## wpiea2022210-print-pdf

## Source details

**Canonical URL:** [wpiea2022210-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022210-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022210-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022210-print-pdf.pdf.json)

---

### Rapid adoption of digital technologies during the pandemic
- The pandemic drove rapid adoption of new digital technologies, including teleconferencing software and e-commerce platforms, enabling hybrid work models and contactless transactions.
- Digitalization improved the ability to work remotely and sell without contact, shielding workers and firms from the pandemic’s negative effects.
- Simple correlations: estimates of output losses are higher in countries with weaker digital infrastructure; job posts fell less and recovered more quickly in digital occupations.

### Research question and empirical approach
- Objective: empirically investigate and quantify whether digitalization improves economic resilience to recessions (typical recessions and COVID-19) at the firm level.
- Sample: firm-level sample consisting of 75 advanced, emerging market and developing economies from 2001Q1 to 2021Q4 (analysis uses various subsamples, e.g., 2001Q1–2021Q1 and 2016Q1–2021Q1).
- Identification: difference-in-differences approach in a local projection setting (Jordà 2005), comparing firms in more digitalized industries to firms in less digitalized industries.
- Baseline digitalization measure: industry-wise Calvino et al. (2018) quartiles (time-invariant in baseline).

### Data — firm-level, recession, and digitalization inputs (key statistics preserved)
- Firm-level data source: S&P Capital IQ; database covers 150 countries from 1950Q1 to 2021Q2; sample restricted to 2001Q1 onwards, leaving 75 countries.
- Panel size: over 24,000 firms for the period 2001Q1 to 2021Q1; total of more than 800,000 observations.
- Revenue measure: total revenue (IQ_TOTAL_REV).
- Intangibles measure: IQ_GW_INTAN as share of IQ_TOTAL_ASSETS; approximately 60% of firms do not report intangibles.
- Recession counts: 231 recessions for advanced economies and 336 for emerging market and developing economies.
- Recession baseline: start of a technical recession defined by two consecutive quarters of negative GDP growth.
- Digitalization measures (pluralist approach):
  - 퐷_r^C (Calvino et al. 2018), time-invariant industry-wise quartile.
  - 퐷_{r,i}^{ICIO} = ∑ α_{rli} ⋅ D_l using OECD ICIO for 66 countries (median 2001–2015).
  - 퐷_{r,i}^{TiVA} = ∑ ζ_{rli} ⋅ D_l using OECD TiVA for 66 countries (median 2001–2018).
  - 퐷_{r,i}^{Int} = median intangibles share in country-industry (median 2001–2021 within firms).
  - 퐷_{r,i}^{TechSkills} from LinkedIn (2016Q1–2022Q1) for 20 industries × 40 countries.
- Table A1.6 descriptive statistics (selected):
  - Calvino: Countries 76; Obs. 3,304,900; Mean .60; Std .26; Min .25; Max 1.
  - ICIO: Countries 56; Obs. 2,861,280; Mean .11; Std .20; Min 0; Max .86.
  - TiVA: Countries 58; Obs. 2,016,718; Mean .083; Std .15; Min 0; Max .76.
  - Intangibles: Countries 76; Obs. 1,301,928; Mean .11; Std .17; Min 0; Max .76.
  - Tech Skills: Countries 38; Obs. 2,254,000; Mean .099; Std .04; Min 0; Max .19.

### Empirical methodology (specification highlights)
- Primary method: Jordà’s (2005) local projection method; horizons k = 0,1,...15 (quarters) for historical recessions; COVID-19 analyses estimated over shorter horizons (up to five quarters).
- Specification 1: unconditional recession effect on firm revenue; dependent Δy_{n,i,t+k} = log difference in revenue over k quarters; recession dummy R_{i,t} denotes beginning of a technical recession.
- Specification 2: heterogeneous effect by digitalization; interaction R_{i,t} × D_m with D_m standardized so μ_j^k interpretable as marginal response for firms in industries one standard deviation above mean digitalization; figures present μ_j^k ×100 in percentage point terms.
- Fixed effects and clustering:
  - Baseline: firm-quarter dummies γ_{nq}^k; country-sector FE α_{is}^k; standard errors clustered by firm and country-time.
  - COVID-19 regressions: replace α_{ist}^k with α_{it}^k and α_{ir}^k; clustering by firm and country-time.
  - LinkedIn industry-level regressions: α_{it}^k and α_{ri}^k; clustered at country level.

### Main empirical findings (exact magnitudes preserved)
- Historical recessions (all recessions since 2001):
  - Average recession associated with reduction in level of revenue by more than 10 percent four quarters after the recession.
  - Reduction of 5 percent after 8 quarters.
  - Heterogeneity by digitalization: firms in industries one standard deviation more digitalized than the mean experience 1 percent lower revenue losses after two years and about 2 percent lower revenue losses and highly statistically significant four years after the recession.
- COVID-19 pandemic recessions (recessions begun on or after 2019Q4; sample 2016Q1 onwards):
  - COVID-19 recessions associated with persistent effects on level of revenue, of up to 20 percent after five quarters.
  - Heterogeneity by digitalization during COVID-19:
    - Firms in industries one standard deviation higher digitalization: almost 4 percent higher revenue four quarters after a recession.
    - Increase to close to 5 percent after five quarters.
    - Effect is statistically significant and more than four times larger than the corresponding effect for average historical recessions after five quarters.
- LinkedIn industry-level labor outcomes (2016Q1–2022Q1, 20 industries, 40 countries):
  - Hiring rates in highly digitalized industries grew almost 3 percent faster than in average industries in the year after the COVID-19 recession.
  - Hiring rates grew almost 4 percent faster after two years.
  - Inflow of workers to highly digitalized industries grew more than 3 percent faster than in other industries in the two years after the shock.
  - Outflow of workers from highly digitalized industries grew 1 percent slower than in other industries in the two years after the shock.
- Summary magnitudes (baseline statements preserved):
  - Four years after a typical recession, firms in industries one standard deviation more digitalized than the mean experience about 2 percent less sale losses relative to firms in industries with an average level of digitalization.
  - One year after the COVID-19 recession, firms in industries one standard deviation more digitalized than the mean experience about 4 percent smaller declines in sales than other firms.

### Reasons reported effects may underestimate true impact
- Digitalization measures are pre-determined and time-invariant to reduce endogeneity, but recessions may induce firms to become more digital, implying a larger overall effect than the differential effect detected.
- Sample consists of publicly listed firms, which are likely more digitalized than firms not included.
- Role of digitalization in mitigating scarring is likely larger in the pandemic context due to greater requirement for remote work and contactless sales.

### Robustness and alternative measures (key outcomes)
- Robust across six robustness categories: i) alternative digitalization measures; ii) alternative recession definitions; iii) alternative samples; iv) placebo tests; v) additional control variables; vi) alternative methodology.
- Alternative digitalization measures (qualitative robustness):
  - Country-industry measures 퐷_{r,i}^{ICIO}, 퐷_{r,i}^{TiVA}, and 퐷_{r,i}^{Int} show firms in more digital country-industries face 1-2 percent less revenue losses four years after recessions (less precisely estimated).
  - Time-varying ICIO and TiVA (contemporaneous and one-year lagged) yield qualitatively robust findings.
  - Pandemic-period TechSkills from LinkedIn corroborate main findings, with caveats for intangibles in some panels.
- Alternative recession definitions:
  - Using start of a currency crisis or banking and currency crisis yields very similar results with marginally larger effects over four years.
- Alternative samples:
  - Excluding year 2020 or excluding 2008 Global Financial Crisis yields similar differential responses.
  - Leave-one-country/region/industry-out tests: baseline results not dependent on any specific country, group, or 2-digit sector.
  - Income-level splits: almost no difference between advanced and emerging economies — digitalization buffers shocks independently of income characteristics.
  - Winsorization sensitivity (0.05 and 5 percent tails): similar results.
- Placebo test:
  - 200 random recession dummies produced coefficients centered on zero; no differential growth in revenues after placebo recessions.
  - Note: Approximately 3% of the original dummy (start of a technical recession) is equal to 1.
- Additional controls:
  - Interactions with financial stress and World Uncertainty Index leave baseline results essentially unchanged; inflation interaction no significant impact.
- Alternative methodology (firm-level and industry-level DiD around 2020Q1):
  - Results similar to baseline: approximately 2 percentage point rise in revenue growth post-pandemic for more digitalized industries; 3 percentage point higher growth in transitions into more digitalized industries; roughly 1 percentage point lower growth in transitions out.

### Tables and figures (selected numeric highlights)
- Table 1 correlations (industry-level medians):
  - ICIO with Calvino: 0.4*.
  - TIVA with Calvino: 0.3*; TIVA with ICIO: 1*.
  - Intangibles with Calvino: 0.4*; Intangibles with ICIO: 0.3*; Intangibles with TIVA: 0.4*.
  - Digital Skills with Calvino: 0.5*; with ICIO: 0.5*; with TIVA: 0.4*; with Intangibles: 0.5*.
  - Note: * Significant at 5% level.
- Figure highlights (descriptive notes):
  - Figure 1: Output losses computed as percent deviation between pre-pandemic (January 2020) and latest GDP projections for 2024.
  - Figure 3: Average recession → >10 percent revenue reduction four quarters after.
  - Figure 5: COVID-19 recession → up to 20 percent revenue reduction after five quarters.
  - Figure 6: COVID-19 differential effect → almost 4 percent higher revenue at four quarters, close to 5 percent at five quarters for highly digital industries.
  - Figure 7: Hiring and transitions (LinkedIn) effects summarized above.

### Annex — sample composition and size (selected country and sector counts)
- Countries included: sample drawn from 75 countries (Annex lists region breakdowns and country counts).
- Table A1.2 (selected country counts):
  - United States: Number of firms 4,740; Obs. 388,680.
  - China: Number of firms 4,077; Obs. 334,314.
  - Japan: Number of firms 3,085; Obs. 252,970.
  - India: Number of firms 2,672; Obs. 219,104.
- Table A1.3 (top sectors by firm counts):
  - Materials: Number of Firms 5,433; Obs. 445,506.
  - Capital Goods: Number of Firms 4,888; Obs. 400,816.
  - Technology Hardware and Equipment: Number of Firms 2,286; Obs. 187,452.
- Table A1.4 revenue summary:
  - Revenue: Count 1,692,161; Mean 5.49; Std 3.58; 25th 3.39; 75th 7.8.
  - Revenue – High Digital (above 75th percentile): Count 264,067; Mean 5.26; Std 3.78; 25th 2.84; 75th 7.81.

### Conclusion — interpretation and policy implications
- Main empirical conclusions:
  - Higher digitalization ex ante is associated with lower medium-term firm revenue losses following recessions.
  - During the COVID-19 pandemic recessions, the role of digitalization was larger: digitalized industries experienced substantially smaller revenue losses and stronger hiring and inflows of workers.
  - Results are robust across multiple digitalization measures, recession definitions, samples, controls, placebo tests, and alternative estimation methods.
- Policy-relevant implication (as emphasized in source):
  - Findings support efforts to promote widespread adoption of digital technologies, beyond forced adoption during the COVID-19 pandemic, because digitalization can help prevent and reduce harmful medium-run effects of economic downturns and improve firm and industry resilience.

*International Monetary Fund — wpiea2022210-print-pdf*

### 5.3 percent below those projected in January 2020 for the same year.

### 5.3 percent below those projected in January 2020 for the same year.

### Rapid adoption of digital technologies during the pandemic
- The pandemic drove rapid adoption of new digital technologies, including teleconferencing software and e-commerce platforms, enabling hybrid work models and contactless transactions.
- Digitalization improved the ability to work remotely and sell without contact, shielding workers and firms from the pandemic’s negative effects.
- Simple correlations: estimates of output losses are higher in countries with weaker digital infrastructure; job posts fell less and recovered more quickly in digital occupations.

### Research question and empirical approach
- Objective: empirically investigate and quantify whether digitalization improves economic resilience to recessions (typical recessions and COVID-19) at the firm level.
- Sample: firm-level sample consisting of 75 advanced, emerging market and developing economies from 2001Q1 to 2021Q4.
- Identification: difference-in-differences approach in a local projection setting, comparing firms in more digitalized industries to firms in less digitalized industries.
- Baseline digitalization measure: industry-wise measure of Calvino et al. (2018), which constructs digitalization quartiles using ICT input shares, the number of robots per employee, the share of ICT specialists in total employment, and the share of turnover from online sales.

### Main empirical findings (exact magnitudes preserved)
- Four years after a typical recession, firms in industries that are one standard deviation more digitalized than the mean experience about 2 percent less sale losses relative to firms in industries with an average level of digitalization.
- One year after the COVID-19 recession, firms in industries that are one standard deviation more digitalized than the mean experience about 4 percent smaller declines in sales than other firms.
- The effect is economically significant and suggests a substantial role for digitalization in improving resilience and mitigating scarring.

### Reasons the reported effects may underestimate true impact
- The analysis focuses on pre-determined and time-invariant measures of digitalization to reduce endogeneity, but recessions may induce firms to become more digital, implying a larger overall effect than the differential effect detected.
- The sample consists of publicly listed firms, which are likely more digitalized than firms not included in the sample.
- The role of digitalization in mitigating scarring is likely larger in the context of the pandemic than with previous recessions because of the greater requirement for remote work and contactless sales.

### Robustness and alternative measures
- Results are robust across a range of digitalization measures:
  - Main measure from Calvino et al. (2018).
  - Own measures of digital input shares constructed using OECD input-output tables.
  - Intangibles shares constructed from balance sheet data.
  - For COVID-19 regressions, data on the relative frequency of digital skills on LinkedIn profiles within each industry.
- Acknowledged measurement limitations: no single measure exhaustively captures all digitalization dimensions (e.g., intangibles share may also reflect copyrights and licensing agreements), but robustness across measures corroborates the role of digitalization.

### Relation to existing literature
- Extends literature on digitalization as a driver of long-run growth and innovation to the role of digitalization during downturns.
- Related empirical findings cited in the source:
  - Alcedo et al. (2022): rise in e-commerce during the pandemic.
  - Pierri and Timmer (2020): unemployment rose less in US states with greater IT adoption pre-pandemic.
  - Bellatin and Galassi (2022): cities with more pre-pandemic digital job postings posted more jobs overall during the pandemic.
  - Comin et al. (2022): technological sophistication positively impacted firms’ sales during the pandemic in three emerging market countries.
  - Abidi et al. (2021): firms using websites, email, cell phones and/or foreign technology experienced smaller sales losses during COVID-19 in 13 Middle East and Central Asia economies.
- Builds on research on automation, telework, and e-commerce and on findings that increased internet or mobile connectivity can increase export sales, reduce price dispersion, and mitigate negative shocks.

*INTERNATIONAL MONETARY FUND*

### Section III outlines our methodology. Section IV presents our results, first from

### wpiea2022210-print-pdf - Section III outlines our methodology. Section IV presents our results, first from

### II. Data — firm-level, recession, and digitalization inputs
- Firm-level data
  - Source: S&P Capital IQ.
  - Coverage: database covers 150 countries from 1950Q1 to 2021Q2; sample restricted to 2001Q1 onwards, leaving 75 countries.
  - Sample restrictions: non-financial corporations only; real estate and insurance excluded; firms with negative assets or debt removed; observations with incorrect sign for revenue, capital expenditure, cash, tangible assets, and interest expenditure set to missing.
  - Winsorization: all firm-level variables winsorized at the 1st and 99th percentiles.
  - Revenue measure: total revenue (IQ_TOTAL_REV).
  - Intangibles measure: intangible assets (IQ_GW_INTAN) as a share of total assets (IQ_TOTAL_ASSETS).
  - Tables: Tables A1.1–A1.4 report country/sector lists and summary statistics (referenced in source).

- Recession and macroeconomic data
  - Baseline recession measure: start of a technical recession defined by two consecutive quarters of negative GDP growth.
  - GDP growth source: Haver Analytics, supplemented by WEO where Haver limited.
  - Recession dummy constructed as first observation of each country recession episode.
  - Counts: 231 recessions for advanced economies and 336 for emerging market and developing economies.
  - Alternative indicators: banking and currency crisis dummies from Global Crises Data of Reinhart and Rogoff (2009).
  - Recession summary statistics: Table A1.5 (referenced in source).

- Digitalization measures (pluralist approach)
  - Baseline: industry-wise digitalization quartile 퐷_r^C (Calvino et al. 2018), time-invariant.
  - Country-industry measures constructed from input-output and value-added:
    - 퐷_{r,i}^{ICIO} = ∑ α_{rli} ⋅ D_l using OECD Inter-Country Input-Output Tables for 66 countries (median across years 2001-2015).
    - 퐷_{r,i}^{TiVA} = ∑ ζ_{rli} ⋅ D_l using OECD Trade in Value Added database for 66 countries (median across years 2001-2018).
  - Intangibles-based measure: 퐷_{r,i}^{Int} = median share of intangibles among firms in a country-industry cell using Capital IQ (median across 2001-2021 within firms, then median across firms).
    - Note: approximately 60% of firms do not report intangibles.
  - Human-capital measure (pandemic period): 퐷_{r,i}^{TechSkills} constructed from LinkedIn profiles (2016Q1 to 2022Q1) for 20 broad industries across 40 countries; measures intensity of tech skills in top 50 skills by occupation weighted by employment shares.
  - Digitalization summary statistics and cross-correlations: Table A1.6 and Table 1 (referenced).

### III. Empirical methodology
- Primary estimation approach: Jordà’s (2005) local projection method.
- Specification 1 (average unconditional recession effect on firm revenue):
  - Dependent variable: Δy_{n,i,t+k} = log difference in revenue of firm n from country i at quarterly date t over k quarters.
  - Recession dummy: R_{i,t} denotes beginning of a technical recession.
  - Controls: firm-quarter dummies γ_{nq}^k; country-sector fixed effects α_{is}^k; leads of recession variable included.
  - Horizons: ∀ k = 0,1,...15 (quarters).
- Specification 2 (heterogeneous effect by digitalization):
  - Interaction: recession variable interacted with digitalization measure D_m.
  - Digitalization in baseline: time-invariant 퐷_r^C to avoid endogeneity.
  - Fixed effects: country-sector-time fixed effects α_{ist}^k in baseline; for pandemic-specific analysis these are replaced with country-time and country-industry fixed effects α_{it}^k and α_{ir}^k.
  - Interpretation: μ_{j}^k is the marginal additional response for firms in industries one standard deviation above the mean digitalization; figures show μ_{j}^k ×100 in percentage point terms.
- Clustering: standard errors clustered by firm and country-time.
- Panel size: over 24,000 firms for the period 2001Q1 to 2021Q1; total of more than 800,000 observations.
- Pandemic-specific adjustments:
  - Drop data before 2016; set R_{i,t} to zero before 2019Q4 to focus on COVID-19 recessions.
  - Since most pandemic recessions occurred in 2020Q1, fixed effects adjusted: drop α_{ist}^k and replace with α_{it}^k and α_{ir}^k.
- LinkedIn industry-level specification (hiring and worker transitions):
  - Industry-level dependent variable Δy_{r,i,t+k} (change in log hiring or transitions over k quarters).
  - Fixed effects: α_{it}^k (country-time) and α_{ri}^k (industry-country).
  - Clustering: clustered at country level.
  - Data: quarterly for 20 industries from 40 countries over 2016Q1 to 2022Q1; approximately 7000 observations.

### IV. Results — Historical recessions and COVID-19 pandemic
- Historical recessions (all recessions since 2001)
  - Revenue dynamics after a recession:
    - Average recession associated with reduction in level of revenue by more than 10 percent four quarters after the recession.
    - Reduction of 5 percent after 8 quarters.
  - Heterogeneity by digitalization (firms in industries one standard deviation more digitalized than the mean):
    - 1 percent lower revenue losses after two years.
    - About 2 percent lower revenue losses and highly statistically significant four years after the recession.

- COVID-19 pandemic recessions (recessions begun on or after 2019Q4; sample restricted to 2016Q1 onwards)
  - Revenue dynamics after a COVID-19 recession:
    - COVID-19 recessions associated with persistent effects on level of revenue, of up to 20 percent after five quarters.
  - Heterogeneity by digitalization during COVID-19:
    - Firms in industries one standard deviation higher digitalization: almost 4 percent higher revenue four quarters after a recession.
    - Increase to close to 5 percent after five quarters.
    - Effect is statistically significant and more than four times larger than the corresponding effect for average historical recessions after five quarters.
  - Labor market reallocation (LinkedIn industry-level outcomes):
    - Hiring rates in highly digitalized industries grew almost 3 percent faster than in average industries in the year after the COVID-19 recession.
    - Hiring rates grew almost 4 percent faster after two years.
    - Inflow of workers to highly digitalized industries grew more than 3 percent faster than in other industries in the two years after the shock.
    - Outflow of workers from highly digitalized industries grew 1 percent slower than in other industries in the two years after the shock.

### V. Robustness checks
- Overview: six robustness categories — i) alternative digitalization measures; ii) alternative recession definitions; iii) alternative samples; iv) placebo tests; v) additional control variables; vi) alternative methodology.

- A. Alternative digitalization measures
  - Country-industry measures:
    - 퐷_{r,i}^{ICIO} (direct digital inputs) and 퐷_{r,i}^{TiVA} (digital share of total input value added) and 퐷_{r,i}^{Int} (intangibles share) used.
    - Qualitative robustness: firms in more digital country-industries face 1-2 percent less revenue losses four years after recessions, though less precisely estimated.
  - Time-varying input-output measures:
    - Contemporaneous and lagged values 퐷_{r,i,t}^{ICIO}, 퐷_{r,i,t}^{TiVA}, and lagged versions used to mitigate endogeneity; main findings qualitatively robust.
  - Pandemic-period checks:
    - 퐷_{r,i}^{TechSkills} from LinkedIn used for shorter period; main findings robust except Panel C using intangibles share, which may reflect pressures from pandemic-era merger activity.

- B. Alternative recession dummies
  - Alternative definitions: start of a currency crisis; start of a banking and currency crisis.
  - Results: very similar to baseline with marginally larger effects over four years (Figure 11).

- C. Alternative samples
  - Excluding year 2020: baseline results remain similar.
  - Excluding 2008 Global Financial Crisis or other systemic banking crises: differential response remains similar and not statistically different from baseline.
  - Leave-one-country/region/industry-out tests: baseline results not dependent on any specific country, group of countries, or 2-digit sector.
  - Income-level splits:
    - Advanced vs Emerging Economies: almost no difference between overall effect and effect by income level — digitalization buffers shocks independently of income characteristics.
  - Winsorization sensitivity:
    - Winsorizing 0.05 and 5 percent tails yields similar results.

- D. Placebo test
  - Created 200 random recession dummies with analogous distribution to original recession dummy.
  - Result: coefficients from random dummies centered on zero; no differential growth in revenues after placebo recessions.
  - Note: Approximately 3% of the original dummy (start of a technical recession) is equal to 1.

- E. Additional controls
  - Interacted digitalization with other macro variables:
    - Financial stress (Ahir et al. 2022): inclusion of interaction leaves baseline results essentially unchanged.
    - World Uncertainty Index (Ahir et al. 2022): inclusion leaves estimated impact essentially identical to baseline.
    - Inflation interaction: no significant impact on main results.

- F. Alternative methodology (difference-in-differences around global pandemic date 2020Q1)
  - Firm-level DiD: Δy_{n,i} = log average quarterly revenue post-pandemic (2020Q2-2021Q1) minus pre-pandemic (2017-2019) regressed on 퐷_r^C with country and sector fixed effects.
  - Industry-level DiD (LinkedIn outcomes): Δy_{r,i} regressed on 퐷_r^C with country fixed effects.
  - Results: similar to baseline — approximately 2 percentage point rise in revenue growth post-pandemic for more digitalized industries; 3 percentage point higher growth in transitions into more digitalized industries; roughly 1 percentage point lower growth in transitions out.

### VI. Conclusion — interpretation and policy implications
- Main empirical findings
  - Higher digitalization ex ante is associated with lower medium-term firm revenue losses following recessions.
  - During the COVID-19 pandemic recessions, the role of digitalization was larger: digitalized industries experienced substantially smaller revenue losses and stronger hiring and inflows of workers.
  - Results robust across multiple digitalization measures, recession definitions, samples, controls, placebo tests, and alternative estimation methods.

- Policy-relevant implication (as emphasized in source)
  - Findings support efforts to promote widespread adoption of digital technologies, beyond forced adoption during the COVID-19 pandemic, because digitalization can help prevent and reduce harmful medium-run effects of economic downturns and improve firm and industry resilience.

*International Monetary Fund — wpiea2022210-print-pdf*

### REFERENCES

### REFERENCES

### Bibliographic citations
- Abidi, N., Herradi, M.E., Sakha, S. (2021). Digitalization and Resilience: Firm-Level Evidence During the COVID-19 Pandemic. International Monetary Fund Working Paper, 22/34.  
- Acemoglu, D., Restrepo, P. (2017). Robots and Jobs: Evidence from US Labor Markets (Working Paper No. 23285). National Bureau of Economic Research. https://doi.org/10.3386/w23285  
- Ahir, H., Bloom, N., and Furceri, D., 2022. "The World Uncertainty Index," NBER Working Papers 29763, National Bureau of Economic Research.  
- Alcedo, J., Cavallo, A., Dwyer, B., Mishra, P., Spilimbergo, A., 2022. E-commerce During Covid: Stylized Facts from 47 Economies (Working Paper No. 29729). National Bureau of Economic Research. https://doi.org/10.3386/w29729  
- Arbatli Saxegaard, E. C., Firat, M., Furceri, D., Verrier, J. (2022). U.S. Monetary Policy Shock Spillovers: Evidence from Firm-Level Data. International Monetary Fund Working Paper  
- Bellatin, A., Galassi, G., (2022). “What COVID-19 May Leave Behind: Technology-Related Job Postings in Canada”. https://doi.org/10.34989/swp-2022-17  
- Bloom, N., Liang, J., Roberts, J., Ying, Z.J. (2015). Does Working from Home Work? Evidence from a Chinese Experiment. The Quarterly Journal of Economics 130, 165–218. https://doi.org/10.1093/qje/qju032  
- Brynjolfsson, E., Hui, X., Liu, M. (2019). Does Machine Translation Affect International Trade? Evidence from a Large Digital Platform. Management Science 65, 5449–5460. https://doi.org/10.1287/mnsc.2019.3388  
- Buera, F. and Karmakar, S. (2019), “Real Effects of Financial Distress: The Role of Heterogeneity“, Bank of England Working Papers, 814.  
- Calvino, F., Criscuolo, C., Marcolin, L., Squicciarini, M., (2018). A taxonomy of digital intensive sectors. https://doi.org/10.1787/f404736a-en  
- Campello, M., Graham, J. R. and Harvey, C. R. (2010), “The Real Effects of Financial Constraints: Evidence from a Financial Crisis”, Journal of Financial Economics, 97(3), 470–487.  
- Choi, S., Furceri, D., Huang, Y., & Loungani, P. (2018). “Aggregate uncertainty and sectoral productivity growth: The role of credit constraints”. Journal of International Money and Finance, 88, 314-330.  
- Choi, S., Furceri, D., Loungani, P., and Shim, M. (2022). “Inflation anchoring and growth: The role of credit constraints”. Journal of Economic Dynamics and Control, 134, 104279.  
- Comin, D.A., Lee, K.M., Cruz, M., Torres, J., Cirera, X., 2022. Technology and Resilience.  
- Dabla-Norris, E., Kinda, T., Chahande, K., Chai, H., Chen, Y., de Stefani, A., Kido, Y., Qi, F., and Sollaci, A., (forthcoming). “Accelerating Innovation and Digitalization in Asia to Boost Productivity”. International Monetary Fund Departmental Paper  
- Dauth, W., Findeisen, S., Südekum, J., and Woessner, N., (2017). “German Robots – The Impact of Industrial Robots on Workers.” CEPR Discussion Paper No. DP12306, Available at SSRN: https://ssrn.com/abstract=3039031  
- Dell Ariccia, G., Detragiache, E., and Rajan, R., (2008). “The real effect of banking crises”. Journal of Financial Intermediation, 17(1), 89-112.  
- Duval, R., Hong, G. H., and Timmer, Y. (2020). “Financial frictions and the great productivity slowdown”. The Review of Financial Studies, 33(2), 475-503.  
- Freund, C., Weinhold, D., (2002). The Internet and International Trade in Services. American Economic Review 92, 236–240. https://doi.org/10.1257/000282802320189320  
- Freund, C., Weinhold, D., (2004). The effect of the Internet on international trade. Journal of International Economics 62, 171–189. https://doi.org/10.1016/S0022-1996(03)00059-X  
- Graetz, G., Michaels, G. (2018). Robots at Work. The Review of Economics and Statistics 100, 753–768. https://doi.org/10.1162/rest_a_00754  
- Haskel, J., Westlake, S., (2018). “Capitalism without capital: the rise of the intangible economy.” Princeton University Press, Princeton.  
- Jack, W., Suri, T., (2014). “Risk Sharing and Transactions Costs: Evidence from Kenya’s Mobile Money Revolution.” American Economic Review 104, 183–223. https://doi.org/10.1257/aer.104.1.183  
- Jensen, R., (2007). “The Digital Provide: Information (Technology), Market Performance, and Welfare in the South Indian Fisheries Sector.” The Quarterly Journal of Economics 122, 879–924. https://doi.org/10.1162/qjec.122.3.879  
- Jordà, O. (2005). “Estimation and Inference of Impulse Responses by Local Projections” American Economic Review, vol. 95(1), pages 161-182, March.  
- Kalemli-Ozcan, S., Laeven, L. and Moreno, D. (2019), “Debt Overhang, Rollover Risk, and Corporate Investment: Evidence from the European Crisis”, ECB Working Paper Series 2241, European Central Bank.  
- Kim, M., Mano, R.C. and Mrkaic, M., 2020. “Do FX interventions lead to higher FX debt? Evidence from firm-level data”. International Monetary Fund Working Paper, 20/197  
- OECD (2021a). “Digital supply-use tables: A step toward making digital transformation more visible in economic statistics.” Paris: OECD. Available at: http://goingdigital.oecd.org/data/notes/No8_ToolkitNote_DigitalSUTs.pdf  
- OECD (2021b). OECD Inter-Country Input-Output Database, http://oe.cd/icio.  
- OECD (2021c). OECD Trade in Value Added Database, https://www.oecd.org/sti/ind/measuring-trade-in-value-added.htm  
- Pierri, N., Timmer, Y., (2020). IT Shields: Technology Adoption and Economic Resilience during the COVID-19 Pandemic.  
- Suri, T., Jack, W., Stoker, T.M., (2012). “Documenting the birth of a financial economy.” Proceedings of the National Academy of Sciences 109, 10257–10262. https://doi.org/10.1073/pnas.1115843109  
- Teulings, C.N., Zubanov, N., (2014). “Is Economic Recovery a Myth? Robust Estimation of Impulse Responses.” Journal of Applied Econometrics 29, 497–514. https://doi.org/10.1002/jae.2333

### Notes on referenced empirical methods and data (as cited)
- Jordà (2005) local projection methods are cited repeatedly for impulse response estimation using firm-level quarterly data for 75 countries (periods vary by figure).  
- Firm-level samples: quarterly data from 75 countries for the period 2001Q1 to 2021Q1 (figures 3, 4, 8, 9, 11) and 2016Q1 to 2021Q1 (figures 5, 6, 10) depending on the analysis.  
- Industry-level samples: quarterly data from 40 countries for the period 2016Q1 to 2022Q1 (figure 7).  
- Standard errors: clustered by firm and country-time for firm-level regressions; clustered at the country level for industry-level regressions (figure 7).  
- Digitalization measures referenced: Calvino et al. (2018) industry-wise digitalization (DrC), ICIO (direct digital inputs), TiVA (total value added from digital inputs), Intangibles share, LinkedIn Digital Skills.  
- Recession and crisis indicators referenced: start of a technical recession (general), start of a technical recession from 2019Q4 onwards (COVID-19 recession), start of banking and/or currency crisis (robustness checks), start of currency crises (robustness checks).

### TABLES AND FIGURES

### Table
- Table 1. Correlation between digitalization measures  
  - Columns/rows: Calvino, ICIO, TIVA, Intangibles, LinkedIn Digital Skills  
  - Correlations (pairwise correlation coefficients; collapsed to industry level by median):  
    - ICIO with Calvino: 0.4*  
    - TIVA with Calvino: 0.3*; TIVA with ICIO: 1*  
    - Intangibles with Calvino: 0.4*; Intangibles with ICIO: 0.3*; Intangibles with TIVA: 0.4*  
    - Digital Skills with Calvino: 0.5*; Digital Skills with ICIO: 0.5*; Digital Skills with TIVA: 0.4*; Digital Skills with Intangibles: 0.5*  
  - Note: * Significant at 5% level.

### Figures (titles and key notes)
- Figure 1: Scarring and Internet Access (Percent)  
  - Source: World Economic Outlook and World Bank Development Indicators.  
  - Note: Output losses are computed as the percent deviation between pre-pandemic (January 2020) and latest GDP projections for 2024.

- Figure 2: Job Posts by Income Group (30-day MA, indexed to January + February 2020)  
  - Source: Indeed.com and authors’ calculations. Sample contains daily job counts of total job posts for 35 countries.

- Figure 3: Unconditional effect of recessions on changes in (log) revenue  
  - Note: Impulse response function based on local projection methods following Jordà (2005) using firm-level quarterly data from 75 countries for the period 2001Q1 to 2021Q1. Regression specification and fixed effects described in text. Dashed and dotted lines are the 68 percent and 90 percent confidence intervals respectively. Standard errors are clustered by firm and country-time.

- Figure 4: Differential Effect of Recession on log Revenue for Highly Digitalized Industries Vs. Average  
  - Note: Uses Calvino et al. (2018) measure of industry-wise digitalization (DrC). Same estimation approach and sample as figure 3. Confidence intervals: 68 percent and 90 percent. Standard errors clustered by firm and country-time.

- Figure 5: Unconditional effect of COVID-19 recessions on log revenue  
  - Note: Local projections using firm-level quarterly data from 75 countries for the period 2016Q1 to 2021Q1. Recession dummy equals 1 at the start of a technical recession from 2019Q4 onwards. Estimated over a five-quarter period. Confidence intervals: 68 percent and 90 percent. Standard errors clustered by firm and country-time.

- Figure 6: Differential Effect of COVID-19 Recession on log Revenue for Highly Digitalized Industries Vs. Average  
  - Note: Uses standardized Calvino et al. (2018) industry digitalization measure (DrC). Sample and estimation period: 2016Q1 to 2021Q1. Confidence intervals: 68 percent and 90 percent. Standard errors clustered by firm and country-time.

- Figure 7: Differential Effect of COVID-19 Recession on log Hiring Rates and Workers Transition for Highly Digitalized Industries Vs. Average  
  - Panels: A. Log Hiring Rate; B. Log Transitions into Industry; C. Log Transitions out of Industry.  
  - Note: Impulse responses based on local projections using industry-level quarterly data from 40 countries for the period 2016Q1 to 2022Q1. Recession dummy equals 1 at the start of a technical recession from 2019Q4 onwards. Standard errors clustered at the country level. Confidence intervals: 68 percent and 90 percent.

- Figure 8: Robustness: Country-Industry Alternative Digitalization Measures  
  - Panels: A. ICIO (Direct digital inputs); B. TiVA (Total value added from digital inputs); C. Intangibles share.  
  - Note: Firm-level quarterly data from 75 countries for the period 2001Q1 to 2021Q1. Alternative digitalization measures Dr,i ∈ {Dr,iICIO, Dr,iTIVA, Dr,iInt} used. Confidence intervals: 68 percent and 90 percent. Standard errors clustered by firm and country-time.

- Figure 9: Robustness: Time-Varying Alternative Digitalization Measures  
  - Panels: A. ICIO (Contemporaneous and One-year Lagged); B. TiVA (Contemporaneous and One-year Lagged).  
  - Note: Firm-level quarterly data from 75 countries for the period 2001Q1 to 2021Q1. Digitalization measures specified as time-varying Dr,i,t ∈ {Dr,i,tICIO, Dr,i,tTIVA, Dr,i,t−1ICIO, Dr,i,t−1TIVA}. Confidence intervals: 68 percent and 90 percent. Standard errors clustered by firm and country-time.

- Figure 10: COVID-19 robustness: Country-Industry Alternative Digitalization Measures  
  - Panels: A. ICIO (Direct digital inputs); B. TiVA (Total value added from digital inputs); C. Intangibles share; D. LinkedIn Digital Skills.  
  - Note: Firm-level quarterly data from 75 countries for the period 2016Q1 to 2021Q1. Recession dummy equals 1 at the start of a technical recession from 2019Q4 onwards. Alternative Dr,m ∈ {Dr,iICIO, Dr,iTIVA, Dr,iInt, Dr,iTechSkills}. Confidence intervals: 68 percent and 90 percent. Standard errors clustered by firm and country-time.

- Figure 11: Robustness: Alternative Recession Measures  
  - Panel A: Start of Banking and Currency Crises; Panel B: Start of Currency Crises.  
  - Note: Firm-level quarterly data from 75 countries for the period 2001Q1 to 2021Q1. Recession dummy equals 1 at the start of a banking and/or currency crisis (panel A) or at the start of a currency crisis (panel B). Confidence intervals: 68 percent and 90 percent. Standard errors clustered by firm and country-time.

*International Monetary Fund — REFERENCES, TABLES AND FIGURES as provided in the source content.*

### ANNEX 1:DATA

### ANNEX 1: DATA

### Sample composition by region and country
- Africa – AFR (3): Botswana, Mauritius, South Africa  
- Middle East and Central Asia – MCD (11): Bahrain, Egypt, Jordan, Kazakhstan, Kuwait, Oman, Pakistan, Qatar, Saudi Arabia, Tunisia, United Arab Emirates  
- Western Hemisphere – WHD (10): Argentina, Brazil, Canada, Chile, Colombia, Jamaica, Mexico, Peru, Trinidad & Tobago, United States  
- Asia & Pacific – APD (17): Australia, Bangladesh, China, Hong Kong, India, Indonesia, Japan, Macau, Malaysia, New Zealand, Philippines, Singapore, South Korea, Sri Lanka, Taiwan, Israel, Thailand, Vietnam, Latvia  
- Europe – EUR (34): Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Norway, Poland, Portugal, Romania, Russia, Serbia, Slovakia, Spain, Sweden, Switzerland, Turkey, Ukraine, United Kingdom

### Table A1.2 — Number of firms and observations by country (selected entries)
- United States: Number of firms 4,740; Obs. 388,680  
- China: Number of firms 4,077; Obs. 334,314  
- Japan: Number of firms 3,085; Obs. 252,970  
- India: Number of firms 2,672; Obs. 219,104  
- Canada: Number of firms 2,213; Obs. 181,466  
- South Korea: Number of firms 1,747; Obs. 143,254  
- Taiwan: Number of firms 1,693; Obs. 138,826  
- Australia: Number of firms 1,356; Obs. 111,192  
- Hong Kong: Number of firms 1,106; Obs. 90,692  
- United Kingdom: Number of firms 870; Obs. 71,340  
- Malaysia: Number of firms 771; Obs. 63,222  
- Thailand: Number of firms 555; Obs. 45,510  
- Sweden: Number of firms 525; Obs. 43,050  
- Poland: Number of firms 522; Obs. 42,804  
- Singapore: Number of firms 471; Obs. 38,622  
- France: Number of firms 467; Obs. 38,294  
- Germany: Number of firms 450; Obs. 36,900  
- Vietnam: Number of firms 412; Obs. 33,784  
- Indonesia: Number of firms 399; Obs. 32,718  
- Israel: Number of firms 322; Obs. 26,404  
- Pakistan: Number of firms 321; Obs. 26,322  
- Turkey: Number of firms 280; Obs. 22,960  
- Brazil: Number of firms 246; Obs. 20,172  
- Italy: Number of firms 220; Obs. 18,040  
- Sri Lanka: Number of firms 183; Obs. 15,006  
- Bangladesh: Number of firms 178; Obs. 14,596  
- South Africa: Number of firms 178; Obs. 14,596  
- Russia: Number of firms 177; Obs. 14,514  
- Switzerland: Number of firms 168; Obs. 13,776  
- Philippines: Number of firms 157; Obs. 12,874  
- Greece: Number of firms 155; Obs. 12,710  
- Egypt: Number of firms 134; Obs. 10,988  
- Norway: Number of firms 129; Obs. 10,578  
- Chile: Number of firms 128; Obs. 10,496  
- Spain: Number of firms 119; Obs. 9,758  
- Finland: Number of firms 117; Obs. 9,594  
- Saudi Arabia: Number of firms 114; Obs. 9,348

(continued)
- Netherlands: Number of firms 105; Obs. 8,610  
- New Zealand: Number of firms 105; Obs. 8,610  
- Mexico: Number of firms 98; Obs. 8,036  
- Peru: Number of firms 87; Obs. 7,134  
- Jordan: Number of firms 83; Obs. 6,806  
- Belgium: Number of firms 75; Obs. 6,150  
- Ireland: Number of firms 71; Obs. 5,822  
- Oman: Number of firms 71; Obs. 5,822  
- Argentina: Number of firms 65; Obs. 5,330  
- Romania: Number of firms 63; Obs. 5,166  
- Kuwait: Number of firms 61; Obs. 5,002  
- Croatia: Number of firms 57; Obs. 4,674  
- Bulgaria: Number of firms 54; Obs. 4,428  
- Colombia: Number of firms 49; Obs. 4,018  
- Austria: Number of firms 45; Obs. 3,690  
- Cyprus: Number of firms 45; Obs. 3,690  
- United Arab Emirates: Number of firms 45; Obs. 3,690  
- Mauritius: Number of firms 44; Obs. 3,608  
- Luxembourg: Number of firms 39; Obs. 3,198  
- Jamaica: Number of firms 37; Obs. 3,034  
- Portugal: Number of firms 36; Obs. 2,952  
- Tunisia: Number of firms 27; Obs. 2,214  
- Lithuania: Number of firms 23; Obs. 1,886  
- Qatar: Number of firms 21; Obs. 1,722  
- Malta: Number of firms 20; Obs. 1,640  
- Hungary: Number of firms 18; Obs. 1,476  
- Bahrain: Number of firms 17; Obs. 1,394  
- Kazakhstan: Number of firms 15; Obs. 1,230  
- Estonia: Number of firms 14; Obs. 1,148  
- Iceland: Number of firms 14; Obs. 1,148  
- Latvia: Number of firms 14; Obs. 1,148  
- Trinidad & Tobago: Number of firms 14; Obs. 1,148  
- Serbia: Number of firms 12; Obs. 984  
- Ukraine: Number of firms 11; Obs. 902  
- Macau: Number of firms 10; Obs. 820  
- Botswana: Number of firms 7; Obs. 574  
- Czech Republic: Number of firms 6; Obs. 492  
- Slovakia: Number of firms 6; Obs. 492

### Table A1.3 — Number of firms and observations by sector (top entries)
- Materials: Number of Firms 5,433; Obs. 445,506  
- Capital Goods: Number of Firms 4,888; Obs. 400,816  
- Technology Hardware and Equipment: Number of Firms 2,286; Obs. 187,452  
- Consumer Durables and Apparel: Number of Firms 2,032; Obs. 166,624  
- Software and Services: Number of Firms 2,027; Obs. 166,214  
- Pharmaceuticals and Biotechnology: Number of Firms 1,833; Obs. 150,306  
- Food, Beverage and Tobacco: Number of Firms 1,800; Obs. 147,600  
- Energy: Number of Firms 1,714; Obs. 140,548  
- Media and Entertainment: Number of Firms 1,398; Obs. 114,636  
- Consumer Services: Number of Firms 1,315; Obs. 107,830  
- Retailing: Number of Firms 1,291; Obs. 105,862  
- Health Care Equipment and Services: Number of Firms 1,287; Obs. 105,534  
- Professional Services: Number of Firms 1,160; Obs. 95,120  
- Transportation: Number of Firms 933; Obs. 76,506  
- Automobiles and Components: Number of Firms 865; Obs. 70,930  
- Utilities: Number of Firms 854; Obs. 70,028  
- Semiconductors: Number of Firms 774; Obs. 63,468  
- Telecommunication Services: Number of Firms 407; Obs. 33,374  
- Food and Staples Retailing: Number of Firms 383; Obs. 31,406  
- Household and Personal Products: Number of Firms 361; Obs. 29,602

### Table A1.4 — Summary statistics by revenue
- Revenue: Count 1,692,161; Mean 5.49; Std 3.58; 25th 3.39; 75th 7.8  
- Revenue – High Digital (above 75th percentile): Count 264,067; Mean 5.26; Std 3.78; 25th 2.84; 75th 7.81

### Table A1.5 — Recession variables summary statistics
- Start of Technical Recession (Source: Haver Analytics and World Economic Outlook): Countries 106; Coverage 1960Q2-2022Q4; Obs. 12,011; Mean 0.05; Std 0.21; Min 0; Max 1  
- Currency crises (converted to quarterly) (Source: Reinhart and Rogoff (2009)): Countries 63; Coverage 1960q2-2014q4; Obs. 7,230; Mean 0.06; Std 0.20; Min 0; Max 1  
- Banking and currency crises (converted to quarterly) (Source: Reinhart and Rogoff (2009)): Countries 63; Coverage 1960q2-2014q4; Obs. 7,230; Mean 0.08; Std 0.27; Min 0; Max 1

### Table A1.6 — Descriptive statistics of digitalization variables
- Calvino et al. (2018) (Source: Calvino et al. (2018)): Countries 76; Obs. 3,304,900; Mean .60; Std .26; Min .25; Max 1  
- ICIO (OECD Inter-Country Input-Output Tables (OECD 2021b)): Countries 56; Obs. 2,861,280; Mean .11; Std .20; Min 0; Max .86  
- TiVA (OECD Trade in Value Added database (OECD 2021c)): Countries 58; Obs. 2,016,718; Mean .083; Std .15; Min 0; Max .76  
- Intangibles (Source: Capital IQ): Countries 76; Obs. 1,301,928; Mean .11; Std .17; Min 0; Max .76  
- Tech Skills (Source: LinkedIn): Countries 38; Obs. 2,254,000; Mean .099; Std .04; Min 0; Max .19

### Table A1.7 — Rank of most digital sectors by digitalization measure (top and bottom examples)
- Calvino (most digital): 1 Wireless Telecommunication Services; 2 Industrial Conglomerates; N – 2 Building Products; N – 1 Water Utilities; N = 60 Road and Rail  
- ICIO (most digital): 1 Semiconductors and Semiconductor Equipment; 2 Technology Hardware, Storage and Peripherals; Textiles, Apparel and Luxury Goods; Food Products; Oil, Gas and Consumable Fuels  
- TiVA (most digital): 1 IT Services; 2 Interactive Media and Services; Road and Rail; Diversified Consumer Services; Oil, Gas and Consumable Fuels  
- Intangibles (most digital): 1 Health Care Technology; 2 Wireless Telecommunication Services; Textiles, Apparel and Luxury Goods; Construction and Engineering; Marine  
- LinkedIn Digital Skills (most digital): 1 Communications Equipment; 2 Software; Electronic Equipment, Instruments and Components; Water Utilities; Paper and Forest Products

### Table A1.8 — Variation of digitalization measures on country-sector-time fixed effects
- Calvino (industry): R-squared from regression on Country-Sector-Time FE 0.024; Observations 2,710,018  
- Intangibles (country-industry): R-squared 0.044; Observations 2,699,932  
- TiVA (country-industry): R-squared 0.070; Observations 2,263,446  
- ICIO (country-industry): R-squared 0.021; Observations 2,606,944  
- Note: R2 obtained from regressing the digitalization measures on baseline country-sector-time fixed effects; substantial variation remains, implying substantial cross-industry within-sector variation in digitalization (note: 60 industries nested within 20 sectors).

*International Monetary Fund. Working Paper No. WP/22/210 — ANNEX 1: DATA*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022210-print-pdf.pdf_