Has COVID-19 induced labor market mismatch? Evidence from the US and the UK

Abstract

This paper studies whether labor market mismatch played an important role for employment dynamics during the COVID-19 pandemic. We apply the framework of Şahin et al. (2014) to the US and the UK to measure misallocation between job seekers and vacancies across sectors until the fourth quarter of 2021. We find that mismatch rose sharply at the onset of the pandemic but returned to previous levels within a few quarters. This implies that, as of late 2021, COVID-19 has not set in motion a large wave of structural reallocation involving significant frictions in the matching process between workers and firms. Consequently, the total loss in employment caused by the rise in mismatch has been even smaller and less persistent during the COVID-19 pandemic than during the Global Financial Crisis. The results are robust to considering alternative definitions of job searchers and to using a measure of “effective” job seekers in each sector.

1. Introduction

The COVID-19 pandemic and the containment measures put in place by governments caused severe disruptions to labor markets all around the world during 2020 and 2021. In the US and the UK, two years after the beginning of the pandemic, labor demand has recovered and is exceptionally high, with vacancies well above early-2020 levels. However, employment has not yet fully recouped the losses from the first months of the pandemic. One possible explanation for this unusual coexistence of sluggish employment and tight labor markets is the heterogeneous impact of the COVID-19 shock, which may have generated significant misalignment between the sectors in which the jobless search for work–such as the hard-hit retail and hospitality industries–and those where most vacancies are–such as the ICT sector and those industries that have benefited from increased digitalization and shifting consumption patterns. This paper assesses the extent to which such mismatch has indeed played a significant role for US and UK labor market dynamics from the onset of COVID-19 until the fourth quarter of 2021.

The unprecedented impact of COVID-19 on labor markets has been the subject of research since very early on in the pandemic (Petrosky-Nadeau and Valletta, 2020). Since past recessions have been aligned with waves of structural transformation and reallocation of employment across sectors (Cortes, Jaimovich, Nekarda, Siu, 2020, Jaimovich, Siu, 2020), some studies have suggested that a similar process could take place in the aftermath of COVID-19 (Barrero, Bloom, Davis, 2020, Basso, Boeri, Caiumi, Paccagnella, 2020). Lending support to this hypothesis, in both the US and the UK, several studies highlight the large disparities in the vulnerability of different sectors and demographic groups to the pandemic and containment measures (Adams-Prassl, Boneva, Golin, Rauh, 2020, Albanesi, Kim, 2021, Cortes, Forsythe, 2020, Cribb, Waters, Wernham, Xu, 2021, Powell, Francis-Devine, Shibata, 2021). Salient dimensions of heterogeneity across sectors during the pandemic, which may portend longer-term shifts, have been the ability to work remotely (Dingel and Neiman, 2020), and the need for in-person interaction (Famiglietti, Leibovici, Santacreu, 2020, Kaplan, Moll, Violante, 2020). However, two years after the pandemic began, it remains unclear how persistent heterogeneity along these characteristics is, whether such shifts have in fact taken place on a large scale, and whether the labor force was able to adjust smoothly to them.

Structural reallocation often entails a period of misalignment between labor supply and labor demand across sectors, which would in turn increase frictions in the process of matching workers with firms. In this paper, we thus examine: (i) whether COVID-19 has generated labor market mismatch, in particular in comparison to the 2008–2009 Global Financial Crisis (GFC), and (ii) to what extent mismatch can explain the coexistence of tight labor markets and sluggish employment recoveries as of 2021Q4 in the US and the UK.

To this end, we apply and extend the approach proposed by Şahin et al. (2014) to measure labor market mismatch and its contribution to employment dynamics in the two countries since the beginning of COVID-19. The framework is intuitive and lends itself well to inspecting labor market developments in the aftermath of the pandemic. The resulting mismatch index reports the fraction of hires that are foregone due to misalignment in the distribution of searchers and vacancies. Job creation would be impaired if the unemployed mostly searched for work in shrinking industries while vacancies in growing sectors remained unfilled. For COVID-19, this could be the case if, for instance, the majority of workers are laid off from contact-intensive jobs while jobs with greater ability to work remotely expand, but workers fail to transition smoothly from the former to the latter.

We extend the framework of Şahin et al. (2014) in several directions that are salient for the COVID-19 recession. First, we compute the baseline measure of mismatch until late 2021, which allows us to compare the developments ensuing the COVID-19 pandemic to the aftermath of the GFC. Second, we consider COVID-specific aspects of heterogeneity by computing mismatch when grouping sectors according to their ability to work remotely and their contact intensity. Third, given the large outflows from the labor force witnessed in the first months of the pandemic, we quantify the implication of mismatch for the employment rate rather than just the unemployment rate. Fourth, for the US we exclude the temporary-layoff unemployed from our baseline measure of job seekers, as their search behavior differs markedly from that of the permanent-layoff unemployed. Finally, motivated by the peculiarities of the COVID-19 recession, such as the unprecedented rise in temporary layoffs in the US and the millions of workers covered by the government’s job protection scheme in the UK, we compute mismatch considering a broad set of alternative pools of job seekers.

Our main result is that, while mismatch grew sharply at the onset of COVID-19 in both the US and the UK, this rise was shorter-lived and, in the case of the US, smaller than during the GFC. Consequently, the employment loss due to the rise in mismatch was smaller during the COVID-19 crisis than during the GFC in both countries. Moreover, we find that mismatch across a broad aggregation of sectors grouped by teleworkability and contact intensity does not overturn this result and, somewhat surprisingly, we find that under this alternative grouping mismatch did not rise in the UK. These results are robust to considering alternative pools of job seekers, such as adding marginally attached or furloughed workers or including temporarily laid-off workers, and also to computing “effective searchers” to account for the possibility that the unemployed may search beyond their original industries. With regards to the second question, on the extent to which labor market mismatch can explain the coexistence of tight labor markets and sluggish employment recoveries as of 2021Q4, we find that sectoral mismatch accounts for about 10 percent of the existing employment gap as of late 2021 in both countries.1

This finding suggests that, at least over 2020–2021, COVID-19 did not set in motion a large wave of structural reallocation involving significant frictions in the matching process between workers and firms. Therefore, the strong heterogeneity in the initial exposure of different sectors to the pandemic likely resulted primarily from the short-run impact of the lockdown measures and contagion risks. As restrictions to economic activity and health concerns receded, labor demand recovered, including in hard-hit industries, and its sectoral composition broadly returned to that of the pre-pandemic period. Reflecting this, those sectors with relatively high vacancy postings by the second half of 2021 turned out to be also those with relatively high numbers of job seekers. This stands in contrast with the GFC, where the downturn was followed by a progressive but eventually persistent contraction of the most affected sectors (manufacturing and construction) and a rise in long-term unemployment for displaced workers.

Absent a persistent rise in mismatch, other forces must be slowing the employment recovery, most likely by dampening labor supply. As discussed by recent studies, candidate explanations with empirical support include a persistent rise in inactivity for older workers (Faria e Castro, 2021, Coibion, Gorodnichenko, Weber, 2020), the increased childcare duties falling on mothers of young children (Albanesi, Kim, 2021, Bluedorn, Caselli, Hansen, Shibata, Tavares, 2021, Fabrizio, Gomes, Tavares, 2021, Furman, Kearney, Powell, 2021), and demands for higher pay and better working conditions particularly for workers in low-wage occupations.2 Our paper contributes to this debate by finding that about 10 percent of the sluggish labor market recovery is attributed to the mismatch as of late 2021.

The rationale for focusing on the US and the UK is two-fold. First, worker-level microdata and series on vacancies by sector are available for both countries with only a short lag, allowing for granular and timely analysis of labor market developments. Second, despite having broadly comparable economic and demographic characteristics, these countries differed substantially in the magnitude of the employment contraction during the first quarters of COVID-19, as shown in Fig. 1 . In the US (left panel), the employment-to-working age population ratio fell by ten percentage points (p.p.) between January and April 2020. In the UK (right panel), the employment fall was more gradual, reaching a maximum of 2 p.p. in 2021Q1 relative to 2019Q4. At least in part, the widely different labor market policies implemented during the pandemic, particularly the greater reliance on job retention schemes in the UK, underpin this difference in employment dynamics. With regard to labor market tightness, however, by the second half of 2021, the US and the UK found themselves in very similar situations: In both countries, after a sharp fall early in the pandemic, the vacancies-to-unemployment ratio, also known as labor market tightness, rose above its pre-COVID level. For the US, this pattern holds true regardless of whether unemployed persons on temporary layoff are excluded or included in the measure of the pool of unemployed. Although labor market tightness has declined more sharply once we include temporary layoffs in our measure of the unemployment pool, both measures visibly rose above their pre-COVID levels by mid-2021. Meanwhile, despite this strong recovery in labor demand, employment growth slowed substantially by the beginning of 2021, leaving an employment rate gap vis-à-vis pre-COVID levels.

Fig. 1.

Employment-to-population and vacancies-to-unemployment ratios. Note: The solid blue line reports the employment-to-population ratio. The dashed red line reports the baseline vacancies-to-unemployment ratio, which for the US is computed excluding temporary-layoff workers. For the US, the long-dashed green line reports the vacancies-to-unemployment ratio using the full measure of unemployment. Sources: JOLTS, US CPS, ONS, UK LFS, and authors’ calculations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Through this country comparison, our analysis also sheds some light on the appropriateness of policies enacted by governments both to contain the initial employment contraction and to support the ensuing recovery. For instance, in the UK the furlough scheme muted the rise in job destruction and thus diminished the number of job seekers. In the US, while the initial spike in unemployment was predominantly due to temporary layoffs (Hall and Kudlyak, 2022), permanent layoffs also grew and remained high throughout 2020. As the number of temporary-layoffs sharply declined starting in May 2020, the share of unemployment consisting of permanent job losers steadily rose over time, presaging long-lasting scars along the lines of previous “jobless recoveries” caused by waves of structural reallocation.3 Despite the different policy approaches, the short-lived nature of the rise in mismatch may have been a contributing factor to the initially quick employment rebound. Moreover, the slowing down of the employment recovery throughout 2021 and the emergence of acute worker shortages in some occupations –like truck drivers– spurred a debate over the need for policies to encourage jobless workers to broaden the set of sectors in which to seek employment.4 However, the concurrently low levels of mismatch partly call into question the need for policies focused on labor reallocation relative to more neutral policies supporting higher labor supply.

Our paper directly contributes to the study of sectoral reallocation in the aftermath of downturns in advanced economies, with a focus on the COVID-19 recession. Cortes et al. (2020) and Jaimovich and Siu (2020) show that the GFC accelerated the decline in manufacturing and clerical jobs in the US, which in turn created a “jobless recovery” as displaced workers were either substituted by labor-saving capital or could not smoothly transition into other sectors. Proposing a new methodology to measure labor market mismatch, the seminal study of Şahin et al. (2014) found that higher mismatch due to sectoral and occupational reallocation accounted for up to one-third of the rise in the unemployment rate after the GFC. Their framework was also applied to the UK by Patterson et al. (2016), focusing on industries, and Turrell et al. (2021), focusing on occupations. Studying the COVID-19 pandemic through the same approach provides a useful point of comparison with the GFC. We thus contribute to this strand of research by showing that, at least by late 2021, COVID-19 had not triggered as dramatic a structural transformation of the labor market as the GFC did. Although labor demand in certain teleworkable industries rose, this increase did not appear to be large enough to cause major frictions in aggregate job creation.

Finally, our work adds to the large number of studies on labor market developments during the pandemic. A non-exhaustive list of those focusing closely on the issue of heterogeneity across demographic groups, sectors, and occupations in the US includes Adams-Prassl et al. (2020); Albanesi and Kim (2021); Coibion et al. (2020a); Cortes and Forsythe (2020); Shibata (2021). Prominent works on the UK include Adams-Prassl et al. (2020); Carrillo-Tudela et al. (2022); Cribb et al. (2021); Görtz et al. (2021); Powell and Francis-Devine (2021). Our paper is closest to Carrillo-Tudela et al. (2022), who inspect job search patterns across industries and occupations using the UK Household Longitudinal Study. They find that job search during COVID-19 broadly shifted towards less-affected occupations and, to a lesser extent, industries. However, workers from contracting sectors were less likely to target and transition to growing ones, thus suggesting that structural reallocation on the back of the pandemic ultimately remained limited.

The rest of this paper is structured as follows. Section 2 describes the data sources we use for the analysis. Section 3 motivates the work through descriptive evidence on the presence of mismatch after the start of COVID-19. Section 4 briefly describes the mismatch framework. Sections 5 and 6 present the main results and the extensions. Section 7 concludes.

2. Data

This section briefly describes the data used for the analysis.

US We use the Current Population Survey (CPS), a nationally representative survey, to calculate the stock of employed and unemployed workers by industry at a monthly frequency between January 2003 and December 2021. We also calculate transition rates between labor market states between two consecutive months using the panel dimension of the CPS. We use the Job Openings and Labor Turnover Survey (JOLTS) data on vacancies and hires for 17 industries based on the North American Industry Classification System (NAICS). Our baseline measure of unemployed persons for the US excludes those who are on temporary layoff. As discussed by Forsythe, Kahn, Lange, Wiczer, 2020, Forsythe, Kahn, Lange, Wiczer, 2022; Hall and Kudlyak (2022), the number of temporary-layoff unemployed workers rose sharply during the first months COVID-19. By definition, temporary-layoff unemployed workers expect to resume their jobs within the foreseeable future and are therefore less likely to be active job seekers. In Section 6, where we consider alternative definitions of searchers, we include temporarily laid off workers in the pool of jobs seekers. For these extensions to the baseline mismatch results, we also calculate corresponding stock and flow variables for individuals that were not in the labor force (NLF) –also known as the inactive–, marginally attached, and inactive for less than a month.5

UK The main data source for the UK is the worker-level quarterly Labour Force Survey (LFS) from 2002Q1 to 2021Q4, in its 2-quarter longitudinal format. This survey is used to obtain the stocks of employed, unemployed, and inactive individuals by industry, as well as the worker flows across labor force states and industries over two quarters. Through other questions asked in the survey, we also derive the stocks and job finding rates of marginally attached workers, those inactive, on-the-job searchers, and furloughed workers. The survey includes a breakdown of industries through the UK 2007 Standard Industrial Classification (SIC 2007), which contains 21 sectors. The Office of National Statistics (ONS) also provides a series of vacancies using the same classification for 18 of these industries over the same time period.6

3. The sectoral dimension of the COVID-19 pandemic

The COVID-19 pandemic and the containment measures enacted by governments constituted a combination of supply-side and demand-side shocks with major heterogeneity and complex spillovers across sectors (Alfaro, Becerra, Eslava, 2020, Guerrieri, Lorenzoni, Straub, Werning, 2020). On the one hand, lockdown mandates fully or partially impeded economic activity in specific industries. On the other hand, fear of contagion directly reduced demand for specific products and services (such as dining out or travel). Ultimately, as amply discussed in numerous studies, a sector’s exposure to the COVID-19 shock was strongly determined by the intensity of person-to-person contacts (Famiglietti et al., 2020) and the ability to perform tasks remotely -also called teleworkability (Dingel and Neiman, 2020). Finally, the asymmetric disruption caused by the pandemic may have set in motion long-term structural adjustments in the economy via several channels. For instance, demand for certain products and services may have fallen or risen permanently. On the production side, firms in certain sectors may have invested in labor-saving technologies, thus decreasing demand for workers.

The heterogeneous nature of the COVID-19 shock can be readily seen through its impact on job destruction across industries during the first months of the pandemic. The dark blue bars in Fig. 2 show the average separation rate for each sector between 2010 and 2019.7 The light grey bars show instead the maximum value of the separation rate during 2020. In both countries, the separation rate rose much more sharply in some industries than in others. The hotel and restaurant, entertainment, and retail trade sectors were among those with the largest increase in separations. Furthermore, the overall rise in job destruction was significantly larger in the US than in the UK, a fact that underpins the milder contraction of employment in the latter during 2020 seen in Fig. 1.

Fig. 2.

Separation rates by industry. Note: The separation rate represents the probability of transitioning from employment to unemployment between two adjacent months (quarters) for the US (UK). The dark blue bars report the average separation rate over 2010–2019. The light grey bar report the maximum separation rate during 2020. Sources: US CPS, UK LFS, and authors’ calculations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The asymmetric nature of the COVID-19 shock may in turn lead to mismatch if workers who were separated from their jobs face limited opportunities of re-employment in comparable jobs due to a shift in labor demand towards other sectors and occupations. At the onset of the COVID-19 crisis, both the US and the UK experienced a misalignment in the composition of labor supply and labor demand. Fig. 3 shows that the correlation between the shares of vacancies and the shares of unemployment across industries fell sharply in early 2020.8 In the US, the contraction was smaller and less persistent than during the GFC. In the UK, the correlation fell more than during the GFC before recovering fast, suggesting a short-lived period of high misallocation.9

Fig. 3.

Correlation of vacancy share and unemployment share. Note: The figure plots the correlation between vacancies shares and unemployment shares across 17 (18) industries at monthly (quarterly) frequency for the US (UK). Sources: JOLTS, US CPS, ONS, UK LFS, and authors’ calculations.

While our analysis focuses on mismatch across industries, workers’ occupations also constituted a salient dimension of heterogeneity in their exposure to the pandemic (Carrillo-Tudela, Comunello, Clymo, Jäckle, Visschers, Zentler-Munro, 2022, Cortes, Forsythe, 2020). Appendix Fig. C.1 shows that the increase in separation rates was also heterogeneous across occupations. Industry and occupation capture different aspects of jobs. While industries classify the ultimate product or service of a job, occupations are defined by the tasks directly performed and the workers’ skills. Since both dimensions were individually relevant for the likelihood of finding a job or of not losing one during the pandemic, measuring mismatch across both industries and occupations would present the most exhaustive analysis of job seekers’ misallocation.10 Due to lack of a long enough time series of vacancies by occupation for the UK, however, in this work we only consider industries to better focus the analysis on comparing mismatch dynamics during the GFC and COVID-19. Appendix C provides a more tentative analysis of occupations for both countries during the pandemic using data collected by Indeed, a large online job posting platform.

Fig. C.1.

Separation rates by occupation. Note: The separation rate shows the probability of transitioning from employment to unemployment between two adjacent months (quarters) for the US (UK). The dark blue bars report the average separation rate over 2010-–2019. The light grey bar report the maximum separation rate during 2020. Occupations are defined based on the 2-digit levels of the US SOC2010 and UK SOC2010, respectively. Sources: US CPS, UK LFS, and authors’ calculations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Framework

This section briefly outlines the framework proposed by Şahin et al. (2014) to measure mismatch between vacancies and job seekers. Our departures from the original framework are the focus on the impact of mismatch on employment, rather than unemployment, and the introduction of inactivity as an additional labor market state. As discussed below, this addition allows for flexibility in adjusting the framework to alternative definitions of job seekers. However, it does not alter the nature of the baseline framework in which unemployed workers are assumed to be the only job seekers.11

General Environment Time is discrete. The economy is formed by a finite number of discrete sectors (industries) indexed by $i=1,\cdots ,\mathbb{I}$. In each period $t$, a unit mass of workers are either employed in a sector ($e_{it}$), unemployed and searching for jobs uniquely in the sector ($u_{it}$), or inactive and not searching ($n_t$), such that $n_t+{\sum }_{i=1}^{\mathbb{I}}(e_{it}+u_{it})=1$.

Firms in each sector post vacancies ($v_{it}$), which can be filled with job seekers through a frictional process. The number of hires ($h_{it}$) resulting from the matching of vacancies and searchers is determined by the matching function $h_{it}={\Phi }_t{\varphi }_{i}m(v_{it},u_{it})={\Phi }_t{\varphi }_i{v}_{it}^{\eta }{u}_{it}^{1-\eta }$ where ${\Phi }_t$ is aggregate matching efficiency, $\eta \in (0,1)$ is the elasticity parameter, constant across sectors, ${\varphi }_i$ is sector-specific matching efficiency that is constant over time. Total hires are the sum of hires across the sectors, and can be expressed as follows:

$$h_t={\Phi }_t{v}_t^{\eta }{u}_t^{1-\eta }\left[\sum _{i=1}^I{\varphi }_i{\left(\frac{v_{it}}{v_t}\right)}^{\eta }{\left(\frac{u_{it}}{u_t}\right)}^{1-\eta }\right]$$ (1)

Planner’s solution Taking the allocation of vacancies ${\left\{v_{it}\right\}}_{i=1}^{\mathbb{I}}$, the total number of job seekers $u_t$, and industry-specific matching efficiencies ${\left\{{\varphi }_i\right\}}_{i=1}^{\mathbb{I}}$ as exogenous, the social planner’s optimal solution maximizes total hires ($h_t^{*}$) by allocating job seekers across sectors to equalize their marginal contribution to total job creation. In other words, the planner chooses ${\left\{u_{it}^{*}\right\}}_{i=1}^{\mathbb{I}}$ such that

$${\varphi }_1{m}_u(u_{1t}^{*},v_{1t})=\cdots ={\varphi }_i{m}_u(u_{it}^{*},v_{it})=\cdots ={\varphi }_{\mathbb{I}}{m}_u(u_{\mathbb{I}t}^{*},v_{\mathbb{I}t}),$$ (2)

subject to ${\sum }_{i=1}^{\mathbb{I}}{u}_{it}^{*}=u_t$, where $m_u(.)$ is the derivative of the matching function with respect to unemployment. This condition is equivalent to equalizing the labor market tightness ${\theta }_{it}^{*}=v_{it}/u_{it}^{*}$, weighted by matching efficiencies ${\varphi }_i$, across sectors.

Given the optimal allocation of job seekers, ${\left\{u_{it}^{*}\right\}}_{i=1}^{\mathbb{I}}$, in Eq.  (2) and using the expression of total hires in the economy in Eq.  (1), the total optimal hires in the economy can be expressed as

$$h_t^{*}={\Phi }_t{v}_t^{\eta }{u}_t^{1-\eta }\left[\sum _{i=1}^I{\varphi }_i{\left(\frac{v_{it}}{v_t}\right)}^{\eta }{\left(\frac{u_{it}^{*}}{u_t}\right)}^{1-\eta }\right],$$ (3)

Mismatch Index Given an optimal allocation ${\left\{u_{it}^{*}\right\}}_{i=1}^{\mathbb{I}}$ and an observed actual allocation ${\left\{u_{it}\right\}}_{i=1}^{\mathbb{I}}$, the level of mismatch can be quantified as the percent deviation of total observed hires, $h_t$ in (1), from the total optimal hires in the economy, $h_t^{*}$ in (3):

$${\mathcal{M}}_t=1-\frac{h_t}{h_t^{*}}=1-\sum _{i=1}^{\mathbb{I}}\left(\frac{{\varphi }_i}{\overline{{\varphi }_t}}\right)\left(\frac{v_{it}}{v_t}\right){}^{\eta }\left(\frac{u_{it}}{u_t}\right){}^{1-\eta },$$ (4)

where $v_t={\sum }_{i=1}^{\mathbb{I}}{v}_{it}$ and $\overline{{\varphi }_t}=\left[{\sum }_{i=1}^{\mathbb{I}}{\varphi }_i^{\frac{1}{\eta }}\left(\frac{v_{it}}{v_t}\right)\right]{}^{\eta }$. The mismatch index in Eq.  (4) can be interpreted as the fraction of hires that are lost due to misallocation relative to the optimal level of hires $h_t^{*}$. This mismatch index captures the additional job creation that would be generated in the economy if a social planner could optimally allocate job seekers across different sector-specific labor markets given the distribution of vacancies and matching efficiencies, as well as the total number of job seekers.

Employment loss due to mismatch To assess the economic significance of mismatch, a counterfactual employment series $e_t^{*}$ is constructed. By rearranging Eqs.  (1) and (4), total hires can be also expressed as $h_t=(1-{\mathcal{M}}_t)\overline{{\varphi }_t}{\Phi }_t{v}_t^{\eta }{u}_t^{1-\eta }$. Then, the actual job finding rate in the economy can be expressed as

$$f_t=\frac{h_t}{u_t}=(1-{\mathcal{M}}_t){\overline{\varphi }}_t{\Phi }_t{\left(\frac{v_t}{u_t}\right)}^{\eta }.$$ (5)

and the optimal job finding rate, $f_t^{*}$, can be expressed as

$$f_t^{*}=\overline{{\varphi }_t}{\Phi }_t{\left(\frac{v_t}{u_t^{*}}\right)}^{\eta }=f_t\frac{1}{1-{\mathcal{M}}_t}{\left(\frac{u_t}{u_t^{*}}\right)}^{\eta }.$$ (6)

The counterfactual job finding rate, $f_t^{*}$, is higher than the observed job finding rate, $f_t$, through two channels. The first channel is the direct positive impact of mismatch on the job finding rate, $\frac{1}{1-{\mathcal{M}}_t}>1$, which arises from a social planner efficiently reallocating available job seekers $u_t$ in the economy across sectors. The second channel, an indirect one, is the feedback effect arising from lower unemployment in the absence of mismatch. The compounded effect of greater job creation in past periods (i.e. $f_s^{*}>f_s,\forall s<t$) implies that $u_t^{*}<u_t$, entailing fewer job seekers for the same number of vacancies and hence higher chances for any job seekers to meet a vacancy independent of the current level of mismatch.

Under the assumption of no mismatch at all times, starting from an initial period $e_0^{*}=e_0$, the series $e_t^{*}$ and the companion series $u_t^{*}$ and $n_t^{*}$ are computed using the laws of motion for each labor market state, as reported in Appendix B.1, where the only change compared to $e_t,u_t,$ and $n_t$ is the job finding rate for job seekers $f_t^{*}=h_t^{*}/u_t^{*}$ instead of the actual job rate $f_t=h_t/u_t$. Transition rates across all other labor market states, including the separation rate from employment to unemployment and inactivity, the transitions from (to) unemployment to (from) inactivity, and from inactivity to employment are maintained equal to the empirical ones. Appendix B.1 contains the full laws of motion and explains how the framework is adjusted to capture alternative definitions of the pool of job seekers.

Once a counterfactual employment series free of mismatch ($e_t^{*}$) is constructed, the employment loss due to mismatch is computed as $(e_t^{*}-e_t)$, representing the deviation of the no-mismatch counterfactual employment rate from its empirical counterpart in percent of the total working age population.12 While $f_t$ is a purely contemporaneous loss in hires at time $t$ only, $(e_t^{*}-e_t)$ is history-dependent, as the employment loss at time $t$ reflects the impact of mismatch from all previous periods propagated through the laws of motion of employment and unemployment.

Estimation For the US, we compute mismatch at monthly frequency from January 2003 to December 2021 on 17 industries. For the UK, we compute mismatch at quarterly frequency from 2002 Q1 until 2021Q4 on 18 industries. For both countries, we follow Şahin et al. (2014) in estimating the sector-specific matching efficiencies ${\varphi }_i$’s through a pooled regression of hires on vacancies and unemployment at the sector level on the pre-GFC period.13 For the computation of (4), we assume $\eta =0.5$ as in the original paper, a value that is also conventionally used in the calibration of theoretical models.

5. Main results: Mismatch during COVID-19

In this section, we present our main findings on mismatch and its contribution to employment dynamics during the pandemic. To contextualize the magnitude of these results, we compare them to mismatch dynamics following the GFC.

Fig. 4 presents our baseline results for the US and the UK in the left and right panels, respectively. The solid blue lines report the mismatch index. Although mismatch rose sharply during the early phase of the COVID-19 crisis in both countries, the spike in the index was short-lived. By September 2021 the index had returned to pre-COVID levels. Comparisons with the GFC period are also insightful to understand the dynamics of mismatch. In the UK, the index reached its highest historical value during the COVID-19 spike, while in the US the peak of the index during the GFC was higher than during COVID-19. Moreover, in both countries the rise of mismatch during the GFC was followed by a more gradual decline than during COVID-19, suggesting more persistent heterogeneity across sectors in the recovery from the GFC.

Fig. 4.

Mismatch index and employment loss due to mismatch. Note: The figure plots the mismatch index (solid blue line) and the resulting employment loss (dashed red line, right y-axis). Results are based on 17 and 18 industries for the US and UK, respectively. Only unemployed workers are included in the pool of job searchers. For the US, this baseline mismatch index excludes unemployed persons who are on temporary layoff. The mismatch index is bounded below and above by 0 and 1. Higher values imply a higher degree of mismatch. The employment loss is reported as a share of the working age population in percentage points. Sources: JOLTS, US CPS, ONS, UK LFS, and authors’ calculations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The dashed lines in Fig. 4 report the employment loss $(e_t^{*}-e_t)$ due to mismatch in percent of the total working age population. Greater mismatch at the onset of COVID-19 implied fewer hires from unemployment and, as a result, a widening employment rate gap vis-à-vis the counterfactual $e_t^{*}$. However, in both the US and the UK, the rise in employment loss from mismatch, although steep, was smaller during COVID-19 than during the GFC.14 ^, 15

In B2, Table B1 quantifies the contribution of mismatch to the loss of employment, comparing the pandemic to the GFC. The table more formally confirms that at the trough of the COVID-19 recession mismatch accounted for a smaller employment loss compared to the trough of the previous downturn. At the trough of the COVID-19 recession and by 2021Q4, mismatch explained 20 and 11 percent of the fall in employment in the US, respectively, and 10 and 13 percent in the UK.16

The finding that mismatch was quantitatively less important during COVID-19 than during the GFC applies to both the US and the UK, but the underlying reasons partially differ. In the US, the rise in mismatch was visibly smaller and more transitory than during the GFC. In the UK, even though the rise in mismatch was unprecedented, it was also very short-lived. Moreover, the low rates of job destruction (Fig. 2) limited the immediate rise of unemployment at the onset of the pandemic. Hence, despite the large spike in mismatch, the very contained number of job seekers meant that there was little scope for mismatch to play a quantitatively important role in aggregate employment dynamics.

Which sectors were the main drivers of aggregate-level mismatch in the two downturns? Fig. 5 shows the deviation of the historical unemployment ($u_{it}$) level in each sector and the period-specific optimal unemployment allocated by the social planner ($u_{it}^{*}$) in each period. Positive deviations imply that for a given industry $i$ unemployment at time $t$ is higher than the optimal number of job seekers. In both the US and the UK, after the GFC the main industries with excess job seekers where manufacturing and construction, while during COVID-19 it was the hotel and restaurant sector. Professional services in the US and health and social services in the UK were the main sectors with an under-supply of job seekers during both downturns. From the figure, it is also visible how the excess supply of construction and manufacturing workers persisted from the beginning of the GFC until the mid-2010s. Meanwhile, the excess supplies and demands during COVID-19 reverted to 2019 values or below within a few quarters.17

Fig. 5.

Total deviations of $u_{it}$ from $u_{it}^{*}$. Note: Each shaded area reports the deviation of the unemployment level for industry $i$ at time $t$ from the social planner’s optimal allocation of job seekers, ($u_{it}-u_{it}^{*}$). A positive (negative) value represents an excess (shortage) of job seekers relative to the optimal allocation. The figure reports individual series for the most quantitatively relevant industries, while the Other-Positive and Other-Negative series combine all the other industries with positive and negative deviations, respectively. Sources: US CPS, JOLTS, ONS, UK LFS, and authors’ calculations.

5.1. Did teleworkability and contact intensity matter?

As discussed earlier, job characteristics such as teleworkability and contact intensity were key determinants of sectors’ exposure to disruptions during the pandemic. We thus ask how salient these sectoral characteristics were for the transitory spike in mismatch of 2020–2021. In other words, was there significant misalignment between labor supply and demand across, say, teleworkable and non-teleworkable sectors as a result of the pandemic?

The baseline specification of mismatch assumes the unemployed search in the sectors where they previously worked. However, some skills relevant for one sector may be easily applicable to other industries, thus broadening the scope of workers' potential job search. For instance, given the increased need for remote work, a job seeker with IT skills acquired in one teleworkable industry could have become a relatively more competitive candidate also in other industries with high ability to telework. Similarly, job seekers with strong abilities for in-person interactions may have seen a drop in demand for those skills in the broader set of industries with high contact intensity. On the other hand, if not all industries with high contact intensity saw a drop in labor demand (e.g., a sustained need for health and home care workers), then the ability of workers to relocate across industries with similar skill contents could have mitigated labor market mismatch.

To examine this issue, we combine the individual industries into 4 groups based on their degree of teleworkability and contact-intensity.18 We then estimate the mismatch index across these four groups, which is plotted in Fig. 6 . While, as explained in Şahin et al. (2014), the level of the index is not directly comparable across sectoral aggregations, the focus of our analysis is on the fluctuations of the index over time within this broader grouping of sectors. In particular, a smaller rise in the index during COVID-19 under this alternative grouping compared to the baseline index would suggest that accounting for the transferability of key skills across sectors further mitigates the quantitative importance of mismatch during the COVID-19 downturn. Such result would suggest that at least some job seekers who faced lower job prospects in their original sectors, such as contact-intensive ones, did benefit from an offset in demand in other sectors with similar production characteristics.

Fig. 6.

Mismatch across teleworkable and contact-intensive industries. Note: The figure reports the mismatch index for the “baseline” cases (dashed grey line) based on 17 (18) industries for the US (UK) and “Contact. int. $\times$ Tele.” version (solid blue line) based on four groups of industries comprising teleworkable $\times$ contact-intensive industries. Only unemployed workers are included in the pool of job searchers. The mismatch index represents the fraction of hires lost due to misallocation between job seekers and vacancies and is bounded below and above by 0 and 1. Higher values imply a higher degree of mismatch. Sources: JOLTS, US CPS, ONS, UK LFS, and authors’ calculations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

In the US (Fig. 6, left panel), mismatch across the teleworkability and contact intensity dimensions (solid line) is lower than baseline mismatch (dashed line) throughout the period 2003–2021, over which it accounted for around 48 percent of the baseline 17-industry-based mismatch index, on average. The mismatch index based on teleworkability and contact intensity was still higher during the GFC than the COVID-19 crisis, but the difference between the two downturns shrinks based on this alternative measure, confirming a unique feature of the COVID-19 shock, namely its impact on in-person interactions. In the UK, the alternative mismatch index is close in value to the baseline index for most of the time sample but it did not rise during the GFC and during the pandemic.19 Overall, the findings suggest that misalignment in labor supply and demand across sectors based on teleworkability and contact intensity played at best a small role during the pandemic, despite these dimensions being unique features of this crisis. Moreover, they do not overturn our main result that employment loss due to mismatch was larger during the GFC than during the COVID-19 crisis.

Although our findings may at first glance seem at odds with evidence on the increasing frequency of remote work, they are actually aligned with recent studies. For instance, Adrjan et al. (2021) show that the possibility to telework is increasingly mentioned in the job descriptions of newly posted online vacancies in many advanced economies, including the US and the UK. However, they find that the rise is almost entirely accounted for by increases in advertised telework within industries rather than by a shift in vacancies towards sectors with high teleworkability. Moreover, sectors with greater ex ante potential for remote work are those experiencing the largest rise in advertised telework. Hence, this process is not likely to generate sectoral mismatch, since it does not entail a shift in the sectoral composition of labor demand.20

6. Extensions and robustness checks

In this section, we present a series of extensions to the baseline results. First, we consider the sensitivity of mismatch to alternative measures of job seekers. Second, we allow for the possibility that the unemployed may be searching in industries other than their previous ones.

6.1. Alternative pools of job searchers

We consider how the baseline measures of mismatch and estimates of employment loss change under alternative definitions of job seekers. The unemployed are not the only ones in the labor market competing for new jobs, although they may do so more intensely than other workers. If the amount of other job seekers –such as those already employed or those not actively searching– vary over time and their sectoral composition differs from that of the unemployed, the baseline estimate would be an incorrect measure of true mismatch.

Inspecting the robustness of our result to broader definitions of searchers is particularly important in the context of COVID-19 given the uncommon labor market flows observed ensuing the COVID-19 pandemic and the establishment of lockdown measures. In particular, flows into inactivity and other states of partial job search were markedly different from previous downturns. Appendix Fig. D.1 shows the transition probabilities from employment to unemployment, inactivity, and marginal attachment. We find that in both the US and the UK, employed workers were much more likely to transition into inactivity during the COVID-19 crisis than the GFC.

Fig. D.1.

Transition probabilities from employment to unemployment, inactivity, and marginal attachment. Note: The solid blue reports the probability rate that an employed worker transitions to unemployment between one period to the next. The long-dashed line reports the transition probability from employment to NLF. The short-dashed line reports the transition probability from employment to marginal attachment (a subset of inactivity). The period of analysis is one month for the US and one quarter for the UK. Sources: US CPS, UK LFS, and authors’ calculations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Additional peculiarities in the labor market adjustment during COVID-19 include the share of temporary layoffs in the US and the large reach of the furlough scheme in the UK. In the US, labor force participation dropped sharply (Coibion et al., 2020b), while an unprecedented fraction of unemployment consisted of “temporary layoffs” (Forsythe, Kahn, Lange, Wiczer, 2020, Shibata, 2021). In the UK, although the drops in employment and labor force participation were substantially milder, the government’s Coronavirus Job Retention Scheme (CJRS) -colloquially known as furlough- protected up to 8.8 million workers in April 2020 (close to 30 percent of employment) from the risk of joblessness (Appendix Fig. A.4).

Fig. A.4.

UK: Daily individual claims for the Coronavirus Job Retention Scheme (CJRS). Note: The plot shows the number of daily claims for the Coronavirus Job Retention Scheme in the UK. Sources: ONS and authors’ calculations.

These unique dynamics may have also entailed very different search behaviors for workers in different labor market states. For instance, it is possible that workers on temporary layoff in the US were effectively not searching for new employment in the anticipation that they would return to their previous jobs. Conversely, furloughed workers in the UK may have looked for other opportunities as they considered the risk that their current jobs might eventually disappear.21 Finally, in both countries, many inactive workers may have fallen into the category of the “marginally attached”: They were discouraged from actively looking for jobs because of the pandemic and the adverse macroeconomic conditions, but they may have been willing and able to take up a new job if the opportunity arose.22

6.1.1. Alternative definitions of job seekers for the US

For the US, we consider four broader definitions of the pool of job searchers. We first (i) include temporary layoffs in the unemployment pool, and then we add one at a time to the full pool of unemployed workers the following groups: (ii) marginally attached workers (iii) those not in labor force for less than 1 month (“NLF $<$ 1 month”), and (iv) all NLF workers.

The left panel of Fig. 7 plots how these different groups of workers evolved over time. The total unemployed pool (labelled “U” in blue) rose much more sharply during COVID-19 than during the GFC, but the increase was more short-lived: After its initial spike in April 2020, unemployment quickly declined. A unique feature of this recession is that the number of temporary-layoff unemployed workers rose to historically high levels and accounted for the majority of the unemployment pool in the early stages of the downturn. The share of temporary layoffs among the unemployed increased by almost 50 percentage points at the onset of the COVID-19 pandemic, from 28.7 percent in March 2020 to 78.2 percent in April 2020. If these workers expect a recall by their previous employers, they might not be actively searching for jobs. We thus excluded the temporary-layoff unemployed workers from the baseline measure of job seekers for the US. However, foreseeing the possibility of their employment hiatus turning into permanent unemployment, it is likely that some temporarily laid off workers also search for new jobs. Hence, in our first exercise, we add them to the pool of job seekers to compute mismatch.

Fig. 7.

US: Alternative groups of job seekers and correlation with vacancies. Note: In the left panel, “U”, “Temp. Layoffs”, “Marg. Att.”, “NLF $<$ 1 month”, and “NLF (RHS)” show the total number of unemployed, unemployed persons that are on temporary-layoffs, marginally attached workers, those that moved to inactivity (NLF) from employment within the past month, and the total NLF as a share of the working-age population. In the right panel, the series show the correlation between vacancies and shares of job seekers for the respective categories. “U” stands for the unemployed pool. “U-Temp. Layoffs” excludes those on temporary layoff from the unemployment pool. “U+Marg. Att.” is the unemployment pool plus marginally attached workers. “U+NLF < 1 month” is the sum of unemployed pool and those who have been inactive (NLF) for less than a month. “U+NLF” is the unemployment pool and the total number of people in inactivity. Sources: JOLTS, US CPS, and authors’ calculations.

As pointed out by Coibion et al. (2020a), COVID-19 also sparked large inflows into the NLF pool, with the non-participation rate increasing by around 7 percentage points between January and April 2020. Accordingly, the number of those who moved directly from employment to non-participation (“NLF $<$ 1 month”) increased to its historically highest level. Marginally attached workers also sharply rose at the onset of the COVID-19 recession.

The right panel of Fig. 7 shows the correlation between vacancies shares and unemployment shares obtained when using the alternative definitions of searchers. All the series except for the full NLF pool show very similar levels and fluctuations of the correlation between vacancies shares and unemployment shares. Once all NLF persons are included, the correlation is slightly higher and fluctuates less throughout the period, particularly during the GFC.

Fig. 8 plots the mismatch indices and the employment loss due to mismatch based on the baseline (top left panel, “Unemployed - Temporary Layoff”) and alternative groups of job searchers. The general pattern that mismatch was lower during COVID-19 than during the GFC holds true for all definitions of job searchers. Both the average level and the fluctuations of the alternative mismatch indices are comparable to the baseline, except for the entire NLF population.

Fig. 8.

US: Mismatch and employment loss for alternative groups of job seekers. Note: “Unemployed - Temp. Layoffs”, “Unemployed”, “Unemployed + Marg. Att.”, “Unemloyed + NLF $<$ 1 month”, and “Unemployed + NLF” show the results for the total number of unemployed excluding those that are on temporary-layoffs, including those on temporary-layoffs, adding one at a time marginally attached workers, those that moved to inactivity (NLF) from employment within the past month, and the total NLF. The solid line reports the mismatch index, while the dashed line reports the resulting employment loss. The mismatch index represents the fraction of hires lost due to misallocation between job seekers and vacancies and is bounded below and above by 0 and 1. Higher values imply a higher degree of mismatch. The employment loss is reported as a share of the working age population in percentage points. Sources: JOLTS, US CPS, and authors’ calculations.

The inclusion of temporary layoffs in the unemployment pool has two effects on labor market dynamics. On the one hand, mismatch increases, implying greater misalignment between vacancies and job seekers. On the other hand, the job finding probability of the full unemployment pool is higher, as those on temporary layoff typically have a higher job finding probability. Through the lens of Eq.  (6), the former channel increases the efficient job finding probability, $f_t^{*}$, relative to $f_t$ by increasing the mismatch index, ${\mathcal{M}}_t$, while the latter increases it in absolute terms due to a higher observed job finding probability, $f_t$, for the total unemployment pool. As a result, employment loss due to mismatch is lower than in the baseline.

The pools including, respectively, the full stock of NLF individuals, the short-term NLF, and the marginally attached show the nuances of considering different margins of inactivity. As the full stock of inactive workers is very large, short-term fluctuations in the flows out of employment into non-employment, such as those due to the business cycle or the pandemic, do not substantially change the sectoral composition of the pool of job seekers. Hence, fluctuations in the mismatch index and the resulting employment loss are significantly more muted than for the unemployment-only case. The difference between the GFC and the COVID-19 recession also becomes less stark.23 Interestingly, besides cyclical fluctuations, the long-run average of the employment loss over 2003–2019 is comparable to the baseline case despite mismatch having a slightly higher long-run average in the former. The job finding probability for inactive workers is only about a third that of the unemployed.24 This leaves less room for mismatch to play a significant role for employment dynamics even though the pool of workers affected by mismatch is now about five times larger. Overall, the NLF constitute a large and more “static” group of workers whose dynamics are less affected by sectoral mismatch with respect to both long-run levels and cyclical fluctuations. As shown by Fig. 8, the period of the COVID-19 recession was no exception.

Even though short-term NLF workers and the marginally attached are subsets of the inactive workers, these two groups are much smaller in size and their composition resembles more that of the unemployed. Hence, when adding these categories, the level and dynamics of the mismatch index are similar to the baseline case. Given the limited size of these two groups in the US, adding them to the pool of job seekers potentially affected by mismatch also does not alter the total employment loss substantially. As of 2021 Q4, the contribution of mismatch to the employment loss was 15 percent when adding the short-term NLF to the pool of job seekers and 11 percent when including the marginally attached.

Overall, considering alternative groups highlights the differences in the nature and business cycle dynamics of different job searchers but does not overturn the original finding that mismatch did not matter as much during the COVID-19 crisis as it did during the GFC in the US.

6.1.2. Alternative definitions of job seekers for the UK

For the UK, we consider five additions to the unemployed in computing the pool of job seekers: (i) marginally attached workers, (ii) inactive workers who have been jobless for less than a year (“NLF $<$ 12 months”), (iii) all inactive workers (“NLF”), (iv) on-the-job searchers (“OJS”), and (v) “furloughed” workers covered by the CJRS.25

The left panel of Fig. 9 shows how these groups of workers evolved since 2002. All series exhibit visible fluctuations following the beginning of COVID-19. Marginally attached and short-term inactive workers rose moderately, while inactive workers rose only mildly. On-the-job searchers contracted sharply for one quarter before returning to pre-pandemic levels. Finally, furloughed workers went from close to 0 to approximately 8 million within one quarter before declining gradually, with a second small spike in 2021Q1.26 The right panel of Fig. 9 reports the correlation of vacancies shares with the expanded definitions of job seekers. In most cases, this correlation is higher than when considering exclusively the unemployed and with more moderate dips during the GFC and COVID-19. The only exception is the “Unemployed + Furlough” group.

Fig. 9.

UK: Alternative groups of job seekers and correlation with vacancies Note: In the left panel, “U”, “Marg. Att.”, “NLF $<$ 12 months”, “OJS” “NLF (RHS)”, and “Furlough (RHS)” show the number of unemployed, marginally attached workers, those that moved to inactivity (NLF) from employment within the last 12 months, those that are engaged in on-the-job search, the total NLF, and furloughed workers as share of the working-age population. In the right panel, the series show the correlations between vacancies and shares of job seekers for the respective categories. “U” stands for the unemployed pool. “U+Marg. Att.” is the unemployment pool plus marginally attached workers. “U+NLF < 12 months” is the sum of unemployed pool and those who have been inactive (NLF) for less than a year. “U+NLF” is the unemployment pool and the total number of people in inactivity. “U+OJS” is the unemployment pool plus on-the-job searchers. “U+Furlough” is the unemployment pool plus furloughed workers. Sources: UK LFS, ONS, and authors’ calculations.

Fig. 10 reports mismatch and employment loss for the five expanded pools of job searchers, and compares them to the baseline case (top left panel). The average level and cyclical dynamics of mismatch differ across groups. In all cases, the average level of the mismatch index exhibits smaller fluctuations over time prior to COVID-19 compared to the “Unemployed” pool. At the onset of COVID-19, mismatch rose for all pools, although with varying relative magnitudes. The evolution of these alternative indices and the resulting employment losses during COVID-19 broadly reflect the fluctuations in the workers in each labor market status plotted in Fig. 9.

Fig. 10.

UK: Mismatch and employment loss for alternative groups of job seekers. Note: “Unemployed”, “Unemployed + OJS”, “Unemployed + Furloughed”, “Unemployed + Marginally Attached”, “Unemployed + NLF”, and “Unemployed + NLF $<$ 12 months” show results for the total number of unemployed, adding one at a time, on-the-job searchers, furloughed, marginally attached workers, those that moved to inactivity (NLF), those who moved to NLF from employment within the last 12 months, respectively. The solid line reports the mismatch index, while the dashed line reports the resulting employment loss. The mismatch index represents the fraction of hires lost due to misallocation between job seekers and vacancies and is bounded below and above by 0 and 1. Higher values imply a higher degree of mismatch. The employment loss is reported as a share of the working age population in percentage points. Sources: UK LFS, ONS, and authors’ calculations.

The results when including the NLF (“Unemployed + NLF” panel) paint a similar story as in the US. The influxes of workers into inactivity during the GFC and the pandemic are relatively small compared to the stock of incumbent inactive workers. Hence, the mismatch index and the resulting employment loss remain almost constant over the two downturns.27

This mitigation effect is smaller when considering the two subsets of the NLF group: the short-term NLF and the marginally attached. Short-term inactivity has historically been almost a-cyclical and between one half and two thirds the size of the unemployed pool. However, it also rose sharply in early 2020 and returned to historical levels by 2021. In this conjuncture, the choice not to search for a job may have been more closely related to the specific impact of the pandemic on a worker’s sector. Hence, mismatch rose sharply in early 2020 in a similar fashion to the unemployment-only case, while it had remained almost flat over the previous two decades. The contribution of mismatch to the employment loss in 2021Q4 was thus 12 percent when including the short-term NLF, which is close to the baseline case. In contrast, marginally attached individuals in the UK constitute a sizeable group, larger in size than the unemployed in most periods except 2009–2014. Interestingly, mismatch approximately doubles over 2002–2019 once the marginally attached are added, suggesting that their sectoral composition and the sector-specific matching efficiency parameters differ markedly from that of the unemployed. Consequently, throughout 2002–2022 the employment loss is about four times higher when adding the marginally attached than in the baseline case. However, similar to the NLF case, mismatch is also less susceptible to cyclical fluctuations and shows only a small uptick during COVID-19. Thus, by 2021Q4 the contribution of mismatch to the rise in employment during the pandemic was only 1 percent.

The “Unemployed+OJS” and “Unemployed+Furlough” groups show spikes in mismatch as large as the baseline. In the “Unemployed+Furlough” case, the spike is also more long-lasting, receding only around mid-2021. This path reflects the second smaller increase in the number of workers covered by the CJRS (Appendix Fig. A.4). The addition of furloughed workers deserves particular consideration for the discussion of labor market policies during the pandemic. While considering all workers covered by the CJRS as potential job seekers might be a strong assumption, some of them likely were. Hence, the fact that the “Unemployed + Furlough” mismatch index exhibits a more persistent rise suggests that the avoidance of job destruction via the CJRS also had potentially beneficial effects for job creation. However, even this alternative mismatch index shows that the spike during COVID-19 was short-lived, implying that these indirect benefits of the CJRS were likely small.

6.2. Effective searchers

A further extension of the baseline model, considered in the original work of Şahin et al. (2014), is the possibility that unemployed workers might search in sectors different from those where they previously worked. This would be very likely in the aftermath of COVID-19, given that some sectors were disproportionately affected by lockdown measures and the pandemic may have triggered a wave of permanent structural reallocation. For instance, workers laid off from the hotel and restaurant sector may have been looking for opportunities in non-contact intensive industries.

If job seekers try to switch sectors, the stock of unemployed who previously worked in a given industry may not be representative of the true extent of competition for jobs in that industry. Consequently, mismatch would also be erroneously measured. To address this issue, Şahin et al. (2014) propose a generalization of their framework where mismatch is computed considering “effective searchers” in each sector. Imposing the minimal assumption that workers who search in their original sector face a probability of finding a job in that sector that is higher by a factor $(1+{\gamma }_t)>1$ compared to industry switchers, Şahin et al. (2014) provide a method to recover the measure of unobserved effective searchers in each industry through empirical unemployment-to-employment flows across industries. These flows, which can be observed both in the US CPS and the UK LFS, partly reflect the (unobserved) measure of “industry switchers” among job seekers. In other words, observing the number of unemployed workers previously employed in industry $j$ who find a job in industry $i$ ($h_{jt}^i$), for all combinations of $j$ and $i$, provides an idea of how many workers from $j$ were really searching somewhere other than $j$, and of how many workers who were searching in $i$ in fact came from some other industry. Hence, effective searchers in $i$ can be defined as ${\widehat{u}}_{it}={\sum }_j{u}_{jt}^i$, where $u_{jt}^i$ is the measure of the industry-$j$ unemployed searching in $i$.28

6.2.1. Effective searchers in the US

Fig. 11 plots the sum of absolute deviations of effective searchers from the unemployed in each sector as a share of total unemployment for the US. This measure provides an idea of how many workers search in industries different from their past one (i.e., “switchers”).29 The series fluctuates over time around its mean of 0.053, implying that around 5 percent of job seekers actually search in a different sector. While the series sharply decreased at the onset of the COVID-19 recession, it soon bounced back, but not to its historical peak.

Fig. 11.

US: Fraction of prospective “switchers” among the unemployed when computing effective searchers by sector. Note: The figure plots the sum of absolute deviations of effective job seekers from the unemployed in each sector, as a share of all the unemployed workers. This series roughly translates as the fraction of unemployed who are searching in an industry different from their previous one. Sources: US CPS, and authors’ calculations.

The left panel of Fig. 12 plots the mismatch index based on effective searchers overlaid against the baseline index. We find that the mismatch index based on effective searchers is lower, implying that, once we account for the fact that some unemployed workers actually search in an industry different from their previous one, there is a lower degree of misalignment between vacancies and unemployment shares.

Fig. 12.

US: Mismatch and employment loss when computing effective searchers by sector. Note: The left and right panels show mismatch indices and corresponding employment losses based on the baseline (dashed line) and the alternative “effective” searchers (solid and long dashed line), respectively. The mismatch index represents the fraction of hires lost due to misallocation between job seekers and vacancies and is bounded below and above by 0 and 1. Higher values imply a higher degree of mismatch. The employment loss is reported as a share of the working age population in percentage points. Sources: JOLTS, US CPS, and authors’ calculations.

Lastly, the right panel of Fig. 12 plots the employment loss due to mismatch based on effective searchers against the baseline. Again, the employment loss due to mismatch is lower once we account for the fact that some unemployed workers search in another industry. In relative terms, the difference in employment loss due to mismatch between the GFC and COVID-19 is smaller based on effective searchers than under the baseline because, as discussed above, there were only limited switches of job searchers across industries at the beginning of the pandemic. However, our baseline result that the COVID-19 recession did not trigger as much mismatch and associated employment loss as the GFC is not overturned after accounting for effective searchers.

6.2.2. Effective searchers in the UK

Fig. 13 reports the sum of absolute deviations of effective job seekers from the unemployed in each sector, as a share of all unemployed workers, for the UK. Although the series fluctuates over time, with some short-lived spikes, its mean value is 0.12, implying that on average 12 percent of job seekers search in a different sector from their previous one. Importantly, after COVID-19 the fraction of “switchers” increased only mildly, suggesting no large-scale adjustment in the sectors that workers target.

Fig. 13.

UK: Fraction of prospective “switchers” among the unemployed when computing effective searchers by sector. Note: The figure plots the sum of absolute deviations of effective job seekers from the unemployed in each sector, as a share of all the unemployed workers. This series roughly translates as the fraction of unemployed who are searching in an industry different from their previous one. Sources: UK LFS, and authors’ calculations.

The left and right panels of Fig. 14 report mismatch and the employment loss computed using effective searchers, respectively, overlaid against the baseline results. Throughout the period 2002–2021, mismatch was lower than in the baseline case, suggesting that decisions to search in new sectors contribute to reducing mismatch, as job seekers attempt to switch towards sectors with higher vacancy-to-unemployment ratios. As a result, the employment loss due to mismatch is also lower compared to the baseline case. Despite the lower pre-COVID values, both mismatch and employment loss rose as much after COVID-19 as in the baseline case. Moreover, the employment loss up to 2021Q4 remains lower than during the GFC.

Fig. 14.

UK: Mismatch and employment loss when computing effective searchers by sector. Note: The left and right panels show mismatch indices and corresponding employment losses based on the baseline (dashed line) and the alternative “effective” searchers (solid and long-dashed line), respectively. The mismatch index represents the fraction of hires lost due to misallocation between job seekers and vacancies and is bounded below and above by 0 and 1. Higher values imply a higher degree of mismatch. The employment loss is reported as a share of the working age population in percentage points. Sources: UK LFS, ONS, and authors’ calculations.

Overall, the main findings are robust to measuring mismatch using an estimate of effective searchers in each sector. In particular, the fact that mismatch rose sharply after COVID-19 suggests that workers did not shift their targeted sectors in large numbers. This result is broadly consistent with evidence by Carrillo-Tudela et al. (2022). These authors find that job seekers (both employed and non-employed) increasingly targeted expanding industries and occupations as the pandemic progressed, indicating a restructuring of the economy and job switching. However, they also note that non-employed workers (more likely to be from previously contracting industries) were less likely to target an expanding industry compared to those employed. Moreover, comparing workers’ search plans reported in the UK Household Longitudinal Survey and empirical workers flows from the UK LFS, they find that the shares of realized transitions into expanding industries and occupations were lower than the shares of targeted transitions. Their results thus point to only limited intention and even more limited ability of unemployed job seekers from hard-hit sectors to move to expanding ones. The divergence between self-reported search intentions and realized transitions can further be interpreted as an element of support for using a worker’s past sector of employment as a proxy for her search behavior, as done in the baseline mismatch index. While workers’ intentions certainly matter, with switches being frequent and quantitatively important, firms most likely do not see workers from other sectors as perfect substitutes for those with a more relevant background.30

7. Conclusion

In this paper, we built upon the mismatch framework proposed by Şahin et al. (2014) to assess whether misalignment between labor supply and demand across industries played an important role for employment dynamics during the COVID-19 crisis in the US and the UK. Our main finding and key contribution to the literature is that, surprisingly, the total loss in employment caused by the rise in mismatch was smaller during the COVID-19 crisis than in the aftermath of the Global Financial Crisis. During the COVID-19 recession, both countries experienced a sharp but short-lived rise in mismatch in the second and third quarters of 2020. The temporary nature of this spike means that mismatch played a quantitatively modest role in slowing down the employment recovery that started in the second half of 2020 for the US and in early 2021 for the UK. This key result is robust to considering broader measures of job seekers and to estimating the “effective searchers” in each sector.

This finding also seems to suggest that COVID-19 did not generate a large-scale structural reallocation involving significant frictions in the matching process between workers and firms, at least as of late 2021. Rather, the large heterogeneity in initial employment declines across industries primarily resulted from lockdown measures and contagion risks. As restrictions on economic activity were lifted and vaccination plans were rolled out, labor demand recovered, including in hard-hit industries, and its sectoral composition largely reverted backed to pre-pandemic patterns.

The absence of a major rise in mismatch raises the issue of which factors accounted for the coexistence of tight labor markets and a sluggish employment recovery in the US and the UK by the fall of 2021. Given that in both countries vacancies reached high levels for historical standards in late 2021, forces inducing a persistent reduction in labor supply likely hold great relevance. Finally, future research could explore more subtle dimensions over which mismatch may play a role. For instance, in line with recent studies on spatial differences in gross worker flows (Kuhn et al., 2021), the increased preference for remote work may have created geographic mismatch as job seekers moved away from high-density areas where vacancies are still primarily located.

Appendix A. Additional Figures

Fig. A.1.

US: Unemployment and vacancy shares by industry: GFC vs COVID-19. Note: The bar charts show unemployment (dark blue) and vacancies (light grey) shares in three different periods for the GFC and the COVID-19 recession, respectively: i) before, ii) period of peak in mismatch, and iii) recovery for the US. The recovery period represents the fall of 2021 for COVID-19 and, for comparability, for the GFC it is chosen as the period with the same time difference from the peak period as the distance between the COVID-19 peak and the recovery period. Sources: JOLTS, US CPS, and authors’ calculations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. A.2.

UK: Unemployment and vacancy shares by industry: GFC vs COVID-19. The bar charts show unemployment (dark blue) and vacancies (light grey) shares in three different periods for the GFC and the COVID-19 recession, respectively: i) before, ii) period of peak in mismatch, and iii) recovery for the UK. The recovery period represents the fall of 2021 for COVID-19 and, for comparability, for the GFC it is chosen as the period with the same time difference from the peak period as the distance between the COVID-19 peak and the recovery period. Sources: UK LFS, ONS, and authors’ calculations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. A.3.

UK: On-the-job search by hours worked last week. Note: “Working at least 1 hour” (solid line) shows the share of job searchers among those employed who worked at least one hour during the previous week. “Working zero hours” (dashed line) shows the share of job searchers among those employed who had worked zero hours in the previous week. Sources: UK LFS and authors’ calculations.

Appendix B. Further details on the mismatch framework

B1. Laws of motion

This section describes the laws of motion used to compute $e_t^{*}$, $u_t^{*}$, $n_t^{*}$. We first outline the baseline case where unemployed workers are the only group comprising the pool of job seekers over which mismatch is computed.

To maintain consistency with the empirical series $e_t$, $u_t$, and $n_t$, all transition rates must be included in the law of motion. For instance, even if inactive workers are not used to compute the mismatch index, empirically there are transitions from inactivity to employment that must be accounted for. We assume that these transitions are exactly the same in the empirical laws of motion and the counterfactual ones, and thus are not affected by mismatch. With the exception of the job finding rate $f_t$, for any two labor market states $j$ and $k$, the transition rate from $j$ to $k$ is denoted as $x_t^{jk}$. E.g., $x_t^{nu}$ is the probability that a worker moves from inactivity to unemployment from time $t-1$ to time $t$.

Note that, in this set-up, inactive workers can transition into employment at the rate $x_t^{ne}$. We account for these transitions in order to maintain comparability with the empirical series. However, unlike the job finding rate for the unemployed, we assume that these transitions are not affected by mismatch.

All transition rates are computed using the microdata for the respective country. The no-mismatch job finding rate $f_t^{*}$ is computed as described in Section 4.

The laws of motion for the baseline case, with only the unemployed workers as job seekers, are as follows:

$$\begin{array}{cc}\hfill e_t^{*}=& (1-x_t^{eu}-x_t^{en})e_{t-1}^{*}+f_t^{*}{u}_{t-1}^{*}+x_t^{ne}{n}_{t-1}^{*}\hfill \\ \hfill u_t^{*}=& (1-f_t^{*}-x_t^{un})u_{t-1}^{*}+x_t^{eu}{e}_{t-1}^{*}+x_t^{nu}{n}_{t-1}^{*}\hfill \\ \hfill n_t^{*}=& (1-x_t^{ne}-x_t^{nu})n_{t-1}^{*}+x_t^{en}{e}_{t-1}^{*}+x_t^{un}{u}_{t-1}^{*}\hfill \end{array}$$

B1.1. General case: Expanded pool of job seekers

For the general case, we assume that $u_t$ represents any group of non-employed job seekers and $n_t$ the non-employed workers that are not actively seeking a job. The generalized pool of job seekers, can thus include other groups besides the unemployed, such as the marginally attached or those who have been inactive for less than one month, in the US case, or one year, in the UK case. Correspondingly, $n_t$ will exclude these additional job seekers. Moreover, in order to accommodate OJS workers and furloughed workers, we denote as ${\omega }_t$ the fraction of employed workers who are also searching for a new job. The total number of job seekers is therefore $s_t=u_t+{\omega }_t{e}_t$.

We assume that OJS workers avoid separation into unemployment and inactivity if they find another job. However, they face the same separation risk if they are not matched to another job. For the no-mismatch counterfactual we assume that ${\omega }_t$ remains unchanged, so that the number of job seekers is $s_t^{*}=u_t^{*}+{\omega }_t{e}_t^{*}$.

The laws of motion are as follows:

$$\begin{array}{cc}\hfill e_t^{*}=& (1-x_t^{eu}-x_t^{en})(1-{\omega }_{t-1}{f}_t^{*})e_{t-1}^{*}+f_t^{*}(u_{t-1}^{*}+{\omega }_{t-1}{e}_{t-1}^{*})+x_t^{ne}{n}_{t-1}^{*}\hfill \\ \hfill u_t^{*}=& (1-f_t^{*}-x_t^{un})u_{t-1}^{*}+x_t^{eu}(1-{\omega }_{t-1}{f}_t^{*})e_{t-1}^{*}+x_t^{nu}{n}_{t-1}^{*}\hfill \\ \hfill n_t^{*}=& (1-x_t^{ne}-x_t^{nu})n_{t-1}^{*}+x_t^{en}(1-{\omega }_{t-1}{f}_t^{*})e_{t-1}^{*}+x_t^{un}{u}_{t-1}^{*}\hfill \end{array}$$

B2. Quantifying the contribution of mismatch to the employment loss

Table B.1 zooms in on the comparison between the GFC and the COVID-19 crisis. It reports the employment rate loss $(e_t^{*}-e_t)$ at specific points in the cycle of each recession, starting from a point shortly before each downturn. We first consider two “trough” points during each downturn: (i) based on the lowest value of the employment rate and (ii) based on the highest level of $(e_t^{*}-e_t)$.31 For the GFC, the “mid-recovery” represents a period in which the employment rate recovered approximately half of the gap from the employment trough to the initial period. For COVID-19, the “mid-recovery” represents the latest available period.

Table B1.

Employment loss due to mismatch during the GFC and the COVID-19 crisis .

		GFC				COVID-19
			(1)	(2)			(3)	(4)
		Date	$e^{*}-e$	${\Delta }_{t-t_0}(e^{*}-e)$		Date	$e^{*}-e$	${\Delta }_{t-t_0}(e^{*}-e)$
US:UxlTemp	Before	2006Q4	0.48	–		2019Q4	0.37	–
	Trough 1: Emp. Fall	2010Q1	1.71	1.22		2020Q2	1.11	0.75
	Trough 2: $e^{*}-e$ Peak	2010Q2	1.71	1.23		2020Q3	1.43	1.07
	Mid-Recovery / Latest	2015Q2	0.48	0.0		2021Q4	0.37	0.0
US:U	Before	2006Q4	0.53	–		2019Q4	0.38	–
	Trough 1: Emp. Fall	2010Q1	1.62	1.09		2020Q2	0.95	0.57
	Trough 2: $e^{*}-e$ Peak	2010Q2	1.67	1.14		2020Q3	1.42	1.04
	Mid-Recovery / Latest	2015Q2	0.55	0.02		2021Q4	0.53	0.14
UK	Before	2007Q4	0.29	–		2019Q4	0.13	–
	Trough 1: Emp. Fall	2010Q1	0.44	0.15		2021Q1	0.25	0.12
	Trough 2: $e^{*}-e$ Peak	2010Q3	0.48	0.18		2021Q2	0.26	0.14
	Mid-Recovery / Latest	2012Q3	0.44	0.15		2021Q4	0.23	0.10

Notes: The table shows the employment loss due to mismatch during the GFC and the COVID-19 crisis. The column labeled $e^{*}-e$ (p.p) shows the percentage points difference in the employment rate between the no-mismatch counterfactual $e^{*}$ and the actual value $e$. Column ${\Delta }_{t-0}(e^{*}-e)$ shows the difference between values at the “Trough” or “Mid-Recovery” points and the “Before” point. For the US, the “UxlTemp” panel shows the values for the baseline case when the unemployed pool is defined to be the unemployed excluding those on temporary layoff while “U” panel shows the total unemployed. Sources: US CPS, JOLTS, UK LFS, ONS, and authors’ calculations.

In the US, in the baseline in which job searchers are defined to be unemployed persons excluding those on temporary layoff, the employment loss due to mismatch was smaller during the trough of the COVID-19 crisis compared to those of the GFC. Based on either definition of the trough, in the COVID-19 downturn the employment rate could have been 1.11 or 1.43 p.p. higher in the absence of mismatch (Column 3), compared to 1.71 and 1.71 p.p. in the GFC (Column 1). Moreover, considering that the employment loss due to mismatch was already higher prior to the GFC than to COVID-19, Columns 2 and 4 focus on the change in $(e_t^{*}-e_t)$ from the pre-downturn period, driven by the rise in mismatch during the downturn. In this case, the increase in the employment loss is also lower during COVID-19 than during the GFC. These patterns hold even when we consider define the total unemployed pool including temporary layoffs (a standard definition of unemployment) as job searchers. In the UK, the employment losses at the troughs are also smaller during pandemic (0.25 and 0.26 p.p. in Column 3) than after the GFC (0.44 and 0.48 p.p. in Column 1). Once again, focusing on changes in the loss (Columns 2 and 4) provides a similar perspective.

With respect to the mid-recovery period, for the US the employment loss in the GFC (0.48 [0.55] p.p.) is higher than the latest available period of data for the COVID-19 recession (0.37 [0.53] p.p. in 2021Q4). However, it is worth noting that the mid-recovery point after the GFC was reached almost 9 years after the pre-downturn period. At that point, as visible in Fig. 4, mismatch had already reverted and reached below pre-GFC levels. Hence, this period may not represent a fully fair comparison with the point of the cycle of 2021Q4 – despite the employment rate still being 2 p.p. below its pre-COVID-19 level. Meanwhile, for the UK, the employment loss at the mid-recovery point is still higher in the GFC than during the pandemic, consistent with the relative differences between the two downturns at the trough.

Appendix C. Mismatch across occupations

Although some occupations are tightly linked to specific industries, others, such as managers, clerical workers, and skilled tradesmen, are found in multiple sectors. The COVID-19 shock disproportionately affected certain sectors (e.g., hotel and restaurants), potentially disrupting all occupations within those sectors. However, occupations whose tasks can be performed remotely may have been broadly shielded from disruption across all sectors. Furthermore, occupations may be more closely associated than industries with specific skills that were in high demand during COVID-19, with the necessity of in-person contact to perform tasks, or with the ability to telework.

Fig. C.1 shows that separation rates rose unevenly across occupations during the pandemic. As discussed in Section 3, industries and occupations often independently mattered for workers’ exposure to job disruptions during the pandemic (Carrillo-Tudela, Comunello, Clymo, Jäckle, Visschers, Zentler-Munro, 2022, Cortes, Forsythe, 2020). Hence, there is value to also computing mismatch across occupations. In this section, we provide some tentative and preliminary evidence on mismatch by occupations since 2018/2019. Unfortunately, lack of a long enough time series for vacancies by occupations in the UK rules out a comparative analysis of mismatch dynamics in the GFC and the pandemic for both countries.

To this aim, we use data from Indeed, a large-scale job posting platform, to compute vacancies by occupation for the US and the UK. We use a version of the Indeed database containing individual job postings at daily frequency, starting in January 2019 for the US and January 2018 for the UK. The US sample contains approximately 100 million observations between January 2019 and September 2021, while the UK sample contains approximately 23 million observations between January 2018 and September 2021. Using the advertised job title, the postings are categorized according to the ISCO-08 classification for the US and the UK-SOC 2010 classification for the UK via a matching algorithm.32 Assuming that one post corresponds to one vacancy (see Adrjan et al., 2021), we compute the number of vacancies posted in each occupation in the first four weeks of each month and subsequently take quarterly averages for the UK. Unfortunately, due to the short span of the database, it is unfeasible to estimate and compare mismatch at the occupational level during COVID-19 and the GFC.

Indeed vacancies and unemployment from the micro-data are aggregated at the 2-digit level of the ISCO-08 for the US and of the UK-SOC 2010 for the UK, which include 24 and 25 different occupation groups, respectively.33 To compute mismatch adjusted by matching efficiency, we assign to each occupation $j$ a ${\varphi }_j$ computed as the average of the ${\varphi }_i$’s of industries weighted by the share of the unemployed of occupation $j$ that report industry $i$ as their past sector of employment in 2019.

Fig. C.2 plots occupation-based mismatch against the baseline industry-based index. For the UK (right plot), the results are qualitatively very similar. Mismatch increased sharply in the second quarter of 2020 -although the increment was smaller in relative terms- and was already back to pre-COVID levels by late 2021. However, for the US, occupation-based mismatch is almost flat during the entire period, implying almost no change in the misalignment of the composition of vacancies and job seekers throughout the pandemic.

Fig. C.2.

Mismatch by occupation. Note: The solid blue lines report the mismatch index across occupations using vacancies from Indeed. The dashed lines show the baseline mismatch index across industries. The mismatch index represents the fraction of hires lost due to misallocation between job seekers and vacancies and is bounded below and above by 0 and 1. Higher values imply a higher degree of mismatch. Sources: US CPS, UK LFS, Indeed, and authors calculations. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Overall, we find that mismatch across occupations exhibited either as short-lived a spike (UK) as mismatch across industries or even did not increase (US). However, these results remain exploratory and several caveats apply. First, the short time span of the analysis prevents us from knowing how mismatch across occupations evolved during the GFC. Hence, there is limited scope to interpret the results without a comparison with previous downturns. Second, although the Indeed dataset can serve as an invaluable high-frequency indicator of labor market developments during the pandemic, it has some limitations in the context of our analysis. The classification of job postings into occupations is done through advanced matching algorithms but is only based on job titles rather than on detailed job descriptions. It may therefore contain some measurement error. Furthermore, while past studies found that fluctuations in the aggregate value of vacancies from Indeed track well the vacancies estimates from official statistics, the stability of that relationship for occupational subgroups has not been thoroughly explored so far. Finally, for the US, the conversion of the occupation information in the CPS from the US SOC 2010 to the ISCO-08 is another potential source of measurement error.

Appendix D. Further details on flows from employment to inactivity

Fig. D.1 reports the probabilities of transitioning from employment to unemployment, inactivity, and marginal attachment. As discussed in the main text, some of these probabilities rose more markedly during COVID-19 than during the GFC. However, mismatch indices computed using broader pools of job seekers do not show visibly larger differences in the behavior of mismatch or in its contribution to job creation across the two downturns compared to the baseline “Unemployment” specification. To gain a deeper understanding, this appendix section breaks down in more detail the inflows into inactivity and the interaction between the NLF pool and vacancies by sector.

To this end, for each downturn and country, we first categorize industries into either large- or small-contraction industries based on the percent decline in vacancies for that industry between the pre-recession peak and the trough of the recession.34 By dividing the industries into those that experienced larger contractions vis-a-vis smaller contractions, we can visually see if larger flows into inactivity contributed to the rise in the mismatch index.

Fig. D.2 plots the movement into inactivity for the set of industries that experienced relatively larger contractions and smaller contractions in vacancies, respectively, over each downturn. Importantly, these sets of industries differ across downturns. Here, instead of dividing the gross worker flows from employment to inactivity by the stock of employed workers in the previous month, we divide the same flows by the sum of unemployed and inactivity from the previous period. This directly measures the contributions of flows from employment to inactivity to the change in the stock of job searchers.

Fig. D.2.

Flows from employment to unemployment and inactivity as share of unemployment and inactivity. Note: The solid (long-dashed) line reports the employed worker transitions to unemployment (inactivity) between one period to the next as a share of the sum of the unemployment + NLF stock from the previous period. The period of analysis is one month for the US and one quarter for the UK. The upper panels refer to the GFC. The lower panels refer to the COVID-19 recession. Sources: US CPS, UK LFS, and authors’ calculations.

If flows into inactivity increased more in large-contraction sectors than small-contraction sectors, then mismatch for the “Unemployed + NLF” group should rise. Based on the figure, however, we do not find evidence for this dynamic. Two observations in particular emerge from the figure. First, in both countries, flows into unemployment did seem to increase more for the large-contraction sectors than for the small-contraction sectors, especially during COVID-19. However, the differences across the two groups of industries are smaller or fully absent for the inflows to NLF group. This suggests that, overall, including the inactivity margin tends to attenuate the differences in the stock of job searchers across sectors, as these inflows tended to be more homogeneous across sectors. Second, due to the large stock of incumbent inactive workers, even in the case of large sectoral disparities in employment outflows, the quantitative relevance of these inflows for mismatch will be more modest under the broader group of job seekers.

Overall, this descriptive analysis suggests that even though inflows into inactivity or marginal attachment were larger during the COVID-19 crisis than during the GFC, they were not necessarily more significant in terms of the mismatch they induced.

Data Availability

• Data will be made available on request.