregressive dynamics (examples):
  - L.ccase coefficients generally positive and significant (e.g., 0.172*** at f1; 0.615*** at f10; 0.399** at f30)
  - L2.ccase, L3.ccase, L4.ccase often negative and significant across specifications
- Time trend (ti me) estimates:
  - f1.ccase: -5.93e-05 (SE 4.33e-05)
  - f10.ccase: -0.000421 (SE 0.000464)
  - f20.ccase: -0.00175* (SE 0.000885)
  - f30.ccase: -0.00193 (SE 0.00120)
- Model fit and sample:
  - R-squared ranges from 0.211 (f1) to 0.676 (f30)
  - Observations decline across horizons (e.g., f1: 9,008; f10: 7,608; f20: 4,198; f30: 1,110)
  - Number of state panels: 31
- Robust standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

### Table A3: Impact on Energy consumption growth difference (selected coefficients)
- Coefficient on el i (effect of effective social distancing) across horizons:
  - f1d.en_gr: -0.0225*** (SE 0.00407)
  - f5d.en_gr: -0.0906*** (SE 0.0136)
  - f10d.en_gr: -0.128*** (SE 0.0205)
  - f11d.en_gr: -0.127*** (SE 0.0213)
  - f20d.en_gr: -0.0612** (SE 0.0263)
  - f21d.en_gr: -0.0516* (SE 0.0267)
  - f30d.en_gr: 0.0195 (SE 0.0283)
- Autoregressive structure:
  - L.en_gr positive and significant at short horizons (e.g., 0.928*** at f1) declining over horizons (e.g., 0.383*** at f10; 0.168*** at f30)
  - L2.en_gr negative and symmetric to L.en_gr at short horizons (e.g., -0.929*** at f1)
- Time trend:
  - f1: 2.28e-05* (SE 1.18e-05)
  - f10: 0.000312*** (SE 6.99e-05)
  - f20: 0.000499*** (SE 8.93e-05)
  - f30: 0.000585*** (SE 0.000103)
- Model fit and sample:
  - R-squared declines from 0.925 (f1) to 0.468 (f10) in early panels; later panels show R-squared around 0.396–0.417
  - Observations and number of states: e.g., f1 observations 10,558; number of states 32

### Table A4: Impact of Effective Social Distancing on other proxies for economic activity

- NO2 Emissions cumulative difference (selected coefficients):
  - el i:
    - f1.n: -0.0384*** (SE 0.0111)
    - f5.n: -0.478*** (SE 0.128)
    - f10.n: -1.427*** (SE 0.403)
    - f20.n: -2.264* (SE 1.210)
    - f30.n: -2.683 (SE 2.428)
  - L.n and L2.n show large symmetric positive and negative coefficients at short horizons (e.g., L.n 0.955*** and L2.n -0.955*** at f1)
  - Time trend estimates increase with horizon (e.g., f1: 0.000154; f10: 0.00373)
  - R-squared values: high at short horizons (0.960 at f1) declining to about 0.765 at f30
  - Number of state panels: 15; observations decline across horizons (e.g., f1: 4,957; f30: 522)

- Total flights difference (selected coefficients):
  - el i:
    - f1d.ff: -0.308*** (SE 0.0501)
    - f5d.ff: -1.000*** (SE 0.158)
    - f10d.ff: -1.410*** (SE 0.216)
    - f20d.ff: -1.208*** (SE 0.219)
    - f30d.ff: -0.455** (SE 0.196)
  - L.f and L2.f display strong autoregressive alternation (e.g., L.f 0.925***, L2.f -0.925*** at f1)
  - Time trend positive and increasing with horizon (e.g., f1: 0.000272*; f10: 0.00163*)
  - R-squared declines from 0.982 (f1) to 0.542 (f10) and lower thereafter
  - Number of state panels: 29; observations decrease across horizons (e.g., f1: 9,559; f30: 719)

### Table A5: Interaction with Health Infrastructure — Confirmed cases cumulative difference by doctor density
- Interaction terms (c.dh#c.eli for High doctor density; c.dl#c.eli for Low doctor density) — selected coefficients:
  - High doctor density (c.dh#c.eli):
    - f1.ccase: -0.0927** (SE 0.0389)
    - f10.ccase: -0.788** (SE 0.320)
    - f20.ccase: -1.911*** (SE 0.517)
    - f30.ccase: -2.031*** (SE 0.630)
  - Low doctor density (c.dl#c.eli):
    - f1.ccase: -0.0261 (SE 0.0316)
    - f10.ccase: -0.144 (SE 0.270)
    - f20.ccase: -0.803* (SE 0.448)
    - f30.ccase: -0.706 (SE 0.570)
- Autoregressive terms and dynamics similar to main specifications (L.ccase positive; L2–L4 negative)
- Time trend examples:
  - f1: -6.31e-05 (SE 5.72e-05)
  - f10: -0.000466 (SE 0.000558)
  - f20: -0.00226** (SE 0.000927)
  - f30: -0.00267** (SE 0.00116)
- Model fit and sample:
  - R-squared increases with horizon (e.g., 0.272 at f1 to 0.753 at f30)
  - Observations: e.g., f1 6,706; f10 5,226; f20 2,926; f30 1,040
  - Number of state panels: 23

### Table A6: Interaction with Share of Services — Energy consumption growth percentage points by service-share
- Interaction terms (c.dh#c.eli for High share of services; c.dl#c.eli for Low share of services) — selected coefficients:
  - High share of services (c.dh#c.eli):
    - f1d.en_gr: -0.0268*** (SE 0.00640)
    - f5d.en_gr: -0.102*** (SE 0.0274)
    - f10d.en_gr: -0.152*** (SE 0.0445)
    - f11d.en_gr: -0.154*** (SE 0.0463)
    - f20d.en_gr: -0.0835* (SE 0.0448)
    - f30d.en_gr: 0.0290 (SE 0.0510)
  - Low share of services (c.dl#c.eli):
    - f1d.en_gr: -0.0146 (SE 0.00916)
    - f5d.en_gr: -0.0666* (SE 0.0352)
    - f10d.en_gr: -0.0873 (SE 0.0584)
    - f11d.en_gr: -0.0844 (SE 0.0614)
    - f20d.en_gr: -0.0280 (SE 0.0612)
    - f30d.en_gr: 0.0177 (SE 0.0486)
- Autoregressive dynamics for energy growth mirror main models (L.en_gr positive; L2.en_gr negative)
- Time trends:
  - f1d.en_gr: 2.54e-05** (SE 1.20e-05)
  - f10d.en_gr: 0.000330*** (SE 6.78e-05)
  - f20d.en_gr: 0.000515*** (SE 8.56e-05)
  - f30d.en_gr: 0.000591*** (SE 9.93e-05)
- Model fit and sample:
  - R-squared declines across short-horizon panels from 0.920 (f1) to 0.440 (f10) in first block; later panels show R-squared around 0.367–0.384
  - Observations and number of states: e.g., f1 observations 10,232; number of states 31

*Source: ANNEX TABLES (wpiea2021279-print-pdf - ANNEX TABLES)*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021279-print-pdf.pdf_