COVID-19 and labour market resilience: evidence from large-scale recruitment behaviour

ABSTRACT We construct a theoretical model to interpret the structural shock from the COVID-19 pandemic and the response of local labour market and industry specialization. The empirical study takes the large-scale online labour market of China to analyse firms’ hiring demand for 20 industries across 380 cities with monthly recruitment data from May 2017 to September 2020. Post-event quantitative analysis on job postings and employer demand highlighted that the pandemic resulted in an unemployment shock and industry- and city-level redistribution of the worker. China’s local job market resilience also revealed a regional imbalance, correlated with pandemic risk, city scale and industry structure.


INTRODUCTION
The COVID-19 pandemic has caused a large shock both to the local and global economies (Albanesi & Kim, 2021;Bartik et al., 2020;Forsythe et al., 2020;Guerrieri et al., 2022;Van Bavel et al., 2020).To curb the macroscopic spread of the virus, China immediately implemented lockdown policies during the first wave in late January 2020, coinciding with the traditional Chinese New Year holiday season, to reduce large-scale population mobility (Allen et al., 2020;Brodeur et al., 2021;Chen et al., 2021;Fang et al., 2020;He et al., 2020;Hu et al., 2022).At the microscopic level, China enforced social distancing and mask-wearing regulations to further restrict virus transmission via respiratory droplets (Chen et al., 2021;Dai et al., 2021;Qiu et al., 2020).These combined measures have led to a significant reduction in the number of confirmed cases and the effective recovery of the Chinese economy during the first two quarters of 2020 (Figure 1a, b).However, the impact of unexpected shocks on the labour market, resulting from the pandemic and its containment policies, is more profound.Industries heavily dependent on offline social activities and transportation have been forced to adapt to this new normal and shift their businesses online.This shift has also accelerated their adoption and diffusion of online technologies.Consequently, the market has further reduced its hiring demand for human-centric offline services, such as dining, transportation and tourism.
In this study, we focus on the impact of COVID-19 on China's job market and the latter's resilience to sudden shocks.Having witnessed a complete cycle of outbreak and recovery, we attempt to extend the literature by studying the response of local labour market and the resilience of regional economy, considering exogenous shocks from the pandemic and lockdown policies at the aggregate level.Since the number of confirmed cases and deaths between the Hubei and non-Hubei regions differed drastically throughout the outbreak period (Figure 1c,d), the pandemic containment and prevention policies were also different in the Hubei and non-Hubei regions.Hence, the study uses the empirical strategy to quantify the response effects by comparing cities inside and outside Hubei province.To interpret the mechanism through which the local labour market responds to the pandemic shock, we constructed a theoretical model and used numerical computation to explore the equilibrium outcome.The empirical study has gathered large-scale online recruitment data from one of the largest online labour markets in China.We obtained their proprietary recorded data on corporate job postings and average offered wages for 20 industries across 380 Chinese prefecture-level cities, at monthly level between May 2017 and September 2020.In addition, we included China's lockdown policy to identify pandemic-induced shifts in job demand, with a three-year duration spanning the pre-and post-COVID periods.The empirical facts indicate the shift due to the pandemic shock in China across different dimensions.We used an event study approach to quantify the impact and economic costs of lockdown policies for curbing the spread of the virus.
Our theoretical framework is drawn from three streams of literature.First, it is drawn from existing studies on the resilience of regional labour markets (Borsekova & Korony, 2022;Clark & Bailey, 2018;Jones et al., 2021;Ray et al., 2017;Reveiu et al., 2022).Resilience is defined as the capacity of a regional economy to resist and recover from external shocks (Giannakis & Bruggeman, 2020;Hu et al., 2022;Martin & Sunley, 2015;Soroka et al., 2020).Departing from previous studies, this paper contributes to the emerging literature by constructing a framework to introduce industrial employment structure change induced by the pandemic and incorporating a pair of indices, resistance and recoverability, to measure resilience (Bailey et al., 2021;Borsekova & Korony, 2022;Christopherson et al., 2010;Grafton et al., 2019;Inoue & Todo, 2019;Martin & Gardiner, 2019;Sensier & Devine, 2020).
Second, another stream of relevant literature is the study of the impact of pandemic shock onto the regional market (Albanesi & Kim, 2021;Bartik et al., 2020;Forsythe et al., 2020;Guerrieri et al., 2022;Van Bavel et al., 2020).The COVID-19 pandemic is associated with large supply shocks which aggravates the recession by public policy response, firm exit and employment destruction (Albanesi & Kim, 2021;Brodeur et al., 2021;Guerrieri et al., 2022).In this line of studies, some researchers have enriched economic models of epidemics and combined macro-level data with simulation to study the positive and normative implications of COVID-19 shock (Auray & Eyquem, 2020;Chen et al., 2021;Eichenbaum et al., 2021;Guerrieri et al., 2022).Some prior research has used COVID-19 pandemic as an exogenous event to show the restriction effect and the market supply shift caused by the lockdown policy on mobility regulation (Allen et al., 2020;Balgová et al., 2022;Chen et al., 2021;Fang et al., 2020;Hu et al., 2022;Qiu et al., 2020;Spiegel & Tookes, 2021).Thus, this study has contributed to quantifying the impacts of the pandemic shock to China's local labour market at the aggregated city level.We also link urban economics theory to the empirical design by post-event analysis to interpret the relationship between local labour market and industry specialization.
Third, this paper also contributes to a growing literature using online job postings (Deming & Kahn, 2018;Helleseter et al., 2020;Kuhn et al., 2020;Kuhn & Shen, 2013;Modestino et al., 2020), and extends to explore the pandemic shock on the Chinese economy and the responses of market resilience (Chen et al., 2021;Hu et al., 2022).The empirical data used in the previous studies is at the macro-level and mainly from statistical yearbooks without frequent local labour market information.There exist some studies to measure the impacts of the COVID-19 pandemic on the local labour market using online posting data (Forsythe et al., 2020;Grabner & Tsvetkova, 2022;Hensvik et al., 2021).However, most of these researches use the US or European microevidence or online WebCrawler's data.While the lack of high-quality data about the Chinese labour market and uniform time-series employment behaviours limits the explanation on epidemic shock on economy, we access the real-time Chinese platform and collect the dynamics of recorded jobs and wages to identify the labour market response to the pandemic.To the best of our knowledge, our paper is the first to quantify impacts of the pandemic on the local labour market using the unique large-scale online employment matching data.
Figure 2 shows the static analysis framework of resilience during the pandemic in the economic cycle, where resistance describes the contractile process and recoverability describes the expansive process of the local job market and industrial structure.The theoretical extension allows us to enrich the conceptualization of resilience by a formal economic model from the literature (Bailey et al., 2021;Borsekova & Korony, 2022;Christopherson et al., 2010;Helpman, 1998;Krugman, 1991;Redding & Sturm, 2008).The empirical extension allows us to enrich the regional market resilience at heterogeneous city-and industry-level from the perspectives of labour market diversification and industrial economy complexity (Azar et al., 2020a;Balland et al., 2019;Balland & Boschma, 2021;Borsekova & Korony, 2022;Frésard et al., 2017;Grabner & Modica, 2021;Hidalgo et al., 2007;Hu et al., 2022;Pylak & Kogler, 2021).
This study found that the pandemic had caused a deep shock to China's job market.Based on our results, aggregate job postings first decreased by 50.2% between December 2019 and February 2020, and then quickly rebounded after March 2020.Notably, this 'V'-shaped recovery was accompanied by the reallocation of human capital in the form of talent migration and skill redistribution across regions and industries.Specifically, we found that more job vacancies diverted from larger cities to smaller ones, exhibiting a decentralization trend contrary to the urban agglomeration that has been occurring in China over the last few decades.Additionally, the responses and countermeasures implemented by industries showed heterogeneity.
By visualizing the two dimensions of resilience for over 300 cities, we generated a comparison of resistance versus recoverability.The results showed that China's local job market resilience exhibited a significant regional imbalance, correlated with pandemic risks, city scale and industry structure.Specifically, resistance and recoverability were negatively correlated with city level.Larger cities are prone to lower resistance and recoverability compared with smaller cities.Using the post-event analysis, we identified and quantified the induced shift in job demand (aggregate postings decrease by 38% during the COVID-19 outbreak month) after eliminating the platform effect of growth and seasonal effects like the Chinese New Year holiday.In addition to its 'V'-shaped curvature, the job market recovery was observed to gradually fade.The entire structure showed a lag of nearly two months, indicating the control of the spread of the virus in China.
We also evaluate the structural impact of the pandemic on Chinese labour market.Theoretically, we establish an economic model to illustrate the relationship of local labour market and industry specialization.Our model shows that pandemic-induced redistribution of local labour market could lead to regional convergence in equilibrium.The results are driven by two potential channels: the transportation cost and the share of tradable good expenditure.On the one hand, the concentration of firms and industries can generate significant agglomeration effects that demand large amount of local labours with industry-specific skills.On the other hand, the production of specialized non-tradable inputs and the supply of local labours can sustain larger regional divergence.Empirically, we introduced two metrics of labour market diversification and industrial economy complexity: the inverse Herfindahl-Hirschman index (HHI) and the economic complexity index (ECI) to evaluate the structural impact of the COVID-19 pandemic.The HHI measures labour market concentration effectively (Azar et al., 2020a(Azar et al., , 2020b;;Berger et al., 2022) and the inverse HHI (the reciprocal of HHI) represents local job market diversification and highlights the mitigation of the pandemic's adverse effects (Grabner & Modica, 2021;Pylak & Kogler, 2021).ECI, which measures the economic complexity of different industries (Diodato et al., 2018;Hausmann et al., 2014;Neffke et al., 2011), has a comparative advantage in capturing industrial changes (Balland et al., 2019;Borsekova & Korony, 2022).To apply ECI to job markets, we improved upon existing algorithms to show (Frésard et al., 2017) that industrial structure has considerably changed post-lockdown, exemplified by shifts in human capital accumulation.The degree of job market diversification increased as a whole, while industrial economic complexity increased for the middle and western regions, but not for the eastern and coastal regions.
The remainder of this paper is organized as follows.Section 2 shows our qualitative analysis and economic model to interpret the potential mechanism of the labour market response to the pandemic shock.Section 3 lists all the data sources and identification strategies.Section 4 highlights the key empirical results of our analysis.Section 5 further proposes two factors to explain the persistence of the labour market shift and presents the concluding remarks.

THEORETICAL FRAMEWORK AND MODEL
In order to explain the structural shock from the pandemic on Chinese labour market, following the theory in urban economics (Helpman, 1998;Krugman, 1991;Redding & Sturm, 2008), we construct an economic model to COVID-19 and labour market resilience: evidence from large-scale recruitment behaviour explore the mechanism of industrial structural change induced by the pandemic.The model interprets the relationship of local labour market and industry specialization.On the one hand, the concentration of firms and industries can generate significant agglomeration effects that demand large amount of local labours with industry-specific skills.On the other, the production of specialized non-tradable inputs and the supply of local labours can support larger regional divergence.The feedback mechanisms could be a positive circularity and ultimately reach a self-fulfilling equilibrium with respect to industry specialization that sensitively depends on initial endowments and geographical constraints (Helpman, 1998;Krugman, 1991;Redding & Sturm, 2008).
In this context, pandemic-induced redistribution of regional labour market will lead to not only a large shock on industry structure in the short term but also a substantial impact on industry specialization in the long-run equilibrium.We then construct a static general equilibrium model to interpret the mechanisms.The model set-up considers two regions labelled 1 and 2. Each region is endowed with production of a non-traded good with quantities h 1 and h 2 , respectively.We assume that the non-traded good is supplied by firms in-elastically across regions.All agents in the economy are assumed to share   Xinguo Yu et al.

REGIONAL STUDIES
the same utility function and form as: where N is the total production varieties of potential traded goods, x j is the differentiated variety indexed by j, and s . 1 is the constant elasticity of substitution among these traded goods.And m is the expenditure share of aggregated traded goods and 1 − m is the expenditure share of non-traded goods.
The model remains L 1 and L 2 to be the labours that are supplied in-elastically to traded goods production in regions 1 and 2, respectively, which is the same setting as Krugman (1991).They may migrate among the regions but the overall quantities are normalized to be the expenditure share of aggregated traded goods m, that is: The model also assumes each variety of traded goods is only produced in a distinct production activity by firms, but the production of aggregated traded goods is footloose and only constrained by: where N 1 and N 2 are the overall quantities of variety plants in regions 1 and 2, respectively.
Without loss of generality, we denote the share of the labour market in region 1 as g = N 1 /N and the relative wage as v = w 1 /w 2 .We further define the diversification of the labour market as f(g) = 1/(g 2 + (1 − g) 2 ) − 1.Note that f [ [0, 1] and is decreasing with the concentration part (g 2 + (1 − g) 2 ).This definition is also consistent with our measurement.
In addition, there exist two types of costs in firms' production of traded goods: labour costs and transportation costs.Indeed, the production of each variety of the traded good has a fixed cost a .0 and a constant marginal cost b .0, which is subjected to economies of scale.Thus, the labour cost of the variety x j is l j = a + bx j .With respect to transportation costs, we assume that they take the iceberg form; specifically, transportation costs t . 1 units of each variety are needed for shipping one unit from one region to another.Therefore, the individual in region 1 needs to pay p 1 for locally produced variety but tp 2 for the imported variety from region 2.
We use the numerical computation to solve the model (see Appendix A in the supplemental data online for details).The following figure shows equilibrium diversification f(g * ) before and after the pandemic (L and L ′ in Figure 3, respectively).Note that the parameter values of the benchmark before pandemic are the same as Helpman (1998).While the counterfactual values of transportation cost t and tradable good share m after the pandemic are set to 10% higher and 10% lower, respectively.
The change of equilibrium diversification f(g * ) shows pandemic-induced redistribution of local labour market and the change of industry specialization.We can find that the equilibrium labour share g * is lower and equilibrium regional diversification f(g * ) increases after the pandemic.This is mainly driven by two potential channels.First, China implemented strict lockdown policy in the pandemic, which significantly increases transportation costs across different regions.Thus, the self-sustaining localized labour market will prevail in this autarkic process.Second, the pandemic inevitably decreases the aggregatelevel productivity and the economies of scale.Thus, the decrease of marginal productivity for labour should be severer in the traditional industrial centres that tend to be large cities.This online labour market is a two-sided searchmatch platform and different from most existing types as bulletin board system in China (Deming & Kahn, 2018;Helleseter et al., 2020;Kuhn & Shen, 2013).The platform has more than 80 million registered workers and 20 million registered employers by the end of 2020.On this platform, recruiters can initiate different positions and post-job descriptions online, while applicants can upload and update their CVs online toward specific desired jobs.Once a final interview and negotiation has been made between the two sides, a tentative agreement and a successful wage offer is formed.The database of this website documents proprietary records for the job postings and dynamic behaviours of recruiters and seekers.Therefore, our data are different from other WebCrawler's studies, and we collaborate with the platform and directly extract data from their database at aggregated level.

EMPIRICAL DESIGN
Meanwhile, our study differs from existing literature in other three respects.First, our labour market data covered not only the main COVID-19 outbreak period in China, but also a longer period before and after its peak, presenting a panoramic picture of the complete recovery cycle.China has maintained the number of daily confirmed cases below 100, and most regions have been at a lowrisk level since April 2020.China's first COVID-19 'wave' was from January to March 2020, and no second 'wave' had occurred during the time that this research was conducted.Second, our hiring postings across various dimensions such as time, city, industry, firm scale, and COVID-19 and labour market resilience: evidence from large-scale recruitment behaviour hiring requirements set the stage for a more granular analysis.Third, by decomposing resilience into resistance and recoverability, we observed both regional and industrial imbalances.Based on similar studies (Boustan et al., 2020;Cajner et al., 2020;Campello et al., 2020;Clay et al., 2019;Fang et al., 2020;Inoue & Todo, 2019;Jia et al., 2020;Qiu et al., 2020), this paper provides a larger number of measures to explore the structural impact of the COVID-19 pandemic on local job markets.

Measurement and variables
First, we conducted a basic descriptive analysis of the relationship between online job postings and COVID-19 cases at the aggregate city and industry level in China.We introduced a resilience index to understand the elasticity mechanism of industries and local job markets.We also analysed the heterogeneity in firm scale and job experience requirements to understand which factors determine resilience.
The concept of resilience has emerged as an important characteristic of regional and industrial economies.However, this metric is not easy to measure, as public emergencies of a global scale, such as the pandemic, are rare.We quantified the resilience of China's job market during the COVID-19 pandemic by using an approach similar to Martin and Gardiner (2019) and decomposed resilience into resistance and recoverability (Giannakis & Bruggeman, 2020;Hu et al., 2022;Martin & Sunley, 2015;Soroka et al., 2020), corresponding to the employment contraction and expansion periods, respectively.In addition, we used the change in employment as the aggregate level benchmark to compare relative resilience at both regional and industrial levels.The algorithms used were as follows: where Res i and Rec i represent the resistance and recoverability of region or industry i, respectively.D t = n t = t denotes the rate of change between times t and n.Y i and Y N are the job postings of region or industry i, respectively, at the aggregate level, |D t = t N t = 0 Y N | measures the absolute rate of change of aggregate postings from December 2019 (t = 0) to the aggregate trough (t N ).Meanwhile, |D t = n N t = t N Y N | measures this value from the aggregate trough (t N ) to the aggregate peak (n N ).Similarly, D t = t i t = 0 Y i and D t = n i t = t i Y i measure i's rate of change of postings from the initial time (t = 0) to the first regional (industrial) trough (t i ), and from the first regional (industrial) trough (t i ) to the first regional (industrial) peak (n i ), respectively.In order to accurately identify peaks and troughs, we process our data using the standard Hodrick-Prescott (HP) filter with seasonality adjustment.Then we identify troughs and peaks using the first minimization or maximization values (for details, see Appendix B in the supplemental data online).This choice is justified by the data showing that most regions or industries' dynamics have only one trough and one peak after December 2019.
Based on equations ( 1) and (2), we make two remarks.First, if i's rate of change of postings is approximated by the aggregate level, both resistance and recoverability converge to zero.Second, if i's rate of change of postings significantly deviates from the aggregate level, both resistance and recoverability will diverge from zero.The values of resistance and recoverability represent the strength of resilience.
To measure regional labour market diversification, we used the reciprocal of HHI, Inverse_HHI.Market diversification as the role of related variety is highly correlated with pandemic resilience (Grabner & Modica, 2021).Intuitively, a diversified market tends to have higher resilience.This is because the strategy of 'not to put all eggs in one basket' can enhance a region's resistance (Grabner & Modica, 2021;Pylak & Kogler, 2021;Zhu et al., 2017).Obviously, if Inverse_HHI is higher, HHI decreases and the market is more dispersed.The HHI of labour market is defined by the emerging literature (Azar et al., 2020a;Berger et al., 2022) as follows: where s i,j is the market share of industry i in the local job market j.To simplify the analysis, we regarded one city as one local job market; therefore, j also represents city j.Time t is the monthly index.If HHI approaches one, then Inverse_HHI is small, the job market diversification of industry i is extremely low.Conversely, if HHI is close to zero, then Inverse_HHI is very large, it means that industry i is completely dispersed.Another measure to represent industrial structure is economic complexity (Balland et al., 2019;Borsekova & Korony, 2022;Hidalgo et al., 2007).In this study, we extended this measure by first introducing the concept of industry specialization to measure each city's degree of specialization in specific employment in industries (cityindustry observations).Industry specialization represents a 'revealed comparative advantage' (Frésard et al., 2017), defined as: where w i,j is the share of industry i's job postings (labour demand) in city j's total postings.w i = 1/J J j=1 w i,j is the average national share of i's postings (labour demand).If SP > 1, industry i's labour demand in city j exceeds the national level.As such, a higher SP value indicates a higher degree of industry specialization.Next, we defined an economic matrix M i,j,t whose elements were one if SP i,j,t .1, and 0 otherwise.Then, we referred to Hausmann's classic method to define diversity and 50 Xinguo Yu et al.

REGIONAL STUDIES
ubiquity as follows: Using the definition of economic complexity, we obtain: where the economic complexity (ECI) is precisely captured by the second largest eigenvalue of M j, j ′ ,t , namely, K t .ECI is the normalized form of K t , expressed as:

Identification strategy
We use the post-event analysis based on the fixed effects model to eliminate the trend effect (platform) and the seasonal effect (Chinese New Year holidays) and then document the statistically significant results.Trend and seasonality effects are central empirical challenges in estimating the impact of the COVID-19 pandemic.The trend results mainly from the stationary increase in total postings on the Internet platform (Figure 4; the red line is the fitted line representing the job postings trend).The seasonal decrease (Figure 4) is obvious as well, owing to the Chinese New Year holidays.
In addition, the posting trend and its seasonality are heterogeneous at a disaggregated level.Alternatively, the strengths of the trend and seasonality vary widely across different industries and regions (Figure 5).Thus, such industry-and region-specific trends and seasonality must be eliminated.
To solve these empirical problems, we designed a postevent regression with fixed effects.First, using the onset of the COVID-19 outbreak as the post-event, we evaluated its impact on the number of online job postings, formulated by the following baseline regression model: where i indicates the industry, and c indicates the city.The time index t represents the month, ranging from January 2018 to September 2020.We use data for January 2018 onwards, because missing values existed for some industries before 2018.Y i,c,t is the growth rate of job postings or the average wage for city c and industry i at time t. a i and l c capture the individual fixed effects to control for the time-invariant factors at industry and city levels, respectively.d t is the month dummy to eliminate seasonal and trend effects.Outbreak 0 refers to the COVID-19 outbreak and the Wuhan lockdown in January 2020, and Outbreak j (j .0) is the specific post-event period.Outbreak j is defined as 1 for month j − 1 in 2020, and 0 otherwise.x i,c,t represents key variables such as average experience, square of average experience (Balgová et al., 2022;Deming & Kahn, 2018), and average industrial scale (Azar et al., 2020a(Azar et al., , 2020b;;Berger et al., 2022;Diodato et al., 2018) in industry i and city c at time t.Experience is measured using actual working years and industrial scale is a dummy variable ranging from 1 to 5, averaged using a firm's job postings weighted by its posting share in the local job market.The variables Outbreak j * x i,c,t are the cross-terms used to explain the mediating effect of key variables.We also added other control variables, such as gross domestic product (GDP), population and hospital numbers.Equation ( 9) allows us to eliminate the seasonal effect and trend effect such that the coefficient of post , p. 00 captures the average response of the job market to the COVID-19 outbreak event, and the coefficients of Outbreak j (j .0) estimate the dynamic post-event hysteresis effect after the event.We further run robustness tests by introducing some new control variables with alternative annual data from the China City Statistical Yearbook.The discussion and results are shown in Appendix C in the supplemental data online.

Resilience of China's local job market during the COVID-19 pandemic
Figure 4 plots the time series of the number of online job postings and wages from May 2017 to September 2020.The labour demand of Chinese firms has been dramatically affected by COVID-19, reflected in the 'demand dent' experienced at the end of December 2019.Although the week-long Chinese New Year holidays (in late January or early February depending on the Chinese lunar calendar) cause regular fluctuations every year, a deeper concave shape was observed beginning January 2020.In the aggregate level, the trough is in January 2020 and the peak is in June 2020 (represented by the grey bar in Figure 4).The concavity is deeper and wider than what can be explained by the seasonal effect alone.Most importantly, the concave area exhibits a rapid rebound post-March 2020.In addition, wages also show a trend opposite to that of the job posting numbers.The wage change may be understood by the labour demand-supply relationship.A sudden decrease in labour supply and demand for low-wage jobs, caused by the pandemic outbreak, justifies the increase in the average wage level.
By comparing the job-posting dynamics and the severity of COVID-19 as measured by newly confirmed cases or other measures provided by the Chinese Centre for Disease Control and Prevention (Figure 1), we observe that the occurrence of the abnormal concavity in Figure 4 coincides with the beginning of the virus spread in China and is mainly explained by the COVID-19 shock, which drove the labour market into an economic recession.We further demonstrate the high correlation between the COVID-19 and labour market resilience: evidence from large-scale recruitment behaviour 51 REGIONAL STUDIES drop in job postings and the COVID-19 outbreak using the fixed effects model with the following specifications.How can the simultaneous occurrence of the job posting decline and COVID-19 be explained?As China was the first country to announce the COVID-19 outbreak, it intuitively follows that the job market response did not precede the pandemic.Rather, the virus-induced panic in economic behaviour and the Chinese government's decisive lockdown policy led to a rapid employment contraction, with an amplitude representing resistance.Furthermore, the concavity in Figure 4 was more persistent and did not rebound until after March 2020, even though the daily count of newly confirmed cases had been below 100 since the end of February 2020.The delay in the  rebound process indicates the recoverability.Resistance and recoverability together capture resilience.The impact of the pandemic and prevention policies are likely to last longer and even cause permanent structural changes to the labour market.

Quantifying the pandemic effect
We conducted post-event causal inference using the fixed effects regression to eliminate the trends and seasonal effects.The estimates of the post-event model are summarized in Table 1 and show that the direct effect of the pandemic outbreak on the growth rate of job postings is −0.380, which is statistically significant at the 0.1% level, implying a large shock to China's online labour market.However, its lagged effects after two months are positive despite the decreasing trend, indicating that China's online job market has experienced a large and progressive revival since March 2020.Noticeably, the rebound is much larger in the second and third months after the outbreak, but relatively smaller in the subsequent periods.
The nationwide growth rate dynamics of job postings are consistent with the COVID-19 spread and economic situation in China.We infer that there are three reasons to explain the rebound in job postings.First, successive prevention and control measures have partially eliminated the negative effects of the pandemic, such as public panic and helplessness.Second, the government's relief policies, particularly tax breaks, have alleviated the financial troubles of hard-hit small and medium-sized enterprises (SMEs).Third, many firms, especially information technology (IT)-related firms, have chosen telecommuting and online recruitment during the pandemic.To summarize, although the first two factors are temporary, they are effective.The third hints at a permanent structural change in the benefitted firms, which is correlated with the resilience of industries.We further inspected the structural redistribution of talent by industry and city levels as follows.

Heterogeneous reaction to COVID-19: industry-level evidence
In addition to the severity of the pandemic, the city scale is highly correlated with pandemic resistance and recoverability.To explain the distinctions between different cities, the determinants of resistance and recoverability must be examined.As basic characteristics of the industry and labour market, resistance and recoverability are naturally determined by industry scales and labour market structures.Thus, industry-level studies can capture the pandemic-induced shifts in the employment market's industrial structure.Figure 5(a) illustrates industry-wise job-posting dynamics, showing visible fluctuations in industries such as IT and hotel and catering.Industries such as transport and logistics have benefited from the pandemic as people have limited choices but to resort to online shopping under the lockdown policy and thus rely heavily on transport, and industries such as public administration and mining show steady development.
Figure 5(b) demonstrates that resistance and recoverability are heterogeneous across different industries.For instance, industries such as agriculture, manufacturing, construction, real estate, environment and mining show positive and higher resistance.However, the service industry, particularly hotel and catering, has negative and lower  COVID-19 and labour market resilience: evidence from large-scale recruitment behaviour 55 REGIONAL STUDIES resistance, which indicates that it suffered the most during the employment contraction period.The resistance values of industries such as finance and education are close to zero at the aggregate level.Furthermore, we also find that industries with low resistance, such as transport, hotel and catering, and finance, are also the three industries with the highest recoverability, indicating economic recovery.In particular, the hotel and catering industry has the lowest resistance and the second-highest recoverability.The transport and finance industries have relatively normal resistance levels.This may be attributed to their earlier adoption of and faster switchover to non-contact technology.For example, finance jobs can be easily switched to the telecommuting mode; in fact, no-touch shopping became a standard in the transport and logistics industry much before COVID-19.
As industries are collections of firms wherein different structural needs are satisfied by different skillsets, we used firm scale to represent industry scale and acquired experiences of talent to define industry structure, as shown in Figure 6(a, b), respectively.In Figure 6(a), we observe that larger firms experienced a smaller shock but also a slower rebound than smaller firms during the pandemic (we used 500 people as the threshold to distinguish between small and large firms, in accordance with China's classification guidelines).Generally, larger scale firms have higher resistance but lower recoverability.Furthermore, an industry with larger firms is likely to have a larger scale and therefore higher resistance but lower recoverability.Intuitively, the density of human capital in an industry can reflect its structure.In this study, we used years of work experience specified in the job description to measure human capital.In Figure 6(b), we see that more years of required work experience translate into a higher resistance.Notably, both the most-experienced (10 years or more) and least-experienced (three years or below) lead to higher recoverability.

Heterogeneous reaction to COVID-19: citylevel evidence
Different cities faced different problems during the pandemic.The Chinese government implemented the strictest and longest lockdown policies in Hubei.As centres and sub-centres of the pandemic, cities in Hubei province, specifically its capital Wuhan, were hit harder than cities in most other provinces.Furthermore, city scale may bring about heterogeneity in resistance and recoverability.A simple but efficient way to look at city-level differences is to use representative cities as benchmarks to capture the main characteristics of resistance and recoverability.These representative cities were chosen based on city scale and the degree of pandemic severity.Figure 7 shows the job-posting dynamics in the chosen cities since May 2017.We set the May 2017 values as the base value (100) for all cities.Therefore, Figure 7 mainly reflects the relative rates of change of job postings in different cities, regardless of their city levels.We observe from Figure 7(a) that Wuhan experienced a deeper and more lasting shock than other cities nationwide.Notwithstanding heterogeneity, all cities exhibit similar shockand-rebound processes.In addition, large-scale first-tier cities, such as Beijing, Shanghai and Guangzhou, show a smaller shock compared with second-tier cities of smaller scales, such as Chengdu, Hangzhou, Wuhan, Xi'an, Zhengzhou and Changsha, indicating that larger cities are more resistant.However, mega cities such as Beijing and Shanghai are prone to smaller rebounds, indicating that recoverability is negatively correlated with city scale.Our results indicate that city scale has a significant effect on resilience, such that larger cities tend to have lower resilience because of lower recoverability and resistance.Although there are other standards to measure city scale, such as population or area, we measured it using GDP.In general, these measures are highly positively correlated.History, policy, geography and other factors determine  Xinguo Yu et al.

REGIONAL STUDIES
city scale.It is noteworthy that the largest and wealthiest Chinese cities -Beijing, Shanghai, Shenzhen and Guangzhouare either located on the east coast or being the capital of the country.The second-tier cities are mainly provincial capitals in the middle or eastern regions, and the third and fourth are mainly ordinary prefecture-level cities.Third, Wuhan, as the epicentre of the pandemic, demonstrated lower resistance and higher recoverability, which is consistent with previous observations, implying that the pandemic affects a city's resistance more than its recoverability.

DISCUSSION AND CONCLUSIONS
Our study has shown, both qualitatively and analytically, that labour market resilience to the pandemic is heterogeneous in regions.In fact, this process is notably and persistently affected by labour redistribution and industry structure.We further empirically introduce two plausible underlying variables: labour market diversification and economic complexity.We find that a more diversified labour market is more resilient to the pandemic through the lens of labour market diversification.Besides, larger cities suffered more due to lower resilience to the pandemic.In addition, we find that regional convergence exactly happened after the pandemic in China, which is consistent with our model.

Labour market diversification
COVID-19 has had a significant impact on the internal structure of local job markets in China.When an industry has a higher local job market diversification, that is, a high Inverse_HHI value, it affects job postings positively on the whole.During the contraction period, diversification increases the industry's resistance and helps the industry recover faster in the recovery period.Above all, a diversified job market is more resilient to sudden shocks.Figure 8 shows the average market diversification of different industries before and after the Wuhan lockdown.Evidently, the level of post-lockdown market diversification is higher compared with that pre-lockdown.The magnitude of the increase varies across different industries, indicating a significant adjustment of the industrial structure.
In Figure 8, we observe that the agricultural industry benefits most from market diversification during the pandemic.However, modern service industries such as IT and finance, benefit much less from the labour market diversification channel.

Economic complexity
COVID-19, as an external shock to the industrial structure, has revealed and altered the persistent regional imbalance.We computed the value differences of the average ECI of China's local job market before and after the Wuhan lockdown, based on the industry-city data, finding COVID-19 and labour market resilience: evidence from large-scale recruitment behaviour 57 REGIONAL STUDIES that the economic complexity of middle and western Chinese cities has increased significantly.In contrast, the ECI of developed cities in eastern China has decreased.To summarize, developed cities with high ECI suffered the most in the pandemic, while developing cities had more potential to increase their economic complexity by making structural labour market changes.Prima facie, China's labour market and economy have undergone a 'V'-shaped recovery.Aggregate job postings first decreased by half between December 2019 and February 2020, and then quickly rebounded post-March 2020.In essence, the COVID-19 pandemic has greatly reshaped local job markets.Pandemicinduced shocks have caused a reallocation of talent and industrial redistribution of skillsets.Job opportunities have changed from larger cities to smaller ones, reflecting greater decentralization.Meanwhile, the responses of industries to the pandemic have been heterogeneous.The metrics of resistance and recoverability independently reflect the contraction and expansion in China's local job markets.Our results suggest the following: first, the resilience of China's local job market is significantly imbalanced across regions and correlated with pandemic severity, city scale, and industry structure; second, resistance and recoverability are negatively correlated at the city level; finally, although larger cities are more prone to lower resistance and recoverability, smaller cities perform less consistently.
The post-event analysis after eliminating the effects of platform growth and seasonal changes indicated that the impact of COVID-19 on industries and employment markets exhibits a pattern of rapid decline, rapid increase, and gradual flattening, which is consistent with and lagged by nearly two months to the pandemic spread in China.We also found a significant change in industrial structure after the lockdown policy implementation, which can be explained by the Inverse_HHI and ECI mechanism, specifically, the increase in China's local job market diversification.The economic complexity of China's local job markets increased for the middle and western regions, but not the eastern and coastal regions.This implies that cities that make better policies for talent reallocation may have an advantage in withstanding future shocks.A key takeaway for other countries is that increasing industrial diversification and economic complexity may be an effective countermeasure to mitigate the impact of pandemics such as the COVID-19.Although we have described resilience of Chinese labour local markets and manifested the structural change by channels of Inverse_HHI and ECI in the pandemic, we still observe a short-term impact from the pandemic and did not discuss more about economic factors which can potentially affect market equilibrium.We need to combine traditional and novel data of job market to explore deeper mechanisms in the future study.

ACKNOWLEDGEMENT
Xinguo Yu and Hengxu Song contributed equally to this article.Xinguo Yu et al.

Figure 1 .
Figure 1.The COVID-19 impact in China: (a) accumulated confirmed COVID-19 cases in China; (b) the shock to China's economy; (c) confirmed proportions for the Hubei and non-Hubei regions; and (d) proportion of deaths in the Hubei and non-Hubei regions.Sources: Chinese Centre for Disease Control and Prevention and Chinese National Bureau of Statistics.

Figure 2 .
Figure 2. Resistance and recoverability of labour market resilience.

Figure 3 .
Figure 3. Labour market share and industry specialization in equilibrium. 48 3.1.DataThis study is based on data from BOSS Zhipin (NAS-DAQ: BZ), a leading online recruitment platform in China.Local pandemic data were collected from the Chinese Centre for Disease Control and Prevention.Other local economic and social data were drawn from the Chinese National Bureau of Statistics and China City Statistical Yearbook.Job posting samples were collected by aggregating the levels for the observation period (May 2017-September 2020), covering 380 prefecture-level cities and 20 industries.

Figure 5 .
Figure 5. Industry-wise local job-posting dynamics and industrial resilience of job market: (a) industry-wise local job-posting dynamics; and (b) industrial resilience of China's job market.

Figure 5 .
Figure 5. Continued Figure 7(b) plots the recoverability versus resistance relationship of all cities and Figure 7(c) lists the recoverability and resistance of 10 representative cities.Our observations are three-fold.First, as reflected by the red dashed line in Figure 7(b), recoverability and resistance are negatively correlated.Second, the scatter points in the first quadrant (the right upper quadrant) and the third quadrant (the left lower quadrant) indicate that resistance and recoverability are both high or both low, respectively (scatters with high resistance and high recoverability can be called cities with high sustainability).

Figure 6 .
Figure 6.Job-posting dynamics by firm scale and required years of work experience: (a) job-posting dynamics by firm scale; and (b) job-posting dynamics by required years of work experience. 56

Figure 7 .
Figure 7. Local job-posting dynamics of representative cities and regional resilience of job market: (a) local job-posting dynamics of representative cities; (b) regional resilience of China's local job market; and (c) China's local job market resilience of representative cities.

Figure 8 .
Figure 8. Market diversification of different industries before and after the Wuhan lockdown. 58

Table 1 .
Continued.: Standard errors in parentheses are clustered at the city-industry level; ***p < 0.001, **p < 0.01, *p < 0.05, dlnY represents the logarithm of postings; lnW means the logarithm of wages.Outbreak_month ¼ 0 captures the direct effect in the month of the COVID-19 outbreak.Outbreak_month > 0 corresponds to the dynamic effects relative to the outbreak.Scale, exp and exp^2 are control variables that indicate the average industrial scale, average experience and square of average experience, respectively, in industry i and city c at time t. Note