## wpiea2023162-print-pdf

## Source details

**Canonical URL:** [wpiea2023162-print-pdf](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023162-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023162-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023162-print-pdf.pdf.json)

---

### 1. INTRODUCTION — pandemic context, methods, and high-level findings
- Pandemic context and initial expectations:
  - Global case rates of Covid-19 have fallen substantially since early 2022.
  - Likelihood of severe symptoms or death has also fallen substantially, but as of mid-2023 the global economy is still in a process of economic recovery from the pandemic.
  - Lockdowns in 2020 led to a sudden contraction in output much larger than that seen in past recessions.
  - Contact-intensive services experienced particularly sharp contractions.
  - January 2022 WEO projections:
    - Level of output in EMs by 2024 would be around 4% below that projected before the pandemic.
    - Level of output in AEs was closer to pre-pandemic trends.
  - Weaker medium-term outlook for EMs reflected lower levels of policy support, additional disruption to education, and reduced access to vaccines.

- Research focus and methods:
  - Top-down analysis across the universe of Emerging Markets with available data; focus on aggregate trends.
  - Examination of both Consensus forecasts and forecasts from the IMF’s WEO.
  - Two complementary approaches:
    1. Document evolving impact of Covid on EM activity and assess responsiveness of scarring estimates using a simple Bayesian framework.
    2. Decompose output losses into employment, capital, and Total Factor Productivity (TFP) and compare with past recessions.

- Key high-level findings:
  - Recovery and forecast responsiveness:
    - Covid had a material and persistent impact on activity, but the recovery has proved stronger and faster than expected.
    - Bayesian estimates of scarring were more sensitive to downside data news early in the pandemic, but less responsive to the string of upside data since.
    - Economic forecasts have taken on too little positive signal from the faster-than-expected recovery; positive data surprises have been treated as transitory rather than as evidence that scarring may be smaller than initially feared.
  - Composition of output losses:
    - Scarring from past shocks was driven mainly by persistently weak productivity, particularly weak TFP.
    - During Covid, a larger than usual portion of output losses has been accounted for by lower employment; the impact on TFP has been smaller.
    - By 2022, around two thirds of the shock to employment had unwound compared to only around a tenth three years after a financial crisis.
  - Medium-term scarring implications:
    - Had positive data surprises been treated as a signal about scarring, projected scarring would be ½pp to 2pp lower than WEO or Consensus estimates.
    - Central case projection: Covid-related scarring to reduce to around -2 to -2½% by 2025.
    - January 2022 WEO / Consensus pre-war projections implied output losses building to -3½ to 4%.

- Caveats and limitations:
  - Aggregated, top-down analysis focused on EMs; results may not translate to LICs.
  - Medium term defined as five years after the start of the pandemic; longer-term impacts (e.g., education disruption) may be underplayed.
  - Analysis concentrates exclusively on Covid and does not isolate other factors; Russia’s war in Ukraine complicates interpretation.

### 2. Evolution of post-Covid GDP losses and scarring estimates — measurement, Bayesian framework, and empirical results
- Measurement approach:
  - Output losses measured as difference between realized data and pre-pandemic January 2020 WEO projections for EMs.
  - Scarring estimated as deviation in the projected level of 2024 GDP relative to pre-pandemic forecasts in January 2020.

- Key aggregate milestones and historical comparisons:
  - Peak negative impact on GDP in EMs reached almost 11% relative to pre-pandemic projections in 2020 Q2.
  - By 2022 Q1, level of GDP in EMs was around 3% below the pre-pandemic forecast; AEs were just under 1½% below trend in 2022 Q1.
  - July 2020 WEO projected level of GDP in EMs would be almost 5½ below trend in 2021.
  - Initial estimate of scarring of 5% in the July WEO was revised to below 4% by Spring 2021, then became more stable and slightly increased.
  - Since January 2022, GDP loss estimates relative to pre-pandemic trend have been compounded by spillovers from Russia’s war in Ukraine.

- Bayesian model for forecast revisions (framework highlights):
  - Data-generating process: quarterly (log) GDP y_t = τ_t + c_t − s, with τ_t deterministic trend, c_t zero-mean transitory cyclical component, s unknown permanent scarring percentage.
  - Cyclical component AR(1): c_t = ρ c_{t−1} + ε_t with ε_t ∼ N(0, σ_ε^2).
  - Priors for scarring normal with mean ŝ and variance σ_0^2.
  - One-quarter-ahead GDP forecast error decomposes into: −(1−ρ)(s−ŝ) + ε.
  - Posterior update for scarring: ŝ_{t+1} − ŝ = − [σ_0^2 / (σ_ε^2 + σ_0^2)] (1−ρ) (y_t − ŷ_{t|t−1}).
  - Sensitivity of scarring revisions to quarterly GDP surprises corresponds to the Kalman gain (proportion of a data surprise perceived as permanent).

- Empirical application to WEO and Consensus forecasts (sample and periods):
  - Consensus forecasts cover 17 EMs (Brazil, Mexico, Chile, Colombia, Peru, India, Indonesia, Malaysia, Philippines, Thailand, Turkey, Poland, Russia, Hungary, China, Argentina, Bulgaria).
  - Periods: Initial Covid (April – October 2020 WEOs); Pre-war recovery (January 2021 – January 2022 WEOs); Post-war recovery (subsequent WEOs).
  - Aggregation reported as median across EMs; PPP-weighting provided for reference.

- Key empirical sensitivities and cumulative surprises:
  - Initial Covid period:
    - Ratio between quarterly data surprise and scarring estimate revisions ~ 1/2 for all EMs in sample.
    - Median sensitivity for WEO: 0.38.
    - Average sensitivity for Consensus: 0.48.
    - Cumulative median data surprise: -13.8% of EM GDP.
    - Median WEO scarring revision: -5.2%.
    - Median Consensus scarring revision: -6.7%.
    - PPP-weighted sensitivity: 0.48 for both Consensus and WEO; cumulative PPP-weighted data surprise = -10.9%; median revision to Consensus and WEO scarring = -5.2%.
  - Pre-war recovery period:
    - Quarterly GDP consistently surprised to the upside; scarring estimates revised up but less strongly than initial downward revisions.
    - Median sensitivity in pre-war recovery: 0.18 for WEO and 0.34 for Consensus.
    - PPP-weighted sensitivities in pre-war recovery: 0.29 for WEO and 0.33 for Consensus.
  - Interpretation:
    - High initial sensitivity reflected extreme uncertainty about trend.
    - Reduced sensitivity in recovery indicates forecasters placed less weight on data surprises as signals about permanent scarring over time.

- Counterfactual using initial-period sensitivity (~1/2):
  - By end of pre-war recovery period, predicted level of scarring ≈ 3½% (PPP-weighted), about 0.5 percentage point less than PPP-weighted WEO or Consensus forecasts.
  - Median predicted level of scarring: -2.75% to -3.5%, around 1½–2pp less than median WEO and Consensus forecasts.
  - By 2021 Q4, PPP-weighted deviation in GDP from trend was 2.6% and for the median economy was 2.9%.

- Decomposition data and methodology (production function and inputs):
  - Sample for decomposition: 21 major EMs (accounting by weight for 35% of world GDP and 70% of the EMs group) including Turkiye, South Africa, Brazil, Chile, Colombia, Mexico, Peru, Indonesia, Malaysia, Philippines, Thailand, Beleraus, Kazakhstan, Russia, China, Serbia, Hungary, Croatia, Poland, Romania.
  - Cobb-Douglas: Y_t = A_t K_t^(1−α) L_t^α with α from PWT 10.0 averaging 0.55 in sample.
  - Labor input decomposed into population (Pop), unemployment rate (UE), labor force participation rate (LFPR).
  - Capital stock via perpetual inventory: K_t = K_{t−1} (1−δ_t) + I_t, with δ_t from PWT and investment from WEO.
  - Baseline adjusts for capital utilization falling in line with employment in 2020 to avoid overstating TFP declines.

- Decomposition results (realized and projected losses):
  - 2020:
    - Just under a third of PPP-weighted average EM GDP loss in 2020 was accounted for by lower employment (driven by both higher unemployment and lower LFPR).
    - Initial contribution from capital small in stock terms; effective capital fell sharply during lockdowns due to lower utilization.
    - TFP accounts for about two fifths of the remaining output losses; decomposition using average hours suggests much of persistent shortfall in TFP reflects lower hours worked rather than lower productivity per hour.
  - 2022 Q1:
    - Contribution from lower employment remained around a third.
    - Contribution of capital due to lower investment built over time to a fifth of output losses in 2022 Q1.
    - TFP accounts for about 40% of the remaining losses.
  - Comparison with past large shocks:
    - Past shocks five years after event: persistent losses largely driven by weak productivity/TFP, with modest employment losses.
    - Covid’s persistent losses are more evenly distributed across labor, capital, and TFP; persistent negative impact on TFP from Covid has been much smaller than in past episodes.
  - January 2022 WEO projection:
    - Projected scarring by 2024 = -4% relative to pre-pandemic trend, driven by weaker capital accumulation and a large decline in TFP.

### 4. IMPLICATIONS FOR THE MEDIUM TERM — outlooks for TFP, capital, employment, and summary scarring
- Outlook for TFP:
  - Past episodes: shocks to TFP in EMs are persistent but do not tend to worsen over time.
  - Covid shock: exogenous and short-lived (suggesting lower persistence) but introduced unique channels to lower TFP and involved public and private debt build-up.
  - Bottom-up TFP effects:
    - Spillovers from weak employment to TFP: using Indonesia estimate that a year spent in unemployment reduced subsequent earnings by 3.5% to calculate TFP loss from reduced employment.
    - As unemployment recovered partially quickly, projected medium-term impact on TFP from elevated unemployment is only -0.1% on average.
    - Including time out of the labor force raises medium-term impact on labor via wages and productivity to -0.2%.
    - Education disruption:
      - UNESCO data: weighted average total loss of 90% of a school year in the sample, or 8.5 months of schooling.
      - Peak proportion of the workforce affected by education disruption close to 40% reached by 2031; by 2025 around half of the affected students have entered the workforce.
      - Psacharopoulos and Patrinos (2018): one year’s lost schooling reduces wages by 9%; mapping wages 1:1 to GDP.
      - Age-earnings adjustment: entrants earn three quarters of the average worker (based on Indonesian data).
    - Bottom-up schooling estimate quantitative outcomes:
      - Peak impact on GDP of -2½% by the 2030s.
      - Impact by 2025 builds to -1%, under half of the long-run effect of 2½%.
    - Combined productivity effects (time out of work + lost education):
      - For 2022, estimated effect is just under –½%.
      - Builds to -1% by 2025.
    - Averaging over various estimates points to a reduction in TFP relative to trend of just under 1% in 2025.

- Outlook for capital:
  - Three channels:
    1. Lost capital so far from reduction in investment relative to trend.
    2. Lower TFP reduces marginal return on capital and optimal capital stock; reduction in optimal capital similar to lost capital so far.
    3. Increased corporate leverage associated with reductions in investment in past crises.
  - IMF (2022) estimate: a 1 percentage point increase in three-year average non-financial corporate credit-to-GDP ratio has a peak impact on level of investment of -2%.
  - BIS data up to 2022 Q3: estimated a 1pp increase in corporate credit-to-GDP ratio relative to pre-Covid average; implies:
    - Peak impact of -2% on investment per year.
    - Cumulative impact of 4.5% after four years.
  - Combined effects produce:
    - Reduction in level of capital of -1.4%, which reduces potential GDP by 0.6%.
    - Around one quarter of capital shortfall comes from higher corporate leverage; remainder from weak early-pandemic investment.
  - Financial distress indicators:
    - Ratio of provisions to nominal GDP has risen slightly by 0.1pp since the start of Covid.
    - Ratio of non-performing loans has fallen.
    - Patterns suggest capital shortfall estimates are unlikely to be materially biased by scrapped/redundant capital from bankrupt businesses as of latest data.

- Outlook for employment:
  - Employment recovery quicker than past shocks but incomplete:
    - Unemployment rate: after sharp rise in 2020, unemployment recovered to pre-Covid levels by end-2022.
    - Level of employment: remained around 1% below its pre-Covid trend in 2022, mostly driven by a drop in labor force participation.
  - Factors behind employment shortfall: shifts from formal to informal sectors, less generous job retention schemes, disrupted internal migration patterns.
  - Compared to past crises:
    - After financial crises, peak shortfall in employment typically recovered by about one third after five years; after other recessions, by about one tenth.
    - By 2022, employment had recovered about two thirds of the peak shortfall from Covid.
  - Projection assumptions:
    - Bottom-up estimates assume no change in trend unemployment given current rates are mostly back to pre-Covid levels.
    - Lower labor participation rates projected to drag employment by 1% and output by 0.5%.

- Summing up — medium-term scarring (key numeric summary):
  - Bottom-up estimates suggest scarring related to Covid may reach 2-2½% in the next few years.
  - This scarring is significant if sustained but smaller than average financial crises or other recessions and somewhat more optimistic than other forecasts.
  - Bottom-up estimate implies only a small improvement relative to level of GDP in 2022 Q1: a 0.3pp improvement from the -2.3% deviation in GDP from trend at start of 2022.
  - Key numeric summary (values from reproduced Table 1):
    - GDP: WEO -3.4; Consensus -3.1; Bottom-up estimate -2.0; Memo: 22 Q1 GDP vs trend -2.3
    - o/w TFP: WEO -1.5; Consensus -0.8; Bottom-up estimate -1.0
    - o/w capital: WEO -0.9; Consensus -0.6; Bottom-up estimate -0.5
    - o/w labor: WEO -1.0; Consensus -0.5; Bottom-up estimate -0.8

- Uncertainties and caveats:
  - Positive recent data surprises point to a more robust recovery, but supply constraints and rising inflation may limit sustainability.
  - Timing of education-related losses affecting productivity is particularly uncertain because effects accrue as cohorts enter the workforce.
  - Recovery assessment complicated by compounding shocks from Russia’s war in Ukraine and by data uncertainty and future revisions.

*Source: wpiea2023162-print-pdf*

### 1.  INTRODUCTION ___________________________________________________________________________ 5

### 1. INTRODUCTION

### Pandemic context and initial expectations
- Global case rates of Covid-19 have fallen substantially since early 2022 as many countries around the world have successfully vaccinated their populations against the virus.
- The likelihood of severe symptoms or death has also fallen substantially, but as of mid-2023 the global economy is still in a process of economic recovery from the pandemic.
- The economic shock of the pandemic was unprecedented in many respects:
  - Lockdowns in 2020 led to a sudden contraction in output much larger than that seen in past recessions.
  - The impact was uneven across sectors and countries; contact-intensive services experienced particularly sharp contractions.
  - The pandemic significantly affected patterns of work and mobility, producing enduring and complex effects on supply chains and labor markets.
- At its onset, the pandemic was widely expected to have a persistent negative impact on output, particularly in Emerging Markets (EMs):
  - The January 2022 WEO projected that the level of output in EMs by 2024 would be around 4% below that projected before the pandemic.
  - The January 2022 WEO projected that the level of output in Aes was closer to pre-pandemic trends.
  - The weaker medium-term outlook for EMs relative to Aes reflected lower levels of policy support, additional disruption to education, and reduced access to vaccines.

### Research focus and methods
- Scope and perspective:
  - Top-down analysis across the universe of Emerging Markets with available data.
  - Focus on aggregate trends rather than individual country forecasts.
  - Examination of both Consensus forecasts and forecasts from the IMF’s WEO.
- Two complementary approaches:
  1. Document how the impact of Covid on activity in EMs has evolved relative to expectations and assess responsiveness of scarring estimates to data using a simple Bayesian framework.
  2. Explore the composition of output losses from Covid so far and compare with past recessions (focus on employment versus Total Factor Productivity (TFP)).

### Key findings (high level)
- Recovery and forecast responsiveness:
  - Covid had a material and persistent impact on activity, but the recovery has proved stronger and faster than expected.
  - Using a Bayesian framework, estimates of scarring were more sensitive to downside data news early in the pandemic, but less responsive to the string of upside data since.
  - Economic forecasts have taken on too little positive signal from the faster-than-expected recovery; positive data surprises have been treated as transitory rather than as evidence that scarring may be smaller than initially feared.
- Composition of output losses:
  - Scarring from past shocks was driven mainly by persistently weak productivity, particularly weak TFP.
  - During Covid, a larger than usual portion of output losses has been accounted for by lower employment; the impact on TFP has been smaller.
  - The latest shortfall in employment reflects weak labor participation rates rather than elevated unemployment.
  - The larger contribution of weak employment reflects Covid’s unique effects on working patterns and labor market flows.
  - By 2022, around two thirds of the shock to employment had unwound compared to only around a tenth three years after a financial crisis.
- Implications for medium-term scarring and projections:
  - Had positive data surprises been treated as a signal about scarring, projected scarring would be ½pp to 2pp lower than WEO or Consensus estimates.
  - Bottom-up estimates that consider a range of persistence for productivity and labor-market shocks also indicate potential for Covid-related output losses to improve.
  - Central case projection: Covid-related scarring to reduce to around -2 to -2½% by 2025.
  - This central case is more optimistic than pre-war projections in the January 2022 WEO or Consensus, in which output losses build to -3½ to 4%.
- Confounding factors:
  - Russia’s war in Ukraine complicates interpretation: war-related spillovers likely weighed on recent activity and may imply additional scarring effects.
  - Parsing the effects of Covid versus the war is difficult; to the extent that pre-war forecasts were based on a more pessimistic outlook for Covid scarring, actual growth may not slow as much as projected.

### Important caveats and limitations
- Aggregation and scope:
  - Analysis is aggregated and top-down across Emerging Markets to highlight broad trends rather than idiosyncratic country channels.
  - Results for EMs may not translate to Lower Income Countries (LICs); LICs started from more vulnerable positions and could afford lower levels of policy support.
- Time horizon:
  - The focus is on the medium term, defined here as five years after the start of the pandemic; this horizon may underplay longer-term impacts such as disruption to education.
- Focus on Covid only:
  - The analysis concentrates exclusively on the impact of Covid and does not consider other important factors that may affect trend growth rates over the medium term.
- Additional note:
  - Growth across EMs exhibits substantial co-movement; a principal component of EM GDP growth rates explains half of the total variation in EM growth rates (results available on request).

*Source: wpiea2023162-print-pdf — 1.  INTRODUCTION*

### 2. The Evolution of Post-Covid GDP Losses and

### 2. The Evolution of Post-Covid GDP Losses and Scarring Estimates

### Overview and measurement approach
- Output losses from Covid are measured as the difference between realized data and the pre-pandemic January 2020 WEO projections for EMs.
- The sample uses the WEO definition of Emerging Markets, excluding Developing Economies, with a smaller sample than the full WEO EM set due to data availability.
- Scarring is estimated as the deviation in the projected level of 2024 GDP relative to pre-pandemic forecasts in January 2020.

### Evolution of losses and forecast revisions
- The negative impact on GDP in EMs reached a peak of almost 11% relative to the pre-pandemic projections in 2020 Q2.
- Before Russia’s war in Ukraine and additional Covid outbreaks in China, the level of GDP in EMs was around 3% below the pre-pandemic forecast in 2022 Q1.
- Recovery in EMs lagged AEs where GDP was just under 1½% below trend in 2022 Q1.
- The July 2020 WEO projected that the level of GDP in EMs would be almost 5½ below trend in 2021.
- By 2022 Q1, the level of GDP was around 3% below trend, more than 2pp stronger than where it was expected to be in the initial phase of the pandemic.
- The initial estimate of scarring of 5% in the July WEO was successively revised to below 4% by Spring 2021, before becoming more stable and even increasing slightly.
- Since January 2022, estimates of GDP losses relative to the pre-pandemic trend have been compounded by the economic spillovers of Russia’s war in Ukraine.

### Bayesian model for forecast revisions (framework and implications)
- Assumed data-generating process: quarterly (log) GDP 푦௧ = 휏௧ + 푐௧ − 푠, where 휏௧ is deterministic trend, 푐௧ is zero-mean transitory cyclical component, and 푠 is the unknown permanent scarring percentage.
- Cyclical component follows AR(1): 푐௧ = 휌푐௧ିଵ + 휀௧ with 휌 < 1 and white noise 휀௧ ∼ N(0, 휎ଶ).
- Priors for scarring follow a normal distribution with mean 푠̂ and variance 휎௦ଶ.
- One-quarter-ahead GDP forecast error decomposes into: −(1−휌)(푠−푠̂) + 휀, so forecast error combines forecast error for scarring and cyclical innovation.
- Bayesian update for scarring estimate (posterior change) is:
  - 푠̂ᇱ − 푠̂ = − [휎௦ଶ / (휎ଶ + 휎௦ଶ)] (1−휌) (푦௧ାଵ − 푦ො௧ାଵ|௧).
- The sensitivity of scarring revisions to quarterly GDP surprises corresponds to the Kalman gain — the proportion of a data surprise perceived as permanent.
- Stylized example: with underlying trend growth 1% per period and a period-5 contraction of 30% perceived 50% permanent/50% cyclical and cyclical persistence 0.8, a period-10 positive 15% surprise revises forecast paths; sensitivity determines how much of the surprise shifts the long-run level.

### Application to WEO and Consensus professional forecasts (empirical findings)
- Scarring measured as deviation in projected level of 2024 GDP relative to January 2020 pre-pandemic forecast.
- Data news measured by difference between Consensus professional forecasts and quarterly GDP growth in the initial release. Consensus forecasts typically average projections from 18-24 forecasters and cover 17 EMs.
- Sample for revisions exercise: 17 EMs (Brazil, Mexico, Chile, Colombia, Peru, India, Indonesia, Malaysia, Philippines, Thailand, Turkey, Poland, Russia, Hungary, China, Argentina, Bulgaria).
- Periods analyzed:
  - Initial Covid period: April – October 2020 WEOs.
  - Pre-war Covid recovery: January 2021 – January 2022 WEOs.
  - Post-war Covid recovery: subsequent WEOs.
- Aggregation: median across EMs by default; PPP-weighting reported for reference.

Key empirical sensitivities and cumulative surprises:
- Initial Covid period:
  - Ratio between quarterly data surprise and scarring estimate revisions ~ 1/2 for all EMs in sample.
  - Median sensitivity for WEO: 0.38.
  - Average sensitivity for Consensus: 0.48.
  - Cumulative median data surprise: -13.8% of EM GDP.
  - Median WEO scarring revision: -5.2%.
  - Median Consensus scarring revision: -6.7%.
  - PPP-weighted sensitivity: 0.48 for both Consensus and WEO; cumulative PPP-weighted data surprise = -10.9%; median revision to Consensus and WEO scarring = -5.2%.
- Pre-war recovery period:
  - Quarterly GDP consistently surprised to the upside; scarring estimates revised up but less strongly than initial downward revisions.
  - Median sensitivity in pre-war recovery: 0.18 for WEO and 0.34 for Consensus.
  - PPP-weighted sensitivities in pre-war recovery: 0.29 for WEO and 0.33 for Consensus.
- Interpretation:
  - High initial sensitivity reflects extreme uncertainty about trend during the initial shock, causing forecasters to place high weight on data news as a signal about the trend.
  - Reduced sensitivity in the recovery suggests forecasters became more confident over time in the trend level of GDP, even though data surprises remained informative and often serially correlated.

Counterfactual exercise:
- Updating October 2020 scarring estimates using initial-period sensitivity (~1/2) to subsequent data surprises yields:
  - By end of pre-war recovery period, predicted level of scarring ≈ 3½% (PPP-weighted), about 0.5 percentage point less than PPP-weighted WEO or Consensus forecasts.
  - Median predicted level of scarring: -2.75% to -3.5%, around 1½–2pp less than median WEO and Consensus forecasts.
  - By 2021 Q4, PPP-weighted deviation in GDP from trend was 2.6% and for the median economy was 2.9% (actual outturns shown in the analysis).

### Accounting for Covid-related output losses — Data and methodology
- Analysis draws on WEO forecasts; sample for decomposition: 21 major EMs (accounting by weight for 35% of world GDP and 70% of the EMs group). Sample includes Turkiye, South Africa, Brazil, Chile, Colombia, Mexico, Peru, Indonesia, Malaysia, Philippines, Thailand, Beleraus, Kazakhstan, Russia, China, Serbia, Hungary, Croatia, Poland, Romania.
- Production function: Cobb-Douglas Y_t = A_t K_t^(1−α) L_t^α, with α (labor share) from PWT 10.0 averaging 0.55 in sample.
- Labor input decomposition: population (Pop), unemployment rate (UE), labor force participation rate (LFPR).
- Capital stock constructed via perpetual inventory: K_t = K_{t−1} (1−δ_t) + I_t, with δ_t from PWT and investment from WEO.
- Adjustment for capital utilization: baseline assumes capital utilization falls in line with employment in 2020 to avoid overstating TFP declines when utilization falls during lockdowns; alternative check uses ILO average hours data.
- Deviations relative to projected trend: Y_dev,t = A_dev,t K_dev,t^(1−α) L_dev,t^α; log-differenced growth accounting yields Δy_t = Δa_t + (1−α) Δk_t + α Δl_t.

### Results — decomposition of realized and projected output losses
- 2020:
  - Just under a third of the PPP-weighted average EM GDP loss in 2020 was accounted for by lower employment (driven by both higher unemployment and lower LFPR).
  - Initial contribution from capital was small in stock terms, but effective capital fell sharply during lockdowns due to lower utilization; accounting for lower utilization increases capital’s contribution to the 2020 loss.
  - TFP accounts for about two fifths of the remaining output losses; the decomposition using average hours suggests much of the persistent shortfall in TFP reflects lower hours worked rather than lower productivity per hour.
- 2022 Q1:
  - Contribution from lower employment remained around a third.
  - Contribution of capital due to lower investment built over time to a fifth of output losses in 2022 Q1.
  - TFP accounts for about 40% of the remaining losses.
- Comparisons to past large shocks:
  - Past shocks five years after event: persistent losses largely driven by weak productivity/TFP, with modest employment losses.
  - Covid’s persistent losses are more evenly distributed across labor, capital, and TFP; the persistent negative impact on TFP from Covid has been much smaller than in past episodes, reducing overall scarring relative to typical past shocks.
- January 2022 WEO projection:
  - Projected scarring by 2024 = -4% relative to pre-pandemic trend.
  - This deterioration is driven by weaker capital accumulation and a large decline in TFP; composition similar to past large shocks where long-term damage primarily reflects weaker TFP.

*International Monetary Fund — 2. The Evolution of Post-Covid GDP Losses and Scarring Estimates*

### 4. Implications for the Medium Term

### 4. Implications for the Medium Term

### Outlook for TFP
- Past episodes suggest shocks to TFP in EMs are persistent but do not tend to worsen over time.  
- Using multipliers from Barrett and others (2021) and a temporary TFP shock in Aguiar and Gopinath (2007), projections imply the recent hit to TFP may be persistent but not deteriorating.  
- The Covid shock differs from past shocks: it was exogenous and short-lived (suggesting lower persistence) but introduced unique channels to lower TFP and involved public and private debt build-up.  
- Bottom-up estimate of TFP effects:
  - Spillovers from weak employment to TFP: evidence on wage scarring implies using the Indonesia estimate that a year spent in unemployment reduced subsequent earnings by 3.5% to calculate TFP loss from reduced employment.  
  - As unemployment recovered partially relatively quickly in most EMs, the projected medium-term impact on TFP from elevated unemployment is only -0.1% on average.  
  - Including time out of the labor force raises the medium-term impact on labor via wages and productivity to -0.2%.  
  - Education disruption calculations:
    - UNESCO data: weighted average total loss of 90% of a school year in the sample, or 8.5 months of schooling.  
    - United Nations and WEO projections: peak proportion of the workforce affected by education disruption is close to 40% reached by 2031; by 2025, around half of the affected students have entered the workforce.  
    - Psacharopoulos and Patrinos (2018): on average, one year’s lost schooling reduces wages by 9%; assuming a production function with wages mapping 1:1 to GDP, a 1% reduction in wages translates into a 1% fall in GDP.  
    - Age-earnings adjustment: Indonesian data indicate earnings of workers aged 20-24 are around a quarter below the average before Covid; assume entrants earn three quarters of the average worker.  
  - Quantitative outcomes from bottom-up schooling estimate:
    - Peak impact on GDP of -2½% by the 2030s.  
    - Impact by 2025 builds to -1%, under half of the long-run effect of 2½%, with the peak in the early 2030s.  
  - Combined productivity effects (time out of work + lost education):
    - For 2022, estimated effect is just under –½%.  
    - Builds to -1% by 2025.  
  - Averaging over various estimates points to a reduction in TFP relative to trend of just under 1% in 2025.

### Outlook for Capital
- Three factors for capital stock impact:
  1. Lost capital so far from reduction in investment relative to trend.  
  2. Lower TFP reduces the marginal return on capital and optimal capital stock; reduction in optimal capital is very similar to lost capital so far.  
  3. Increased corporate leverage associated with reductions in investment in past crises.
- IMF (2022) finding used: a 1 percentage point increase in three-year average non-financial corporate credit-to-GDP ratio has a peak impact on the level of investment of -2%.  
- BIS data up to 2022 Q3: estimated a 1pp increase in the corporate credit-to-GDP ratio relative to the pre-Covid average. This implies:
  - Peak impact of -2% on investment per year.  
  - Cumulative impact of 4.5% after four years.  
- Combining effects produces:
  - Reduction in the level of capital of -1.4%, which reduces potential GDP by 0.6%.  
  - Around one quarter of the capital shortfall comes from higher corporate leverage; remainder from existing reduction in capital stock due to weak early-pandemic investment.  
- Financial distress indicators:
  - Ratio of provisions to nominal GDP has risen slightly by 0.1pp since the start of Covid.  
  - Ratio of non-performing loans has fallen.  
  - These patterns suggest estimates of the capital shortfall are unlikely to be materially biased by scrapped/redundant capital from bankrupt businesses as of the latest data.

### Outlook for Employment
- Employment recovery in EMs has been quicker than past shocks but remains incomplete.
  - Unemployment rate: after a sharp rise in 2020, unemployment recovered to pre-Covid levels by end-2022.  
  - Level of employment: remained around 1% below its pre-Covid trend in 2022, mostly driven by a drop in labor force participation.  
- Factors behind employment shortfall include shifts from formal to informal sectors, less generous job retention schemes, and disrupted internal migration patterns.  
- Compared to past crises:
  - After financial crises, peak shortfall in employment typically recovered by about one third after five years; after other recessions, by about one tenth.  
  - By 2022, employment had recovered about two thirds of the peak shortfall from Covid.  
- Projection assumptions and implications:
  - Bottom-up estimates assume no change in trend unemployment given current rates are mostly back to pre-Covid levels.  
  - Lower labor participation rates are projected to drag employment by 1% and output by 0.5%.

### Summing up (Medium-term scarring)
- Bottom-up estimates suggest scarring related to Covid may reach 2-2½% in the next few years.  
- This scarring is significant if sustained but smaller than average financial crises or other recessions and somewhat more optimistic than other forecasts.  
- The bottom-up estimate implies only a small improvement relative to the level of GDP in 2022 Q1: a 0.3pp improvement from the -2.3% deviation in GDP from trend at the start of 2022.  
- Key numeric summary (Table 1 reproduced):
  - GDP: WEO -3.4; Consensus -3.1; Bottom-up estimate -2.0; Memo: 22 Q1 GDP vs trend -2.3  
  - o/w TFP: WEO -1.5; Consensus -0.8; Bottom-up estimate -1.0  
  - o/w capital: WEO -0.9; Consensus -0.6; Bottom-up estimate -0.5  
  - o/w labor: WEO -1.0; Consensus -0.5; Bottom-up estimate -0.8
- Uncertainties and caveats:
  - Positive recent data surprises point to a more robust recovery, but supply constraints and rising inflation may limit sustainability of resilience.  
  - Timing of education-related losses affecting productivity is particularly uncertain because effects accrue as cohorts enter the workforce.  
  - Recovery assessment is complicated by compounding shocks from Russia’s war in Ukraine and by data uncertainty and future revisions.

*Source: 4. Implications for the Medium Term — wpiea2023162-print-pdf*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023162-print-pdf.pdf_