Urban labour market resilience during the Covid-19 pandemic: what is the promise of teleworking?

ABSTRACT The emergence of the Covid-19 pandemic caused immense labour market turbulences and a large-scale turn towards teleworking. This paper contributes to the understanding of how teleworking shapes regional economic outcomes by focusing on labour market resilience in US cities during the 2020 Covid-19 emergency. It examines employment and labour demand and finds that the share of teleworkable jobs is linked to stronger employment resilience, and enhances labour demand resilience during the onset of the pandemic and in smaller cities. The paper discusses possible mechanisms and policies that can help leverage the promise of teleworking for resilient labour markets.


INTRODUCTION
The Covid-19 pandemic is like no other in recent history.Globalization made it possible for a local health emergency to spread quickly across the globe, forcing governments to implement restrictions and lockdowns: in the United States almost the entire population was affected by state or local lockdown measures in April 2020 (Baek et al., 2020).The economic shock that came with the pandemic resulted in an abrupt recession with rapid and devastating effects on labour markets; the US unemployment rate soared to over 14% in April 2020, 1 while labour demand collapsed by 44% between February and April 2020 (Forsythe et al., 2020).To maintain business continuity, many firms shifted to telework, where employees whose work tasks could be performed remotely continued to work from the safety of their home (Brynjolfsson et al., 2020).Teleworking, that is, performing one's employment duties remotely, became a predominant mode of work for many occupations where physical presence was not strictly required (Organisation for Economic Co-operation and Development (OECD), 2021).Many observers agree that 'teleworking is here to stay', with a potential of turning local labour markets into national or even global ones, at least for some occupations (Maselli, 2020;OECD, 2020a).
Much of the academic research about teleworking discusses the productivity impacts, work-life balance, and other employer-or employee-centred outcomes (Choudhury et al., 2021;George et al., 2021;Zhang et al., 2020).Policy discussion, in contrast to the academic one, often focuses on the national or regional outcomes of teleworking, such as its potential effects on agglomerations and the promise of telework for reviving less central regions.Creating better conditions for remote work can attract digital workers, increase the tax base of a region and expand demand for services creating jobs in the sector.These benefits would be particularly welcome and felt in places outside of large agglomerations, such as remote regions or smaller cities.
With all the recent hype around teleworking, however, it is important to collect and carefully evaluate empirical evidence on the actual effects of remote work on economic outcomes before formulating specific policy interventions.This paper contributes to the knowledge base about the regional economic impacts of teleworking.It focuses on labour market performance in US cities during 2020, the year when the pandemic started.The paper investigates the relationship between the share of teleworkable jobs before the pandemic and labour market performance in metropolitan statistical areas (MSAs).It applies the resilience concept and uses data on total MSA employment and labour demand collected from online job postings to derive labour resilience indicators.The paper also investigates the possible mechanism of the observed positive relationship and offers policy-relevant discussion on the promise of teleworking for labour market resilience in US cities.
The empirical analysis finds support for the idea that the concentration of teleworkable jobs in a city can enhance the resilience of local labour markets.We show that employment resilience benefited from teleworkability throughout 2020 and especially in smaller cities with up to 500,000 residents.While the effect of teleworkability on labour demand resilience is theoretically ambiguous, the analysis demonstrates a positive link during the first wave of Covid-19 (March-June 2020) and in smaller cities.
The positive relationship between teleworkability and labour market resilience likely works through two mechanisms.On the one hand, jobs that can be done remotely are less likely to be shed during a health emergency, as they can continue their activities even when normal operations from an office become impossible.On the other, the steady earnings flow to teleworking professionals was likely generating demand for services and supported jobs, as a positive link between teleworkability and labour demand resilience is stronger for non-teleworkable occupations.
These expected indirect income effects (when additional services demand from higher paid remote workers stimulates local employment) seem to be an important part of the ongoing efforts by regions and countries to promote teleworking.Our analysis offers some preliminary (short-term) evidence that such expectations might be justified.Thus, the primary policy implication that follows from this paper is that regions should invest in teleworking capabilities, such as improved broadband infrastructure and digital education and skills (Clancy, 2020).This can support regional resilience also in the face of potential future disruptions linked to climate change or new episodes of healthcare emergencies, which might require employees to work remotely.

THE COVID-19 PANDEMIC AND US LABOUR MARKETS
The Covid-19 shock had an immediate impact on US labour markets.Commonly, the health of labour markets is measured by employment levels or changes, but research on past recessions shows that firm hiring is a dominant factor driving employment declines during downturns (Shimer, 2012).This means that during a crisis such as the Covid-19 pandemic, analysing both employment and hirings offers a more comprehensive picture.
The evidence for labour market performance after the Covid-19 emergency hit is still accumulating.Available indications suggest that disruptions to the US economy happened simultaneously at many levels, but not all sectors and regions were affected equally; employment losses have been concentrated disproportionately in low-wage and face-to-face jobs (Cajner et al., 2020;Cortes & Forsythe, 2020).In contrast, high-skill and high-wage jobs were less likely to be shed during the great lockdown, although they do show a large decline in labour demand (Campello et al., 2020;Forsythe et al., 2020;Tsvetkova et al., 2020).
For example, Tsvetkova et al. (2020) report that the number of online vacancy announcements contracted the most in non-tradable service occupations, particularly those involving face-to-face interactions, such as food preparation and serving.Other occupations, including those in healthcare, transportation and construction, experienced only a small decline or even an increase in demand.Using firm level data, Campello et al. (2020) document that small and large firms reduced their hiring by over 50% and 30-40%, respectively, and that hiring cuts affected high-skill jobs more than for low-skill jobs.
Spatially, labour markets were stronger affected in urban areas compared with their rural counterparts.In cities, as follows from an analysis of the US Current Population Survey Covid-19 supplement, adults were more likely not to be paid for missed hours or be unable to work due to the spread of the virus.At the same time, urban dwellers were more likely to work remotely (Brooks et al., 2021).An analysis of aggregate data suggests that employment losses were higher in MSAs than in nonmetro areas, with the highest losses in MSAs with more than 5 million inhabitants (Cho et al., 2020).This pattern of stronger negative impacts in larger places is also confirmed for hiring: using online job postings data, Tsvetkova et al. (2020) find that labour demand declined more in MSAs with a population larger than 500,000.
Our data confirm these observations (Figure 1).Larger MSAs (above 500,000 residents) experienced a bigger relative drop in both vacancies and total employment during 2020.After an initial drop, the comparative gap was stable across the two metropolitan size groups for employment, but it expanded for online job postings.
Labour market performance with respect to major disruptions is often examined through the concept of resilience, which generally refers to the responsiveness of systems to shocks or changes (Christopherson et al., 2010).Martin (2012) identifies four different dimensions of resilience: resistance: the degree of sensitivity of a regional economy to a recessionary shock; recovery: the speed and magnitude of the recovery; reorientation: the adaptation of a regional economy; and renewal: the resumption of the previous growth path or a shift to a new one.The Great Recession, which followed the 2008 financial crisis, triggered an academic focus on the resilience (resistance and recovery) of US labour markets (Deller & Watson, 2016;Doran & Fingleton, 2018;Grabner & Modica, 2021;Han & Goetz, 2015).Yet, past empirical research may only provide limited guidance on regional labour market resilience in the context of the Covid-19 pandemic, as the magnitude and the nature of this shock is different along multiple dimensions.To examine resilience with respect to the Covid-19 shock, we focus on the resistance and (initial) recovery elements of resilience and consider two important components of labour market performance, total employment and hiring; the latter is approximated by online job postings.Labour demand (number of internet job postings) is a relatively small component of the labour market (the number of online vacancy announcements is only around 2% of the total number of jobs) but are an important forward-looking measure of the developments in the labour markets.Data on job postings are more detailed, which allows for additional analyses that can shed light on the mechanisms of a relationship between the two variables of interest in this study, teleworkability (discussed in the next section) and labour market resilience.

TELEWORKABILITY AND THE COVID-19 PANDEMIC
The past crises impacted economies through channels such as the supply of capital (the 2008 financial crisis) or technology (industrial revolutions).In contrast, the Covid-19 pandemic hit the human capital component of the production process (Campello et al., 2020).To protect lives and curb the spread of Covid-19, state and local officials in the United States introduced stay-at-home orders and other restrictions.Baek et al. (2020) document that by 4 April 2020, nearly 95% of the US population was under stay-at-home orders.Along restrictions and concerns about infections in the workplace came another immense change affecting the labour marketa largescale shift towards remote work or teleworking.
Teleworking refers to work arrangements under which individuals work from a physically distant location (not where collaborating co-workers are located).The term is often used synonymously with remote work, smart work, telecommuting or working from home (Clancy, 2020).While the phenomenon is not new, the Covid-19 crisis forced firms to introduce telework on a large scale (OECD, 2020b).Fortunately, a considerable share of jobs are teleworkable, which made a widespread adoption of telework possible.Dingel and Neiman (2020) find that 37% of jobs in the United States can be performed remotely, but with a significant variation across industries.Teleworkable jobs tend to concentrate in high-education, highskill and high-wage industries, and thus account for 46% of all US wages.Firm survey data likewise suggest that the adoption of telework was much more common in industries with better educated and better paid workers (Bartik et al., 2020).According to Brynjolfsson et al. (2020), 35% of those who used to commute to work switched to teleworking as the pandemic began.The authors show that this switch was related to the incidence of Covid-19 but most changes were already in place by April 2020.
The dynamics of job postings were clearly distinct for teleworkable and non-teleworkable occupations after the first shock of the Covid-19 pandemic in 2020.While the initial drop in demand for both types of occupations was comparable, recovery was slower and weaker for the jobs that can be performed remotely (Figure 2).This is in line with other authors' observations, who report that labour demand in teleworkable and high-skill occupations contracted more (Campello et al., 2020;Forsythe et al., 2020;Tsvetkova et al., 2020), while demand for certain non-teleworkable occupations such as healthcare and transportation experienced stable or increasing demand.

CONCEPTUAL MODEL: THE LINK BETWEEN TELEWORKABILITY AND LABOUR MARKET RESILIENCE
How could the ability to telework influence local labour market resilience during Covid-19?As follows from Figure 3, the relationship is not straightforward, since multiple processes are unfolding simultaneously and their links to the outcomes of interest (resilience of labour market) would depend on the industrial and occupational composition.While Figure 3 is not intended to depict all the intricacies of the heterogeneity, 2 it offers the main channels and differentiates between teleworkable and non-teleworkable occupations in the outcomes.Given that the negative effects from the spread of the virus and of the decrease in economic activity and demand for goods and services that followed affected more readily non-teleworkable occupations, concentration of teleworkable jobs in a local economy could serve as a cushion able to soften the blow of the crisis.While mitigating the devastating effects of the pandemic on the labour markets, teleworkability likely turned into a factor directly supporting local demand for goods and services and, hence, for workers. 3 If one abstracts from the industrial and occupational heterogeneity in the effects and focuses on the scenarios that more likely, the share of jobs in a location that can be performed remotely, is expected to be positively linked to employment performance ceteris paribus.This is so because restrictions on movement and stay-at-home orders would have minor ability to disrupt the working process provided employees have the needed technology and infrastructure (personal computers and good-quality internet connection) at home.When it comes to labour demand, however, the effect is ambiguous.On the one hand, teleworkability can have a negative effect on labour demand, as teleworkable jobs are less likely to be shed during the crisis and, combined with overall reduced demand and economic activity, this would translate in lower need for new hires.Forsythe et al. (2020) show that teleworkable occupations experienced a smaller employment loss and fewer unemployment insurance claims but a similar decline in job postings compared with non-teleworkable occupations.On the other hand, teleworkability can also be positively linked to labour demand through a local demand effect.Teleworkable jobs tend to pay higher wages compared with the non-teleworkable ones (Brussevich et al., 2020;Dingel & Neiman, 2020) and, thus, are likely to contribute to local demand for goods and services supporting labour demand indirectly.Indeed, Tsvetkova et al. (2020) show that teleworkability in US MSAs was positively related to the number of job postings in the first semester of 2020.
In this paper we investigate the ability of teleworking to soften the negative impact on urban labour markets during the year when the pandemic hit.We study both direct and indirect effects (interaction with Covid-19related restrictions) and separately explore a possibility that local teleworkable jobs support labour demand.

Time period and Covid-19 waves
In this paper we focus on labour market resilience in the first months of the pandemic, from March to December 2020.While the choice of this timeframe is motivated by data availability, the period is well-suited to answer our research question and to investigate whether teleworkability in the US MSAs was linked to greater resilience of the labour markets.Especially the initial unfolding of the pandemic was an extreme and unexpected emergency.It brought about very high uncertainty, which rapidly translated in changes in labour markets with greater volatility in hiring choices (Campello et al., 2020).During the first  Simone Maria Grabner and Alexandra Tsvetkova wave of Covid-19, almost the entire US population was at some point affected by the lockdown measures (Baek et al., 2020).Moreover, Brynjolfsson et al. (2020) document that the transitions to remote work were mainly completed by April, with little changes in the consecutive waves despite changing pandemic and economic conditions throughout 2020.
To further explore the link and to allow for a variation in the relationship between teleworkability and labour market resilience over time, we separately analyse three Covid-19 waves that can be observed during 2020: from March to June, from July to September and from October to December (Figure 4).

Dependent variable: labour market resilience based on employment and online job vacancy data
We aim to understand the short-term reaction of labour markets to an external shock, and therefore opt for a commonly used index of labour market resilience that captures resistance and recovery.The measure we use is based on Lagravinese (2015) and Martin et al. (2016).It compares performance of each labour market during a period of study in 2020 with the corresponding period in 2019, last pre-pandemic year (equation 1), and benchmarks each MSA's performance to the average performance across all MSAs combined (equation 2).The resultant index is centred around 0, where a positive value suggests that an MSA is more resilient than the average and a negative value implies that an MSA is less resilient than the average.Such approach is widely used in studies of resilience (Cainelli et al., 2019;Giannakis & Bruggeman, 2017): (1) where BM m is the benchmark reaction in period m, and refers to the average change in labour market performance (LMP) indicator across all MSAs (i), which is then used to calculate the resilience of labour market in MSA i in period m.We calculate the resilience index for three separate periods according to Covid-19 waves (March-June, July-September and October-December).
To reach a more complete representation of the dynamics unfolding during the first months of the pandemic, 4 we use two metrics to capture the health of metropolitan labour markets: first, total monthly employment; and second, labour demand as proxied by online job vacancy postings.
Vacancy data were provided by Burning Glass Technologies (BGT), currently EMSI Burning Glass, a company that scrapes up-to-date job postings from online job boards and company websites.Scraped information is saved at a vacancy level and has information on each vacancy's date of posting, location, industry and occupation, as well as other characteristics such as the name of on employer and required skills.
While the use of this data source in academic and policy research is on the rise (e.g., Cammeraat & Squicciarini, 2020;Forsythe et al., 2020;Hershbein & Kahn, 2018), there are obvious advantages and limitations of using these data.In general, the advantages of the data set are granularity, timeliness and sample size.From the perspective of the organization of economic activity in space and for the purposes of this paper, the strength of the data is the granularity of location and, most importantly, occupation data, which are classified at the six-digit Standard Occupational Classification (SOC) level.
Yet, there are several well-documented limitations of this data source that need to be kept in mind.The first limitation is that announcements collected by BGT are a subset of job openings as they by design cover only online postings.Labour demand in industries that are less likely to advertise online (e.g., construction or agriculture) may be higher than what can be inferred from the online vacancy ads.A corollary of this is that the BGT data tend to have more postings for larger cities and for higher skilled occupations.Nevertheless, the use of the internet to hire workers is expanding, particularly in the United States where online job advertisements are very common.The distribution of online job postings across occupations in US MSAs was found to be comparable with that found in the occupational employment statistics (Hershbein & Kahn, 2018) and it is hoped that using the last full year of data available to the authors at the time of writing further mitigates this concern.
While the employment data are available only at the MSA level without any additional details, we take advantage of the granularity of the online vacancy data, first, to better understand the trends in the metropolitan labour demand for teleworkable and non-teleworkable occupations during the pandemic and, second, to better fit our models where we are able to remove the influence of the occupation-specific fixed effects.To achieve the former, we use six-digit occupation teleworkability index developed by Dingel and Nieman (2020) to calculate resilience of labour demand in MSAs separately for teleworkable and non-teleworkable occupations.For the latter, we calculate the resilience index for MSA-occupation pairs for 22 occupational groups. 5 Table 1 shows summary statistics for the five different dependent variables.As can be seen, the two metrics, employment and labour demand, indeed seem to capture differing dynamics in the labour market with labour demand showing quicker adjustments, as evidenced by the changes over time and by the magnitude of the standard deviation.Figure 5 shows the maps of the resilience index during the first Covid-19 wave (March-June 2020).For maps of resilience values during the subsequent waves, see Figures A1 and A2 in Appendix A in the supplemental data online.

Explanatory variable: teleworkability
The 2018 share of teleworkable jobs in an MSA (teleworkability) is the main explanatory variable.The MSA-level measure is provided by Dingel and Neiman (2020) and is widely used in the related literature (e.g., Bartik et al., 2020;Cho et al., 2020;Tsvetkova et al., 2020).The authors first classify detailed occupations (at the six digits of the SOC developed by the US Bureau of Labor Statistics) by the feasibility of remote work based on surveys of worker experience.In the next step, they merge this classification with the 2018 MSA occupational employment counts to get the share of all jobs in an MSA that can be performed remotely.Table 2 lists the top 10 and bottom 10 MSAs in terms of teleworkability in 2018.The shares vary widely across MSAs, ranging from over 50% in California-Lexington Park (MD) and San Jose-Sunnyvale-Santa Clara (CA) to less than 20% in The Villages (FL).

Control variables
All models include a set of controls, which are meant to factor out the likely influence of a range of policy, epidemic, economic and social factors that can be linked to the resilience of labour markets.
The presence and severity of restrictions (as well as their enforceability and the willingness of the citizens to follow the guidelines) which aimed at curbing the spread of Covid-19 should be of great relevance to the performance of the labour markets in cities.In the United States, usually states were imposing such restrictions, which resulted in a heterogenous landscape of Covid-19 response policies.This is related to the federal system of government, but also to rising political contestations (Hale et al., 2020), where the stringency and adherence to restrictions was often a matter of political preferences (Rothwell & Makridis, 2020).To account for the severity of the restrictions, we use the Oxford Covid-19 Government Response Stringency Index from the Covid-19 government response tracker, which was developed and

REGIONAL STUDIES
maintained by the Blavatnik School of Government at the University of Oxford (Hale et al., 2021).This stringency index measures the intensity of restrictions based on nine response indicators.Among those are school closures, workplace closures, the cancellation of public events, restrictions on gatherings, closing public transport and travel bans.The stringency index ranges from 0 to 100 and is reported as a daily series at the state level, which we aggregate into an average over a given wave of the pandemic.For multi-state metropolitan areas, we assigned the state where most of the MSA's population resides.
Further controls include: the spread of Covid-19; MSA population density to account for agglomeration effects (Cho et al., 2020;Duranton & Kerr, 2015); population and wage growth and the unemployment rate to control for pre-crisis economic trends (Martin & Sunley, 2015); related and unrelated variety describe the local economic structure in terms of diversity and relatedness (Grabner & Modica, 2021;Xiao et al., 2017) and education (Doran & Fingleton, 2015).
Table 3 describes all dependent and independent variables and indicates data sources.For summary statistics of the control variables and a correlation matrix of the

Empirical models
To investigate the link between the share of teleworkable employment in an MSA and labour market resilience as hypothesized in the conceptual model subsection, we estimate an empirical model presented as follows: where Resilience i m stands for the resilience index in MSA i in period m, with period referring to one of the Covid-19 waves (March-June, July-September and October-December 2020); Teleworkability i stands for the pre-pandemic share of teleworkable employment in an MSA as provided by Dingel and Neiman (2020); and the vector Controls i contains control variables, which are discussed in Table 3.All annual control variables are fixed in the last year (or a multi-year period) for which data are available. 6Equation ( 3) is estimated at the level of MSAs for both employment and labour demand and at the MSA-occupation level for labour demand only.In the latter case, resilience is calculated at MSA-occupation (22 occupational groups) level and the model is augmented with occupation-level fixed effects to account for potential occupation-specific differences in the link between teleworkability and resilience in labour demand (corresponding subscripts are omitted for simplicity).
The hypothesized positive link between the share of teleworkable jobs and labour market resilience can be impacted by the restrictions in place.Thus, as a robustness check, we also estimate equation ( 4), which adds an interaction term between the share of teleworkable jobs pre-pandemic and the severity of the anti-Covid-19 restrictions.To mitigate the problem of multicollinearity, which routinely arises in such regressions, the teleworkability and the severity of restrictions variables were demeaned by subtracting the mean value of the variable.All subscripts and superscripts have the meaning identical to equation (3): To further investigate the relationship between teleworkability and labour market resilience in the US MSAs, we repeat the analysis by estimating equations ( 3) and ( 4) separately for teleworkable and non-teleworkable occupations (defined at six-digit SOC code level by Dingel & Neiman, 2020) and separately for smaller (up to 500,000 residents) and larger (above 500,000 residents) MSAs.

Main analyses
There are several mechanisms at work that can link teleworkability to the two components of the metropolitan labour market resilience as discussed in the conceptual model section.The link between the share of teleworkable employment and total employment is likely to be positive, while the link to the labour demand can be both positive and negative.Table 4 shows estimation results for the main explanatory variables and three models, resilience index based total employment, total vacancies in an MSA and vacancies in an MSA-occupation pair.Each model is separately estimated by time period.Full estimation results are provided in the online Appendix.
Table 4 shows that indeed there is a positive association between the pre-pandemic share of jobs that can be performed remotely, and labour market resilience approximated by employment.The coefficient is positive and statistically significant in all three time periods.When it comes to vacancies, regardless of the specification, a positive relationship is detected only during the first Covid-19 wave, the most profound one in terms of labour market collapse.The size of the coefficient shows that an MSA with one more percent of teleworkable jobs tended to have their resilience index higher by 0. To probe the possible mechanism of the link between teleworkability and labour demand, we estimate the vacancy models separately for teleworkable and non-teleworkable occupations by period (Table 5).The results suggest that the positive association between the share of teleworkable jobs and online job postings reported in Table 4 is through an indirect income mechanism.If teleworkable jobs were less likely to be shed during the pandemic and they, on average, tend to be higher paid, local demand for goods and services that remote workers generate is likely to translate into more job openings in non-teleworkable occupations keeping everything else constant.
In the next step of the analysis, we explore potential geographical differences in labour market resilience in US MSAs. Figure 1 suggests that there were differences in labour market performance across smaller and larger MSAs.We follow the grouping of Figure 1 and divide all Metropolitan Statistical Areas into those below 500,000 residents (small) and those above (large).Table 6 presents estimation results and shows that the positive association between teleworkability and labour market resilience is more likely to be detectable in small MSAs.It is true for both the employment and vacancies metrics.There is also some variation over time with the link being Urban labour market resilience during the Covid-19 pandemic: what is the promise of teleworking?2529 REGIONAL STUDIES statistically significant also for large MSAs during the second Covid-19 wave (July-September 2020).In this period, the relationship between the share of jobs that can be performed remotely and the resilience of employment indicators is positive and statistically significant at the 0.05 level, but it is negative for the resilience of job postings.It appears that the negative mechanism hypothesized in the conceptual model subsection outweighs the positive one.Indeed, the strength of the negative mechanisms would be expected to be greater in regions with more teleworkable jobs, such as larger cities.
The final set of results for the main model is shown in Table 7. Differences in the link between teleworkability and labour market resilience measured by online job vacancies are probed in this specification.The results generally confirm observations from the previous tables.A positive relationship tends to be detectable in the beginning of the pandemic, in small MSAs and it appears stronger for non-teleworkable occupations.
Overall, our analysis supports the idea that concentration of teleworkable jobs in an MSA can enhance the resilience of local labour markets.For employment, teleworkability can offer protection as the activities do not need to be discontinued (and jobs shed) when normal operations from the office become impossible.For vacancies, while we cannot specifically test for it, the results are in line with a hypothesized demand mechanism.The positive link between the share of teleworkable jobs and labour demand during the study period is more consistently observed for non-teleworkable occupations.Since jobs that can be performed remotely, on average, pay higher wages compared with jobs that require physical presence (Brynjolfsson et al., 2020;Dingel & Neiman, 2020) and they were less likely to be shed (Cajner et al., 2020;Cortes & Forsythe, 2020), we conclude that the ability of teleworkable employees to sustain local demand for goods and services likely supported metropolitan labour markets in the beginning of the pandemic.Yet, the regression analysis also shows that teleworkability is not a panaceait can be unrelated or even negatively related to the resilience of labour demand in teleworkable occupations.Given the evidence that the number of job postings contracted more in teleworkable occupations (Figure 2) (Forsythe et al., 2020), this is not unexpected.It may, however, have negative implications for post-pandemic recovery if demand for teleworkable occupations is not picking up (Campello et al., 2020).
In terms of control variables, it is worth highlighting that the spread of Covid-19 (measured by the daily average number of cases per 10,000 residents) is not statistically significant in almost all specifications, a result that also Forsythe et al. (2020) and Tsvetkova et al. (2020) observe.The severity of restrictions tends to be positively related to the resilience of metropolitan labour markets.This might reflect the fact that introducing and following the restrictions in the United States tended to be linked more to political inclinations of governors, mayors and the population at large than to the gravity of the

July-September
October-December

March-June
July-September October-December

March-June
July-September Note: ***Significant at the 0.01 level; **significant at the 0.05 level; and *significant at the 0.1 level.All models include population density, population growth, wages growth, unemployment rate, unrelated industrial variety, related industrial variety and share of the population with bachelor's degree or higher as controls.Robust standard errors are shown in parentheses.

2530
Simone Maria Grabner and Alexandra Tsvetkova health situation.On average, if better-performing places were more likely to introduce restrictions, which on the surface would seem plausible, the coefficient would pick up this situation.We also find that both related and unrelated industrial variety tended to have a persistent negative association with labour market resilience.Previous literature mostly found a positive effect of related variety on regional economic resilience (Cainelli et al., 2018;Xiao et al., 2017), however metropolitan areas appear to benefit less from related variety (Grabner & Modica, 2021).Moreover, in the short-term, like in our analysis, related variety appears to exacerbate the demand shock by potential contagion effect across related industries (Acemoglu et al., 2013).More educated MSAs tended to have lower labour market resilience during the studied periodlikely a reflection that larger cities, where the average level of educational attainment is higher, were hit harder in the beginning of the pandemic.Population density was negatively linked to the resilience of the metropolitan labour markets measured by employment but positively to the resilience measured by vacancies.This could offer an additional indirect support to the hypothesized income mechanisms behind the link between teleworkability and labour demand.Other significant coefficients are in line with expectations.MSAs with greater levels of unemployment performed worse, while faster growing cities tended to have more resilient labour markets.

Additional analyses
A series of additional analyses were performed to explore the sensitivity of the reported results to alternative specifications. 8Table 8 reports results for a full sample model, which additionally includes an interaction between the share of teleworkable employment and the anti-Covid-19 restrictions (for the full models, see Appendix A the supplemental data online).The main results in Table 8 are in line with the ones reported previously.For labour demand resilience, however, Table 8 reveals the likely presence of the interaction effects.In MSAs with more severe restrictions, the positive association between teleworkability and job postings tends to be of smaller magnitude, at least in the beginning of the pandemic.

CONCLUSIONS
Teleworking is heralded as a potential key to labour market resilience during pandemics caused by a contagious disease.
For less central regions and smaller cities, teleworking can offer hope for additional growth, both in terms of population and economic performance.The attention to the regional economic effects of teleworking so far seems stronger in policy discussion that in the academic one.Policy makers seem attuned to the idea of advancing economic prospects of their regions through teleworking.As the modalities of work are changing, teleworking indeed might offer an additional lever, especially to the amenityrich places.Yet, the expectations of the positive effects should rely on the results of empirical research, which is able to inform corresponding policy design.This paper is a contribution to building the knowledge base on the link between teleworking and regional economic outcomes.In particular, it investigates the relationship between the pre-pandemic share of jobs that can be performed remotely and labour market resilience in the US MSAs during 2020, the year when the pandemic started.
Our knowledge on the link between teleworking and economic performance is naturally limited.Mass adoption of remote work is a very recent phenomenon triggered by the Covid-19 pandemic and the many restrictions imposed by the governments to curb its spread.Research on the past crises, while providing many useful insights into the factors related to regional economic resilience, are only partially able guide us now, as the ongoing crisis has different origins.
We find a spatially uneven pattern of labour market resilience across MSAs and that the pre-crisis concentration of  Urban labour market resilience during the Covid-19 pandemic: what is the promise of teleworking?

REGIONAL STUDIES
teleworkable jobs may partly explain this pattern.Our analysis documents a generally positive link between the teleworkability of a local economy and its labour market resilience during the Covid-19 pandemic when it comes to employment.The positive link between teleworkability and labour demand resilience, on the other hand, is dependent on several other factors.First, it is observable mostly during the first wave of the pandemic, when the labour markets were hit unexpectedly and most profoundly.In the subsequent periods, the link mostly disappears.Second, it appears that smaller cities were able to benefit from the teleworkability of their economies as the relationship is usually statistically insignificant in larger cities.Finally, positive and statistically significant coefficients emerge more often in the models focusing on the non-teleworkable occupations.This implies that the overall positive link between teleworkability and labour market resilience (at least in the labour demand part) seems to stem from the indirect income effects where generally higher wages of remote workers could have supported local labour markets.
The results of this paper bring up several considerations important for regional policy.First, as the paper does not establish a causal relationship, additional analyses are needed to better understand the effects of teleworking on labour markets in regions.The effects will likely vary in different types of places and a more precise knowledge of how and where teleworkability can contribute to economic performance is needed to design more efficient policies.Our analysis already offers insight in this veinwe document a consistent positive link between teleworkability and labour market resilience measured by both employment and labour demand in smaller cities.Second, although we are unable to probe the mechanisms behind the observed relationship for employment, the evidence from labour demand seem to suggest that income effects do exist.In other words, the presence of (usually higher paid) remote workers in an economy is associated with greater demand for non-teleworkable occupations.
If regions choose to pursue teleworking as a part of their economic development strategy, the following would be of importance for the success.A comprehensive coverage of a reliable and high-quality information technology (IT) infrastructure becomes a must.Regions can strengthen conditions for teleworking by investing in IT infrastructure, providing support for remote work to firms and employees as necessary and by investing in digital education and skills (OECD, 2020b).Telework is not only an effective tool to continue business operation during a pandemic but can also act as an emergency response in the case of extreme weather events or other scenarios where commuting would not be considered safe.Moreover, teleworking is likely here to stay (OECD, 2020a).Bartik et al. (2020) uncover that more than one-third of US firms, which switched to telework during Covid-19, believe that telework will remain common at their company even after the pandemic.Telework may also provide several public benefits such as enhanced aggregate productivity, geographical dispersion of employment and reduced carbon emissions if fewer commutes would be necessary (Clancy, 2020).Thus, regional policies which enhance teleworkability can prepare regions to be more resilient in front of a variety of potential crises as well as support economic development in general.
The analysis in this paper can be further extended and improved as more data become available.The first goal would be to establish causality and to investigate the heterogeneity of the effects depending on the specific regional conditions.Another important improvement would be to use a more comprehensive and up to date measures of teleworkability and labour market indicators, which can include, in addition to vacancies and employment, firings and other types of separations, unemployment, labour market participation rates among other relevant metrics.Lastly, the widespread adoption and continued use of telework will likely have comprehensive repercussions on the distribution of economic activity within and across geographies (Clancy, 2020;Delventhal et al., 2021) which requires additional attention to this quite new phenomenon in order for regions and cities to be able to maximize the opportunities opened by these processes and mitigate the possible risks.

Figure 1 .
Figure 1.Monthly vacancies (left) and employment (right) in 2020 (indexed to January) in small and large metropolitan statistical areas (MSAs).Sources: Authors' calculations based on EMSI Burning Glass data (vacancies); and US Bureau of Labor Statistics (BLS) (employment).

Figure 2 .
Figure 2. Monthly vacancies in 2020 (indexed to January) across teleworkable and non-teleworkable occupations.Source: Authors' calculations based on the EMSI Burning Glass data; and Dingel and Neiman (2020) occupation classification.

Figure 3 .
Figure 3. Conceptual model of the developments in the labour market and the role of teleworking. 2524

Figure 5 .
Figure 5. Map of metropolitan labour market resilience during the first Covid-19 wave (top map, vacancies; bottom map, employment).Sources: Authors' calculations based on data from EMSI Burning Glass (vacancies); and US Bureau of Labor Statistics (BLS) (employment).
02 when based on employment, by 0.05 when based on aggregate MSAlevel vacancies and by 0.06 when based on vacancies in MSA-occupation groups during the first wave.Examples of such differences are Chattanooga, TN-GA and Madison, WI for employment; New Bern, NC and Chicago-Naperville-Elgin, IL-IN-WI for aggregate MSA vacancies as well as Buffalo-Cheektowaga-Niagara Falls, NY and Philadelphia-Camden-Wilmington, PA-NJ-DE-MD. 7

Table 1 .
Summary statistics for the resilience index.

Table 2 .
Metropolitan statistical areas (MSAs) with the highest and the lowest shares of teleworkable jobs.
Note: North American Industry Classification System; SOC, Standard Occupational Classification.

Table 4 .
Estimation results for all sample, no interaction.

Table 5 .
Estimation results for teleworkable and non-teleworkable occupations, no interaction.Significant at the 0.01 level; **significant at the 0.05 level; and *significant at the 0.1 level.All models include population density, population growth, wages growth, unemployment rate, unrelated industrial variety, related industrial variety and share of population with bachelor's degree or higher as controls.Robust standard errors are shown in parentheses.Urban labour market resilience during the Covid-19 pandemic: what is the promise of teleworking?2531

Table 6 .
Estimation results for all sample by size, no interaction.
Note: ***Significant at the 0.01 level; **significant at the 0.05 level; and *significant at the 0.1 level.All models include population density, population growth, wages growth, unemployment rate, unrelated industrial variety, related industrial variety and share of population with bachelor's degree or higher as controls.Robust standard errors are shown in parentheses.

Table 7 .
Estimation for teleworkable and non-teleworkable occupations, by size, no interaction.Significant at the 0.01 level; **significant at the 0.05 level; and *significant at the 0.1 level.All models include population density, population growth, wages growth, unemployment rate, unrelated industrial variety, related industrial variety and share of population with bachelor's degree or higher as controls.Robust standard errors are shown in parentheses.Significant at the 0.01 level; **significant at the 0.05 level; and *significant at the 0.1 level.All models include population density, population growth, wages growth, unemployment rate, unrelated industrial variety, related industrial variety and share of population with bachelor's degree or higher as controls.Robust standard errors are shown in parentheses.