## wpiea2022120-print-pdf

## Source details

**Canonical URL:** [wpiea2022120-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022120-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022120-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022120-print-pdf.pdf.json)

---

### Major findings
- Imports of goods fell by much less than predicted by the demand model, while imports of services fell by much more than predicted.
- International spillovers from lockdowns were economically large but short-lived and concentrated in the early phase of the pandemic:
  - Compared to a counterfactual with no lockdown restrictions, trade partners’ lockdowns explain up to 60 percent of the observed decline in imports in the first five months of 2020.
  - The elasticity of trade to partners’ lockdown intensity becomes insignificant after the initial phase as trade begins to recover despite only marginal relaxation of lockdown stringency.
- Countries whose trade partners had better health preparedness experienced larger prediction errors (goods imports fell by less than predicted), indicating partner policies mattered.
- The pandemic induced a rotation in demand from services towards goods:
  - Greater pandemic intensity in a country is associated with smaller declines in goods imports relative to standard import demand model predictions.
  - Services imports fell by much more than predicted in countries where tourism services comprised a large share of total imports.

### Methodology overview
- Country-level import demand model estimated on a sample of 127 countries over 1985–2019 following Bussière et al., 2013 to predict import growth in 2020 and obtain prediction errors.
- Import regressions estimated separately for goods (τ=g) and services (τ=s) with specification:
  - ∆lnMτmt = πm + βτ Dm ∆lnDτmt + βτ Pm ∆lnPmt + ετmt, ∀τ∈{g,s}
- Import-intensity adjusted demand Dτmt: harmonic weighted average of domestic demand components (private consumption C, government consumption G, investment I, exports X) with weights from the Eora Global Supply Chain Database.
- Model performance to 2019 was good; in 2020:
  - Services model predicted about -8%, observed -25% (prediction error series shows services error in 2020 at 0.2 log-points).
  - Goods model predicted -10% vs observed -6% (model slightly overpredicts fall).

### International spillover identification and empirical setup
- Bilateral product-level gravity model estimated at country-pair-industry-month level:
  - M_meit = exp[β Stringency Index_et + δ Controls_met + α_mei + γ_mit] + ε_meit.
- Key regressor: exporter-time-varying Oxford COVID-19 Government Response Stringency Index (monthly average).
- Controls: new COVID-19 cases and deaths per month (per million), number of new export restrictions and number of removed export barriers (Global Trade Alert).
- Fixed effects:
  - α_mei: exporter-importer-industry fixed effects.
  - γ_mit: importer-industry-time fixed effects (absorb importer-product demand variation).
- Estimation: Poisson pseudo-maximum likelihood (PPML); standard errors clustered at the exporter level.
- Main dataset: Trade Data Monitor monthly bilateral goods imports at HS6 from January 2020 to June 2021; analysis dataset: 98 importing countries × 163 exporting countries; 15,880 country pairs; 4,652,840 unique industry-exporter-importer corridors.
- Industry aggregation: HS6 aggregated into about 300 industries to match industry upstreamness (Antràs et al., 2012).

### Baseline results and magnitude
- Baseline semielasticity estimate: about -0.15 (a one percentage point increase in the stringency index associated with a 0.15 percent reduction in imports).
- Temporal dynamics:
  - Strongest effects in the first 5 months of 2020, gaining strength in February and March 2020, declining and becoming insignificant in June 2020.
  - Smaller but significant effect in Spring 2021.
- First-half 2020 (2020H1) semielasticity:
  - Point estimate more than twice the whole-sample estimate; counterfactual exercise implies containment policies accounted for about 60 percent of the observed fall in imports between January and May 2020.
  - Counterfactual calculation reported: effective fall in imports = January value (96.5) − May value (72.5) = 24; fall in counterfactual without containment = 100 − 90.3 = 9.7; lockdowns account for (24 − 9.6)/24 = 59.7 percent of the actual import decline.

### Heterogeneous effects
- Fiscal response:
  - Spillovers significantly larger when exporting partners implemented relatively small discretionary fiscal responses (bottom quartile) versus larger fiscal responses (Table 4, column 7).
- Teleworkability:
  - Spillover effect more than twice as strong for imports from countries in the bottom quartile of teleworkable jobs compared to those with a high share of teleworking.
- Industry and GVC exposure:
  - Effects stronger in GVC-intensive industries (automobiles, electronics, textiles and garments, medical goods), especially electronics, than in non GVC-intensive industries.
  - Position along value chain matters:
    - Negative effect of stringency measures is dampened for industries further upstream (e.g., metals and minerals).
    - Negative effect is stronger for downstream industries (e.g., transportation and textiles).
- Exporters’ health preparedness:
  - Import-weighted Global Health Security Index positively associated with better-than-expected goods import performance (goods coefficient 0.00518***).

### Full-fledged gravity model (upstreamness interactions)
- Augmented specification includes Stringency Index × Upstreamness (Antràs et al., 2012) and exporter-time fixed effects μ_et to account for multilateral resistance.
- Findings:
  - Without exporter-time fixed effects, one standard deviation of upstreamness (SD = 0.82) reduces the supply effect of lockdown by about 20 percent.
  - With exporter-time fixed effects, the differential effect by upstreamness remains statistically significant and similar in magnitude.
  - Downstream industries are more exposed to lockdown-induced supply disruptions.

### Robustness checks and alternative measures
- Alternative containment measure: workplace closure index (discrete 0–3) replicates baseline results; workplace closure and stringency indices are highly correlated.
- Robustness across samples:
  - Results robust to dropping China; not China-specific.
  - Semielasticity becomes smaller when emerging markets and Asian countries are excluded; higher when advanced economies and European countries are excluded.
- Clustering:
  - Main results robust to clustering standard errors at the exporter-month level; significance unaffected and estimated standard errors are—if anything—smaller.

### Model performance, prediction errors, and drivers
- Country-by-country import demand regressions (127 countries, 1985–2019) produce fitted values accurate to 2019; 2020 residuals reveal pandemic shocks not captured by standard import demand variables.
- Key summary statistics from demand coefficients:
  - Demand coefficients: Mean 1.2955; Median 1.3468; Interquantile range [0.95,1.54].
  - Services demand coefficients: Mean 1.3472; Median 1.0932; Interquantile range [0.68,1.76].
  - Goods demand coefficients: Mean 1.3305; Median 1.3574; Interquantile range [0.99,1.69].
  - Price coefficients: Total Mean -0.2230; Services Mean -0.2931; Goods Mean -0.2030.
- Residual regressions (prediction errors in 2020) report:
  - Goods COVID-19 cases coefficient: 0.00812** (0.00408).
  - Stringency index coefficients: Total 0.00172* (0.000895); Goods 0.00229** (0.00103).
  - Mobility coefficients: Goods -0.00363*** (0.00117).
  - Trade partners’ health preparedness (import-weighted GHS Index) goods coefficient: 0.00518*** (0.00183).

### Key empirical coefficient values (selected)
- International spillover Poisson estimates (stringency index coefficients):
  - Column (1): -0.00141***.
  - Column (2): -0.00149***.
  - Column (3): -0.00183***.
  - Column (4): -0.00160***.
  - Column (5): -0.00182***.
  - Column (6) (2020H1 sample): -0.00307***.
  - Column (7): -0.00062***.
- Reported semielasticities (converted):
  - Column (1) semielasticity -0.1413.
  - Column (2) semielasticity -0.1494.
  - Column (3) semielasticity -0.1828.
  - Column (4) semielasticity -0.1594.
  - Column (5) semielasticity -0.1818.
  - Column (6) semielasticity -0.3068.
  - Column (7) semielasticity -0.0616.
- Interaction coefficients (selected):
  - Stringency index × Small fiscal response: -0.00211***.
  - Stringency index × Large fiscal response: -0.00146***.
  - Stringency index × Low telework: -0.00126***.
  - Stringency index × High telework: -0.00058***.
  - Stringency index × Electronics: -0.00312***.
  - Stringency index × Upstreamness: 0.00039* and 0.00057*** across specifications.
- Workplace closings coefficients (selected):
  - Column (1): -0.02831*** (0.010).
  - Column (3) (2020H1): -0.07762*** (0.021).
  - Workplace closings × Low telework: -0.02177* (0.012).
  - Workplace closings × Upstreamness: 0.00801*** (0.003) in one specification.

### Policy-relevant implications
- Lockdowns imposed by trade partners can generate large international costs via trade channels, concentrated in the initial months of the pandemic.
- Country characteristics that mitigate domestic supply disruption (larger discretionary fiscal responses, higher teleworkability, stronger health preparedness) reduce adverse spillovers to trade partners or amplify import resilience.
- GVC exposure and industry position along value chains are key determinants of vulnerability to partner-country lockdowns; policies to bolster supply chain resilience should consider these dimensions.
- The rotation from services to goods and the unusually large fall in services imports highlight the need to consider sectoral demand shifts when designing trade and domestic support policies during pandemics.

*International Trade Spillovers from Domestic COVI-19 Lockdowns, Working Paper No. WP/2022/120*

### 2020.  Imports of goods fell by much less than predicted (and those of services much more than

### 2020.  Imports of goods fell by much less than predicted (and those of services much more than predicted)

### Major findings
- Imports of goods fell by much less than predicted by the demand model, while imports of services fell by much more than predicted.
- International spillovers from lockdowns were economically large but short-lived and concentrated in the early phase of the pandemic:
  - Compared to a counterfactual with no lockdown restrictions, trade partners’ lockdowns explain up to 60 percent of the observed decline in imports in the first five months of 2020.
  - Moving past the initial phase, the elasticity of trade to partners’ lockdown intensity becomes insignificant as trade begins to recover despite only marginal relaxation of lockdown stringency.
- Countries whose trade partners had better health preparedness experienced larger prediction errors (goods imports fell by less than predicted), suggesting that policies implemented in partner countries in response to COVID-19 mattered.
- The pandemic was accompanied by a rotation in demand from services towards goods:
  - The greater the intensity of the pandemic in a given country, the smaller the decline in goods imports relative to predictions of the standard import demand model.
  - Services imports fell by much more than would be predicted by demand alone in countries for which tourism services comprised a large share of total imports.

### Methodology overview
- A country-level import demand model (based on Bussière et al., 2013) was estimated on a sample of 127 countries over 1985–2019 to predict import growth in 2020 and obtain country-level prediction errors.
- Import regressions were estimated separately for goods (τ=g) and services (τ=s) with the specification:
  - ∆lnMτmt = πm + βτ Dm ∆lnDτmt + βτ Pm ∆lnPmt + ετmt, ∀τ∈{g,s}
- Import-intensity adjusted demand Dτmt is a harmonic weighted average of domestic demand components (private consumption C, government consumption G, investment I, exports X) with weights computed from the Eora Global Supply Chain Database.
- The model performs well in sample up to 2019 but fails to predict the large fall in services trade in 2020 (model predicts about -8%, observed -25%) and slightly overpredicts the fall in goods trade (10% predicted vs 6% observed fall).

### International spillover identification
- A bilateral product-level gravity model with country-industry time-varying fixed effects was used to isolate supply shock spillovers from lockdowns by comparing imports of the same product from partners with different containment policy severities while absorbing demand through importer-product-time fixed effects.
- This approach assumes changes in importer demand for the product follow the same pattern across partners, so variation in partner containment severity captures supply spillovers.

### Heterogeneity in spillovers
- Spillovers are stronger from partners that were less able to deploy large discretionary fiscal measures to mitigate the pandemic.
- Spillovers are more than twice as strong from partner countries less able to rely on remote working (teleworkability, Dingel and Neiman 2020) compared to those with a higher share of teleworkable jobs.
- Effects are stronger in GVC-intensive industries (automobiles, electronics, textiles and garments, medical goods), especially electronics, than in non GVC-intensive industries.
- Position along the value chain matters:
  - Negative effect of stringency measures is dampened for industries further upstream (like metals and minerals products).
  - Negative effect is stronger for downstream industries (like transportation and textiles).
  - Interaction with upstreamness (Antràs et al., 2012) and exporter-time fixed effects helps control for multilateral resistance and better identify differential effects across industries.

### Related evidence and context
- Country-level and firm-level studies find similar supply-chain propagation and short-lived spillovers:
  - Heise (2020): U.S. firms sourcing from China pre-pandemic saw imports fall 15% more than comparable firms sourcing elsewhere.
  - Lafrogne-Joussier et al. (2022): French firms relying on Chinese inputs experienced a 7% greater decline in overall imports following the Chinese lockdown.
  - Pimenta et al. (2021): Portuguese firms show a progressive adaptation with lockdown effects turning insignificant in the second half of 2020.
  - Majune and Addisu (2021): Kenya’s trading partners’ lockdowns associated with imports falling by 23% on average after measures.
  - Arenas et al. (2022): Philippines experienced a -57% drop in imports compared to pre-pandemic levels.
- Product-level analyses (e.g., Cerdeiro and Komaromi 2020; Espitia et al. 2022) document downstream propagation, short-lived supply spillovers, and stronger effects for sectors relying on imported inputs or mobility declines.

### Model performance and drivers of prediction errors
- Aggregate predictions from country-by-country regressions yield accurate fitted values up to 2019; for 2020:
  - Services model predicted about -8% but observed -25% (prediction error series shows services error in 2020 at 0.2 log-points, “off-the-charts”).
  - Goods: model predicted -10% vs observed -6% (model slightly overpredicts fall).
- Residuals capture changes in preferences or supply shocks not captured by standard price indexes; the pandemic produced such shocks.
- Regression results (Table 2 referenced) indicate:
  - Countries with more severe pandemic indicators (more cases, more stringent measures, less mobility) show better-than-expected goods import growth.
  - The ability of countries to increase goods imports above expected amounts was associated with partners’ health preparedness as measured by the Global Health Security Index (import-weighted average).

### Policy-relevant implications
- Lockdowns imposed by trade partners can generate large international costs via trade channels, concentrated in the initial months of the pandemic.
- Country characteristics that mitigate domestic supply disruption (larger discretionary fiscal responses, higher teleworkability, stronger health preparedness) reduce adverse spillovers to trade partners or amplify import resilience.
- GVC exposure and industry position along value chains are key determinants of vulnerability to partner-country lockdowns; policies to bolster supply chain resilience should consider these dimensions.
- The observed rotation from services to goods, and the unusually large fall in services imports, highlight the need to consider sectoral demand shifts when designing trade and domestic support policies during pandemics.

*Excerpted and summarized from the provided IMF content unit.*

### 3.1    Modeling international spillovers

### 3.1    Modeling international spillovers

### Empirical setup and data
- Estimated gravity equation at country-pair-industry-month level: M_meit = exp[β Stringency Index_et + δ Controls_met + α_mei + γ_mit] + ε_meit.
- Dependent variable: bilateral imports of products in industry i (M_meit) by importer m from exporter e in month t.
- Key regressor: exporter country time-varying Oxford COVID-19 Government Response Stringency Index (monthly average).
- Controls include: number of new COVID-19 cases and deaths per month (per million), number of new export restrictions and number of removed export barriers from Global Trade Alert (quarterly, country-pair level).
- Fixed effects:
  - α_mei: country-pair-industry fixed effects (controls for time-invariant bilateral and industry-specific factors).
  - γ_mit: importer-industry-time fixed effects (absorb unobserved time-varying heterogeneity across importers and industries).
- Estimation method: Poisson pseudo-maximum likelihood (PPML) following Silva and Tenreyro and implemented as in Correia et al. Standard errors clustered at the exporter level.
- Main dataset:
  - Trade Data Monitor (TDM) monthly bilateral goods imports at HS6 from January 2020 to June 2021.
  - Original sample: 99 importing countries × 196 exporting countries.
  - Analysis dataset: 98 importing countries and 163 exporting countries; total of 15,880 country pairs and 4,652,840 unique industry-exporter-importer trade corridors.
- Industry aggregation: HS6 aggregated into about 300 industries to match industry upstreamness (Antràs et al., 2012) via BEA concordance.

### Identification and interpretation
- Parameter of interest: β — effect of trade partners’ containment policies on domestic imports; negative β interpreted as supply-channel impact via reduced exporter supply.
- Import demand controlled through γ_mit under assumption that proportional change in country-specific demand for products in a given industry-month is the same across source countries.
- Caveat: stringency index may correlate with other exporter-level changes; addressed with controls for export restrictions and COVID-19 incidence.

---

### 3.2    Baseline Results

- Baseline findings (Table 4, Figure 4):
  - Negative and significant association between partners’ stringency index and domestic imports across specifications.
  - Semielasticity estimate: about -0.15, i.e., a one percentage point increase in the stringency index is associated with a 0.15 percent reduction in imports.
- Robustness of β:
  - Stable moving from importer-time fixed effects to importer-industry-time fixed effects.
  - Robust to controls for exporter COVID-19 cases and deaths per capita and changes in export restrictions (separately and jointly).
- Temporal dynamics (monthly split of β):
  - Strongest effects in the first 5 months of 2020, gaining strength in February and March 2020, declining and becoming insignificant in June 2020.
  - Smaller but significant effect in Spring 2021, coincident with Delta variant spread.
  - Stringency index remained persistently high over the period, suggesting adaptation rather than relaxation of policies.
- First-half 2020 estimate (column 6):
  - Semielasticity more than twice the whole-sample estimate.
  - Counterfactual exercise using this point estimate implies containment policies accounted for about 60 percent of the observed fall in imports.
  - Calculation cited: effective fall in imports = January value (96.5) − May value (72.5) = 24; fall in counterfactual without containment = 100 − 90.3 = 9.7; lockdowns account for (24 − 9.6)/24 = 59.7 percent of the actual import decline.

---

### 3.3    Heterogeneous Effects of Lockdowns

- Heterogeneity by fiscal response:
  - Fiscal response measure: IMF data on COVID-19 related discretionary fiscal measures announced/taken between January and June 2020, scaled by GDP (includes above-the-line and below-the-line measures; June 2020 vintage).
  - Semielasticity of imports to stringency is significantly larger when trade partners implemented relatively small fiscal responses (bottom quartile) compared to those with larger fiscal responses (Table 4, column 7; Figure 6).
- Heterogeneity by teleworkability:
  - Teleworkability measured using Dingel and Neiman (2020) cross-country data.
  - Baseline model on restricted sample (column 8) shows a negative and significant spillover.
  - Spillover effect more than twice as strong for imports from countries in the bottom quartile of teleworkable jobs compared to those with a high share of teleworking (column 9; Figure 6).
- Heterogeneity by industry / GVC intensity:
  - GVC-intensive goods defined as inputs and finished goods in automotive industries, electronics, textiles and garments, and medical goods (together about 24 percent of global goods trade).
  - Column 10 (Table 4) and Figure 6: effect of lockdowns stronger in GVC-intensive industries, especially electronics, than in non-GVC-intensive industries.
  - Interpretation: imports in GVC-intensive industries are relatively more exposed to supply-chain disruptions due to lockdowns.

---

### 3.4    A Full-Fledged Gravity Model

- Augmented specification (Bartik-style interaction):
  - M_meit = g(β Stringency Index_et × Upstream_i + δ Controls_eit + α_mei + γ_mit + μ_et + ε_meit).
  - Upstream_i: industry upstreamness (average distance from final use) from Antràs et al. (2012), applied uniformly using U.S. I-O based measures.
  - Includes exporter-time fixed effects μ_et to account for multilateral resistance.
- Findings (Table 5):
  - Without exporter-time fixed effects, lockdown negative effect is dampened in upstream industries and stronger in downstream industries; one standard deviation of upstream index (SD = 0.82) reduces the supply effect of the lockdown by about 20 percent (column 1).
  - With exporter-time fixed effects (column 2), the differential effect across industries by upstreamness remains statistically significant and similar in magnitude.
  - Interpretation: downstream (more final-use oriented) industries are more exposed to lockdown-induced supply disruptions.

---

### 3.5    Robustness

- Alternative measure of containment policies:
  - Results robust to using workplace closure index (discrete 0–3 scale) from Oxford COVID-19 Government Response indicators; index is one of 8 components of the Stringency Index.
  - Workplace closure index and stringency index are highly correlated and show similar evolution over time.
- Additional robustness exercises:
  - Sensitivity checks on choice of variables, sample and methodology confirm main patterns (details referenced in text).

*Italic: Source — wpiea2022120-print-pdf - 3.1    Modeling international spillovers (IMF).*

### 0.82 and a regression of the stringency index against the workplace closings index with month

### 0.82 and a regression of the stringency index against the workplace closings index with month and country fixed effects

### Empirical estimates and key coefficients
- Regression of the stringency index against the workplace closings index with month and country fixed effects gives a coefficient equal to 13.3 (s.e.=0.56).
- Given that average value of the upstreamness indicator is equal to 2.07, the coefficient on stringency (measured at the average level of upstreamness) is equal to −0.00234+0.00039∗2.07=−0.0015.
- A one standard deviation increase in upstreamness increases the coefficient by 0.00039∗.81=0.0003, which correspond to about one fifth of teh average effect (.0003/.0015=0.20).
- A percentage point increase in the stringency index is associated with a 0.15 percent reduction in imports.
- Up to 60 percent of the observed fall in imports between January and May 2020 can be explained by lockdowns.

### Heterogeneity and robustness findings
- Using the workplace closings measure replicates baseline results: more stringent containment policies in workplaces instituted by trade partners are associated with a significant decline in imports (columns 1-3).
- The negative effect is significantly stronger for:
  - trade partners which implemented a smaller discretionary fiscal response to the COVID-19 pandemic (column 5),
  - trade partners with a low capacity to work remotely (column 6).
- The negative effect is larger for goods upstream in the value chain (column 8).
- Robustness across country groups:
  - Dropping specific country groups (by income and region) one at a time from the set of exporters shows the significance of the spillover effect is robust to alternative samples.
  - The semielasticity of the stringency index becomes smaller when emerging markets and Asian countries are excluded, consistent with larger impacts in the first phase of the crisis when Asian countries were disproportionately affected.
  - The semielasticity is higher when advanced economies and European countries are excluded, suggesting containment policies in Europe (which occurred later) had weaker spillover effects.
- Role of China:
  - Re-estimating the baseline model dropping China shows results on international spillovers of lockdowns are not China-specific and remain economically meaningful excluding China.
  - Smaller spillover effects in February and March and larger ones in April and May could indicate that initial global trade spillovers were driven by lockdowns in China, while lockdowns in other regions mattered more in Spring 2020.
- Clustering:
  - Table A3 reports main results estimated by clustering the standard error at the exporter-month level.
  - The significance of the findings is not affected, and the estimated standard errors are—if anything—smaller.

### Interpretation and broader conclusions
- Trade dynamics during the pandemic:
  - Goods trade fell sharply in the early stages of the pandemic, but rebounded quickly.
  - Trade in services had not yet recovered as of mid-2022.
- Role of standard vs. non-standard drivers:
  - Standard factors of import demand observed in 2020—components of domestic demand and relative import prices—are consistent with weaker than observed demand in goods imports and stronger than observed demand in services imports.
  - Feeding standard factors into a simple model of import demand estimated on historical data suggests non-standard drivers of imports came to the fore during the pandemic.
- Spillovers and mechanisms:
  - Pandemic containment policies in partner countries played an important role beyond those countries’ borders: international spillovers from supply disruptions due to lockdowns were sizable and negative.
  - Spillover effects were (i) short-lived, fading out starting in the second half of 2020, (ii) stronger for goods belonging to GVC-intensive industries, and (iii) mitigated if exporting countries were more conducive to teleworking and implemented large discretionary fiscal measures in response to the pandemic.
- Demand shifts:
  - Where the pandemic was more intense, goods imports rebounded faster than predicted by the import demand model, consistent with an induced shift in preferences away from services (domestic and imported) towards goods.

*Source: wpiea2022120-print-pdf - 0.82 and a regression of the stringency index against the workplace closings index with month and country fixed effects (IMF)*

### References

### References

### Key empirical findings from figures and tables
- Import demand model (country-by-country estimates, 127 countries):
  - Demand coefficients: Mean 1.2955; Median 1.3468; Interquantile range [0.95,1.54].
  - Services demand coefficients: Mean 1.3472; Median 1.0932; Interquantile range [0.68,1.76].
  - Goods demand coefficients: Mean 1.3305; Median 1.3574; Interquantile range [0.99,1.69].
  - Price coefficients: Total Mean -0.2230; Services Mean -0.2931; Goods Mean -0.2030.
  - Price coefficient medians: Total -0.1913; Services -0.2266; Goods -0.1390.
  - Number of countries: 127.

- Residual analysis (prediction errors in 2020 regressed on pandemic-relevant variables):
  - Total COVID-19 cases coefficient: 0.00321 (standard error (0.00360)); standardized coefficient 0.0574.
  - Services COVID-19 cases coefficient: -0.00639 (0.0123); standardized coefficient -0.0416.
  - Goods COVID-19 cases coefficient: 0.00812** (0.00408); standardized coefficient 0.121**.
  - Stringency index coefficients: Total 0.00172* (0.000895); Services 0.00177 (0.00231); Goods 0.00229** (0.00103).
  - Mobility coefficients: Total -0.00236** (0.000918); Services -0.00183 (0.00330); Goods -0.00363*** (0.00117).
  - Observations vary by specification: 125, 121, 99 as reported.
  - Adjusted R-squared reported across specifications: e.g., 0.038, 0.049, 0.074, and negatives indicated for some columns.

- Trade partners’ health preparedness (import-weighted Global Health Security Index):
  - Total coefficient 0.00213 (0.00212); standardized coefficient 0.0964.
  - Services coefficient -0.00520 (0.00546); standardized coefficient -0.08540.
  - Goods coefficient 0.00518*** (0.00183); standardized coefficient 0.195***.
  - Observations: 122; Adjusted R-squared: 0.002, -0.001, 0.036.

- International spillover effect of lockdowns (Poisson pseudo-maximum likelihood estimates):
  - Stringency index coefficients by column:
    - Column (1): -0.00141***.
    - Column (2): -0.00149***.
    - Column (3): -0.00183***.
    - Column (4): -0.00160***.
    - Column (5): -0.00182***.
    - Column (6) (2020H1 sample): -0.00307***.
    - Column (7): -0.00062***.
  - Additional interaction coefficients (selected):
    - Stringency index x Small fiscal response: -0.00211***.
    - Stringency index x Large fiscal response: -0.00146***.
    - Stringency index x Low telework: -0.00126***.
    - Stringency index x High telework: -0.00058***.
    - Stringency index x Automotive: -0.00169**.
    - Stringency index x Electronics: -0.00312***.
    - Stringency index x Medical: -0.00246***.
    - Stringency index x Textiles: -0.00243***.
    - Stringency index x non-GVC industries: -0.00118***.
  - Sample sizes and features:
    - Observations range: e.g., 23,594,169 (col 1), 23,531,808 (col 2), 21,787,468 (cols 3 and 5), 6,118,735 (col 6), 23,256,563 (col 7), 14,764,840 (cols 8-9).
    - Fixed effects: Exporter-importer-industry FE present in all columns; importer-month FE and importer-industry-month FE vary by column; industry-month FE present.
  - Reported semielasticities:
    - Column (1) semielasticity -0.1413.
    - Column (2) semielasticity -0.1494.
    - Column (3) semielasticity -0.1828.
    - Column (4) semielasticity -0.1594.
    - Column (5) semielasticity -0.1818.
    - Column (6) semielasticity -0.3068.
    - Column (7) semielasticity -0.0616.

- Spillovers across industry upstreamness (Table 5):
  - Stringency index: -0.00234*** (column 1).
  - Stringency index x Upstreamness: 0.00039* (column 1) and 0.00057*** (column 2).
  - Observations: 23,531,808 (both columns).
  - Fixed effects: Exporter-importer-industry FE and Importer-industry-month FE included; exporter-month FE varied.

- Workplace closings index (alternative to stringency index) and interactions (Table 6):
  - Workplace closings coefficients (selected):
    - Column (1): -0.02831*** (0.010).
    - Column (2): -0.02840*** (0.010).
    - Column (3) (2020H1 sample): -0.07762*** (0.021).
    - Column (5): -0.00885 (0.006).
    - Column (6): -0.04202*** (0.016).
  - Other coefficients and interactions:
    - Number of removed export restrictions: -0.00309* (0.002) in one specification.
    - Workplace closings x Low telework: -0.02177* (0.012).
    - Workplace closings x High telework: -0.00814 (0.006).
    - Workplace closings x Upstreamness: 0.00622 (0.004) and 0.00801*** (0.003) across columns.
  - Observations across columns: 23,531,808; 21,787,468; 6,118,735; 14,764,840; 23,256,563; 14,764,840.
  - Fixed effects: Exporter-importer-industry FE and Importer-industry-month FE present in main specifications; exporter-month FE included in column 8; industry-month FE present.

### Data, samples, and estimation notes
- Primary indexes and data sources referenced:
  - Oxford COVID-19 Government Response Stringency Index (Hale et al., 2021).
  - Workplace closings index (component of Oxford stringency measures).
  - COVID-19 cases and deaths, mobility: Our World in Data.
  - Bilateral imports data: IMF Direction of Trade Statistics and CPB World Trade Monitor for aggregate quarterly series.
  - Trade Data Monitor (TDM) used for goods import microdata; representativeness compared to DOTS reported in supplementary figures.
  - Teleworkability index from Dingel and Neiman (2020).
  - Global Health Security Index used for trade partners’ health preparedness (import-weighted averages).
  - Estimation methods: Poisson pseudo-maximum likelihood for gravity-style trade models (equation 5 and 6); country-by-country regressions for import demand model estimated on 1985-2019; panel regressions and fixed effects as specified per table.
- Temporal coverage:
  - Samples for spillover estimations span 2020:m1-2021:m6; selected specifications restrict to first six months of 2020 (2020H1).
  - Import demand model estimated on 1985-2019 data; predictions and residuals analyzed for 2020 deviations.
- Robustness and heterogeneity analyses:
  - Results reported for splits by fiscal response (small vs large discretionary fiscal measures between January and June 2020), teleworkability (low vs high), GVC-intensive industries (Automotive, electronics, medical, textiles) vs non-GVC-intensive industries, and upstreamness.
  - Sub-sample robustness excludes country groups (AE, EM, LIDC, AFR, APD, EUR, MCD, WHD) in supplementary figures.

*Source: wpiea2022120-print-pdf - References*

### 2019.  Standardized coefficients represent the number of standard deviation changes in the dependent variables

### Table A3: The international spillover effect of lockdowns: Alternative clustering

### Estimation method and sample
- Estimation method: Poisson pseudo-maximum likelihood.
- Stringency index: Oxford COVID-19 Government Response Stringency Index (Hale et al. (2021)).
- Sample period: 2020:m1-2021:m6; column 3 restricted to the first six months of 2020 (2020H1).
- Columns 5-6 limited to exporting countries for which the teleworkability measure (Dingel and Neiman (2020)) is available.
- The coefficient of the stringency index is estimated separately for:
  - Trade partners which have announced and implemented high and low discretionary fiscal measures in response to the pandemic between January and June 2020 (fiscal spending is scaled by GDP and the sample is split along the first quartile of the distribution of fiscal spending over GDP).
  - Trade partners with high and low values of the teleworkability index (the sample is split along the first quartile of the distribution of the index).
- Standard errors are robust to heteroscedasticity; standard errors in parenthesis are clustered at the exporter-month level in the table note.
- Significance notation: ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.
- TW = teleworkability.

### Key coefficient estimates (columns 1–8)
- Stringency index:
  - Column (1): -0.00149*** (0.000)
  - Column (2): -0.00182*** (0.000)
  - Column (3): -0.00307*** (0.001)
  - Column (4): -0.00062*** (0.000)
  - Column (5): -0.00234*** (0.001)
- Covid cases per million, lagged:
  - Column (2): 0.00002 (0.000)
- Covid deaths per million, lagged:
  - Column (3): -0.00051 (0.001)
- Number of new export restrictions:
  - Column (4): 0.00917 (0.008)
- Number of removed export restrictions:
  - Column (4): -0.00199 (0.002)
- Interaction terms (columns 5–8):
  - Stringency index x Small fiscal response: -0.00201*** (0.000)
  - Stringency index x Large fiscal response: -0.00146*** (0.000)
  - Stringency index x Low telework: -0.00126*** (0.000)
  - Stringency index x High telework: -0.00058*** (0.000)
  - Stringency index x Upstreamness:
    - Column (7): 0.00039** (0.000)
    - Column (8): 0.00057*** (0.000)

### Observations, fixed effects, and sample variants
- Observations:
  - Column (1): 23,531,808
  - Column (2): 21,787,468
  - Column (3): 6,118,735
  - Column (4): 14,764,840
  - Column (5): 23,256,563
  - Column (6): 14,764,840
  - Column (7): 23,531,808
  - Column (8): 23,531,808
- Fixed effects included (per column, all marked Y except where noted):
  - Exporter-importer-industry FE: Y (all columns)
  - Importer-industry-month FE: Y (all columns)
  - Exporter-month FE: N for columns 1–7; Y for column (8)
  - Industry-month FE: Y (all columns)
- Sample labels:
  - Column (1): All
  - Column (2): All
  - Column (3): 2020H1
  - Column (4): TW
  - Column (5): All
  - Column (6): TW
  - Column (7): All
  - Column (8): All

### Notes on interpretation
- Standardized coefficients represent the number of standard deviation changes in the dependent variables associated to one standard deviation change in the variable of interest.
- Standard errors in parenthesis are robust to heteroscedasticity.
- The table reports results of equation 5 (columns 1-6) and equation 6 (columns 7-8).

*International Trade Spillovers from Domestic COVI-19 Lockdowns, Working Paper No. WP/2022/120*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022120-print-pdf.pdf_