Modelling the Differing Impacts of Covid-19 in the UK Labour Market.

This article studies the impact of the Covid-19 pandemic on graduates and non-graduates in the United Kingdom. We construct a DSGE model with search frictions that is designed around key features of the UK labour market and simulate the model using an array of shocks, designed to mimic the impact of the Covid-19 pandemic. We show that our relatively simple macroeconomic model can describe the impact of the pandemic on output, employment and wages. Our results show that the impact of the pandemic on employment and wages was more severe for non-graduates than for graduates, and that up to 5 million jobs would have been lost in the first wave of the pandemic in the absence of the Job Retention Scheme.

Introduction

This article models the response of the labour markets for graduates and non‐graduates in the United Kingdom to the Covid‐19 pandemic. The crisis has had a major impact. At the peak of the pandemic, only around 50% of workers were at work; two‐thirds of employers made use of the Job Retention Scheme (JRS) and up to 10 million workers were furloughed (Leslie, 2020). GDP fell by 20% in April 2020, and was 5% lower in 2020 compared with 2019. Across 2020, employment fell by 1%–3% ONS (2021a).

We address three main questions. First, can this extreme turbulence be explained using a relatively simple macroeconomic model? Second, how did the pandemic affect the labour market experience of different types of workers in the United Kingdom and what is the likely adjustment process as the economy recovers? And, third, what was the impact of the JRS, adopted by the UK Government in March 2020 as the core of their economic policy response to the pandemic?

We address these questions using a DSGE model with labour market frictions, which we simulate using a series of shocks that are constructed to mimic the effects of the pandemic. We adapt the models of Krause and Lubik (2007), Thomas and Zanetti (2009), Blanchard and Gali (2010), Ravenna and Walsh (2011), Zanetti (2011) and Mumtaz and Zanetti (2015)1to allow for three key features of the UK labour market, which we document below using data from the Labour Force Survey. First, the labour market experience of graduates and non‐graduates differs markedly; graduates typically have higher wages and greater job security than non‐graduates (ONS, 2017a). Second, many graduates are employed in ‘non‐graduate’ occupations (see also ONS, 2017a). And third, there is substantial movement of workers between jobs, with most hires coming from the employed rather than the unemployed (see also ONS, 2017b). These features imply that non‐graduates are especially vulnerable to severe shocks such as the Covid‐19 pandemic, and highlight the complexities of the UK labour market that our analysis needs to take account of.

Reflecting these features, our model distinguishes between graduates and non‐graduates and between ‘high productivity’ and ‘low productivity’ occupations. Both graduates and non‐graduates can be employed in both types of occupation, but with differing productivities2. As we document below, the UK labour market is characterized by complex flows of workers between high productivity and low productivity occupations and between these and non‐employment. We generate worker flows to match these data through job search by employed and non‐employed workers, and job destruction.

We model the impact of Covid‐19 on the UK labour market as a series of simultaneous adverse shocks. Baqaee and Farhi (2020) and Maria del Rio‐Chanona et al. (2020) use a similar approach for the United States. Related work includes Fornaro and Wolf (2020) and Guerrieri, Lorenzoni and Straub (2020), who conceptualize the transmission of adverse supply shocks into adverse demand shocks. Other models (e.g. Eichenbaum, Rebelo and Trabandt, 2020; Mihailov, 2020) incorporate simple Susceptible‐Infectious‐Recovered epidemiological processes into DSGE models. Our simulations of the impact of the pandemic are mainly driven by adverse shocks to aggregate demand and supply, where the latter reflects the impact of workers who are furloughed through the JRS or working from home, as well as social distancing. Reflecting the JRS, we model wage shocks that reduce the cost of labour to firms. We also model shocks to job destruction. Productivity and wage shocks are specific to the job and not the worker and so are the same for graduates and non‐graduates in the same occupation.

In our baseline scenario, we show that our simple model can match the experience of the UK economy during the pandemic. We focus on three key features of the UK experience: (i) there were deep reductions in output, but relatively little change to employment; (ii) most of the reduction in employment came from non‐graduates; and (iii) aggregate real wages increased in the first half of 2021. We are able to match the behaviour of GDP throughout 2020–21. There is some uncertainty about changes in employment during the pandemic; data from the Labour Force Survey suggest a fall of 1%–2%, while the Workforce Jobs Survey suggests a larger fall, of around 3%. Our simulated values for employment follow the values from the Workforce Job Survey across the pandemic. Our simulations also match the disproportionate impact on non‐graduate employment, even though our shocks do not distinguish between graduates and non‐graduates. This is because the pandemic had a larger impact on lower productivity occupations, where non‐graduate employment is concentrated. And our model explains the rise in the aggregate wage largely through a composition effect that reflects the shift of employment towards higher‐paid graduates3.

To assess the JRS, we note that in the absence of the scheme, aggregate demand would have been lower, but productivity would have been higher (since workers would not have been away from the workplace on furlough), and there would have been no wage subsidies to firms. All these effects imply lower employment. We analyse the impact of the JRS, by considering a range of scenarios. In the first, we assume that the scheme had a larger impact on productivity than on aggregate demand, so there is a small additional fall in aggregate demand but a large improvement in productivity. In the second, we assume that the JRS had a larger impact on aggregate demand than on productivity, so there is a larger additional fall in aggregate demand but a smaller improvement in productivity. These scenarios give very different impacts on output but similar impacts on employment. This implies that we can be more confident about the impact of the JRS on employment than on output. We find that employment would have fallen by a further 14%–17% in the absence of the JRS. Since up to 29% of jobs were furloughed in the second quarter of 2020, our results suggest that between 48% and 58% of these furloughed jobs would have been lost without the JRS, amounting to between 4.18 and 5.05 million jobs.

We also consider two additional scenarios. In the first, we take the most benign case for what employment would have been in the absence of the JRS. We assume that in the absence of the JRS, there would have been no additional impact on aggregate demand, and productivity would have remained at the prepandemic level4. Here, employment would have fallen by an additional 11% without the JRS, giving a lower bound of 40% for the proportion of furloughed jobs that would have been lost without the JRS. In the second, at the other extreme, we construct a scenario in which, in the absence of the JRS, there would have been a very large negative impact on aggregate demand, and productivity would have been the same as in our baseline scenario. In this case, employment would have fallen by 30%, so all the jobs that were furloughed under the JRS would instead have been lost. However, these additional scenarios are not plausible. This gives us moderate confidence in our conclusion that in the absence of the JRS, employment would have fallen by an additional 4–5 million jobs. Given that UK unemployment was around 1.4 million in 2020Q1, and considering that the US unemployment rate rose from around 4% to close to 15% at the onset of the pandemic, this figure is perhaps not unreasonable.

The remainder of the article is structured as follows. In section II, we document the labour markets for graduates and non‐graduates in the United Kingdom, both before and during the pandemic. Section III outlines our model and section IV explains how we calibrate the steady‐state of our model using the data outlined in section II. Section V outlines how we model the pandemic and presents and discusses our results. It then analyses what the likely outcomes would have been if the JRS were not in place. Section VI concludes by summarizing our model, highlighting the assumptions we have made and discusses possible ways ahead for research on this topic.

The labour market for graduates and non‐graduates in the United Kingdom

We analyse the impact of the pandemic on the employment of graduates and non‐graduates in different occupations and the transitions of workers between jobs. We begin by documenting key facts on these in the period before the pandemic, in order to inform the design and calibration of our model. To do this, we use data from the longitudinal component of the UK Labour Force Survey5. Individuals participate in the survey for 15 months. We use the longitudinal survey for 2018Q4–2019Q4 to derive measures of employment and transitions in the prepandemic period and the longitudinal survey for 2019Q4–2020Q4 to derive these measures during the pandemic.

The prepandemic labour market

We use responses from the same individuals across 2018Q4–2019Q4, using only respondents who are aged above 15 and below 65, to construct measures of employment and labour market flows in the period immediately prior to the pandemic; we use these data to calibrate our model and to provide a baseline against which to assess the impact of the pandemic. In our prepandemic sample, an average of 33.8% of respondents are graduates in this period6.

We distinguish between high productivity and low productivity occupations; we define a high productivity occupation as corresponding to groups (1)–(3) in the UK Standard Occupational Classification (SOC) (ONS, 2020) and a low productivity job as corresponding to groups (4)–(9)7. Based on our data, we make two assumptions that will shape our modelling approach. First, we analyse the numbers in work, combining employees and the self‐employed. This is because the structure of employment for the employed and self‐employed is similar. This assumption enables us to mitigate some of the issues around measurement of employment during the pandemic that have arisen with the LFS and other surveys. Second, we found little difference in the behaviour of the unemployed and the inactive during the pandemic, and so combine these into one category8. This greatly reduces the complexity of our model9. Figure 1 documents the prepandemic labour market in the United Kingdom. of respondents aged between 16 and 65 were graduates employed in high productivity occupations and 6.1% of respondents were graduates employed in low productivity occupations. 5.2% of respondents were unemployed or inactive graduates. Similarly, 15.9% and 28.9% of respondents were non‐graduates in high and low productivity occupations, respectively, and 21.4% of respondents were unemployed or inactive non‐graduates10. These data reveal important features of the UK labour market that we build into the design of our model. In particular, while most graduates are employed in high productivity occupations and most non‐graduates are employed in low productivity occupations, a substantial minority are not. And a non‐employed worker is much more likely to be a non‐graduate.

Figure 1

Employment in the United Kingdom, 2018Q4–2019Q4

Our data also enable us to track transitions of workers between categories and construct estimates of transition rates. These are presented in Table 1. We calculate quarter‐to‐quarter transition rates between 2018Q4 and 2019Q4, and report averages of these. Across this period, the average proportion of graduates employed in high productivity occupations who were still employed in these occupations in the following quarter was 0.956, while the average proportions who moved to lower productivity occupations or to non‐employment in the following quarter were 0.024 and 0.020 respectively. The corresponding proportions for non‐graduates were 0.917, 0.055 and 0.028 respectively. The average proportion of graduates employed in low productivity occupations who were still employed in these occupations in the following quarter was 0.854, while the average rates of transition to higher occupations or to non‐employment were 0.107 and 0.040 respectively. The corresponding average transition rates for non‐graduates were 0.935, 0.033 and 0.032 respectively. The average proportion of graduates who were unemployed or inactive in one quarter and remained so in the following quarter was 0.904, while the average proportions who gained employment in high or low productivity occupations were 0.053 and 0.043 respectively. The corresponding proportions for unemployed or inactive non‐graduates were 0.945, 0.008 and 0.047 respectively.

TABLE 1

Labour market transitions in the United Kingdom, 2018Q4–2019Q4

(i) Graduate transitions	High productivity	Low productivity	Non‐employed
High productivity	0.956	0.024	0.020
Low productivity	0.107	0.854	0.040
Non‐employed	0.053	0.043	0.904

Notes: The table shows the proportion of workers who were in the labour market state in the left hand column in one quarter and in the labour market state in the rows at the top in the following quarter; results are averaged across 2018Q4–2019Q4.

Source: Labour Force Survey Longitudinal Survey.

These data reveal several important features of the UK labour market. First, there is more movement between different occupations than between employment and non‐employment; both graduates and non‐graduates are more likely to move into a high productivity occupation from a low productivity occupation and vice versa, than from non‐employment. Second, there are multiple routes into occupations11. Third, graduates are more likely to remain in high productivity occupations than non‐graduates, but are less likely to remain in low productivity occupations. Fourth, non‐employed non‐graduates are less likely to move into any type of employment than non‐employed graduates.

The impact of the pandemic on the UK labour market

Employment

Table 2 documents the evolution of employment across 2020Q1–2020Q4. Based on these data, Figure 2 shows changes in employment between 2020Q1–2020Q2, 2020Q2–2020Q3 and 2020Q3–2020Q412. The largest change at the onset of the pandemic in 2020Q2 was the loss of more than 300,000 non‐graduates in low productivity occupations, most of whom moved into non‐employment. The following quarter saw a large movement of workers out of high productivity occupations, into low productivity occupations13. The final quarter of 2020 saw a movement of graduates out of non‐employment into low productivity occupations and a further movement of non‐graduates out of high productivity occupations and into low productivity occupations and non‐employment. Two main changes stand out here. Twice as many non‐graduates than graduates became non‐employed. And employment of non‐graduates in high productivity occupations fell by over 250,000 workers14.

TABLE 2

Employment in the United Kingdom, 2020Q1–2020Q4

	2020Q1	2020Q2	2020Q3	2020Q4
Graduates in high productivity occupations	0.246	0.247	0.244	0.245
Graduates in low productivity occupations	0.060	0.060	0.062	0.065
Unemployed or inactive graduates	0.057	0.058	0.063	0.061
Non‐graduates in high productivity occupations	0.156	0.156	0.152	0.148
Non‐graduates in low productivity occupations	0.289	0.279	0.280	0.281
Unemployed or inactive non‐graduates	0.196	0.200	0.199	0.200

Source: Labour Force Survey Longitudinal Survey.

Figure 2

Employment changes in the United Kingdom, 2020Q1–2020Q4

How reliable are these data? Concern has been expressed that data from the Labour Force Survey understate the loss of employment during the pandemic (Leaker, 2021). LFS data imply a loss of 570,000 jobs during the pandemic, whereas tax data show that the number of workers on payroll and paying tax fell by around 800,000. Data from the Workforce Jobs Survey (WJS) support a larger fall in employment, showing a loss of employment of around 3% across 2020. One factor behind the relatively small reduction in employment in ONS data is that around 500,000 workers changed their reported status from self‐employed to employed, possibly to enable claims for support under the JRS (ONS, 2021). Our approach in this article mitigates this effect, as our data include employed and self‐employed workers. The data in Table 2 are consistent with the evidence in ONS (2021a), showing a large move out of employment in low productivity occupations at the onset of the pandemic and reductions in higher productivity employment towards the end of 2020. On balance, we will use the evidence in Table 2 on the impact of the pandemic on different types of employment. But this should be treated with some caution.

Furloughing

The longitudinal data that we use do not contain information on which workers have been furloughed under the Job Retention Scheme15. But data on the percentages of individuals reporting working zero hours give a useful indication of the extent of furloughing across different occupations and types of workers. Figure 3 shows the proportions of different types of workers who report working zero hours across 2018Q4–2020Q4. The proportions of non‐graduates employed in low productivity occupations who report zero hours jump from a prepandemic average of around 10% to close to 40% in 2020Q2; this figure then falls, but remains above 20% at the end of 2020. The proportion of graduates employed in low productivity occupations who report zero hours also jump, to more than twice the prepandemic mean. And the proportion of non‐graduates employed in high productivity occupations who report zero hours jumps to almost three times above its prepandemic mean in 2020Q2. These latter figures decline but remain well above historical averages by the end of 2020. By contrast, the proportion of graduates in higher productivity occupations who report zero hours increased by a lesser amount and had returned to close to normal values by the end of 2020. These data are consistent with other evidence on the impact of the JRS (e.g. Tomlinson, 2020; Gustafsson and McCurdy, 2020) that suggests that furloughing was concentrated in lower productivity occupations. The data on zero hours worked in Figure 3 are also consistent with data on total hours of work reported in ONS (2021a).

Figure 3

Percentage of workers reporting zero hours in the United Kingdom, 2018Q4–2020Q4

The model

In this section, we outline our model. Full details and derivations are in the Online Appendix.

Overview

The economy is composed of households, wholesale firms, retail firms, the government and the Central Bank. Household members are either graduates or non‐graduates. Wholesale firms employ graduates and non‐graduates in high and low productivity occupations to produce wholesale goods that are sold to retail firms who use them to produce retail goods. Retail goods are sold to households and the government. The wholesale goods market is competitive, but the retail goods market is imperfectly competitive. The government collects taxes and purchases retail goods. The Central Bank sets the interest rate on the financial asset that households use to smooth consumption over time.

The labour market

There are graduates and non‐graduates16. Defining graduates and non‐graduates who are not in employment as 17and and defining graduates in high and low productivity occupations as and , respectively, we have . Defining non‐graduates in high and low productivity occupations as and , then .

All workers search for jobs. Non‐employed graduates and non‐graduates search for jobs in high and low productivity occupations with respective intensities , and and . Graduates and non‐graduates employed in low productivity occupations search for high productivity jobs with respective intensities and , while workers employed in high productivity occupations search for low productivity jobs with respective intensities and . Our assumption that workers search with fixed intensity is in line with the recent literature, including Moscarini and Postel‐Vinay (2017) and Faccini and Melosi (2020)18. Job matches for each occupation and worker type are given by for and , where is the number of vacancies posted19, is search for employment and is the matching efficiency. Search for graduate jobs is given by and . Similar expressions describe search for non‐graduate jobs.

Our model has multiple job‐finding rates and rates of job destruction, which all vary over time and between occupations. We denote these as for , and . For example, the rates at which non‐employed graduates find high and low productivity jobs are and , respectively. And the job‐finding rate for non‐graduates employed in low productivity occupations who are searching for a high productivity job is . There are four rates of job destruction, given by for and . We assume ; reflects systematic differences in rates of job destruction between different occupations, while is a shock that captures the different impacts of the pandemic on job destruction in different occupations. We summarize these worker flows in Figure 4 and we will calibrate the different rates of job‐finding and job destruction in steady‐state using the data in Table 1.

Figure 4

The labour market dynamics of the model

At the beginning of each period, vacancies are filled, workers find new jobs and job separation occurs20. Then, wages are set, followed by production. Firms set vacancies and workers search for jobs. This results in new job matches at the start of the next period.

Wholesale firms

There is a continuum of identical wholesale firms with the production function for and , where . The parameters reflect systematic differences in the productivities of different workers in different occupations, while the terms are occupation‐specific shocks that capture the differential impact of the pandemic on the productivities of different occupations, through furloughing, working from home, and other factors. Employment dynamics are given by for and .

Wholesale firms post vacancies for different types of workers and bargain over their wages. Defining , and , for and , as the real wages, employment and vacancy posting costs of different types of workers in different occupations, we can express the objective function of the representative wholesale firm as where is the stochastic discount factor, is the price of wholesale goods and is the price of retail goods. The optimality conditions for each type of worker and occupation are for and , where is the marginal cost of hiring21(see the Online Appendix)22

Aggregate demand

Household utility is where is consumption and is a preference shock. The household budget constraint is where is the consumption price index, is the opportunity cost of employment, is the nominal price of bonds, is profit received from ownership of firms and is a lump‐sum tax. All graduates in high productivity jobs earn the same wage, all graduates in low productivity jobs earn the same wage, and the same applies to non‐graduates. The household chooses consumption and bond purchases to maximize utility subject to their budget constraint. This gives the Euler equation The real interest rate . Aggregate demand is where is government expenditure23. The Central Bank sets the interest rate using the simple Taylor rule where is inflation, is the output gap, and is a monetary policy shock.

Wage determination

Wages for each occupation and worker type are determined through bargaining between firms and the relevant workers. We incorporate real wage rigidity, which we model following Krause and Lubik (2007) and Faia (2008). Wages are for and , where is the wage implied by bargaining, is the steady‐state value of this wage, captures real wage rigidity. During the pandemic, the JRS, among other things, provided a wage subsidy. We model this using the occupation‐specific wage subsidy shocks , which creates a wedge between the cost of labour to firms and the wages received by workers.

Nash bargaining over wages is analysed in the Online Appendix. The bargained wage for graduates in high productivity occupations is where and are the bargaining power of graduates in high‐ and low productivity occupations, respectively, and . The bargained real wage for graduates in high productivity occupations depends on their marginal revenue product and the marginal cost of hiring replacement workers (proportional to ). It also depends on the cost of hiring graduates in low productivity occupations (proportional to ) as this affects the outside option of graduates in high productivity occupations. As the Online Appendix documents, the expressions for the bargained wages of different types of workers and occupations are similar, with all bargained wages driven by marginal revenue products, the cost of hiring replacement workers and the ease with workers can find alternative employment in other occupations.

Retail firms

There is a continuum of identical retail firms, with production function where are purchases of wholesale goods. Retail firms re‐optimize price with probability ; firms that do not re‐optimize increase price by a proportion of the previous inflation rate. The Phillips Curve is where the retail marginal cost is equal to the cost of purchasing the intermediate good.

Calibration

We calibrate our model in two stages. First, we calibrate the model in steady‐state; we do this to match the data on the UK labour market in the prepandemic period, presented in section II. Then, we calibrate the aggregate demand side of the model and other factors that are only relevant out of steady‐state. We do this by drawing on previous UK evidence on these.

Since the impact of the pandemic was so large, we can assume that the UK labour market was close to steady‐state in the periods immediately before the onset of the crisis. As a result, we treat the data in Figure 1 and Table 1 as steady‐state values. Based on Figure 1, we set , , , , and as calibration targets. From Table 1, we also set , , and as calibration targets for the rates of job destruction. We set , , and as calibration targets for graduate transitions and , , and as calibration targets for non‐graduate transitions. Our calibration strategy ensures these targets are hit exactly.

Our calibration targets imply values for worker hires. For example, hiring of graduates into high productivity occupations from non‐employment and from low productivity occupations are and , respectively. Similar calculations give values for the other six flows of hires in our model. These values in turn enable us to calibrate the parameters determining job search. We normalize the search effort of non‐employed workers to unity, so and . We also calibrate . The remaining parameters can then be inferred from the data using our assumption that the hiring rates of different groups are proportional to relative search intensities. For example, since , we calibrate the search intensity of graduates in low productivity occupations as . Similar calculations give , and . These results show that workers in low productivity occupations search harder for a high productivity job than do the non‐employed: for graduates, the intensity of search by a worker in a low productivity occupation for a high productivity occupation is four times larger than the intensity of a non‐employed worker; for non‐graduates, it is 10 times larger. In addition, the search intensity of workers in low productivity occupations for high productivity jobs is much larger than the search intensity of workers in high productivity occupations for low productivity jobs: for both graduates and non‐graduates, the intensity of search by workers in low productivity occupations for a high productivity job is seven times larger than the intensity of search by workers in high productivity occupations for a low productivity job.

In the next step, we solve for the values of labour market tightness and matching efficiency ( and for and ) that satisfy the optimality conditions in (5), the matching functions in (1) and the values for employment, labour market transitions, hires and job search obtained above. To do this, we assume a small productivity increment to having a degree by calibrating , , and . We assume that it is more expensive to post vacancies for high productivity occupations and set and . We assume that all workers in the same occupation have the same bargaining power and that workers in high productivity occupations have more bargaining power; we use and . And we assume that for all occupations and workers24. We also calibrate the discount factor as and the mark‐up as , and follow Faccini et al. (2013) in calibrating . Based on these values, we obtain values for labour market tightness and matching efficiency. Using these, we then solve for the implied values of wages and the marginal cost of hiring.

Finally, we calibrate the parameters that determine aggregate demand, monetary policy and the Phillips curve. We assume , implying logarithmic utility from consumption. We follow Faccini et al. (2013) in calibrating the parameters of the Taylor Rule as and . We follow Murray (2012) in calibrating real rigidity as for all four wages and in calibrating the slope of the Phillips curve as . Our calibration choices are summarized in Tables 3 and 4.

TABLE 3

Calibration choices (1)

(i) Graduates	ζ	α	A	ϑ	γ
High productivity occupations	0.5	0.5	1.1	0.5	0.2
Low productivity occupations	0.5	0.5	0.75	0.2	0.1

TABLE 4

Calibration choices (2)

Parameter	Interpretation	Source/target	Value
β	Discount factor	Faccini et al. (2013)	0.995
η	Risk aversion	Faccini et al. (2013)	1
φ_π	Mon Pol response to inflation	Faccini et al. (2013)	1.48
φ_y	Mon Pol response to output	Faccini et al. (2013)	0.31
b	Opp cost of employment	Faccini et al. (2013)	0.58
κ	Slope of Phillips curve	Murray (2012)	0.10
ϕ^π	Inflation persistence	Murray (2012)	0.85
ϕ	Real wage rigidity	Authors' calibrations	0.95

Modelling the pandemic

Modelling the pandemic using shocks

We model the Covid‐19 pandemic and the policy measures taken to mitigate this as a series of simultaneous shocks. To model the impact of these, we write the linearized representation25 of our model as for , where is a vector containing the endogenous variables of the model at time ; is a vector containing the shocks that we use to represent the pandemic and policy measures; , and are matrices and is a matrix; these contain structural parameters of the model, calibrated as described in the previous section. We assume the pandemic began in 2020Q2, when ; we assume the UK economy was in steady‐state (relative to the major disruption that followed) in 2020Q1. We use a deterministic simulation of the model, showing the response of the endogenous variables to the shocks in . We specify the shocks in period , as described below. We model the shocks after this date as autoregressive processes, so for , where indexes the shock; so the behaviour of the shock over time is characterized by the incidence in 2020Q2 and the persistence parameter. We begin by specifying a baseline simulation to show the likely impact over 2020–2023. We then consider alternative scenarios.

Calibration of the pandemic shocks

Our calibration of supply shocks reflects data on the impact of the Covid‐19 pandemic on productivity via workers being on furlough and working from home. We use three parameters, which vary between high‐ and low‐productivity occupations, to capture the impact of these factors. A proportion , for of workers are furloughed and hence do not contribute to production. A proportion work from home. These workers contribute to production, but are less productive than those who continue at the workplace; the relative loss of productivity of working from home is denoted as . From (2), this implies that output during the pandemic is , for and , and so the supply shock for different occupations is for .

There are not sufficient data to calibrate these individual parameters with any precision, but we are able to determine the relative size of the supply shocks for high and low productivity occupations with more confidence. Considering furloughing, data on uptake of the JRS by occupation have not been released, but data on uptake by industry show that this was heavily concentrated in sectors with high proportions of lower productivity employment, such as retail, hospitality and construction (Tomlinson, 2020; Hopson and Wilkinson, 2021). This suggests that was substantially larger than . For working from home, using data from the LFS and the Understanding Society survey (ISER, 2021), Gustafsson and McCurdy (2020) report that close to 50% of workers in high productivity occupations were working from home at the height of the pandemic, compared with slightly above 10% for lower productivity occupations. This is supported by our evidence in Figure 3. Information on the relative productivity of working from home is not yet available, but high productivity occupations appear better able to sustain output when working from home (Costa Dias et al., 2020). This suggests that and were higher than their counterparts for lower productivity occupations. This evidence implies that the impact of the pandemic on the productivity of low productivity occupations was much worse than the impact on high productivity occupations. We operationalize this by assuming that the reduction in productivity for low productivity occupations is larger than for higher productivity occupations. As we summarize in column (i) Table 5, in our baseline simulation we assume that productivity for low productivity occupations falls by 23% at the onset of the pandemic, while productivity for low productivity occupations falls by 17%.

TABLE 5

Calibration of Shocks for the Baseline Simulation and Job Retention Scheme Scenarios

Variable	Interpretation	(i) Baseline	(ii) JRS:1	(iii) JRS:2	(iv) JRS:3	(v) JRS:4
A^H	Productivity: High prod jobs	↓17%	↓6%	↓14%	0%	↓17%
A^L	Productivity: Low prod jobs	↓23%	↓8%	↓18%	0%	↓23%
w^H	Wage: High prod jobs	↓15%	0%	0%	0%	0%
w^L	Wage: High prod jobs	↓25%	0%	0%	0%	0%
Y	Aggregate demand	↓11%	↓15%	↓22%	↓11%	↓33%
π^{H,U}	Job destruction: High prod jobs	0%	↑5%	↑5%	↑5%	0%
π^{L,U}	Job destruction: Low prod jobs	↑5%	↑10%	↑10%	↑10%	↑5%

To calculate the impact of the JRS on wages, we note that firms paid only part of the wage for workers on furlough. Denoting the proportion of the wage of furloughed workers paid by firms as for , the wage paid by the representative wholesale firm, relative to the prepandemic wages, is , for and , and so the shock to wages during the pandemic is . In this case, the evidence suggests a stronger impact on the wage costs of lower productivity workers. This is because, as discussed above, these workers were more likely to be furloughed, and because the Job Retention Scheme paid a larger proportion of the wages of these workers 26. Based on this, our baseline simulation assumes that the wage shock for low productivity occupations is 66% larger than the shock for high productivity occupations. As column (i) of Table 5 shows, in our baseline simulation, we assume that wage costs fall at the onset of the pandemic by 25% in low productivity occupations and by 15% in high productivity occupations.

To model the impacts of the pandemic on aggregate demand, we calibrate so that the interest rate decreases by 250 basis points in 2020Q227. We also calibrate an 11% reduction in aggregate demand via the shock . Finally, we calibrate a small 5% increase in job destruction shock for low productivity jobs, , at the onset of the pandemic. This shock brings some of the loss of low productivity jobs forward from 2020Q3 to 2020Q2, but otherwise has little effect. We do not model an increase in job destruction in high productivity jobs in our baseline simulation because reductions in employment are achieved through reduced hiring.

The baseline scenario

The results of our baseline simulation are shown in Figure 5. Our simple model is able to match the experience of the UK economy during the pandemic well. Our combination of demand and supply shocks generates large reductions in output, with only small reductions in employment, close to what was observed in the United Kingdom in 2020. Movements in employment are similar to evidence from the WJS. Strikingly, we are able to match the larger fall in the employment of non‐graduates, compared with graduates, even though the shocks are specific to occupations rather than the education status of workers. This is because non‐graduate employment is more concentrated in lower productivity occupations. Finally, our simulated model matches the increase in real wages in 2021; it does this through a composition effect, reflecting the shift in employment towards graduates during the pandemic.

Figure 5

The baseline model This figure compares simulated impulse response functions from our model with UK data. The simulations use the shocks in column (i) of Table 6

Examining the mechanisms that underlie these results, we note that the contraction in demand at the onset of the pandemic reduced output. From the production function in (2), the negative supply shock ensures that a deep fall in output is associated with a smaller fall in employment. There was also a reduction in vacancies28. The reduction in employment of non‐graduates documented in Table 2 is achieved through reduced hiring. Figure 2 shows that employment in low productivity occupations fell at the onset of the pandemic, whereas employment in high productivity occupations began to fall in the following quarter.

Evaluating the job retention scheme

At this point, we have addressed our first two research questions, showing that the impact of the pandemic can be explained using a relatively simple macroeconomic model and highlighting the stronger impact of the pandemic on non‐graduate workers. Our final research question concerns the role of the JRS. The JRS can be analysed through the production functions in (2) and the optimality conditions in (5). The JRS had three important impacts. First, the JRS boosted aggregate demand. In the second quarter of 2020, close to 9 million employed workers were furloughed under the scheme (ONS (2021b). Since many of these workers might otherwise have become unemployed, the JRS was a major support to aggregate demand in this period. Second, the JRS led to large negative shocks to productivity, as furloughed workers remained in employment but did not contribute to production29. And third, the JRS provided wage subsidies to firms that reduced the cost of workers on furlough. From (5), these wage subsidy shocks partially offset the adverse effects of the supply shocks, implying a smaller reduction in the marginal cost of hiring, and hence smaller reduction in vacancies and hiring than would otherwise have occurred30.

To illustrate the impact of the JRS, we simulate what the impacts of the pandemic on the UK economy might have been without it. In the absence of the JRS, the aggregate demand shock would have been more severe. The adverse supply shock would have been less severe as no worker would have been on furlough31. From (2), these imply a steeper fall in employment. This in turn implies a more prominent role for job destruction shocks, as the steeper fall in employment cannot be generated through lower hiring alone. The removal of the large wage subsidy implies an increased cost of labour to firms. From (12), the impact on the wage is a priori unclear, as the probable reduction in labour market tightness is offset by a smaller reduction in productivity; but one might expect take‐home wages paid to workers to be lower without the JRS. We analyse these different effects using a range of scenarios.

We first consider two scenarios designed to encompass a ‘reasonable’ range of possible effects of the JRS. Scenario (1) captures the case where the JRS had a larger impact on productivity than on aggregate demand. We assume the reduction in aggregate demand in the absence of the JRS would have been quite small: aggregate demand falls by 15% in scenario (1) compared with a fall of 11% in our baseline. But we assume that the productivity of high and low productivity occupations would have fallen by only 6% and 8% in scenario (1), compared with 17% and 23% in the baseline. Scenario (2) captures the opposite case, where the JRS had a larger impact on aggregate demand than on productivity. In this scenario, the fall in aggregate demand is twice as severe as in the baseline. But we assume that the declines in productivity for high and low productivity occupations are 14% and 18%, much closer to the baseline than scenario (1). In both scenarios, we also calibrate a 10% increase in the rate of job destruction in low productivity occupations and a 5% increase in the rate of job destruction in high productivity occupations. These scenarios are summarized in columns (ii) and (iii) of Table 5.

The results are shown in Figure 6. As the top‐left panel shows, the scenarios have very different impacts on output. The much smaller reduction in productivity in scenario (1) offsets the steeper reduction in aggregate demand, leading to a smaller reduction in output compared with the baseline. Scenario (2) has a more severe reduction in aggregate demand, with a smaller improvement in productivity; in this case, the fall in output is similar to the baseline. In both scenarios, the subsequent rise in output is slower than with our baseline, highlighting the role of the JRS in accelerating the recovery.

Figure 6

The Job Retention Scheme (1). This figure compares simulated scenarios to our baseline simulation. Scenario 1 uses the shocks in column (ii) of Table 5. Scenario 2 uses the shocks in column (iii) of Table 5. Scenario 1 relative to Baseline: Small additional fall in aggregate demand. Large improvement in productivity. Scenario 2 relative to Baseline: Large additional fall in aggregate demand. Small improvement in productivity.

By contrast, as the top‐right panel shows, the two scenarios have similar implications for employment. With scenario (1), employment falls by 17%, while it falls by close to 20% with scenario (2). This narrow range implies that we can be more confident about the impact of the JRS on employment than on output. Since employment fell by around 3% during the pandemic, these results imply that employment would have fallen by a further 14%–17% in the absence of the JRS. These simulations enable us to estimate the number of furloughed jobs that would have been lost without the JRS. Estimates from the ONS Business Insights and Conditions Survey suggest that up to 29% of jobs were furloughed in the second quarter of 202032. Our results suggest that between 48% and 58% of these furloughed jobs would have been lost without the JRS33. Since 8.7 million workers were furloughed in April 2020, this implies that the JRS prevented the loss of between 4.18 and 5.05 million jobs.

The bottom‐left panel of Figure 6 shows that the great majority of jobs that would have been lost without the JRS are in lower productivity occupations. We find that employment in high productivity occupations would have fallen by around 8%–10%, while employment in low productivity occupations would have fallen by 24%–30%. This highlights how the disproportionate effect of the pandemic on low productivity employment, shown in Figures 2 and 6, would have been more severe in the absence of the JRS. The bottom‐right panel of Figure 6 shows the implications of this for aggregate wages. There is a rise of 5%–8% in aggregate wages. This is entirely due to a composition effect, which arises as the share of low productivity occupations in employment falls sharply in these scenarios.

We next analyse the lower bound for the number of furloughed jobs that would have been lost without the JRS. To obtain this, we consider a scenario that only includes the negative impacts of the JRS on employment. Scenario (3) assumes that aggregate demand would have fallen by 11% in the absence of the JRS, the same as the baseline scenario, implying that the JRS had no impact on aggregate demand. And we assume that all reductions in productivity during the pandemic were due to the scheme, so productivity would have remained at its prepandemic level in the absence of the JRS. This is summarized in column (iv) of Table 5. As Figure 7 shows, employment falls by 14% in this scenario, only slightly less than in Figure 6. This implies an estimated lower bound of for the proportion of the jobs furloughed under the JRS that would otherwise have been lost. So even under the most benign assumptions about the impact of the pandemic, the JRS still prevented a large proportion of furloughed jobs from being lost. However, this benign scenario is not plausible: some loss of productivity would have occurred without the JRS, due to working from home; and the JRS clearly had a role in supporting aggregate demand. We conclude that our estimated lower bound of 50% for the proportion of furloughed jobs that would have otherwise been lost is relatively credible.

Figure 7

The Job Retention Scheme (2) This figure compares simulated scenarios to our baseline simulation. Scenario 3 uses the shocks in column (iv) of Table 5. Scenario 4 uses use the shocks in column (v) of Table 5. Scenario 3 relative to Baseline: No change in aggregate demand. Very large improvement in productivity. Scenario 4 relative to Baseline: Very large additional fall in aggregate demand. No change in productivity.

Finally, in scenario (4), the reduction in aggregate demand is three times as severe as in the baseline, but productivity remains at the level of our baseline scenario. This is summarized in column (v) of Table 5. Figure 7 shows that in this scenario employment falls by 30%. This implies that all of the jobs furloughed under the JRS would otherwise have been lost. However, scenario (4) is highly implausible: it assumes that the withdrawal of 29% of employed workers from the workplace would have had no impact on productivity but that the impact on aggregate demand would have been extremely severe. The implausibility of this scenario implies that a conclusion that the JRS prevented the loss of all furloughed jobs is not credible.

Conclusions

This article has analysed the differing impacts of the Covid‐19 pandemic on graduates and non‐graduates. We have highlighted how the crisis has exacerbated structural differences in the UK labour market as job loss has disproportionately affected non‐graduates. We have argued that the JRS had a major impact on the UK labour market, preventing around 48%–58% of jobs that were furloughed under the scheme from being lost. The large majority of these jobs would have been in lower productivity occupations and largely been done by non‐graduates.

Research into an ongoing event of the scale and rarity of the Covid‐19 pandemic must be treated with caution. This article makes a series of strong assumptions. First, there is no uncertainty. Second, we assume the pandemic does not affect the steady‐state or change structural relationships. Third, there are limited margins of adjustment for firms since we do not include hours worked. Fourth, we use a linearization of the model34, even though the pandemic moves the economy some distance away from the steady‐state. Further research on these issues using alternative approaches would be useful. For example, there may be a role for models that incorporate labour market institutions (e.g. Pissarides, 2000; Thomas and Zanetti, 2009; Zanetti, 2011). And including a richer set of margins of adjustment may well yield additional interesting insights. Our results show that a standard DSGE model with labour market frictions can give useful insights into the impact of the pandemic and can be useful for policy evaluation. But our results are dependent on a specific model that makes a series of assumptions. Given the unprecedented scale and nature of the crisis, it is important that the impact of the pandemic on the labour market is also analysed using alternative approaches, which address the impact of uncertainty. We hope to extend our analysis to address these issues in subsequent work.

Appendix S1. Supplementary Material

Click here for additional data file.