Long COVID? Prospects for economic scarring from the pandemic

Abstract

This paper examines the potential persistent effects (scarring) of the COVID‐19 pandemic on the economy and the channels through which they may occur. Our findings from a broad set of historical recessions confirm that recessions are associated with persistent output losses and that the greatest scarring has occurred following financial crises. The amount of scarring following pandemic and epidemic recessions in the sample is in between that of typical recessions and financial crises. Results on the channels show that the productivity channel is important, as all types of recessions have been followed by persistent losses to total factor productivity.

Abbreviations

AEs

advanced economies

EMDEs

emerging market and developing economies

IMF

International Monetary Fund

MERS

Middle East respiratory syndrome

SARS

severe acute respiratory syndrome

TFP

total factor productivity

1. INTRODUCTION

The COVID‐19 pandemic led to a severe global recession that is unique in many ways. The contraction in 2020 was very sudden and deep compared to previous global crises, even as the policy response in many countries was swift and sizable. The pandemic crisis also stands out for its differential impacts across sectors and countries, complex channels of transmission, and high uncertainty about the recovery path. The extent of scarring (persistent damage to supply potential) ¹ following the recession differs across countries, as the health crisis interacts with countries' economic structures (such as the importance of “high‐contact” sectors, where people are in close proximity) and varying policy responses.

The change in economic forecasts following the pandemic suggest a sizeable amount of scarring is expected. Forecasts based on IMF (2022) envision output losses, relative to pre‐pandemic projections, at about 3% for the world economy by 2024, with much more adverse effects in emerging market and developing economies relative to advanced economies (AEs) (red bars in Figure 1). At the same time, the lasting damages over a comparable period from the global financial crisis (GFC) were larger, at almost 8.7% for the world (blue bars in Figure 1). A key feature of the pandemic is that systemic financial instability was avoided, lessening some of the adverse effects.

FIGURE 1.

Medium‐Term Output Losses (percent difference from precrisis forecast). Bars show the percent difference in real GDP 4 years after the crisis and anticipated GDP for the same period prior to the crisis for the indicated group. For the COVID‐19 crisis, it compares the April 2022 WEO vintage forecast for 2024 versus that from the January 2020 vintage (prior to the pandemic). For the global financial crisis (GFC), it compares the April 2013 vintage for 2012 versus the October 2007 vintage (prior to the start of the US recession at end‐2007). AEs, advanced economies; EMDEs, emerging market and developing economies; EMEs, emerging market economies; LICs, low‐income countries. Source: International Monetary Fund, World Economic Outlook; and authors' calculations

Nevertheless, the atypical features of the current crisis—its severity, differential impacts, complex transmission, and high uncertainty—make assessment of the economic effects of COVID‐19 challenging. This paper aims to shed light on the potential main channels of scarring post‐COVID‐19 and implications for the medium‐term outlook. We ask what can we learn about prospects for scarring from historical experience with recessions? What are the most relevant channels in the current setting (productivity, labor, capital)? We draw lessons from previous recessions including those associated with past pandemics and epidemics, financial crises, natural disasters, and violent conflict outbreaks.

Previous literature has found that output losses following recessions are persistent, particularly after financial crises, with differential impact across country groups. Cerra and Saxena (2008) find that currency crises lead to permanent output losses 10 years after onset, with more adverse impacts for middle‐ and low‐income countries, and that banking crises or concurrent twin crises have even more adverse effects. Moreover, Blanchard et al. (2015) find that recessions in general, and also those associated with financial crises and oil price increases, are often followed not only by a lower output level, but also lower growth, implying that the scarring effect increases over time. Ball (2014) likewise points to significant scarring following the GFC, with an adverse effect on output growth. Abiad et al. (2009) and Chen et al. (2019) also document larger output losses following banking crises in general and the GFC, respectively, stemming from lasting declines in capital per worker, total factor productivity (TFP), and employment.

Several recent studies focus on the economic impact of past pandemics and epidemics. Jordà et al. (2020), who study six European economies over centuries, starting with the Black Death in the 14th century, find that macroeconomic effects of pandemics persist for decades, leading to a decline in real interest rates, indicating a disproportionate effect on the labor force relative to other factors of production such as land. Ma et al. (2020) study GDP, unemployment, and trade following six modern health crises, and find that, following the initial decline, the bounce‐back in output is rapid, but remains below pre‐recession level 5 years after the shock. Emmerling et al. (2021) and Cuesta Aguirre and Hannan (2021) come to similar conclusions based on the analysis of five health crises in the 21st century, and posit that the COVID‐19 pandemic is likely to lead to significant scarring. Barro et al. (2020) attempt to disentangle the effects of the Spanish flu and WWI deaths ² and find that GDP per capita declined by 6% as the result of the pandemic, which was on par with the 8.4% decline associated with the war. The aforementioned studies utilize local projections (Jordà, 2005) or vector autoregressions (VAR) to identify the impact of a given event (such as a pandemic or a financial crisis) on the economy. Using a different approach, Eichengreen et al. (2021) study the factors that affect the length of the recovery following a recession by first identifying recession episodes (267 recessions across 39 countries) and classifying them into (i) supply‐shock or demand‐shock driven, (ii) global or local, (iii) financial crisis or normal recessions. The authors conclude that negative supply and demand shocks during the COVID‐19 pandemic are likely to lead to a prolonged recovery.

Our main contributions to the literature on the economics effects of recessions are twofold. First, we conduct a comprehensive analysis of past recessions, using a broader sample of 586 recession episodes from 115 countries over 1957–2019, and differentiate between different types of crises (past pandemics and epidemics, financial crises, natural disasters, and violent conflict outbreaks). Some previous studies have compared different types of financial crises and recent studies focus on the COVID‐19 recession. However, we employ a unified framework in which all types of recessions are analyzed within the same regression via interaction terms, allowing us to account for potential co‐occurrence of several types of crisis events. Of the sample of 586 recessions, 108 coincided with a pandemic or epidemic (see Appendix Table a2 for the list of pandemics and countries affected). Of these 108 recessions, 34 co‐occurred with a financial crisis, 20 with a natural disaster, and 2 with a violent conflict. Second, we study the channels through which persistent damage occurs, by analyzing the effects of recessions on the supply‐side components of GDP. Previous studies have explored the channels of impact following only financial crises. Understanding the channels of impact following both “typical” recessions, which are not coincident with a particular kind of crisis, as well as following pandemics and other crises provides a better framework for understanding the potential for scarring post‐COVID‐19 recessions.

Our findings from the broad set of historical recessions confirm that recessions are associated with persistent output losses and that the greatest scarring has occurred following financial crisis recessions. The amount of scarring following the pandemic and epidemic recession in the sample is in between that of typical recessions and financial crises. Given that the COVID‐19 crisis is global and more severe than those previous pandemics, however, the amount of scarring is likely to be greater. The policy response to the COVID‐19 pandemic was unprecedented, with large fiscal policy responses in AEs in particular (IMF, 2021a), suggesting the impact may be more akin to typical recessions in these countries. Results on the channels of scarring show that that the productivity channel is particularly important, as all types of recessions have been followed by persistent losses to TFP.

The rest of the paper is organized as follows: Section 2 describes the data used in the analysis, Section 3 looks at the impact of past recessions on aggregate output and the channels of impact, also differentiating recessions by their depth and duration, and Section 4 concludes.

2. DATA

The historical analysis relies on the Penn World Table (PWT) 10.0 database (Feenstra et al., 2015; Inklaar & Timmer, 2013), from which we draw on data for real GDP per capita (at constant prices in 2017 US dollars) that we use to identify recession episodes and to quantify the aggregate impact of those recession episodes on the economy. We also look at the supply‐side channels of scarring (capital, labor, and productivity) using PWT data on capital stock (per person engaged), number of persons engaged (as employment‐population ratio), and TFP.

Recession episodes and the corresponding peaks and troughs of the cycle are identified using the Harding and Pagan (2002) algorithm on annual real GDP per capita, with a window of 1 year, minimum phase length of 1 year, and minimum cycle length of 2 years. While the standard approach for business cycle dating is typically done using quarterly data, the use of annual data allows for the identification of cycles for a larger sample of countries, in particular including developing economies for which quarterly data is often not available. Recessions identified using this approach for the United States match those reported by the NBER.

Recessions are further classified by co‐occurrence of a particular type of a crisis, namely: a financial crisis, an epidemic or pandemic, a disaster, or a violent conflict. Each recession can be associated with several types of crises, or with no crisis, in which case it is referred to as a “typical” recession (which is 202 out of 586 recessions). We look at crisis events in the year when they occur, and to further associate a recession with a crisis we check whether the crisis event has occurred within [t − 2; t + 2] of the recession. This is to account for cases when a crisis occurs at the end of the calendar and could not yet affect economic activity in that year but has an impact the following year, or has a slow ramp‐up (e.g., a disease outbreak might be initially reported only in a few isolated locations, and spread to other places later), or there are delays in reporting (e.g., initial cases of a disease might be not diagnosed, but with better monitoring in the following years the presence of the disease in the country is noticed).

The incidence of financial crises is taken from Laeven and Valencia (2018) for the period going back to 1970, and Reinhart et al. (2016) for years prior to 1970. In both cases, financial crises include banking crises, currency crises, and sovereign debt crises. Past modern epidemics and pandemics (and outbreaks) include the Hong Kong flu, SARS, H1N1, MERS, Ebola and are identified for countries in which cases have been reported (see Appendix Table a2 for the list of pandemics and countries affected). ³ Disasters are identified using the Emergency Events Database (EM‐DAT) when a country in a given year has experienced disasters that led to damages exceeding 1% of GDP or affected 5% of population (including deaths). Finally, a country is defined as being in conflict if in a given year there are battle‐related deaths that exceed 100 people per one million population (Novta & Pugacheva, 2021).

Throughout the text, countries are classified into AEs and EMDEs. Country list and samples are provided in Appendix Table a1.

3. ANALYSIS OF HISTORICAL RECESSIONS

3.1. Aggregate impact

This section looks at the aftermath of previous recessions, distinguishing between more typical downturns and those associated with financial crises, epidemics or pandemics, violent conflicts, or natural disasters, to get a sense of how long‐lived their effects have been and the supply‐side channels (capital, labor, and productivity) through which they occur.

The analysis of the impact of a recession relies on local projections (Jordà, 2005) to estimate the dynamic effects of the various types of recessions. This approach employs a set of regressions to estimate the impact of current covariates on future outcomes at different horizons. Local projections have been advocated by Montiel Olea and Plagborg‐Møller (2021) and Plagborg‐Møller and Wolf (2021), among others, as a flexible and robust alternative to vector autoregressions (VAR). The specification is based on the following equation:

$$y_{i,t+h}-y_{i,t-1}={\beta }_1^h{D}_{i,t}+\sum _{E\in \left\{types\right\}}\left[{{\beta }_2^{E,h}{E}_{i,t}+\beta }_3^{E,h}{D}_{i,t}\ast E_{i,t-2,t+2}\right]$$

$$+{\phi }_1^h{X}_{i,t}+{\mu }_i^h+{\theta }_t^h+{\epsilon }_{i,t}^h$$ (1)

in which ($y_{i,t+h}-y_{i,t-1}$) represents cumulative growth in log points in real GDP per capita (or another dependent variable) at different horizons (h = 0,…7), where h = 0 represents the contemporaneous effect; $D_{i,t}$ is a dummy for recession onset (first year after the peak); $E_{i,t}$ is a dummy for occurrence of a crisis event for each of the following types: financial crisis, an epidemic or pandemic, a disaster, or a conflict; the interaction terms $D_{i,t}\ast E_{i,t-2,t+2}$ capture different types of crisis events that happened within t − 2 to t + 2 of a given recession (see further details on the timing in the Data section); $X_{i,t}$ is a set of controls that includes two lags of the dependent variable's growth rate, one lag of log GDP in constant US dollars, and two lags of credit‐to‐GDP ratio; ${\mu }_i^h$ and ${\theta }_t^h$ are country and year fixed effects (dummy variables that take the value of 1 for a given country, and zero otherwise; and dummy variables that take the value of 1 for a given year, and zero otherwise) that control for time‐invariant country characteristics and time‐specific common global shocks, respectively. The impact of a typical recession is given by ${\beta }_1^h$, and the impact of a recession associated with a crisis event $E$ is given by ${\beta }_1^h+{\beta }_2^{E,h}+{\beta }_3^{E,h}$. Regressions are estimated separately for each horizon on a fixed sample. Thus, the number of observations, countries, and recession episodes is the same at all horizons and across all dependent variables. In all regressions, the left‐hand‐side variable has been winsorized at 0.5/99.5 percentiles to mitigate the effect of outliers.

The estimation results are presented in Table 1 columns 1–5, and depicted in Figure 2 panel 1. The coefficients show the cumulative impact of a recession relative to the baseline, thus the return of the impulse response to zero signifies that the dependent variable has recovered to its pre‐recession level. While the path of output differs by the type of recession, the estimates are negative and mostly statistically significant across all horizons, indicating that recessions are associated with permanent output losses, on average.

TABLE 1.

Medium‐term output losses and channels of impact

	Real GDP per capita					Total factor productivity
Recession type:	Typical	Past pandemic/Epidemic	Financial crisis	Disaster	Conflict	Typical	Past pandemic/Epidemic	Financial crisis	Disaster	Conflict
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
h = 0	−4.180***	−3.905***	−5.740***	−4.512***	−6.812***	−3.376***	−2.919***	−4.543***	−3.965***	−5.627***
	(0.362)	(0.540)	(0.496)	(0.459)	(1.965)	(0.328)	(0.524)	(0.408)	(0.468)	(1.537)
h = 1	−4.750***	−4.555***	−8.830***	−4.455***	−10.450***	−3.251***	−2.182***	−6.194***	−3.433***	−7.606***
	(0.495)	(0.813)	(0.737)	(0.650)	(2.703)	(0.433)	(0.768)	(0.590)	(0.606)	(2.215)
h = 2	−4.830***	−5.554***	−9.170***	−3.719***	−10.078***	−3.135***	−2.417**	−5.853***	−2.522***	−7.256**
	(0.640)	(1.145)	(0.917)	(0.841)	(3.220)	(0.534)	(1.056)	(0.756)	(0.835)	(2.820)
h = 3	−4.964***	−6.697***	−9.565***	−3.579***	−9.486***	−2.923***	−3.445***	−5.536***	−2.330**	−5.856**
	(0.711)	(1.231)	(1.118)	(0.946)	(3.393)	(0.579)	(1.191)	(0.877)	(0.955)	(2.808)
h = 4	−4.157***	−6.553***	−9.320***	−2.462**	−8.586**	−2.298***	−3.267**	−5.027***	−1.329	−5.532
	(0.816)	(1.668)	(1.238)	(1.079)	(4.125)	(0.689)	(1.536)	(0.960)	(1.109)	(3.421)
h = 5	−3.990***	−7.536***	−9.820***	−1.984	−6.754	−2.316***	−4.126**	−5.277***	−1.052	−3.824
	(0.875)	(1.872)	(1.295)	(1.287)	(5.291)	(0.745)	(1.850)	(1.009)	(1.263)	(4.483)
h = 6	−3.680***	−8.476***	−10.066***	−1.752	−6.099	−2.271***	−4.930**	−5.145***	−1.075	−2.999
	(0.904)	(1.984)	(1.412)	(1.462)	(5.564)	(0.763)	(2.070)	(1.124)	(1.378)	(4.565)
h = 7	−2.960***	−8.718***	−9.950***	−1.131	−4.407	−1.737**	−5.074**	−4.862***	−0.562	−1.288
	(0.937)	(2.151)	(1.423)	(1.569)	(5.350)	(0.804)	(2.218)	(1.154)	(1.382)	(4.502)
Number of observations	4341					4341
Number of countries	115					115
Number of recessions	586					586
R 2 (for h = 0)	0.38					0.32
R 2 (for h = 7)	0.59					0.53

	Capital stock per worker					Employment‐population ratio
	Typical	Past pandemic/Epidemic	Financial crisis	Disaster	Conflict	Typical	Past pandemic/Epidemic	Financial crisis	Disaster	Conflict
Recession type:	(11)	(12)	(13)	(14)	(15)	(16)	(17)	(18)	(19)	(20)
h = 0	0.279	0.435	0.166	−0.561**	−0.587	−0.659***	−1.057***	−0.997***	−0.045	−0.199
	(0.199)	(0.313)	(0.272)	(0.243)	(0.671)	(0.170)	(0.233)	(0.224)	(0.180)	(0.348)
h = 1	0.337	0.838	−0.444	−1.289***	−0.898	−1.270***	−2.133***	−1.828***	−0.101	−0.549
	(0.289)	(0.660)	(0.482)	(0.459)	(1.296)	(0.218)	(0.454)	(0.328)	(0.287)	(0.629)
h = 2	−0.017	0.119	−1.661**	−2.196***	−2.010	−1.345***	−2.224***	−1.908***	0.232	−0.395
	(0.420)	(0.919)	(0.713)	(0.717)	(1.714)	(0.279)	(0.559)	(0.429)	(0.441)	(1.127)
h = 3	−0.407	−1.560	−2.696***	−2.752***	−2.613	−1.328***	−1.174	−1.953***	0.536	−0.396
	(0.528)	(1.227)	(0.905)	(0.947)	(2.151)	(0.309)	(0.803)	(0.485)	(0.568)	(1.309)
h = 4	−0.790	−1.639	−3.714***	−3.149***	−3.014	−1.084***	−1.274	−1.853***	0.824	−0.284
	(0.632)	(1.299)	(1.084)	(1.178)	(2.740)	(0.298)	(0.792)	(0.534)	(0.681)	(1.509)
h = 5	−1.187	−2.025	−4.478***	−3.353**	−3.594	−0.838***	−1.195	−1.943***	1.042	0.083
	(0.731)	(1.453)	(1.253)	(1.415)	(3.386)	(0.316)	(0.817)	(0.608)	(0.791)	(1.616)
h = 6	−1.366	−1.946	−5.122***	−3.318**	−4.391	−0.655*	−1.469*	−2.081***	1.136	0.645
	(0.845)	(1.552)	(1.458)	(1.646)	(3.894)	(0.344)	(0.812)	(0.657)	(0.880)	(1.912)
h = 7	−1.282	−2.714	−5.858***	−3.257*	−4.334	−0.565	−1.124	−1.990***	1.289	0.518
	(0.934)	(1.714)	(1.632)	(1.855)	(4.368)	(0.402)	(0.839)	(0.687)	(0.982)	(2.152)
Number of observations	4341					4341
Number of countries	115					115
Number of recessions	586					586
R 2 (for h = 0)	0.57					0.22
R 2 (for h = 7)	0.65					0.37

Note: The reported coefficients represent the impact of a recession associated with a particular crisis (β ₁ for typical recessions, and β ₁ + β ₂ + β ₃ for other types of recessions, as per Equation (1)). The dependent variables are cumulative growth of real GDP per capita, total factor productivity, capital per worker, employment‐population ration in the horizon year h after a recession. Regressions are estimated separately for each horizon. All regressions include interaction terms for recession types (financial crisis, pandemic, disaster, conflict, or typical recession that occurred due to other reasons) and controls for two lags of the dependent variable's growth rate, one lag of log GDP per capita (in constant US dollars), and two lags of credit‐to‐GDP ratio, country and year fixed effects. Past modern pandemics or epidemics include Hong Kong flu, SARS, H1N1, MERS, Ebola (see Appendix Table a2 for more details). Standard errors are clustered at the country level.

*p < 0.1; **p < 0.05; ***p < 0.01.

Source: Authors' calculations.

FIGURE 2.

Medium‐Term Output Losses and Channels of Impact (percentage points). The solid lines represent the estimated cumulative impulse response functions and shaded areas represent 90 percent confidence intervals. Time since the shock (in years) on the x‐axis. Past modern pandemics and epidemics include Hong Kong flu, SARS, H1N1, MERS, Ebola (see Appendix Table a2 for more details). Sources: Penn World Table 10.0; and authors’ calculations.

Recessions associated with financial crises lead to more negative outcomes (column 3), as has also been shown in the previous literature (Cerra & Saxena, 2008). The path of output after past modern epidemic or pandemic recessions (column 2) is in between that of typical recessions and financial crisis recessions. However, the COVID‐19 crisis is global and more severe than those previous pandemics. The impact of natural disasters (column 4) and violent conflict (column 5) is likewise negative and severe on impact, with effects persisting for several years following the crisis; in later horizons, the effect remains negative but no longer statistically significant, which could be attributed to the positive effects of post‐disaster reconstruction efforts and sample limitations as data for fragile states is often not available. In the following analysis, due to space considerations and our focus on the effects of past pandemics or epidemics and associated recessions, we skip the presentation of results on the impact of natural disasters and violent conflict, for which the findings in general are consistent with the literature.

3.2. Depth and duration

Drawing on the observation that the COVID‐19 crisis is characterized by its unprecedented depth, and will differ in how long it lasts across country groups, with faster recovery projected in AEs (see IMF, 2021a), each recession episode is further characterized by its depth (defined as the loss in real GDP per capita between the peak and the trough in percentage terms) and duration (defined as the number of years between the peak and the trough). In the sample, past recession durations range between 1 and 10 years, with 60% of recessions lasting one year and 90% of recessions lasting not more than three years for both AEs and EMDEs. We define the depth of a recession as the loss between the peak and the first year of the recession, to ease the comparison across recessions of different duration. Under this definition, the median recession is associated with a 2.2% decline in per capita output in the first year. Recessions are classified as high (low) depth when they fall above (below) the median loss.

Our analysis of the differential effects of recession depth and duration is based on a modified version of regression Equation (1) that includes interaction terms for recessions of (1) high depth and 1 year duration, (2) low depth and 1 year duration, (3) high depth and more than a year duration, (4) low depth and more than a year duration. The interaction terms are included for all recession types. Table 2 shows the estimated coefficients for short‐duration typical recessions of different depth. Overall, deep recessions—those with a greater initial impact—result in greater scarring, as expected. However, controlling for the initial depth of the recession, shows that recoveries proceed differently in AEs and EMDEs. In AEs, there is a rebound following deep recessions and no permanent output loss after several years (column 3 and red line in Figure 3, panel 2). Emerging market and developing economies, however, experience protracted downturns and permanent losses, on average (column 5 and red line in Figure 3, panel 3). ⁴

TABLE 2.

Medium‐term output losses by recession depth (short duration)

	World		AEs		EMDEs
Recession type:	High depth	Low depth	High depth	Low depth	High depth	Low depth
	(1)	(2)	(3)	(4)	(5)	(6)
h = 0	−6.116***	−2.755***	−6.717***	−2.337***	−5.916***	−2.887***
	(0.665)	(0.337)	(1.181)	(0.362)	(0.824)	(0.514)
h = 1	−4.359***	−2.321***	−6.309***	−2.503***	−3.616***	−2.072***
	(0.806)	(0.426)	(1.349)	(0.588)	(0.968)	(0.576)
h = 2	−3.957***	−1.787***	−4.452***	−2.378***	−3.573***	−1.526*
	(1.114)	(0.599)	(1.547)	(0.661)	(1.293)	(0.910)
h = 3	−4.824***	−1.572**	−4.130**	−2.519***	−4.789***	−1.300
	(1.301)	(0.739)	(1.825)	(0.884)	(1.537)	(1.121)
h = 4	−3.388*	−1.383	−3.146	−2.510**	−3.216	−1.233
	(1.731)	(0.904)	(1.982)	(1.104)	(2.001)	(1.382)
h = 5	−5.024**	−0.652	−2.847	−2.010*	−5.369**	−0.454
	(1.975)	(1.031)	(2.078)	(1.187)	(2.242)	(1.577)
h = 6	−5.930**	−0.244	−2.028	−2.388*	−6.897**	−0.013
	(2.370)	(1.024)	(1.861)	(1.330)	(2.798)	(1.491)
h = 7	−4.847**	0.531	−1.366	−2.343*	−5.780**	0.816
	(2.330)	(1.147)	(1.823)	(1.366)	(2.732)	(1.707)
Number of observations	4341		1337		2999
Number of countries	115		34		81
Number of recessions	586		146		439
R 2 (for h = 0)	0.40		0.63		0.38
R 2 (for h = 7)	0.59		0.71		0.61

Note: The dependent variable is cumulative growth of real GDP per capita in the horizon year h after a recession. Regressions are estimated separately for each horizon. All regressions include interaction terms for recession types (financial crisis, pandemic, disaster, conflict) and recession depth (high and low depth recessions are split based on the median loss, with separate interaction terms for recessions that last only 1 year and those that last longer than 1 year), as well as controls for two lags of the dependent variable's growth rate, one lag of log GDP per capita (in constant US dollars), and two lags of credit‐to‐GDP ratio, country and year fixed effects.

*p < 0.1; **p < 0.05; ***p < 0.01.

Source: Authors' calculations.

FIGURE 3.

Medium‐Term Output Losses by Recession Depth (Short Duration) (percentage points). Figure shows the results for “typical” recessions. The solid lines represent the estimated cumulative impulse response functions and shaded areas represent 90 percent confidence intervals. High and low‐depth recessions are split based on the median per‐capita output loss. Short duration recessions last not more than 1 year. Time since the shock (in years) on the x‐axis. Source: Penn World Table 10.0; and authors' calculations.

3.3. Channels of impact

Previous literature suggests that permanent damage to an economy's supply potential following a recession can occur through a number of channels. ⁵ First through the labor channel, as unemployment may remain higher even after the recession (Blanchard & Summers, 1986) and could result in a smaller labor force as discouraged workers exit. Human capital accumulation and future earnings can be affected by skill deterioration during extended periods of unemployment, delayed labor market entry for young workers, and negative effects on educational achievement in the longer term. ⁶ Second through the capital channel, as weak investment could result in both slower physical capital accumulation and slower technology adoption that hampers productivity growth. Greater scarring through the physical capital channel could also materialize as the result of capital being stranded and corporate debt buildup constraining future investment (IMF, 2021b). Lastly, productivity could also be permanently affected by the loss of firm‐specific know‐how as a result of bankruptcies and their spillovers (Bernstein et al., 2019), the effects of a decline in research and development and innovation during a recession, and an increase in resource misallocation (Adler et al., 2017; Furceri et al., 2021).

Focusing on the supply‐side channels, we look at the components of the Cobb‐Douglas production function. We estimate Equation (1) with each of TFP, capital per worker, and the employment‐population ratio as the dependent variable, to show the impact of recessions on each of these three components. The results for the World are presented in Table 1. The analysis shows that medium‐term losses in GDP per capita for typical recessions can be primarily attributed to losses in TFP (column 6). Employment per capita also declines before recovering somewhat in the medium‐term (column 16). For financial crisis recessions, there is significant, persistent damage to all factors: TFP (column 8), capital‐to‐worker ratio (column 13), and employment per capita (column 18), consistent with the findings of Abiad et al. (2009). For the GFC, Adler et al. (2017) found the subsequent widespread decline in TFP growth was the main contributor to output losses relative to the precrisis trend.

Tables 3 and 4, and Figure 4 report impulse response functions for AEs and emerging market and developing economies separately. For typical recessions and financial crises, the channels of impact are broadly the same across country groups, except that employment per capita losses play a role in AEs, on average, and not EMDEs. In modern era epidemics and pandemics, productivity losses were the main contributor to output losses in both AEs and EMDEs.

TABLE 3.

Medium‐term output losses and channels of impact: Advanced economies

	Real GDP per capita			Total factor productivity			Capital stock per worker			Employment‐population ratio
Recession type:	Typical	Past pandemic/Epidemic	Financial crisis	Typical	Past pandemic/Epidemic	Financial crisis	Typical	Past pandemic/Epidemic	Financial crisis	Typical	Past pandemic/Epidemic	Financial crisis
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
h = 0	−3.442***	−2.624***	−4.562***	−2.101***	−2.344***	−2.641***	0.729***	0.266	0.899***	−1.307***	−0.305	−2.056***
	(0.456)	(0.593)	(0.771)	(0.374)	(0.518)	(0.547)	(0.225)	(0.342)	(0.295)	(0.208)	(0.314)	(0.449)
h = 1	−4.575***	−3.942***	−6.935***	−2.176***	−3.087***	−3.183***	1.137***	0.869	1.101	−2.392***	−0.884*	−3.935***
	(0.574)	(1.049)	(1.123)	(0.444)	(0.821)	(0.884)	(0.375)	(0.574)	(0.714)	(0.318)	(0.464)	(0.836)
h = 2	−4.360***	−4.836***	−8.293***	−1.741***	−3.829***	−2.971***	0.721	0.764	0.568	−2.584***	−0.998	−5.105***
	(0.622)	(1.485)	(1.327)	(0.410)	(1.174)	(0.916)	(0.578)	(0.870)	(1.320)	(0.379)	(0.682)	(1.184)
h = 3	−4.310***	−5.754***	−9.948***	−1.766***	−4.786***	−3.627***	0.173	0.146	−0.058	−2.529***	−0.689	−6.062***
	(0.697)	(1.411)	(1.451)	(0.500)	(1.212)	(1.205)	(0.757)	(1.160)	(1.749)	(0.403)	(0.730)	(1.212)
h = 4	−4.053***	−5.422***	−11.328***	−1.650***	−4.478***	−4.095***	−0.487	−0.312	−1.085	−2.242***	−0.545	−6.559***
	(0.798)	(1.503)	(1.673)	(0.527)	(1.194)	(1.478)	(0.903)	(1.521)	(1.975)	(0.467)	(0.802)	(1.265)
h = 5	−3.559***	−5.898***	−10.759***	−1.224*	−4.718***	−3.134*	−0.865	−0.567	−2.011	−2.094***	−0.558	−7.056***
	(0.902)	(1.657)	(1.977)	(0.608)	(1.363)	(1.598)	(1.188)	(1.853)	(2.317)	(0.576)	(0.830)	(1.313)
h = 6	−3.611***	−5.137***	−10.311***	−1.302*	−3.847**	−2.776	−1.125	−0.721	−3.446	−1.965***	−0.364	−6.723***
	(0.905)	(1.797)	(2.294)	(0.646)	(1.475)	(1.804)	(1.320)	(2.278)	(2.601)	(0.644)	(0.972)	(1.342)
h = 7	−3.448***	−2.790	−8.605***	−0.900	−1.629	−1.113	−1.304	−1.451	−5.050*	−1.960***	0.228	−6.056***
	(0.869)	(2.036)	(2.635)	(0.689)	(1.780)	(2.018)	(1.398)	(2.468)	(2.852)	(0.688)	(1.059)	(1.388)
Number of observations	1337			1337			1337			1337
Number of countries	34			34			34			34
Number of recessions	146			146			146			146
R 2 (for h = 0)	0.60			0.38			0.68			0.48
R 2 (for h = 7)	0.70			0.47			0.72			0.41

Note: The dependent variables are cumulative growth of real GDP per capita, total factor productivity, capital per worker, employment‐population ration in the horizon year h after a recession. Regressions are estimated separately for each horizon. All regressions include interaction terms for recession types (financial crisis, pandemic, disaster, conflict, or regular recession that occurred due to other reasons) and controls for two lags of the dependent variable's growth rate, one lag of log GDP per capita (in constant US dollars), and two lags of credit‐to‐GDP ratio, country and year fixed effects. Past modern pandemics or epidemics include Hong Kong flu, SARS, H1N1, MERS, Ebola (see Appendix Table a2 for more details). Standard errors are clustered at the country level.

*p < 0.1; **p < 0.05; ***p < 0.01.

Source: Authors' calculations.

TABLE 4.

Medium‐term output losses and channels of impact: Emerging market and developing economies

	Real GDP per capita			Total factor productivity			Capital stock per worker			Employment‐population ratio
Recession type:	Typical	Past pandemic/Epidemic	Financial crisis	Typical	Past pandemic/Epidemic	Financial crisis	Typical	Past pandemic/Epidemic	Financial crisis	Typical	Past pandemic/Epidemic	Financial crisis
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
h = 0	−4.438***	−4.487***	−5.834***	−3.961***	−3.341***	−4.907***	0.065	0.300	0.110	−0.312	−1.163***	−0.791***
	(0.469)	(0.741)	(0.612)	(0.395)	(0.742)	(0.498)	(0.266)	(0.444)	(0.321)	(0.216)	(0.311)	(0.228)
h = 1	−4.702***	−4.749***	−8.661***	−3.759***	−2.144*	−6.616***	−0.074	−0.075	−0.720	−0.632**	−1.827***	−1.213***
	(0.604)	(1.086)	(0.894)	(0.514)	(1.096)	(0.703)	(0.378)	(0.975)	(0.555)	(0.248)	(0.629)	(0.274)
h = 2	−4.940***	−5.719***	−8.448***	−3.843***	−2.162	−6.124***	−0.410	−1.037	−1.886**	−0.648*	−1.955**	−1.051***
	(0.808)	(1.628)	(1.094)	(0.654)	(1.538)	(0.910)	(0.575)	(1.384)	(0.821)	(0.356)	(0.790)	(0.346)
h = 3	−5.316***	−6.684***	−8.259***	−3.676***	−3.001*	−5.457***	−0.767	−3.066	−2.748**	−0.682	−0.672	−0.933**
	(0.889)	(1.779)	(1.314)	(0.718)	(1.771)	(1.017)	(0.733)	(1.860)	(1.043)	(0.425)	(1.216)	(0.396)
h = 4	−4.490***	−6.027**	−7.193***	−2.898***	−2.349	−4.494***	−1.017	−2.665	−3.459***	−0.569	−1.069	−0.732
	(1.077)	(2.476)	(1.412)	(0.914)	(2.362)	(1.080)	(0.874)	(1.965)	(1.276)	(0.410)	(1.165)	(0.473)
h = 5	−4.532***	−7.007**	−7.497***	−3.054***	−3.403	−4.857***	−1.424	−2.851	−3.983***	−0.389	−0.908	−0.651
	(1.117)	(2.770)	(1.390)	(0.972)	(2.870)	(1.043)	(0.995)	(2.205)	(1.477)	(0.426)	(1.178)	(0.543)
h = 6	−4.258***	−8.356***	−7.782***	−3.001***	−4.799	−4.865***	−1.633	−2.359	−4.163**	−0.225	−1.356	−0.900
	(1.104)	(2.897)	(1.476)	(0.969)	(3.229)	(1.147)	(1.152)	(2.343)	(1.724)	(0.476)	(1.142)	(0.632)
h = 7	−3.463***	−9.754***	−7.980***	−2.449**	−6.208*	−4.978***	−1.558	−2.757	−4.465**	−0.115	−1.072	−0.948
	(1.212)	(3.116)	(1.487)	(1.064)	(3.462)	(1.150)	(1.263)	(2.575)	(1.918)	(0.559)	(1.199)	(0.690)
Number of observations	2999			2999			2999			2999
Number of countries	81			81			81			81
Number of recessions	439			439			439			439
R 2 (for h = 0)	0.36			0.34			0.55			0.18
R 2 (for h = 7)	0.60			0.56			0.66			0.39

Note: The dependent variables are cumulative growth of real GDP per capita, total factor productivity, capital per worker, employment‐population ration in the horizon year h after a recession. Regressions are estimated separately for each horizon. All regressions include interaction terms for recession types (financial crisis, pandemic, disaster, conflict, or regular recession that occurred due to other reasons) and controls for two lags of the dependent variable's growth rate, one lag of log GDP per capita (in constant US dollars), and two lags of credit‐to‐GDP ratio, country and year fixed effects. Past modern pandemics or epidemics include Hong Kong flu, SARS, H1N1, MERS, Ebola (see Appendix Table a2 for more details). Standard errors are clustered at the country level.

*p < 0.1; **p < 0.05; ***p < 0.01.

Source: Authors' calculations.

FIGURE 4.

Medium‐Term Output Losses and Channels of Impact: Across advanced economies (AEs) and Emerging Market and Developing Economies (percentage points). The solid lines represent the estimated cumulative impulse response functions and shaded areas represent 90 percent confidence intervals. Time since the shock (in years) on the x‐axis. Past modern pandemics and epidemics include Hong Kong flu, SARS, H1N1, MERS, Ebola (see Appendix Table a2 for more details). AEs, advanced economies; EMDEs, emerging market and developing economies. Source: Penn World Table 10.0; and authors' calculations.

The impact of the COVID‐19 pandemic could be even larger than suggested by the analysis of past recessions. From the labor side, some high‐contact sectors may shrink permanently. Moreover, widespread school closures have occurred across countries, with disproportionately adverse impacts on schooling in low‐income countries and those less prepared to switch to virtual learning. Productivity‐decreasing resource mismatches from the COVID‐19 crisis, across sectors and occupations, may likewise be larger than in previous crises, depending on how permanent the asymmetric losses are. ⁷ Productivity could also be negatively affected by a decline in competition, if the market power of large companies increases due to small business closures in high‐contact sectors and even more broadly. ⁸ At the same time, the pandemic has spurred increased digitalization and innovation in production and delivery processes, likely helping to offset the adverse productivity shock in some countries, as others lack the prerequisite widespread and reliable connectivity (Njoroge & Pazarbasioglu, 2020).

4. CONCLUSIONS

The historical record suggests that most recessions leave persistent scars—largely through lower productivity growth and, particularly in the case of financial crises, slower capital accumulation. There is high uncertainty around the current outlook, over both the short and medium term. The extent of scarring following COVID‐19 also depends on factors unique to a pandemic‐driven downturn and inherently hard to predict: the path of the pandemic (whether transmission of new variants outpaces vaccinations) and the scale of activity disruptions from restrictions needed to lower transmission. Moreover, repeated shocks have beleaguered the global economy as it began to recover from the pandemic recession and high uncertainty surrounds the outlook.

At the same time, the relative financial stability following the COVID‐19 shock so far is encouraging, as the greatest scarring in the past has occurred in recessions associated with financial crises. The expected losses are lower than what was seen during the GFC, consistent with the swift policy response that supported incomes and helped contain financial sector disruptions. However, emerging market and developing economies, in particular, are expected to have deeper scars than AEs (Figure 1), partly reflecting their greater sectoral exposure to the pandemic shock and more muted policy response (Das et al., 2022).

The picture of divergent recoveries, with a larger likelihood and extent of scarring in many of the same countries that have limited ability to provide further fiscal support, suggests a challenging path ahead. Experience from past recessions underscores the importance of avoiding financial distress as the COVID‐19 policy response evolves. To prevent scarring that could result from future financial instability, measures that support credit provision should be maintained while ensuring balance sheet resilience and adequate buffers. To maximize the use of limited fiscal space, policymakers should tailor their responses, targeting support to the most‐affected sectors and firms. Policies that reverse the setback to human capital accumulation, boost job creation, and facilitate worker reallocation will be key to addressing long‐term output losses, supply chain issues, and the rise in inequality. Policies to promote competition, innovation, and technology adoption would also lift productivity growth and boost investment.

Supporting information

Supplementary Material 1

TABLE A1.

Economies Included in the Analysis

Advanced Economies (N = 34)

Australia; Austria; Belgium; Canada; Cyprus; Czech Republic; Denmark; Estonia; Finland; France; Germany; Greece; Hong Kong SAR; Iceland; Ireland; Israel; Italy; Japan; Korea; Luxembourg; Macao SAR; Malta; Netherlands; New Zealand; Norway; Portugal; Singapore; Slovak Republic; Slovenia; Spain; Sweden; Switzerland; United Kingdom; United States

Emerging Market and Developing Economies (N = 81)

Angola; Argentina; Armenia; Bahrain; Barbados; Benin; Bolivia; Botswana; Brazil; Bulgaria; Burkina Faso; Burundi; Cameroon; Central African Republic; Chile; China; Colombia; Costa Rica; Croatia; Côte d'Ivoire; Dominican Republic; Ecuador; Egypt; Eswatini; Fiji; Gabon; Guatemala; Honduras; Hungary; India; Indonesia; Iran; Iraq; Jamaica; Jordan; Kazakhstan; Kenya; Kuwait; Kyrgyz Republic; Lao P.D.R.; Lesotho; Malaysia; Mauritania; Mauritius; Mexico; Moldova; Mongolia; Morocco; Mozambique; Namibia; Nicaragua; Niger; Nigeria; Panama; Paraguay; Peru; Philippines; Poland; Qatar; Romania; Russia; Rwanda; Saudi Arabia; Senegal; Serbia; Sierra Leone; South Africa; Sri Lanka; Sudan; Tajikistan; Tanzania; Thailand; Togo; Trinidad and Tobago; Tunisia; Turkey; Ukraine; Uruguay; Venezuela; Zambia; Zimbabwe

Source: Authors’ compilation.

TABLE A2.

List of Pandemics/epidemics/outbreaks

Ebola (2014)

Guinea; Italy*; Liberia; Mali; Nigeria*; Senegal*; Sierra Leone*; Spain*; United States*; United Kingdom*

MERS (2012)

Algeria; Austria*; China*; Egypt*; France*; Germany*; Greece*; Iran*; Italy*; Jordan*; Korea*; Kuwait*; Lebanon; Malaysia*; Netherlands*; Oman; Philippines*; Qatar*; Saudi Arabia*; Thailand*; Tunisia*; Turkey*; United Arab Emirates; United Kingdom*; United States*; Yemen

H1N1 (2009)

Afghanistan; Albania; Algeria; Angola*; Antigua and Barbuda; Argentina*; Aruba; Australia*; Austria*; Azerbaijan; Bahamas, The; Bahrain*; Bangladesh; Barbados*; Belarus; Belgium*; Belize; Bhutan; Bolivia*; Bosnia and Herzegovina; Botswana*; Brazil*; Brunei Darussalam; Bulgaria*; Cabo Verde; Cambodia; Cameroon*; Canada*; Chile*; China*; Colombia*; Congo, Democratic Republic of the; Costa Rica*; Croatia*; Cyprus*; Czech Republic*; Côte d'Ivoire*; Denmark*; Djibouti; Dominica; Dominican Republic*; Ecuador*; Egypt*; El Salvador; Estonia*; Eswatini*; Ethiopia; Fiji*; Finland*; France*; Gabon*; Georgia; Germany*; Ghana; Greece*; Grenada; Guatemala*; Guyana; Haiti; Honduras*; Hong Kong SAR*; Hungary*; Iceland*; India*; Indonesia*; Iran*; Iraq*; Ireland*; Israel*; Italy*; Jamaica*; Japan*; Jordan*; Kazakhstan*; Kenya*; Kiribati; Korea*; Kosovo; Kuwait*; Kyrgyz Republic*; Lao P.D.R.*; Latvia; Lebanon; Lesotho*; Libya; Lithuania; Luxembourg*; Macao SAR*; Madagascar; Malawi; Malaysia*; Maldives; Malta*; Marshall Islands; Mauritius*; Mexico*; Micronesia; Moldova*; Montenegro, Rep. of; Morocco*; Mozambique*; Myanmar; Namibia*; Nauru; Nepal; Netherlands*; New Zealand*; Nicaragua*; North Macedonia; Norway*; Oman; Pakistan; Palau; Panama*; Papua New Guinea; Paraguay*; Peru*; Philippines*; Poland*; Portugal*; Puerto Rico; Qatar*; Romania*; Russia*; Samoa; Saudi Arabia*; Serbia*; Seychelles; Singapore*; Slovak Republic*; Slovenia*; Solomon Islands; South Africa*; Spain*; Sri Lanka*; St. Kitts and Nevis; St. Lucia; St. Vincent and the Grenadines; Sudan*; Suriname; Sweden*; Switzerland*; Syria; Taiwan Province of China; Tanzania*; Thailand*; Timor‐Leste; Tonga; Trinidad and Tobago*; Tunisia*; Turkey*; Tuvalu; Uganda; Ukraine*; United Arab Emirates; United Kingdom*; United States*; Uruguay*; Vanuatu; Venezuela*; Vietnam; West Bank and Gaza; Yemen; Zambia*; Zimbabwe*

SARS (2003)

Australia*; Canada*; China*; France*; Germany*; Hong Kong SAR*; India*; Indonesia*; Ireland*; Italy*; Korea*; Kuwait*; Macao SAR*; Malaysia*; Mongolia*; New Zealand*; Philippines*; Romania*; Russia*; Singapore*; South Africa*; Spain*; Sweden*; Switzerland*; Taiwan Province of China; Thailand*; United Kingdom*; United States*; Vietnam

Hong Kong flu (1968)

Argentina*; Australia*; Brazil*; Bulgaria; Canada*; Chile*; China; Czech Republic; Finland; Germany*; Hong Kong SAR; Hungary; Iceland*; India*; Indonesia; Iran*; Japan*; Malaysia*; Netherlands*; New Zealand*; Philippines*; Poland; Singapore*; Slovak Republic; South Africa*; Sri Lanka*; Sweden*; Taiwan Province of China; Thailand*; United Kingdom*; United States*; Uruguay*; USSR (Armenia; Azerbaijan; Belarus; Estonia; Georgia; Kazakhstan; Kyrgyz Republic; Latvia; Lithuania; Moldova; Russia; Tajikistan; Turkmenistan; Ukraine; Uzbekistan); Vietnam

Note: Countries marked with *are included in the regression sample given availability of data on real GDP per capita, total factor productivity, capital stock per worker, and employment‐population ratio.

Sources: Incidence of the Hong Kong flu is taken from Cockburn et al. (1969), SARS from the World Health Organization (2003), H1N1 from flucount.org (2009), MERS from European Centre for Disease Prevention and Control (2015), Ebola from the World Health Organization (2016).

Barrett, P. , Das, S. , Magistretti, G. , Pugacheva, E. & Wingender, P. (2022) Long COVID? Prospects for economic scarring from the pandemic. Contemporary Economic Policy, 1–16. Available from: 10.1111/coep.12598