## wpiea2020284-print-pdf

## Source details

**Canonical URL:** [wpiea2020284-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020284-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020284-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020284-print-pdf.pdf.json)

---

### Abstract and key findings
- World trade contracted dramatically during the global economic crisis induced by the COVID-19 pandemic.
- The paper quantifies the causal effect of supply spillovers from lockdowns using a novel dataset of daily bilateral seaborne trade and a shift-share identification strategy that leverages geography-induced cargo delivery lags.
- Main empirical findings:
  - Strong but short-lived supply spillovers of lockdowns through international trade.
  - In a hypothetical case where all of a country’s suppliers went from no lockdown to a full lockdown, the preferred estimate implies a more than 20 percentage points drop in the country’s seaborne import growth.
  - The estimated spillover effect is especially large and statistically significant in the early stages of the crisis—explaining about 10 percent contraction of world trade in February-March—but becomes statistically insignificant later in the sample.
  - Evidence is suggestive of downstream propagation of countries’ lockdowns through global supply chains; during the early stages of the crisis both direct and indirect supply-chain effects are marginally significant and economically sizeable, and the hypothesis of joint non-significance of direct and indirect effects is strongly rejected.
- JEL Classification Numbers: F10, F14, F42, I18, R40
- Keywords: COVID-19, lockdowns, trade, spillovers, supply chains

### Identification strategy and intuition
- Research design: shift-share framework treating foreign lockdown stringency as shifters and pre-COVID import shares as shares, augmented by geography-induced travel-time lags to exploit delivery delays in seaborne trade.
- Illustrative timing:
  - China imposed lockdown restrictions on January 23rd. Travel-time distributions from China to Korea show most trips take between one to three days; Korea’s lockdown exposure rose very soon and Korean import growth fell significantly.
  - Modal travel time from China to the U.S. West Coast is of around two weeks; U.S. import growth also fell significantly as China’s January 23rd containment measures affected the region’s lockdown exposure.
- The geography-induced lag is unique to seaborne trade and enables identification beyond standard shift-share variation.

### Empirical specification (summary)
- Dependent variable: year-on-year import growth M̂it for country i on day t.
- Foreign lockdown exposure LLEXit constructed as:
  - LLEXit = Σj wij · lj,t−d(i,j)
  - where lj,t is stringency of lockdown measures in country j on day t, wij are pre-COVID import weights (Σj wij = 1), and d(i,j) is travel time in days from j to i.
- Baseline regression:
  - M̂it = γt + αi + β·LLEXit + X′it δ + εit
  - γt and αi are time and country fixed effects; Xit includes controls.
- Controls in preferred specification:
  - domestic lockdown stringency;
  - (change in) domestic cases in ratio to population;
  - (change in) domestic deaths in ratio to population;
  - country fixed effects;
  - time fixed effects.

### Endogeneity concerns and mitigation
- Main concerns:
  - Proximity to virus hotspots correlates with both higher lockdown exposure (via shorter travel times) and worse local health conditions that could depress import demand independently.
  - Pre-COVID trade links may be endogenous to geographic proximity, which also relates to virus spread.
- Mitigation:
  - Daily frequency and geographic disaggregation reduce likelihood confounders operate at same speed as shipping lags.
  - Inclusion of domestic COVID intensity measures and domestic lockdown stringency as controls.
  - Country fixed effects and time fixed effects.
  - Use of import growth rates rather than levels to address gravity-based endogeneity related to distance.

### Sources of identification
- Time-series variation at daily frequency: evolution of lockdown policies ljt.
- Cross-sectional variation in pre-COVID trade shares: heterogeneity in wij.
- Cross-sectional variation in travel times: heterogeneity in d(i,j) giving differential timing of exposure.

### High-frequency seaborne trade dataset and measurement
- Data source and processing:
  - AIS messages (position, speed, draught) converted to port-to-port voyage and trade volume estimates using CKLS methods.
  - Random forest classifier trained on U.S. vessel entry records; draught-based volume estimates.
  - Raw AIS messages collected between January 1st 2015 and June 30th 2020; down-sampled to hourly frequency. Raw AIS data collected by MarineTraffic.
- Aggregation and scope:
  - Bilateral data aggregated to country-pair level except U.S., where ports grouped into U.S. West and U.S. East.
  - Focus on non-commodity trade (general cargo, container ships, and vehicle carriers).
- Noise reduction:
  - Steps applied to reduce large abrupt jumps in daily import growth rates for small countries with infrequent ship arrivals.
- Practical advantage:
  - Daily bilateral trade estimates allow exploitation of travel-time heterogeneity to identify causal supply spillovers that monthly trade data could not resolve.

### Data coverage and concentration (2015–2019, non-commodity imports by weight)
- The top 50 countries account in aggregate for 87 percent of the estimated global non-commodity imports by weight.
- Top ranked importers (Millions of metric tons; cumulative world share):
  - 1 China 571.08 8.3
  - 2 Singapore 320.6 13.0
  - 3 Korea 287.4 17.2
  - 4 United Kingdom 287.3 21.4
  - 5 US East 255.9 25.1
  - 6 Japan 249.4 28.7
  - 7 Netherlands 213.3 31.8
  - 8 Malaysia 212.7 34.9
  - 9 Hong Kong SAR 194.0 37.7
  - 10 United Arab Emirates 189.4 40.5
- Routes:
  - Total active routes detected over 2015-2019: 7,580.
  - The top 50 routes (0.66% of all routes) account for 30 percent of non-commodity trade by weight.
  - Top routes (millions of metric tons):
    - 1 from CN to KR 153.5 2.0
    - 2 from SG to CN 146.7 3.8
    - 3 from CN to JP 89.6 5.0
    - 4 from KR to CN 87.3 6.1
    - 5 from HK to CN 81.4 7.1
    - 6 from SG to MY 74.6 8.1
    - 7 from CN to USW 66.1 8.9
    - 8 from NL to GB 66.0 9.8
    - 9 from SG to HK 61.7 10.6
    - 10 from MY to CN 60.7 11.4
  - Of the top 10 routes, 8 correspond to intra-Asia trade; only one is fully outside Asia (Netherlands to UK); one is China to U.S. West Coast.

### Travel times and identification implications
- Empirical distribution of country-to-country travel times: around 93% of all country-to-country voyages take place within a 30-day window.
- Travel-time treatment:
  - Travel time treated as random variable with pmf f_ij(d) = P[d(i,j)=d].
  - Average lockdown stringency experienced by i from j on day t: l̅_ij,t = Σ_{d=0}^∞ f_ij(d) l_{j,t-d}.
  - Aggregate lockdown exposure: LEX_i,t = Σ_j w_ij l̅_ij,t.
  - Empirical distribution of d(i,j) truncated by dropping values below the 10th percentile and above the 90th percentile.
- Controls measured using empirical distribution of d(i,j) and import weights so controls reflect conditions at time incoming ships set sail.

### Lockdown stringency and disease data
- Lockdown stringency index: simple average of nine ordinal indicators (eight closures/containment indicators and one public information campaigns health measure), each taking integer values up to 4; normalized to range from 0 to 100 (Hale et al., 2020).
- For China, an updated province-aggregated stringency index (Zhang, forthcoming) used.
- Daily confirmed COVID cases and deaths from Hale et al. dataset, originally compiled by the European Centre for Disease Control.

### Main empirical results — effect on import growth (Table 3)
- Domestic and foreign lockdowns measured 0 (no restrictions) to 100 (strictest possible lockdown).
- Key point estimates (standard errors in parentheses); significance levels: * 5%, ** 1% and, *** 0.1%. Standard errors clustered by country:
  - Column (1): Domestic lockdown coefficient = -0.105*** (0.0228)
  - Column (2): Foreign lockdowns coefficient = -0.158*** (0.0255)
  - Column (3): Domestic lockdown coefficient = 0.0970 (0.0690); Foreign lockdowns = -0.261** (0.0781)
  - Column (4): Domestic lockdown = 0.0864 (0.0775); Foreign lockdowns = -0.284** (0.105)
  - Column (5): Domestic lockdown = -0.0304 (0.0720); Foreign lockdowns = -0.246*** (0.0682)
  - Column (6) [preferred specification, includes domestic and foreign disease controls]: Domestic lockdown = -0.0181 (0.0750); Foreign lockdowns = -0.226** (0.0715)
    - D.Domestic cases = -0.122 (0.414)
    - D.Domestic deaths = -6.406 (4.065)
  - Column (7): Foreign lockdowns = -0.210** (0.0767); D.Foreign cases = -0.702 (0.755); D.Foreign deaths = 6.082 (9.721)
  - Example constants: column (6) constant = 10.94** (3.237)
  - Example observations: 8869, 8869, 8869, 8698, 8698, 8698, 7718, 771
- Interpretation:
  - Going from no restrictions (0) to a full lockdown (100) in the local economy associated with 10.5 percentage point lower import growth (column (1)).
  - All suppliers simultaneously going from no restrictions to full lockdown associated with 15.8 percentage point lower import growth (column (2)).
  - Preferred specification (column (6)): all partners going from no lockdown to a full lockdown leads to a fall of 22.6 percentage points in import growth.
  - Foreign disease intensity (confirmed cases and deaths) does not significantly affect import growth once lockdown exposure is included, suggesting government-imposed lockdowns capture supply disruptions rather than voluntary fear-driven behavior by suppliers.

### Time-varying spillovers and aggregate interpretation
- Preferred coefficient of -0.226 is time-varying:
  - Figure 4 (60-day rolling windows) shows the effect of foreign lockdowns was around -0.4 between mid-February and late March, economically large and statistically significant.
  - By April, the effect dissipates in size and significance.
- Possible explanations for vanishing spillovers:
  - Frictions that initially prevented rapid reallocation were large enough initially.
  - Absence of effect later could reflect general equilibrium reallocation (supply chains reconfigured) or later lockdowns being qualitatively different and less disruptive.
  - Non-linearities: supply disruptions may occur only after a tipping point in lockdown stringency.
  - Anecdotal evidence suggests firms began reconfiguring supply chains within 2-3 months, potentially reducing spillovers.
- The paper leaves disentangling these channels to future research.

### Counterfactual world trade estimates (Section 3)
- Translation from estimated effect on import growth to import-volume differences:
  - Counterfactual import volumes M_i,t^c = M_i,t − M_i,t−365 · β̂ · LLEX_i,t.
- Using β̂ = −0.4 and the country-level, time-varying lockdown-exposure measures, a counterfactual series for world import volumes was constructed.
- Key empirical finding:
  - Supply disruptions due to lockdowns, on average, reduced global seaborne imports in February-March 2020 by 10 percent.
  - China’s lockdowns contributed about 4 percentage points to that reduction.

### Illustrative example and indirect supply-chain effects
- Three-port illustrative world: China, Korea, and the U.S.
  - Travel times: China→Korea = 3 days; China→U.S. = 14 days; Korea→U.S. = 11 days.
  - U.S. import growth modeled as function of direct exposure LEX_US,1 and indirect exposure LEX_US,2.
  - Specification example:
    - M̂_US,t = β1 LEX_US,1 + β2 LEX_US,2 + ε_US,t.
    - Direct exposure example: LEX_US,1 = w_US,K l_K,t−11 + w_US,C l_C,t−14.
    - Indirect exposure defined recursively using partners’ exposures lagged by travel times.
- Operational assumptions to estimate higher-order effects:
  - K = 2 (include only up to one step of indirect effects).
  - w_i,j^2 = w_i,j^1 (indirect weights equal direct pre-COVID import-share weights).
  - Implicit zero processing time of intermediate inputs in illustrative example; when estimating, impose homogeneous processing lags and explore plausible lags.
- Potential bias:
  - Measurement error from equal-weight and homogeneous-lag assumptions could bias estimates. If classical, point estimates and t-statistics biased downwards.
- Empirical results on indirect effects:
  - Panel (a) of Figure 6 (full sample): both direct and indirect coefficients have expected sign across processing lags; indirect coefficient exhibits U-shaped pattern; indirect exposure not statistically significant in full sample for any lag.
  - Focusing on initial stages through March 31:
    - U-shaped pattern of indirect spillovers more pronounced; maximum effect around two-week processing lag.
    - At two-week processing lags both direct and indirect foreign lockdown exposures are marginally statistically significant:
      - Direct exposure p-value: 0.061
      - Indirect exposure p-value: 0.068
    - Column (4) in Table 4 (14-day processing lag) yields economically sizeable point estimates on both direct and indirect lockdown exposures.
    - An F-test cannot reject equality of direct and indirect coefficients (p-value: 0.3972).
    - The null that both coefficients are equal to zero is strongly rejected (p-value: 0.0011).
  - Table 4 excerpt (indirect spillovers measured with 14-day processing lag) entries include:
    - Foreign lockdowns coefficients: -0.226**, -0.171, -0.433**, -0.351 (with standard errors shown in the table).
    - L14.Indirect lockdowns coefficients: -0.163, -0.837 (with standard errors shown).
    - Observations: 877180, 854312, 3626 across columns.
    - Sample ends: June 30 (full), March 31 (subsample).

### Conclusions
- Shift-share research design on high-frequency daily bilateral seaborne trade data traces origins of the trade collapse during the COVID-19 Great Lockdown.
- Main conclusions:
  - Lockdowns and supply disruptions played a significant role in the trade collapse at the beginning of the crisis.
  - Countries with stronger trade links to and closer geographic proximity to countries under heavy lockdowns experienced larger and faster import contractions.
  - Evidence for indirect spillovers through global supply chains exists, but these effects were present only during the first 2-3 months of the pandemic; thereafter demand effects likely dominated.
- Two hypothesized explanations for vanishing supply spillovers over time:
  - Frictions were resolved within the timeframe and firms flexibly adapted supply chains, including finding new suppliers.
  - Later lockdowns (e.g., in the West) were intrinsically different from early lockdowns imposed in Asia and Europe.

*Source: WP/20/284, Supply Spillovers During the Pandemic: Evidence from High-Frequency Shipping Data (Sections I–III).*

### Section 1

### Supply Spillovers During the Pandemic: Evidence from High-Frequency Shipping Data

### Abstract and key findings
- World trade contracted dramatically during the global economic crisis induced by the COVID-19 pandemic.  
- The paper quantifies the causal effect of supply spillovers from lockdowns using a novel dataset of daily bilateral seaborne trade and a shift-share identification strategy that leverages geography-induced cargo delivery lags.  
- Main empirical findings:
  - Strong but short-lived supply spillovers of lockdowns through international trade.
  - In a hypothetical case where all of a country’s suppliers went from no lockdown to a full lockdown, the preferred estimate implies a more than 20 percentage points drop in the country’s seaborne import growth.
  - This estimated spillover effect is especially large and statistically significant in the early stages of the crisis—explaining about 10 percent contraction of world trade in February-March—but becomes statistically insignificant later in the sample.
  - Evidence is suggestive of downstream propagation of countries’ lockdowns through global supply chains; during the early stages of the crisis both direct and indirect supply-chain effects are marginally significant and economically sizeable, and the hypothesis of joint non-significance of direct and indirect effects is strongly rejected.
- JEL Classification Numbers: F10, F14, F42, I18, R40
- Keywords: COVID-19, lockdowns, trade, spillovers, supply chains

### Identification strategy and intuition
- Research design: a shift-share framework that treats foreign lockdown stringency as shifters and pre-COVID import shares as shares, augmented by geography-induced travel-time lags to exploit delivery delays in seaborne trade.
- Illustrative cases:
  - China imposed lockdown restrictions on January 23rd. Travel-time distributions from China to Korea show most trips take between one to three days; Korea’s lockdown exposure rose very soon and Korean import growth fell significantly.
  - Modal travel time from China to the U.S. West Coast is of around two weeks; U.S. import growth also fell significantly as China’s January 23rd containment measures affected the region’s lockdown exposure.
- The geography-induced lag is unique to seaborne trade and enables identification beyond standard shift-share variation.

### Empirical specification (summary)
- Dependent variable: year-on-year import growth M̂it for country i on day t.
- Foreign lockdown exposure LLEXit constructed as:
  - LLEXit = Σj wij · lj,t−d(i,j)
  - where lj,t is the stringency of lockdown measures in country j on day t, wij are pre-COVID import weights (Σj wij = 1), and d(i,j) is travel time in days from j to i.
- Baseline regression:
  - M̂it = γt + αi + β·LLEXit + X′it δ + εit
  - γt and αi are time and country fixed effects; Xit includes controls.
- Controls included in preferred specification:
  - domestic lockdown stringency;
  - (change in) domestic cases in ratio to population;
  - (change in) domestic deaths in ratio to population;
  - country fixed effects;
  - time fixed effects.

### Endogeneity concerns and mitigation
- Main concerns:
  - Proximity to virus hotspots correlates with both higher lockdown exposure (via shorter travel times) and worse local health conditions that could depress import demand independently.
  - Pre-COVID trade links may be endogenous to geographic proximity, which also relates to virus spread.
- Mitigation strategy:
  - Daily frequency and geographic disaggregation reduce likelihood that confounders operate at the same speed as shipping lags.
  - Inclusion of domestic COVID intensity measures and domestic lockdown stringency as controls.
  - Country fixed effects to absorb time-invariant regional idiosyncrasies.
  - Time fixed effects to control for common global trends (e.g., rising uncertainty, tightening financing conditions).
  - Use of import growth rates rather than levels to address gravity-based endogeneity related to distance.

### Sources of identification (three channels)
- Time-series variation at daily frequency: evolution of lockdown policies ljt.
- Cross-sectional variation in pre-COVID trade shares: heterogeneity in wij.
- Cross-sectional variation in travel times: heterogeneity in d(i,j) giving differential timing of exposure.

### High-frequency seaborne trade dataset and variable construction
- Coverage and method:
  - The dataset uses AIS (Automatic Identification System) radio messages emitted by cargo ships for navigational safety; messages include position, speed, draught, etc.
  - CKLS (Cerdeiro, Komaromi, Liu and Saeed, 2020) methods convert over one billion AIS messages into port-to-port voyage and trade volume estimates using spatial clustering to detect ports, a random forest classifier trained on U.S. vessel entry records to distinguish trade visits, and draught-based volume estimates.
  - Raw AIS messages were collected between January 1st 2015 and June 30th 2020.
  - While most ships send AIS messages with a frequency of 2-10 seconds, the data used are down-sampled to the hourly frequency. The raw AIS data were collected by MarineTraffic.
- Aggregation and scope:
  - Bilateral data are aggregated to the country-pair level except for the U.S., where ports are grouped into U.S. West and U.S. East due to very different travel times.
  - Focus on non-commodity trade (general cargo, container ships, and vehicle carriers).
- Noise reduction:
  - Small countries that receive ships infrequently can show large and abrupt jumps in daily import growth rates; the dataset applies steps to reduce such noise.
- Practical advantage:
  - High-frequency (daily) bilateral trade estimates allow exploitation of travel-time heterogeneity to identify causal supply spillovers, which monthly trade data could not resolve.

*Source: WP/20/284, Supply Spillovers During the Pandemic: Evidence from High-Frequency Shipping Data (Sections I–III).*

### Section 2

### Section 2

### Data and high-frequency measurement
- Use of AIS-based ship-movement data aggregated to country-level for the top 50 importing countries; results based on the top 50 importing countries.
- A 7-day moving average of the daily trade estimates is applied to reduce noise in daily arrivals.
- Country-level aggregated trade volume estimates can be visualized and downloaded from the COMTRADE AIS monitoring platform (as noted in the source).
- Raw-AIS data sample starts on January 1st, 2015, but estimates are censored before April 1st, 2015 to avoid start-point estimation problems.
- Daily growth rates for 2020 are calculated by taking the average of the same dates in the previous three years (2017-19) as the base period.

### Dataset coverage and top importers (2015-2019, non-commodity imports by weight)
- The top 50 countries account in aggregate for 87 percent of the estimated global non-commodity imports by weight.
- Top ranked importers (Millions of metric tons of imports; cumulative world share):
  - 1 China 571.08 8.3
  - 2 Singapore 320.6 13.0
  - 3 Korea 287.4 17.2
  - 4 United Kingdom 287.3 21.4
  - 5 US East 255.9 25.1
  - 6 Japan 249.4 28.7
  - 7 Netherlands 213.3 31.8
  - 8 Malaysia 212.7 34.9
  - 9 Hong Kong SAR 194.0 37.7
  - 10 United Arab Emirates 189.4 40.5
- The top 50 regions cover economies across Africa, America, Asia and Europe.
- Notes: Table shows the top 50 countries ranked by estimated volume of non-commodity imports over 2015-2019 based on AIS data.

### Routes and concentration
- The dataset detects a total of 7,580 active routes over 2015-2019.
- The top 50 routes (0.66% of all routes) account for 30 percent of non-commodity trade by weight.
- Top routes by millions of metric tons of cargo (rank, route, millions):
  - 1 from CN to KR 153.5 2.0
  - 2 from SG to CN 146.7 3.8
  - 3 from CN to JP 89.6 5.0
  - 4 from KR to CN 87.3 6.1
  - 5 from HK to CN 81.4 7.1
  - 6 from SG to MY 74.6 8.1
  - 7 from CN to USW 66.1 8.9
  - 8 from NL to GB 66.0 9.8
  - 9 from SG to HK 61.7 10.6
  - 10 from MY to CN 60.7 11.4
- Of the top 10 routes, 8 correspond to intra-Asia trade; only one is fully outside Asia (Netherlands to UK); one is China to U.S. West Coast.

### Travel times and implications for identification
- Empirical distribution of country-to-country travel times: around 93% of all country-to-country voyages take place within a 30-day window.
- Virtually all world seaborne trade is shipped and delivered within the month; daily trade data are crucial to identify effects of high-frequency policy changes like lockdowns.
- Travel-time treatment in the model:
  - Travel time between two countries treated as a random variable with estimated probability mass function f_ij(d) = P[d(i,j)=d].
  - Average lockdown stringency experienced by country i from j on day t: l̅_ij,t = Σ_{d=0}^∞ f_ij(d) l_{j,t-d}.
  - Aggregate lockdown exposure of country i: LEX_i,t = Σ_j w_ij l̅_ij,t.
  - Empirical distribution of d(i,j) truncated by dropping values below the 10th percentile and above the 90th percentile.
- Control variables are measured using the empirical distribution of d(i,j) and import weights so controls reflect conditions at the time incoming ships set sail.

### Lockdown stringency and disease-spread data
- Lockdown stringency index sourced from Hale, Angrist, Kira, Petherik, Phillips and Webster (2020); constructed as a simple average of nine ordinal indicators (eight closures/containment indicators and one public information campaigns health measure), each taking integer values up to 4; normalized to range from 0 to 100.
- For China, an updated province-aggregated stringency index (Zhang, forthcoming) is used to capture within-country heterogeneity.
- Daily series of confirmed COVID cases and confirmed deaths are from the Hale et al. dataset, originally compiled by the European Centre for Disease Control.

### Main empirical results (Table 3) — effect on import growth
- Domestic and foreign lockdowns measured 0 (no restrictions) to 100 (strictest possible lockdown).
- Key point estimates (standard errors in parentheses); significance levels: * 5%, ** 1% and, *** 0.1%. Standard errors clustered by country.
  - Column (1): Domestic lockdown coefficient = -0.105*** (0.0228)
  - Column (2): Foreign lockdowns coefficient = -0.158*** (0.0255)
  - Column (3): Domestic lockdown coefficient = 0.0970 (0.0690); Foreign lockdowns = -0.261** (0.0781)
  - Column (4): Domestic lockdown = 0.0864 (0.0775); Foreign lockdowns = -0.284** (0.105)
  - Column (5): Domestic lockdown = -0.0304 (0.0720); Foreign lockdowns = -0.246*** (0.0682)
  - Column (6) [preferred specification, includes domestic and foreign disease controls]: Domestic lockdown = -0.0181 (0.0750); Foreign lockdowns = -0.226** (0.0715)
    - D.Domestic cases = -0.122 (0.414)
    - D.Domestic deaths = -6.406 (4.065)
  - Column (7): Foreign lockdowns = -0.210** (0.0767); D.Foreign cases = -0.702 (0.755); D.Foreign deaths = 6.082 (9.721)
  - Constant terms vary across specifications (e.g., column (6) constant = 10.94** (3.237))
  - Observations reported (examples): 8869, 8869, 8869, 8698, 8698, 8698, 7718, 771
  - Time FE and Country FE included in later specifications as noted.
- Interpretation:
  - Going from no restrictions (0) to a full lockdown (100) in the local economy associated with 10.5 percentage point lower import growth (column (1)).
  - All suppliers simultaneously going from no restrictions to full lockdown associated with 15.8 percentage point lower import growth (column (2)).
  - Preferred specification (column (6)): all partners going from no lockdown to a full lockdown leads to a fall of 22.6 percentage points in import growth.
  - When both domestic and partners’ lockdown stringencies are controlled for, foreign lockdown exposure remains large and negative while domestic lockdown becomes statistically insignificant in some specifications.
  - Foreign disease intensity (confirmed cases and deaths) does not significantly affect import growth once lockdown exposure is included, suggesting government-imposed lockdowns capture supply disruptions rather than voluntary fear-driven behavior by suppliers.

### Time-varying spillovers and aggregate interpretation
- Preferred coefficient of -0.226 is time-varying:
  - Figure 4 (60-day rolling windows) shows the effect of foreign lockdowns was around -0.4 between mid-February and late March, economically large and statistically significant.
  - By April, the effect dissipates in size and significance.
- Possible explanations for the vanishing spillovers:
  - Shift-share design relies on frictions that prevent rapid reallocation; observed effect early in sample implies frictions were large enough initially.
  - Absence of effect later could reflect either general equilibrium reallocation (supply chains reconfigured) or later lockdowns being qualitatively different and less disruptive.
  - Non-linearities: supply disruptions may occur only after a tipping point in lockdown stringency.
  - Early Asian lockdowns may have been qualitatively different from subsequent lockdowns elsewhere.
  - Anecdotal evidence suggests firms began reconfiguring supply chains within 2-3 months (redundancy, buffers), potentially reducing spillovers over time.
- The paper leaves disentangling these channels (reallocation vs. different nature of later lockdowns vs. non-linearities) to future research.

*Source: wpiea2020284-print-pdf - Section 2*

### Section 3

### wpiea2020284-print-pdf - Section 3

### Counterfactual world trade estimates
- The estimated effect of foreign lockdowns on import growth rates, 훽훽̂퐿퐿퐿퐿푖푖푖푖, translates into an estimated effect in differences as 푀푀푖푖푖푖 −365 훽훽̂퐿퐿퐿퐿푖푖푖푖, where 푀푀푖푖푖푖 −365 is the base-period import volume.
- Counterfactual import volumes in country 푖푖, 푀푀푖푖푖푖푐푐, are given by:
  - 푀푀푖푖푖푖푐푐 =푀푀푖푖푖푖 −푀푀푖푖푖푖 −365 훽훽̂퐿퐿퐿퐿푖푖푖푖.
- Using 훽훽̂ =−0.4 and the country-level, time-varying lockdown-exposure measures, a counterfactual series for world import volumes was constructed.
- Key empirical finding:
  - Supply disruptions due to lockdowns, on average, reduced global seaborne imports in February-March 2020 by 10 percent.
  - China’s lockdowns contributed about 4 percentage points to that reduction.

### Illustrative example of indirect supply-chain effects
- Three-port world used to illustrate direct and indirect effects: China, Korea, and the U.S.
  - Travel times: China→Korea = 3 days; China→U.S. = 14 days; Korea→U.S. = 11 days.
  - China exports to Korea and U.S.; Korea exports only to U.S.; U.S. does not export.
- U.S. import growth (푀푀�푈푈푈푈,푖푖) modeled as a function of:
  - Direct lockdown exposure 퐿퐿퐿퐿푈푈푈푈,푖푖1
  - Indirect lockdown exposure 퐿퐿퐿퐿푈푈푈푈,푖푖2
  - Other factors 휀휀푈푈푈푈,푖푖
- Specification for the illustrative example:
  - 푀푀�푈푈푈푈,푖푖 = 훽훽1 퐿퐿퐿퐿푈푈푈푈,푖푖1 + 훽훽2 퐿퐿퐿퐿푈푈푈푈,푖푖2 + 휀휀푈푈푈푈,푖푖.
  - Direct exposure example: 퐿퐿퐿퐿푈푈푈푈,푖푖1 = 푤푤푈푈푈푈,퐾퐾퐾퐾퐾퐾 푙푙퐾퐾퐾퐾퐾퐾,푖푖−11 + 푤푤푈푈푈푈,퐶퐶퐶퐶퐶퐶 푙푙퐶퐶퐶퐶퐶퐶,푖푖−14.
  - Indirect exposure defined recursively:
    - 퐿퐿퐿퐿푈푈푈푈,푖푖2 = 푤푤푈푈푈푈,퐾퐾퐾퐾퐾퐾 퐿퐿퐿퐿퐾퐾퐾퐾퐾퐾,푖푖−11 1 + 푤푤푈푈푈푈,퐶퐶퐶퐶퐶퐶 퐿퐿퐿퐿퐶퐶퐶퐶퐶퐶,푖푖−14 1.
- Interpretation:
  - 훽훽1 measures direct spillover effects of lockdowns.
  - 훽훽2 measures effects transmitted through global supply chains.

### Specification, assumptions, and results
- General empirical model (as presented):
  - 푀푀�푖푖푖푖 = 훾훾푖푖 + 훼훼푖푖 + ∑훽훽푘퐿퐿퐿퐿푖푖푖푖푘퐾푘=1 + 퐗퐗푖푖푖푖′ 훿훿 + 휀휀푖푖푖푖.
  - Direct exposure: 퐿퐿퐿퐿푖푖푖푖1 = ∑푤푤푖푖푗푗1 푙푙푗푗,푖푖−푑푑(푖푖,푗푗).
  - Higher-order exposures: 퐿퐿퐿퐿푖푖푖푖푘푘 = ∑푤푤푖푖푗푗푘푘 퐿퐿퐿퐿푗푗,푖푖−푑푑(푖푖,푗푗)푘푘−1 for k=2,...,K, with ∑푤푤푖푖푗푗푘푘 =1.
- Three identification/operational assumptions made to take model to data:
  - K = 2 (include only up to one step of indirect effects beyond direct exposure).
  - 푤푤푖푖푗푗2 = 푤푤푖푖푗푗1 (indirect weights equal direct pre-COVID import-share weights due to lack of higher-order input-linkage data).
  - Implicitly assume zero processing time of intermediate inputs in the illustrative example; when estimating, impose homogeneous processing lags across countries and explore plausible lags with the data.
- Discussion of potential bias:
  - Measurement error from equal-weight and homogeneous-lag assumptions could bias estimates.
  - If error is classical errors-in-variables, point estimates and t-statistics would be biased downwards.
- Empirical results:
  - Panel (a) of Figure 6 (full sample) shows both direct and indirect coefficients have expected sign across processing lags; indirect coefficient exhibits U-shaped pattern as processing lag increases; indirect exposure is not statistically significant in full sample for any lag considered.
  - Focusing on the initial stages through March 31:
    - U-shaped pattern of indirect spillovers more pronounced; maximum effect around two-week processing lag.
    - At two-week processing lags both direct and indirect foreign lockdown exposures are marginally statistically significant:
      - Direct exposure p-value: 0.061
      - Indirect exposure p-value: 0.068
    - Column (4) in Table 4 (14-day processing lag) yields economically sizeable point estimates on both direct and indirect lockdown exposures.
    - An F-test cannot reject equality of direct and indirect coefficients (p-value: 0.3972).
    - The null that both coefficients are equal to zero is strongly rejected (p-value: 0.0011).
  - Table 4 excerpt (indirect spillovers measured with 14-day processing lag) key entries:
    - Foreign lockdowns coefficient examples: -0.226**, -0.171, -0.433**, -0.351 (with associated standard errors shown in the table).
    - L14.Indirect lockdowns coefficients: -0.163, -0.837 (with standard errors shown).
    - Observations: 877180, 854312, 3626 across columns as presented.
    - Sample ends: June 30 (full), March 31 (for subsample).
    - Significance notation: * 5%, ** 1% and, *** 0.1% significance. Clustered standard errors.
- Overall interpretation:
  - Results are suggestive of downstream propagation of countries’ lockdowns through global supply chains, particularly during the first 2-3 months of the pandemic.

### Concluding remarks
- The paper implements a shift-share research design on high-frequency daily bilateral seaborne trade data to trace origins of the trade collapse during the COVID-19 Great Lockdown.
- Main conclusions:
  - Lockdowns and supply disruptions played a significant role in the trade collapse at the beginning of the crisis.
  - Countries with stronger trade links to and closer geographic proximity to countries under heavy lockdowns experienced larger and faster import contractions.
  - Evidence for indirect spillovers through global supply chains exists, but these effects were present only during the first 2-3 months of the pandemic; thereafter demand effects likely dominated.
- Two hypothesized explanations for vanishing supply spillovers over time:
  - Frictions were resolved within the timeframe and firms flexibly adapted supply chains, including finding new suppliers.
  - Later lockdowns (e.g., in the West) were intrinsically different from early lockdowns imposed in Asia and Europe.

*Source: wpiea2020284-print-pdf - Section 3*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020284-print-pdf.pdf_