## wpiea2022149-print-pdf

## Source details

**Canonical URL:** [wpiea2022149-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022149-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022149-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022149-print-pdf.pdf.json)

---

### Introduction — context and motivation
- Early 2020: exponential COVID-19 case and death growth led to border, school, and workplace closures and widespread social distancing.
- The COVID-19 recession produced an unprecedented loss in output, greater than during the global financial crisis.
- Policymakers used a range of tools; fiscal policy played a key role supporting health care and assisting businesses and households.

### Contribution, methodology, and scope
- Scope: investigates fiscal multipliers during health crises (EVD, SARS, H1N1, MERS-CoV, COVID-19) using cross-country annual data.
- Empirical method: Jorda’s (2005) local projection (LP) methodology, with linear and state-dependent panel LPs and one-step IV multiplier estimation following Ramey and Zubairy (2018).
- Sample: 91 countries, annual data 1980–2020 (excludes countries with population < 1 million).
- Key dependent variables and scaling:
  - Real GDP growth scaled by lagged real GDP: (Yi,t+h − Yi,t−1)/Yi,t−1.
  - Real government expenditure growth scaled by lagged real GDP: (Gi,t+h − Gi,t−1)/Yi,t−1.

### Data and identification
- Fiscal variables: total public expenditure, public consumption, public investment from WEO and CEIC Global Economic Monitor.
- Disease outbreak data: WHO; epidemic dummy = 1 if case number per thousand population > median (health-crisis-specific thresholds).
- Transmission channel proxies:
  - Uncertainty: World Pandemic Uncertainty Index (Ahir, Bloom and Furceri, 2018).
  - Suppressed demand: gross private savings from WEO (excess private savings = deviation from trend).
  - Supply bottlenecks: number of air passengers (ICAO) and container traffic in ports (UNCTAD), detrended.
- Baseline identification: assumes no contemporaneous response of government expenditures to macro aggregates (Blanchard and Perotti, 2002) at annual frequency.
- Alternative identification: WEO forecast errors of public expenditures used as robustness (sample note: some WEOs released in September instead of October included).

### Section VI — transmission channels (heightened uncertainty, supply bottlenecks, suppressed demand)
- Heightened uncertainty
  - Theoretical expectation: uncertainty can dampen fiscal policy effectiveness; stimulus may become more effective once uncertainty subsides.
  - Empirical evidence is mixed: some studies find lower responsiveness during high uncertainty, others find fiscal policy eases tight financial conditions coinciding with uncertainty.
- Supply bottlenecks
  - Pandemic containment and disrupted supply chains reduced labor, investment, and availability of goods and services.
  - Demand-stimulating measures may be less effective when inadequate supply is the downturn source.
  - Re-opening associated with surges in consumption, output, and inflation due to pent-up demand.
- Suppressed demand
  - Spike in unemployment and income-at-risk led to declines in consumer confidence and household consumption.
  - Transfers can stimulate demand but may be less effective when sectors are closed or constrained; evidence that 2020 stimulus checks were mostly saved in areas with strict lockdowns.

### Main empirical results — baseline and disaggregation
- Baseline aggregate multipliers (total public expenditure, Blanchard and Perotti identification):
  - Contemporaneous multiplier (year T): 0.4.
  - Cumulative multiplier in the medium term: 0.5-0.6.
- State-dependent total public expenditure (Equation 4) — pandemic vs non-pandemic:
  - Pandemic state:
    - Contemporaneous multiplier: 0.5.
    - Cumulative multiplier first two years after the shock: 0.7.
  - Non-pandemic state:
    - Contemporaneous multiplier: 0.38.
    - Cumulative multiplier first two years after the shock: 0.55.
  - Statistical test: largest difference at t = 1 with p-value 0.12 (not statistically different overall sample for that comparison).
- Public consumption multipliers (Figure 2):
  - Medium-term multiplier lower; stands at 0.32 in the third year after the shock.
  - Pandemic-state consumption multiplier reaches 0.9 in the first year after the shock.
  - Multipliers in pandemic and non-pandemic states are statistically different.
- Public investment multipliers (Figure 3):
  - Investment multipliers are significantly larger in both states.
  - Multipliers during pandemics exceed 2 one year after the shock and are statistically higher than non-pandemic multipliers.

### Heterogeneity by country group
- Advanced economies:
  - Fiscal multipliers significantly larger during pandemics; about double one year after the shock, with the pandemic multiplier close to 1 versus about 0.5 in non-pandemic times (paper reports nearly 1 in advanced economies one year after the shock).
- Developing economies:
  - Fiscal multipliers reach about 0.6 one year after the shock.
  - No statistically significant difference between pandemic and non-pandemic states for developing economies.
  - Possible interpretation: spending inefficiencies could worsen during a pandemic, constraining effectiveness.

### Transmission channels — empirical contributions and relative importance
- Estimation extends LPs with interaction terms for channel proxies (Equations 5 and 6); focus on p-values testing pandemic vs non-pandemic multiplier differences after including proxies.
- Including proxies (uncertainty, excess private savings, detrended air passengers, detrended container traffic) brings pandemic and non-pandemic multipliers closer together for total expenditures, consumption, and investment.
- Tests of difference: p-values increase markedly after adding proxies; at h = 0 and h = 1 tests very confidently reject statistical differences where previously present — suggesting these channels can underpin larger fiscal multipliers during health crises.
- Relative importance (Table A.3 highlights):
  - Uncertainty: largest immediate impact at onset; positive Shock´Uncert.*IIt coefficient at h=0 is 0.113***.
  - Excess private savings: positive Shock*Exc. Save*IIt coefficients (h=0 0.854; h=1 0.877*; h=2 1.653**) imply fiscal policy is more effective when private consumption impaired.
  - Container traffic: Shock*Contain. traffic*IIt shows negative coefficients at h=1 -0.199***, h=2 -0.284**, h=3 -0.348*** (Table A.3), interpreted as capturing demand deficiencies more than supply constraints in the regression patterns.
- Non-pandemic regime: multipliers virtually unchanged by inclusion of channel proxies.

### Robustness and additional exercises
- Including 2020 in the baseline sample yields qualitatively and quantitatively similar results.
- Results robust to adding controls: additional lags, domestic credit growth, current account balance.
- Debt-to-GDP interaction:
  - High debt levels tend to decrease fiscal policy effectiveness in line with literature.
  - This weakening is not observed in the pandemic regime; fiscal response effectiveness is less constrained by public debt during health crises.
- Fiscal rules:
  - Presence of at least one fiscal rule at pandemic onset associated with more effective fiscal spending (Shock * FiscalRule * IIt: T=0 0.522**), consistent with a credibility channel.
- Alternative identification using WEO forecast errors (sample 2002–2019, 49 countries) confirms main findings: short- and medium-term fiscal multipliers are higher in the pandemic regime and differences are statistically significant.

### Policy implications and conclusions
- Fiscal multipliers are larger in the immediate years following the onset of a health crisis.
- Baseline numerical benchmarks:
  - Contemporaneous multiplier: 0.4 in year T.
  - Medium-term cumulative multiplier: 0.5-0.6.
- Advanced economies: cumulative multiplier in the pandemic rises to almost 1 one year after the shock versus about 0.4 in non-pandemic regime.
- Transmission channels (uncertainty, suppressed demand, supply bottlenecks) can explain differences between pandemic and non-pandemic multipliers; controlling for them removes statistical differences.
- Policy takeaway: growth impact of fiscal stimulus during health crises could be larger and longer lasting than often assumed — potentially at least 50 percent larger one year after large-scale fiscal stimulus compared to expectations based on non-pandemic multipliers.

### Selected annex summary statistics and notable table highlights
- Sample and series summary (selected):
  - Sample observations vary by series (examples): GDP at 2014p (mn. USD) Observations 1947; Total Government Expenditure Observations 2295; Private Consumption Expenditure Observations 2771.
  - Dummy Variable for Deaths Above Median: Mean 0.01, Std . Dev. 0.11, Observations 3380.
  - Government Debt (% of GDP) reported in annex with formatting anomalies in source but retained as presented.
- Table A2 (transmission channel t-tests):
  - Uncertainty: Pandemic Mean 5.247, Non-pandemic Mean 0.182, P-value 0.000.
  - Excess private saving (mill.): Pandemic Mean -0.400.154, Non-pandemic Mean 0.093, P-value 0.000.
  - Air passengers (detrended): Pandemic Mean -0.9280.019, Non-pandemic Mean 0.014, P-value 0.000.
  - Container traffic (detrended): Pandemic Mean -3.7804.132, Non-pandemic Mean 5.111, P-value 0.000.
- Table A3 (selected coefficients, total expenditure, Shock interactions):
  - Shock´Uncert.*IIt: h=0 0.113***; h=1 0.011; h=2 -0.054; h=3 -0.158.
  - Shock*Exc. Save*IIt: h=0 0.854; h=1 0.877*; h=2 1.653**; h=3 0.812.
  - Shock*Psgr.*IIt: h=3 33.084**.
  - Shock*Contain. traffic*IIt: h=1 -0.199***; h=2 -0.284**; h=3 -0.348***.
- Table A4 (fiscal position and credibility highlights):
  - Shock * Debt * (1−IIt): T=0 -0.302*** (suggesting debt reduces multiplier outside pandemics at T=0).
  - Shock * FiscalRule * IIt: T=0 0.522** (suggesting fiscal rules enhance immediate effectiveness during pandemics).

*Source: Introduction, Section VI, Main Results, Annex Tables and Figures — wpiea2022149-print-pdf*

### Introduction............................................................................................................

### Introduction

### Context and motivation
- In early 2020 the world faced exponential growth in cases and deaths from COVID-19, prompting closures of borders, schools, and workplaces and widespread social distancing.
- The recession resulting from the COVID-19 pandemic produced an unprecedented loss in output, greater than during the global financial crisis.
- Policymakers deployed a range of tools, with fiscal policy playing a key and central role through support to the health care sector and assistance to businesses and households.

### Literature review and theoretical background
- State-dependent fiscal multipliers: Auerbach and Gorodnichenko (2012) show larger fiscal multipliers during recessions or periods of slack; subsequent studies corroborate these findings.
- Interaction with monetary policy and interest rates: multipliers tend to be larger when interest rates are near the zero lower bound or when monetary policy accommodates government spending.
- Cross-country determinants: Ilzetzki, Mendoza, and Végh (2013) find larger multipliers in industrial countries, under fixed exchange rates, and in closed economies.
- Recent literature on health-crisis macro policy:
  - Eichenbaum, Rebelo and Trabandt (2020) and Glover et al. (2020) embed epidemiology in real business cycle models to study optimal health policy responses.
  - Elenev, Landvoigt and Nieuwerburgh (2020) find fiscal support to distressed firms prevented corporate bankruptcies (without computing multipliers).
  - Auerbach et al. (2021b) estimate larger employment multipliers during 2020 lockdowns in US states, particularly with less stringent stay-at-home orders.
  - Bayer et al. (2020) find US fiscal multipliers around 1.5 for targeted transfers such as unemployment insurance and around 0.25 for untargeted lump-sum transfers during COVID-19.
  - Faria-e-Castro (2021) using a TANK model calibrated to US data: lump-sum transfers multiplier of 0.65; government consumption multiplier of about 1.25.
  - Wilson (2020) suggests multipliers were around 1.5 during the health crisis.
  - Guerrieri et al. (2022) present a theory with lower fiscal multipliers (below one) in presence of COVID-19 type shocks that shut specific sectors, versus larger multipliers (above one) for supply shocks affecting all segments.
- Empirical estimates at the zero lower bound: examples include multipliers ranging from 1.5 to 2.5 for Japan (Miyamoto, Nguyen, and Sergeyev, 2018) and around 1.5 for the United States (Ramey and Zubairy, 2018).

### Contribution and main findings of this paper
- Scope: investigates fiscal multipliers during episodes of health crises (e.g., COVID-19) using cross-country estimates and assessing potential transmission channels.
- Methodology: based on Jorda’s (2005) local projection methodology.
- Key empirical findings:
  - Cumulative fiscal multipliers one year after a health crisis are about twice larger than during normal times, particularly in advanced economies.
  - During health crises, multipliers associated with public investments are significantly larger than those associated with public consumption.
  - Higher debt levels tend to decrease the effectiveness of fiscal policy in normal times, while public debt seems to be less of a concern for an effective fiscal response during a health crisis.
  - The presence of a fiscal rule can further enhance the output effect of a fiscal expansion at the onset of a pandemic, most likely via a credibility channel.
- Transmission channels investigated: uncertainty, suppressed demand, and supply bottlenecks.
  - Controlling for these channels, fiscal multipliers in the pandemic and non-pandemic regimes are no longer statistically different.

### Structure of the paper
- Section II: discusses potential channels through which health crises could impact output and how fiscal policy could cushion these effects.
- Section III: presents the data.
- Section IV: discusses the empirical strategy.
- Section V: presents the baseline results.

*Source: Introduction, wpiea2022149-print-pdf*

### Section VI investigates the potential role of transmission channels through which the pandemic could impact

### wpiea2022149-print-pdf - Section VI investigates the potential role of transmission channels through which the pandemic could impact

### Health Crises and Economic Output: Potential Transmission Channels
- The section focuses on three categories of transmission channels active during the COVID-19 pandemic: heightened uncertainty, supply bottlenecks, and suppressed demand.
- Heightened Uncertainty
  - A rise in uncertainty could dampen the stimulative effect of fiscal policy because economic agents postpone hiring and investment.
  - Theoretical work suggests policy stimulus can become more effective once uncertainty subsides (Bloom, 2014; Bloom et al., 2018).
  - Empirical evidence is mixed:
    - Alloza (2017) finds the response of output to fiscal stimulus is low or insignificant during periods of high uncertainty, defined as unusually high implied stock market volatility.
    - Berg (2019) finds the response of output to fiscal stimulus increases with the level of uncertainty using a business uncertainty measure from firm-level data; the author argues fiscal policy helps ease tight financial conditions coinciding with uncertainty.
- Supply Bottlenecks
  - Early pandemic containment reduced in-person interactions, disrupted workplaces, caused large drops in employment, weaker investment, and worsened supply capacities.
  - The pandemic caused longer lasting scarring and disruptions in international supply chains in production and shipping, reducing availability of goods and services (Bonadio, et al., 2020; Zhang, 2020; Mahajan and Tomar, 2021; Lafrogne-Joussier, Martin and Mejean, 2021).
  - Theoretical work shows demand-stimulating measures tend to be less effective when inadequate supply is the source of the downturn (Ghassibe and Zanetti, 2021; Jo and Zubairy, 2021).
  - Auerbach, Gorodnichenko, and Murphy (2021) find fiscal stimulus has a reduced effect in presence of restrictions, reflecting muted consumption responses.
  - Re-opening is associated with a surge in consumption, output, and inflation due to pent-up demand.
  - Recent empirical studies show fiscal stimulus has been less effective in presence of containment measures (Coibion, Gorodnichenko, and Weber, 2020c; Auerbach, et al., 2021b; Brunet, 2021).
- Suppressed Demand
  - The pandemic and associated spike in unemployment led to a sharp decline in affected households’ income, increased income at risk, and higher household uncertainty.
  - Consumer confidence and household consumption expenditures dropped rapidly at the onset of the health crisis.
  - Fiscal measures such as cash transfers can stimulate demand (Eggertsson and Krugman, 2012; Kaplan and Violante, 2014; Ghassibe and Zanetti, 2021), but their impact can be reduced in a health crisis.
  - Guerrieri et al. (2020) highlight fiscal transfers could be less effective in sectors closed or constrained by containment measures.
  - U.S. experience: stimulus checks in 2020 were mostly saved in areas with strict lockdowns (Coibion et al. 2020a), while regions with less stringent lockdowns experienced significant employment increases in response to higher fiscal spending (Auerbach et al. 2021b).

### Data
- Sample: 91 countries with annual data spanning from 1980 to 2020.
  - Note: The sample excludes small countries with a population of less than 1 million.
- Output and fiscal variables: total public expenditure, public consumption, and public investment from the IMF’s World Economic Outlook (WEO) database and CEIC’s Global Economic Monitor dataset.
- Disease outbreak data: World Health Organization; health crises included: Ebola Virus Disease (EVD), Severe Acute Respiratory Syndrome (SARS), Influenza A (H1N1), Middle East Respiratory Syndrome Coronavirus (MERS-CoV), and Coronavirus Disease 2019 (COVID-19).
  - Case numbers per thousand population were aggregated at annual frequencies for each health crisis to create an epidemic/pandemic dummy variable taking value 1 when case number per thousand population is greater than the median value, and 0 otherwise (health crisis specific case number thresholds).
- Measures for transmission channels:
  - Uncertainty: World Pandemic Uncertainty Index from Ahir, Bloom and Furceri (2018).
  - Suppressed demand: gross private savings from the WEO.
  - Supply bottlenecks: number of air passengers from the International Civil Aviation Organization (ICAO) and container traffic in ports from UNCTAD.
- Fiscal rule dummy: captures presence or not of the rule (Schaechter et al. 2012) to gauge fiscal credibility.
- Alternative identification: WEO forecast errors of public expenditures used to identify unexpected expenditure shocks and provide an alternative method to estimate fiscal multipliers.
  - Note: For certain years the WEO was released in September instead of October which were included.
- Measure of uncertainty: derived from text mining methods (Ahir, Bloom, and Furceri, 2018).

### Empirical Strategy
- Method: Impulse responses produced by local projections (LP) as introduced by Jorda (2005).
  - Advantages of LP cited:
    1. Not constrained by the curse of dimensionality inherent to VARs (Ramey, 2016).
    2. Better captures nonlinearities (Auerbach and Gorodnichenko, 2012).
    3. Avoids misspecification errors and biases compounded at each horizon when using VAR (Jorda 2005).
    4. Facilitates accommodation of state dependency (Auerbach and Gorodnichenko 2013).
- Estimated models:
  - Linear and state-dependent panel local projections as specified in Equation (1) and Equation (2) with country fixed effects 훼훼푖푖ℎ and time fixed effects 푐푐푡푡ℎ.
  - Dependent variables 푍푍푖푖,푡푡 are either real GDP or real government expenditure; control variables include one lag of real GDP growth, one lag of government expenditure growth and the lagged output gap.
  - Driscoll and Kraay (1998) standard errors reported to correct for potential heteroskedasticity, autocorrelation, or correlated errors across countries.
- Scaling and dependent variable definitions:
  - Use variable definition of Hall (2009) and Owyang, Ramey, and Zubairy (2013) and scale dependent variables by lagged real GDP.
  - Two dependent variables:
    - (푌푌푖푖,푡푡+ℎ − 푌푌푖푖,푡푡−1)/푌푌푖푖,푡푡−1 where 푌푌푖푖,푡푡 is real GDP.
    - (퐺퐺푖푖,푡푡+ℎ − 퐺퐺푖푖,푡푡−1)/푌푌푖푖,푡푡−1 where 퐺퐺푖푖,푡푡 is real government expenditure.
- Identification:
  - Baseline identification assumes no contemporaneous response of government expenditures to macroeconomic aggregates, following Blanchard and Perotti (2002).
  - Rationale: Though proposed for quarterly data, authors note its prior application at annual or semi-annual frequency and advantages of annual data (wider country coverage, noninterpolated fiscal data historically unavailable, identified shocks do not reflect timing changes of spending).
  - Robustness: Results shown using forecasting errors to identify unpredictable innovations to government spending.
- Multiplier estimation:
  - One-step IV methodology of Ramey and Zubairy (2018) used to calculate cumulative fiscal multipliers and confidence bands.
  - Equation (3) instruments cumulative future government spending growths with current period government spending 퐺퐺푖푖,푡푡 to isolate variation due to current fiscal shock; estimated coefficient 푚푚ℎ is the cumulative multiplier at horizon ℎ.
  - State-dependent one-step IV (Equation (4)) instruments pandemic and non-pandemic regime cumulative government spending growths by 퐼퐼푖푖,푡푡 × 퐺퐺푖푖,푡푡 and (1−퐼퐼푖푖,푡푡) × 퐺퐺푖푖,푡푡 respectively.
    - Estimated coefficients 푚푚푃푃,ℎ and 푚푚푁푁푃푃,ℎ are the cumulative multipliers at horizon ℎ in the pandemic and non-pandemic regimes respectively.

### Main Results (overview presented in section)
- Focus and structure:
  - Presents main results first for total public expenditure, then disaggregates to public consumption and investment.
  - Analyzes advanced and developing countries separately.
  - Investigates roles of uncertainty, supply bottlenecks, and suppressed demand transmission channels.
- Baseline sample excludes 2020 (year of COVID-19); results are qualitatively and quantitatively similar when 2020 is included.
- Baseline findings (summary as introduced; detailed figures and coefficients reported in the paper):
  - Using total public expenditure and the Blanchard and Perotti (2002) identification, expansionary fiscal policy has a positive and significant effect on output.
  - The left panel of Figure 1 shows estimated 푚푚ℎ coefficients from Equation (3); the estimated cumulative multiplier is reported in the main text/figures (see paper for numeric multiplier paths).

*IMF Working Paper — Fiscal Multipliers During Pandemics — Section VI (Health Crises and Economic Output: Potential Transmission Channels; Data; Empirical Strategy; Main Results)*

### 0.4 in the year of the shock,  and 0.5-0.6   in the medium term. T he right panel of Figure 1 s

### Fiscal Multipliers During Pandemics

### Baseline findings and comparisons between pandemic and non-pandemic states
- Estimated baseline effect of expansionary fiscal policy:
  - Contemporaneous multiplier (year T): 0.4.
  - Cumulative multiplier in the medium term: 0.5-0.6.
- Total public expenditures — state-specific estimates (from Equation 4, right panel):
  - Pandemic state:
    - Contemporaneous multiplier: 0.5.
    - Cumulative multiplier in the first two years after the shock: 0.7.
  - Non-pandemic state:
    - Contemporaneous multiplier: 0.38.
    - Cumulative multiplier in the first two years after the shock: 0.55.
- Statistical tests:
  - The largest difference between estimated multipliers occurs at t = 1, but a p-value of 0.12 suggests the estimated fiscal multipliers between pandemic and non-pandemic states are not statistically different (overall sample) for that comparison.
- Public consumption (Figure 2):
  - Fiscal multiplier is lower over the medium term and stands at 0.32 in the third year after the shock.
  - Multipliers in pandemic and non-pandemic states are statistically different; the pandemic-state consumption multiplier reaches 0.9 in the first year after the shock.
- Public investment (Figure 3):
  - Investment multipliers are significantly larger in both pandemic and non-pandemic states.
  - Multipliers during pandemics exceed 2 one year after the shock and are statistically higher than multipliers in non-pandemic states.

### Heterogeneity by country group
- Advanced economies:
  - Fiscal multipliers are significantly larger during pandemics, particularly one year after the shock when the multiplier is close to 1, about double the multiplier in the non-pandemic state.
- Developing economies:
  - Fiscal multipliers reach about 0.6 one year after the shock and are not statistically different between pandemic and non-pandemic states.
  - Possible interpretation: spending inefficiencies could worsen during a pandemic and constrain the effectiveness of fiscal policy.

### Transmission channels (uncertainty, suppressed demand, supply bottlenecks)
- Proxies used:
  - An uncertainty index related to pandemic events.
  - Excess private savings (deviation from trend) as proxy for suppressed demand.
  - Number of air passengers (detrended) as proxy for mobility.
  - Container traffic in ports (detrended) as proxy for transportation/production bottlenecks.
- Estimation approach:
  - Extended specifications (Equations 5 and 6) include the proxies and their interactions with the shock; focus on p-values for differences between pandemic and non-pandemic multipliers after including proxies.
- Key empirical findings:
  - Including the proxies shifts the estimated multipliers in the two states noticeably closer together for total expenditures, consumption, and investment (Figure 6).
  - P-values associated with tests of differences increase markedly after adding proxies; at h = 0 and h = 1 the tests very confidently reject a statistical difference between the multipliers in the two states in cases where a difference was previously established—suggesting the tested transmission channels could underpin larger fiscal multipliers during health crises.
- Relative importance of channels (Table A.3 summary):
  - Uncertainty has the largest impact on effectiveness of fiscal policy immediately at the onset of a health crisis; a positive coefficient implies higher uncertainty is associated with a higher fiscal multiplier.
  - Excess private savings (positive triple interaction) implies fiscal policy is more effective when private consumption is impaired due to a health crisis.
  - Below-trend container traffic in ports (positive triple interaction) increases multipliers during and after pandemics; this proxy may capture deficiencies in demand more than supply bottlenecks.
- Additional observation:
  - Multipliers in the non-pandemic regime show virtually no differences compared to normal-regime multipliers estimated without the proxies.

### Robustness and additional exercises
- Extending estimation period to include 2020 (height of COVID-19) yields qualitatively and quantitatively similar multipliers to baseline results.
- Results robust to adding controls such as additional lags of existing control variables, domestic credit growth, and the current account balance.
- Debt-to-GDP interaction:
  - High levels of debt tend to decrease the effectiveness of fiscal policy in line with the literature.
  - This weakening effect is not found in the pandemic regime, consistent with priority given to health responses and immediate economic support at the onset of COVID-19.
- Fiscal rules:
  - Presence of at least one fiscal rule at the onset of the pandemic is associated with fiscal spending being more effective at boosting aggregate activity, suggesting fiscal credibility can help anchor medium-term expectations (noting many countries relaxed rules or activated escape clauses).
- Alternative identification:
  - Using public expenditure forecast errors (October WEO) to identify unexpected expenditure shocks (sample restricted to 2002-2019 and 49 countries) confirms main findings: short- and medium-term fiscal multipliers are higher in the pandemic regime and differences between regimes are statistically significant.

### Conclusions and policy implications
- The paper shows fiscal multipliers are larger in the immediate years following the onset of a health crisis.
- Baseline estimates:
  - Contemporaneous multiplier: 0.4 in year T.
  - Cumulative multiplier in the medium term: 0.5-0.6.
- Pandemic vs non-pandemic comparisons:
  - In advanced economies, cumulative multiplier in the pandemic rises to almost 1 one year after the pandemic shock, compared to about 0.4 in the non-pandemic regime.
- Transmission channels (uncertainty, suppressed demand, and supply bottlenecks) can explain differences in pandemic and non-pandemic fiscal multipliers.
- Policy implication:
  - Growth impact of fiscal stimulus packages during health crises and pandemics could be larger and longer lasting than often assumed.
  - The large fiscal support deployed at the onset of the COVID-19 pandemic could have a larger impact than in non-pandemic times—potentially at least 50 percent larger one year after the large-scale fiscal stimulus compared to expectations based on non-pandemic fiscal multipliers.

*IMF Working Paper: Fiscal Multipliers During Pandemics*

### References

### wpiea2022149-print-pdf - References

### References: scope and thematic focus
- Bibliography covers empirical and theoretical literature on fiscal multipliers, fiscal policy effectiveness, uncertainty, pandemics and macroeconomic transmission, including:
  - Studies of fiscal multipliers in recessions and expansions (Auerbach and Gorodnichenko; Ilzetzki, Mendoza, and Végh; Ramey; Christiano, Eichenbaum, and Rebelo).
  - Pandemic- and COVID-19-specific macroeconomic analysis (Eichenbaum, Rebelo, and Trabandt; Bonadio et al.; Chetty et al.; Faria-e-Castro; Fornaro and Wolf).
  - Transmission channels and state dependence (Bloom; Bloom et al.; Caggiano et al.; Ghassibe and Zanetti; Shen and Yang).
  - Open-economy, trade, and supply-chain aspects (Bonadio et al.; Forni and Gambetti; Lafrogne-Joussier et al.; Mahajan and Tomar).
  - Datasets and methodological references (Davoodi et al., Fiscal Rules Dataset; Jordà on local projections; Driscoll and Kraay for covariance estimation).

### Annex — Table A1: Summary Statistics (selected series)
- Dummy Variable for Deaths Above Median: Mean 0.01, Std . Dev. 0.11, Minimum 0.00, Maximum 1.00, Observations 3380
- Gross Domestic Product at 2014p (mn. USD): Mean 904760.80, Std . Dev. 2188649, Minimum 3411.64, Maximum 19700000, Observations 1947
- Gross Domestic Product at current prices (mn. USD): Mean 568781.20, Std . Dev. 1688126, Minimum 568.21, Maximum 21400000, Observations 2756
- Log of GDP at current prices: Mean 4.55, Std . Dev. 0.20, Minimum 1.84, Maximum 6.18, Observations 2321
- Log of GDP at 2014p: Mean 12.36, Std . Dev. 1.67, Minimum 8.13, Maximum 16.80, Observations 1947
- Total Government Expenditure: Mean 142167.30, Std . Dev. 406480.50, Minimum 136.84, Maximum 4625097, Observations 2295
- Gov. Gross Nominal Fixed Capital Formation (mn. USD): Mean 32935.11, Std . Dev. 130854.70, Minimum -30393.352229804, Maximum 2538
- Gov. Consumption Expenditure, current price (mn. USD): Mean 94461.13, Std . Dev. 266176.30, Minimum 0.002973918, Maximum 2789
- Tax revenue (mn. USD): Mean 125604.40, Std . Dev. 329061.70, Minimum 15.86, Maximum 3411509, Observations 1957
- Consolidated Fiscal Balance (mn. USD): Mean -18773.3590847.98, Std . Dev. -1471297254848, Minimum 2050
- Government Debt (% of GDP): Mean 0.510.34, Std . Dev. 0.022.01, Minimum 1943
- Government Debt (mn. USD): Mean 548445.701853147, Std . Dev. 293.5923200000, Minimum 1787
- Private Consumption Expenditure, current price (mn. USD): Mean 325973.80, Std . Dev. 1045287, Minimum 0.00, Maximum 14400000, Observations 2771
- Private Nominal Saving and Investment Saving (mn. USD): Mean 40800000, Std . Dev. 315000000, Minimum -10410976530000000, Observations 2845
- Total Imports (mn. USD): Mean 123085.10, Std . Dev. 260540.80, Minimum 478.262537730, Observations 2484
- Current Account Balance (mn. USD): Mean 247.95, Std . Dev. 57578.74, Minimum -816646.00420568.50, Observations 2769
- Domestic Credit, Y-o-Y growth (%): Mean 17.55, Std . Dev. 142.29, Minimum -4079.334093.33, Observations 1961
- Unemployment Rate (%): Mean 7.98, Std . Dev. 5.56, Minimum 0.0538.40, Observations 2389
- Consumer Price Index, Period Avg., Y-o-Y growth (%): Mean 38.03, Std . Dev. 359.28, Minimum -8.5311749.63, Observations 2937
- Population (mn. persons): Mean 59.90, Std . Dev. 176.54, Minimum 1.311410.08, Observations 3322
- Tourist arrival (persons): Mean 10100000, Std . Dev. 15600000, Minimum 700.00190000000, Observations 2220
- GDP Deflator: Mean 96.68, Std . Dev. 19.08, Minimum 6.29482.49, Observations 2321
- GDP Deflator, Y-o-Y growth (%): Mean 11.36, Std . Dev. 52.55, Minimum -26.871489.50, Observations 2321

### Annex — Figures (descriptive notes)
- Figures A1–A4 and A5 present cumulative fiscal multipliers (public consumption and fixed capital formation) for advanced and non-advanced countries and for robustness with forecast-error identification.
  - Black line: average (non-state dependent) fiscal multiplier.
  - Red line and blue lines: multipliers in pandemic and non-pandemic states respectively.
  - Shaded areas and dashed lines: corresponding 95% Driscoll and Kraay confidence intervals.
- Figure A5: Top panel shows forecast error identification results; bottom panel shows baseline specification constrained to match forecast error identification sample.

### Table A2: Transmission channel summary statistics and t-tests (mean differences)
- Uncertainty: Pandemic Mean 5.247, Std. Dev. 10.210; Non-pandemic Mean 0.182, Std. Dev. 2.110; P-value 0.000
- Excess private saving (mill.): Pandemic Mean -0.400.154, Std. Dev. 0.003; Non-pandemic Mean 0.093, Std. Dev. 0.000; P-value 0.000
- Air passengers (detrended): Pandemic Mean -0.9280.019, Std. Dev. 0.001; Non-pandemic Mean 0.014, Std. Dev. 0.000; P-value 0.000
- Container traffic (detrended): Pandemic Mean -3.7804.132, Std. Dev. 0.300; Non-pandemic Mean 5.111, Std. Dev. 0.000; P-value 0.000
- Note: Uncertainty is the World Pandemic Uncertainty index. Excess private savings is the deviation of real private savings over GDP from its trend. Air passengers is the deviation of the number of air passengers over population from its trend. Container traffic is the deviation of the container traffic in ports in one hundred thousand over its trend. Trends obtained via the HP filter.

### Table A3: Transmission Channels Regressions (Total Expenditure, All Countries) — selected coefficient highlights
- Estimation: Output of Equation 6 via the 1-step IV method of Equation 4. Fixed-effects and control variables omitted. Robust standard errors in parenthesis. Stars denote significance at 1, 5 and 10%.
- Key coefficient patterns (horizons h=0 to h=3):
  - Uncert.*IIt: h=0 0.001*, h=1 0.002**, h=2 0.002**, h=3 0.003**
  - Uncert.*(1−IIt): coefficients near zero, with h=3 0.001**
  - Exc. Save*IIt: h=0 0.051**, h=1 0.063**, h=2 0.064, h=3 0.067
  - Exc. Save*(1−IIt): h=0 0.019, h=1 0.037**, h=2 0.040*, h=3 0.037
  - Psgr.*IIt and Psgr.*(1−IIt): generally smaller and mostly insignificant across horizons
  - Cont. traff.*IIt and Cont. traff.*(1−IIt): small coefficients, generally insignificant
  - Shock´Uncert.*IIt: h=0 0.113***; h=1 0.011; h=2 -0.054; h=3 -0.158
  - Shock*Exc. Save*IIt: h=0 0.854; h=1 0.877*; h=2 1.653**; h=3 0.812
  - Shock*Psgr.*IIt: h=0 19.884; h=1 20.466; h=2 13.388; h=3 33.084**
  - Shock*Contain. traffic*IIt: h=1 -0.199***; h=2 -0.284**; h=3 -0.348***
  - R-squared by horizon: h=0 0.707, h=1 0.783, h=2 0.817, h=3 0.842
- Variable definitions: IIt is the pandemic dummy; Shock is the fiscal policy shock; Uncert. is the Pandemic Uncertainty Index; Exc. Save is detrended private savings; Cont. Traff. is detrended container traffic; Psgr is detrended number of air passengers.

### Table A4: The Role of Fiscal Position and Credibility (selected coefficients)
- Estimation: Output of Equation 6 via the 1-step IV method of Equation 4. Fixed-effects and control variables omitted.
- Panel (A): Debt-to-GDP ratio (coefficients at horizons T=0 to T=3)
  - Debt * IIt: T=0 0.006, T=1 0.038***, T=2 0.056***, T=3 0.080***
  - Debt * (1−IIt): T=0 0.009, T=1 0.039***, T=2 0.070***, T=3 0.096***
  - Shock * Debt * IIt: T=0 0.012, T=1 -0.122, T=2 -0.285, T=3 -0.11
  - Shock * Debt * (1−IIt): T=0 -0.302***, T=1 -0.095, T=2 0.010, T=3 -0.087
- Panel (B): Fiscal Rule (dummy = 1 if a country has a fiscal rule in place)
  - FiscalRule * IIt: T=0 -0.002, T=1 -0.003, T=2 -0.002, T=3 0.004
  - FiscalRule * (1−IIt): T=0 -0.003, T=1 -0.004, T=2 -0.005, T=3 -0.002
  - Shock * FiscalRule * IIt: T=0 0.522**, T=1 0.119, T=2 -0.019, T=3 -0.2
  - Shock * FiscalRule * (1−IIt): T=0 0.198, T=1 0.233, T=2 0.395, T=3 0.336

*wpiea2022149-print-pdf - References (canonical PDF content).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022149-print-pdf.pdf_