The economic impacts of COVID-19 and city lockdown: Early evidence from China

Abstract

As the first major developing country heavily struck by the COVID-19 pandemic, China adopted the world's most stringent lockdown interventions to contain the virus spread. Using macro- and micro-level data, this paper shows that both the pandemic and lockdown policies have had negative and significant impacts on the economy. Gross regional product (GRP) fell by 9.5 and 0.3 percentage points in cities with and without lockdown interventions, respectively. These impacts represent a dramatic recession from China's average growth of 6.74% before the pandemic. The results indicate that lockdown explains 2.8 percentage points of the GDP loss. We also document significant spill-over effects of the pandemic in adjacent areas but no such effects of lockdown. Reduced labor mobility, land supply, and entrepreneurship are among the most significant mechanisms underpinning the impacts of the pandemic and lockdown. Cities with higher share of secondary industry, higher traffic intensity, lower population density, lower internet access, and lower fiscal capacity suffered more. However, these cities seem to have recovered well from the recession and quickly closed the economic gap in the aftermath of the pandemic and city lockdown. Our findings have broader implications for the global interventions in pandemic containment.

1. Introduction

With the wide spread of the COVID-19, the pandemic has led to significant health and economic losses worldwide. Many countries experienced significant economic slowdown and are still struggling to recover from the COVID pandemic. Various non-pharmaceutical intervention measures, including social distancing, extension of public holiday, school closure, stay-at-home orders, large-gathering bans, non-essential business closures, transportation restrictions, and even more drastic measures such as complete city lockdown, have been adopted. However, the stringency of these restrictions varies significantly across countries and regions. Concerned about possible impacts on the economy, many governments are reluctant to implement the most drastic interventions or remove restrictions as soon as the pandemic is considered under control. An important and challenging question is how much of the impacts resulted from the pandemic versus government-imposed interventions (Goolsbee and Syverson, 2020). These impacts may also depend on the structure of the economy. A sound understanding of such heterogeneity will provide vital and timely input to the challenging policy tradeoff between intervening to reduce disease transmission and maintaining economic growth.

However, gaging and disentangling the impacts of the pandemic and interventions on economic growth are always challenging because traditional data are rarely reported at high-enough frequencies and at the treatment levels (e.g. city, county) where most intervention measures are implemented (Kong and Prinz, 2020). Most recent studies examined the impact of the pandemic using indirect proxies, such as online search data on unemployment insurance claims (Kong and Prinz, 2020; Forsythe et al., 2020; Brodeur et al., 2021a), mobility variations from Facebook (Bonaccorsi et al., 2020), or mobile phone records data on customer visits to businesses (Goolsbee and Syverson, 2020; Castells-Quintana et al., 2021), and job vacancy postings (Forsythe et al., 2020; Campello et al., 2020).

This paper employs quarterly data for China's prefectural-and-above (PAA) level cities to evaluate the impacts of COVID-19 and intervention polices on cities’ gross regional product (GRP). We use city-level confirmed cases and deaths of COVID-19 to measure the severity of the pandemic, and focus on the most stringent intervention – city lockdown. This paper takes the advantage of the spatial and temporal variation of the pandemic and city lockdown and uses a difference-in-difference approach to quantify the causal effects. Specifically, we compare the before-and after variation and cross-city variation of the outcome variable.

The analyses yield three sets of results. First, the results show that both the pandemic and lockdown policies have negative impacts on economic growth. An increase of confirmed cases and deaths of the COVID-19 by 1% led to a decline of gross regional product (GRP) by 0.024 and 0.022 percentage point, while city lockdown resulted in a GRP loss of 2.8 percentage points.1 The results document significant spillovers from the pandemic but no spillover effects for city lockdown. The findings also show that COVID-19 had a continuing but much reduced economic impact during the immediate post-pandemic periods whereas city lockdown had no dynamic impact beyond the intervention period.

Second, we examined the main mechanisms through which the pandemic and city lockdown affected the economy. The analyses of microeconomic activities show that reduced input supplies including labor, land and entrepreneurship are among the most significant channels. We find that the pandemic, and city lockdown in particular have much larger impacts on these microeconomic activities than on overall economy.

Third, the study also examined the heterogeneous effects of the pandemic and lockdown on cities with different characteristics. Analyses of the recession period show that cities with higher share of secondary industry and higher passenger traffic intensity suffered more from the pandemic and lockdown policies indicating larger impacts on manufacturing and mobility. The results also show larger economic decline in less developed regions with lower population density, lower internet access, and lower fiscal capacity. However, results from the post-pandemic periods also show that the economy has recovered well from the recession. Although those cities hit harder in the pandemic still lag behind, they seem to have recovered faster and quickly closed the economic gap.

The remainder of the paper is organized as follows. The next section provides a brief description of the COVID-19 background and lockdown interventions in China and a review of related literature elaborating on where our work makes new contributions. Section 3 describes the data. The empirical strategy is presented in Section 4. Section 5 presents and discusses the results. The last section concludes the paper.

2. Background and related literature

2.1. COVID-19, city lockdown and economic growth in China

As the first country heavily struck by the COVID-19 pandemic, China has seen its economy shrink by 6.8% in the first quarter of 2020 quarter on quarter. In Hubei province, which has the largest number of confirmed cases and deaths of COVID-19 among all Chinese provinces, the gross provincial product fell by 39.2%. China responded with fast and stringent interventions to avoid catastrophic virus transmission. Many interventions were implemented at the city level. Almost one-third of Chinese cities were locked down in the first quarter of 2020. The Chinese governments collected and reported quarterly city-level socioeconomic data (Au and Henderson, 2006) as well as confirmed COVID-19 cases and death data, making Chinese cities an excellent case to study the economic impacts of the pandemic and related interventions.

China's strict interventions proved to be effective. By the end of the first quarter, most lockdown restrictions were repealed. The number of additional local confirmed cases of COVID-19 declined to almost zero in the second quarter. While many economies are still struggling in the economic downturn of the pandemic, the Chinese economy seems to have recovered rapidly with 3.2%, 4.9% and 6.5% growth in the second, third and fourth quarters in 2020 and 8.1% growth in 2021. This also provide us a unique opportunity to examine the recovery process of a major pandemic-hit economy.

2.2. Related literature

This paper contributes to the rapidly growing literature studying the economic impacts of COVID-19 and related interventions (see Brodeur et al. (2021b) for a review of this literature). Many existing studies focus on the impacts on the job market (Adams-Prassl et al., 2020; Forsythe et al., 2020; Rojas et al., 2020; Bartik et al., 2020; Hensvik et al., 2020; Gupta et al., 2020; Couch et al., 2020; Montenovo et al., 2020; Baek et al., 2020; Kong and Prinz, 2020; Lin et al., 2020; Green and Loualiche, 2020; Crossley et al., 2020; Binder, 2020; Ascani et al., 2021). Another strand of literature examines the impacts on income distribution and consumer behavior (Goolsbee and Syverson, 2020; Chetty et al., 2020; Brewer and Gardiner, 2020; Carvalho et al., 2021). A few other studies use various economic approaches to simulate the macroeconomic impacts and calibrate the effects of potential policies (Atkeson, 2020; Altig et al., 2020; Eichenbaum et al., 2020; Krueger et al., 2020; Jordà et al., 2020; Gregory et al., 2020; Aum et al., 2021; Capello and Caragliu, 2021; Yang et al., 2021; Yilmazkuday, 2021; Dweck et al., 2022). The present paper extends the literature in several ways.

First, most existing studies on the economic impacts of the pandemic focus on the developed countries, particularly the USA. Few efforts have been made to evaluate the pandemic and intervention effects in developing countries. The economic structure, medical treatment capacity and interventions taken in developing countries can differ greatly from developed countries. China makes an excellent case. As the largest developing country with perhaps the most sophisticated economic structure, China is the first major economy heavily struck by the COVID-19 pandemic. However, the severity of the pandemic and its associated economic impacts vary significantly across different regions of the country. There has been an emerging literature on the social and environmental effects of the COVID-19 in China (Fang et al., 2020; He et al., 2020; Matthew et al., 2020; Liu et al., 2020; Ding et al., 2022). Rigorous micro and macro studies on the economic effects of the pandemic are surprisingly limited (Zhang et al., 2020; Dai et al., 2021a).

Second, this paper examines the economy-wide impacts. Compared with indirect proxies such as mobile phone records or online search data, the economy-wide economic growth data provides straightforward and more comprehensive account of the economic impacts. A couple of studies have examined the economic impact of the pandemic in China. Dai et al. (2021a) provides micro evidence of the impacts on small and medium-sized enterprises. To our knowledge, our study is among the first to conduct an economy-wide economic analysis. Zhang et al. (2020) also investigates the macro-economic impact of the COVID-19 in China but largely focuses on agri-food system.

Third, we address the often-asked question about the relative impacts resulted from the pandemic versus government-imposed interventions (Goolsbee and Syverson, 2020). The analysis separately identifies the economic effects of the pandemic and the city lockdown interventions. More importantly, this is done by controlling for possible spillover effects from both the pandemic and city lockdowns. As many Chinese cities adopted complete lockdown restrictions, a study of China's city lockdown also complements the existing studies of interventions in the developed countries that are typically much less stringent. We also explore how these effects differ across cities with different economic structure, which could be of interest for making tailored intervention as well as recovery policies.

Fourth, China is the first major economy that has seen swift and strong post-pandemic recovery. Differing from the existing studies which mostly focus on the pandemic and intervention effects during the economic recession, the present study examines both the decline and recovery processes.

3. Data

Data for the empirical analyses presented in this section were collected from various official statistical publications and public database. Using this dataset, we constructed a quarterly data set for 296 Chinese PAA level cities over the period from the first quarter of 2019 to the third quarter of 2020.

3.1. Outcome variables

We use quarterly GRP as the outcome variable to quantify the economy-wide impact of the pandemic and city lockdown. The GRP data for 296 cities from the first quarter of 2019 to the third quarter of 2020 were collected manually from the local governmental websites and deflated to the constant price in the first quarter of 2019.2 Table 1 presents the detailed variable definition and summary statistics of the main variables for China and for the lockdown and the non-lockdown cities separately.

Table 1.

Summary statistics.

	Lockdown		Non-lockdown
Variable	Mean	Std. Dev.	Mean	Std. Dev.	Description
Confirmed cases	119.88	2026.18	5.27	20.56	Number of confirmed cases
Death cases	7.25	154.83	0.05	0.33	Number of deaths
GRP	148,263	188,797	71,647	65,805	Gross regional product (Million RMB)
Death_neighbor	41.96	296.28	0.62	5.92	Number of deaths in neighboring cities
Firms_new	9920	14,705.91	5138	6921.306	Number of new registered firms
Investment	33,690.71	47,150.74	17,616.61	25,151.51	Capital of new registered firms (Million RMB)
Share of 2nd industry	43.41	8.82	43.26	9.82	Share of second industry in GRP(%)
Passenger traffic intensity	1.01	0.62	1.90	5.25	Ratio of passenger traffic to population
Population density	630.95	831.80	483.02	489.44	Population density (People per square kilometer
Internet access	2.08	2.04	1.35	1.04	Number of Internet users(Million)
Fiscal capacity	0.09	0.03	0.08	0.03	Ratio of fiscal revenue to GRP
Income level	77,035.21	42,677.56	64,791.42	29,277.59	GRP per capita (RMB)
Population mobility	214.16	131.07	200.16	122.03	Ratio of migrant population to local population (per 1000 people)
Doctors	2.84	1.82	3.01	2.29	Doctors per 1000 people
Hospital beds	5.17	3.34	5.35	4.03	Hospital beds per 1000 people
R&D intensity	0.02	0.02	0.75	4.13	Ration of R&D expenditure to government fiscal expenditure
Trade openness	0.22	0.27	0.19	0.29	Ratio of total import and export value to GRP
Industrial structure	0.96	0.33	0.96	0.37	Ratio of the added value of the secondary industry to the tertiary industry
Distance to Wuhan	777.57	463.94	802.08	349.37	Distance to Wuhan (km)
FDI intensity	0.02	0.01	0.02	0.02	Ratio of FDI to GRP
Number of observations	616		616
Land	138.78	160.34	99.76	130.52	Land supply area (hectare)
Number of observations	528		528
Moveout	1.53	2.29	1.11	1.43	Baidu out-migration index
Movein	1.51	2.02	1.12	1.30	Baidu in-migration index
Number of observations	13,200		13,200

Notes: The upper panel of the table presents the mean and standard deviation for each variable at either city-quarter or city-year level, focusing on the comparison between the treated lockdown cities and the matched control non-lockdown cities. There are 88 cities in the treated group and 88 in the control for the period 2019.Q1–2020.Q3. The quarterly GRP data were collected manually from the local governmental websites and deflated to constant price in the first quarter of 2019. The data on confirmed cases and deaths of COVID-19, city characteristics, Baidu migration index, and land supply area are sourced, respectively, from “Resources for COVID-19″, The China City Statistic Yearbook 2019, Baidu Migration Platform, China Land Market Network, and the State Administration for Market Regulation database.

We also use real-time inter-city migration data from the internet services company Baidu, total land supply data, and new firm registration data as outcome variables in the mechanism analysis. The migration data is based on real-time location records for every smartphone using the company's mapping app and other location services. The Baidu migration dataset covers all Chinese cities 24 days before and 51 days after the Spring Festivals in 2019 and 2020, which correspond to the period between January 12 and March 27 in 2019, and the period between January 1 and March 15 in 2020. Baidu out-migration and in-migration index are sourced from Baidu Migration Platform.3 Total land supply include land supply for all commercial, residential, industrial and public infrastructure uses. The data are sourced from China Land Market Network 4 and are aggregated to city-month level. New firm registration data are sourced from the State Administration for Market Regulation database that covers the universe of registered firms in China, which provides details of each registered firm, includes registration time, location, capital and shareholders. New firm numbers and capital are aggregated to city-quarter level in the regression.

3.2. City heterogeneity

We consider five dimensions of city heterogeneity: industry structure (the share of second industry in GRP), passenger traffic intensity (the ratio of passenger traffic to population), population density, internet access, and fiscal capacity (the ratio of fiscal revenue to GRP). To explore how the impacts of COVID-19 and city lockdown differ by each of these dimensions, we split the sample of cities into two groups using the medium value of each characteristic variable and perform analyses for each of the groups. The data on city characteristics are sourced from The China City Statistic Yearbook 2019.

3.3. City lockdown and confirmed COVID-19 cases and deaths

The data on confirmed cases and deaths of COVID-19 is sourced from “Resources for COVID-19″.5 The data was originally collected from http://www.dxy.cn/, China's authorized publishing platform for COVID-19 cases and deaths. The stringency and terms of city lockdown differ greatly across cities. Following He et al. (2020), we identify a lockdown city if it imposes all three preventive measures including: 1) bans on nonessential commercial activities in people's daily lives; 2) bans on any types of gathering by residents; and 3) restrictions on public and private transportation. Lockdown information was manually collected from local governmental announcements. Among the total of 296 cities, 88 cities are identified as the locked-down cities. It is worth noting that all the city lockdown interventions were imposed in the first quarter of 2020 but then repealed by the end of the first quarter. Fig. 1 shows the spatial distribution of the lockdown and non-lockdown cities.

Fig. 1.

The spatial distribution of lockdown cities and non-lockdown cities.

Fig. 2 is the scatter plot of city-level economic growth and the logarithm of total number of infected cases and deaths in the first quarter of 2020. There appear obvious downward trends between economic growth and infected cases and death. Fig. 3 plots the distribution of GRP by quarter in the lockdown and non-lockdown cities. The negative impacts of the pandemic and quick recovery also appear significant. The distribution of the negative impact on the lock-down cities in the first quarter of 2020 seems a bit wider than that of the non-lockdown cities. However, these simple descriptive illustrations are unable to unpack the causal impacts of the pandemic and lockdown interventions

Fig. 2.

Pandemic and economic growth in the 1st Quarter 2020.

Fig. 3.

Economic growth by quarter: lockdown and non-lockdown cities.

3.4. Control variables

We control for city-level heterogeneity including hospital beds, doctors, R&D expenditure as a share of government fiscal expenditure, trade openness, and per capita income. The data was collected from the China City Statistical Yearbook for the pre-treatment year of 2018. We also control for population mobility (the ratio of migrant population to local population) and collected the data from the most recent population census in 2010. All control variables are interacted with quarter dummies in the regressions.

4. Empirical method

4.1. Difference-in-differences (DID) strategy

The DID approach has become popular in empirical economics since the work by Ashenfelter (1978) and Ashenfelter and Card (1985). The classical setting in the DID analysis is one where outcomes are observed for units observed in one of two groups, in one of two time periods. Only one group in the second time period are exposed to a treatment. To gain unbiased estimates, the average gain over time in the non-exposed (control) group is subtracted from the gain over time in the exposed (treatment) group. This double differencing removes biases in second period comparisons between the treatment and control group that could be the result from permanent differences between those groups, as well as biases from comparisons over time in the treatment group that could be the result of time trends unrelated to the treatment (Angrist and Pischke, 2009). There have been many extensions for this approach (see Angrist and Pischke (2009) and Lechner (2011) for reviews). As in our case, the first confirmed case of COVID-19 in China was identified at the end of 2019, but city lockdown and most confirmed cases and deaths were concentrated in the first quarter of 2020.6 The spatial and temporal variations allow us to employ a difference-in-differences (DID) model to quantify the impacts of the pandemic and city lockdown on economic growth. The specification takes the form:

$$Y_{it}=\beta Pandemi{c}_{it}+\gamma Lockdow{n}_{it}+\alpha X_{it}+{\mu }_i+{\eta }_t+{\epsilon }_{it}$$ (1)

where $Y_{it}$ is Log(GRP) in city i at quarter t. $Pandemi{c}_{it}$ denotes the log form of the number of confirmed COVID-19 cases ($\text{Log}\left(Case{s}_{it}\right)$) or deaths ($\text{Log}\left(Death{s}_{it}\right)$) in city i at quarter t. To facilitate taking logarithm, we added 0.01 to reported $Case{s}_{it}$ and $Death{s}_{it}$. $Lockdow{n}_{it}$ is a dummy taking the value of 1 if for lockdown cities and 0 for non-lockdown cities. Thus, the coefficient of interest$\beta$ is expected to capture the effect of the pandemic while $\gamma$ capture the impact of city lockdown. $\beta$ and $\gamma$ are expected to be negative, as production activities were restricted in the cities with pandemic and lockdown policies.

Both restriction measures such as lock-down and pandemic (infected cases and deaths) itself can have independent influences on the economy. The specific mechanisms have been studied with solid empirical support. Lockdown can restrict mobility and reduce economic activity. Pandemic itself can have similar impacts. Even in absence of lockdown restrictions, increased number of cases and deaths could result in declining consumer demand and economic activity (especially in the service sector) due to fears of virus spread and contagion (see Kong and Prinz, 2020; Goolsbee and Syverson, 2020; Keane and Neal, 2021). For example, using American mobile phone usage data, Goolsbee and Severson (2020) found that legal restrictions explain only a small proportion of the fall in the overall consumer traffic, while individual choices under the fears of infection were far more important. In addition, traffic started to drop before the legal orders were in place and were highly influenced by the number of deaths reported in the country. Consumers shift from busier, more crowded stores toward smaller, less busy stores to avoid infection. Similarly, using the unemployment insurance claims data in the US, Kong and Prinz (2020) found that nonpharmaceutical interventions (NPIs) during the COVID-19 pandemic can only explain a small proportion of the unemployment. Most of the short-run increase in unemployment insurance claims during the pandemic was likely due to other factors, such as the decline of consumer spending constrained by health concerns. A number of other studies (E.g. Bartik et al., 2020b; Murray and Olivares, 2020; Chetty et al., 2020), have also documented that economic activity began its steep decline prior to the introduction of NPIs and that it has not recovered in states that have relaxed their restrictions. All these studies suggest that the direct impacts of the pandemic should be disentangled from the governmental interventions, such as city lock-down and state-at-home orders. Because the severity of the pandemic (cases/deaths) is highly correlated with city lockdown, dropping either from the right-hand side would imply a biased estimate. To include both not only addresses OVB but allows for unbiased identification of separate effects of lockdown restrictions and the pandemic.

$X_{it}$ is a vector of controls as previously described. The city fixed effects, ${\mu }_i$ which are a set of city-specific dummy variables, capture time-invariant city characteristics such as geographical location, short-term industrial and economic structure, income and natural endowments. The quarter fixed effects, ${\eta }_t$, are a set of dummy variables that account for shocks that are common to all cities at quarter t, such as public holiday extension, macroeconomic conditions and restricted express services, and ${\epsilon }_{it}$ is the error term. The standard errors are clustered at the city level.

4.2. Selecting the control regions using propensity score matching (PSM)

An accurate evaluation of the policy effects depends on a credible control group that has similar socioeconomic status and satisfies parallel trend and stable unit treatment value assumption (SUTVA). Chinese cities differ greatly in many characteristics (such as geographical location, industrial structure, medical treatment capacity, and population mobility etc.). Therefore, we use the PSM approach developed by Rosenbaum and Rubin (1983) to select a control group from the 208 non-lockdown cities for the 88 lockdown cities. We use a logistic regression and data from 2018 on five matching variables to estimate the propensity score. The matching variables include industrial structure, hospital beds, distant to Wuhan, FDI intensity, and population mobility. These variables ensure that the control group have a socioeconomic status similar to that of the treatment cities. We match treatment and control cities through 1:1 nearest neighbor matching without replacement. This provides us with the same number of control cities for the 88 treatment cities. For robustness, we also provide the results using the full sample and alternative numbers of nearest neighbor.

Table A1 in the appendix presents the mean difference for each of the five covariates between the treatment and control groups before and after matching, along with p-values for the t-statistics. We observe apparent lack of balance between the treatment and control counties before matching. After matching, none of the differences between the treatment and matched control counties are statistically significant. For example, the difference in distance to Wuhan between the treatment and control groups drops from −37.3 (p-value=0.01) before matching to −4.8 (p-value=0.694) after matching.

5. Results

5.1. Baseline results using matched sample

Table 2 shows the results from Specification (1) using different pandemic indicators and sample periods. Columns 1, 3, 5 and 7 use data from the first quarter of 2019 to the first quarter of 2020. Data for the second and third quarter of 2020 are dropped in these regressions. Although the number of new cases declined to almost zero after the first quarter of 2020 and city lockdowns were also repealed by the end of the first quarter, there may be lagged impacts of the pandemic and lockdowns, making cities in the second and third quarters of 2020 contaminated controls in a DID setting. China celebrates the Spring Festival in the first quarter and celebration activities can extend well beyond the official 7-day holiday period. The GDP in the first quarter is typically lower than other quarters. This may bias the estimates of the impacts of the pandemic and lockdown interventions, both also occurring in the first quarter in 2020. We control for such quarterly heterogeneity using quarterly fixed effects. Alternatively, in Columns 2, 4, 6 and 8, we use data including only the first quarters of 2019 and 2020. In Columns 1–4, we only consider the pandemic (cases or deaths). However, one may be concerned that these estimates are biased upwards if cities with more severe pandemic are also likely to adopt more stringent city lockdown which also affects the economy. In Columns 5–8, the specifications include both the pandemic and lockdown intervention.

Table 2.

The impacts of pandemic and city lockdown on GRP^†.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Variables	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)
Log(Cases)	−0.036⁎⁎⁎	−0.026⁎⁎⁎			−0.033⁎⁎⁎	−0.024⁎⁎⁎
	(0.009)	(0.007)			(0.008)	(0.006)
Log(Deaths)			−0.032⁎⁎⁎	−0.023⁎⁎⁎			−0.030⁎⁎⁎	−0.022⁎⁎⁎
			(0.005)	(0.004)			(0.005)	(0.004)
Lockdown					−0.050⁎⁎⁎	−0.030⁎⁎	−0.048⁎⁎⁎	−0.028⁎⁎⁎
					(0.019)	(0.012)	(0.018)	(0.011)
Constant	15.747⁎⁎⁎	15.502⁎⁎⁎	15.850⁎⁎⁎	15.492⁎⁎⁎	15.844⁎⁎⁎	15.542⁎⁎⁎	15.940⁎⁎⁎	15.532⁎⁎⁎
	(0.263)	(0.113)	(0.212)	(0.088)	(0.248)	(0.108)	(0.207)	(0.088)
Observations	880	352	880	352	880	352	880	352
R-squared	0.993	0.998	0.993	0.998	0.993	0.998	0.993	0.998
Cov*Q.Dummy	YES	YES	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES	YES	YES
Quarter FE	YES	YES	YES	YES	YES	YES	YES	YES
Period	2019.Q1–2020.Q1	2019.Q1, 2020.Q1,	2019.Q1–2020.Q1	2019.Q1, 2020.Q1,	2019.Q1–2020.Q1	2019.Q1, 2020.Q1,	2019.Q1–2020.Q1	2019.Q1, 2020.Q1,

^†Robust standard errors clustered at the city level and reported in the parentheses;.

^⁎,.

^⁎⁎and.

^⁎⁎⁎denote significance at 10%, 5% and 1%.

The estimates using different pandemic measures, model specifications and data samples vary only slightly. The estimated coefficients across all specifications provide robust evidence that the pandemic and city lockdown had negative and statistically significant impact on the economy. As the results using confirmed cases or deaths are similar, following Kong and Prinz (2020) and Goolsbee and Syverson (2020), we focus on results using deaths as the indicator of the pandemic in following discussions.

5.2. Robustness

For robustness, we also estimated the results using the full unmatched sample. Similarly, Table 3 only reported the results using the specifications including both the pandemic and lockdown intervention for brevity. We first provide the results using full unmatched sample in Columns 1 and 2. The results are similar to our baseline analysis shown in Table 2.7 Nevertheless, we also examined the sensitivity of our results to samples generated from alternative matching algorithms. We alter to the 2-, 4-, and 6-nearest neighbor matching with replacement. The coefficients of interest remain statistically significant with a very minor change in magnitude from our baseline estimates (Table 2), indicating that our results are not sensitive to the choice of matching algorithms. Similarly, results using data from the first quarter of 2019 to the first quarter of 2020 are presented in Columns 1, 3, 5 and 7, and those using only first quarters of 2019 and 2020 presented in Columns 2, 4, 6 and 8. Taken together, bias due to lack of overlap is not a concern in our case, which is perhaps not surprising given we are using a very dense sample.

Table 3.

Robustness check using different samples ^†.

	Unmatched full sample		2-Nearest neighbor		4-Nearest neighbor		6-Nearest neighbor
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)
Log(Deaths)	−0.028⁎⁎⁎	−0.017⁎⁎⁎	−0.031⁎⁎⁎	−0.022⁎⁎⁎	−0.030⁎⁎⁎	−0.020⁎⁎⁎	−0.030⁎⁎⁎	−0.020⁎⁎⁎
	(0.005)	(0.003)	(0.005)	(0.004)	(0.005)	(0.004)	(0.005)	(0.004)
Lockdown	−0.053⁎⁎⁎	−0.036⁎⁎⁎	−0.055⁎⁎⁎	−0.030⁎⁎⁎	−0.055⁎⁎⁎	−0.032⁎⁎⁎	−0.055⁎⁎⁎	−0.032⁎⁎⁎
	(0.016)	(0.010)	(0.017)	(0.010)	(0.017)	(0.010)	(0.016)	(0.009)
Constant	15.692⁎⁎⁎	15.119⁎⁎⁎	15.898⁎⁎⁎	15.471⁎⁎⁎	15.818⁎⁎⁎	15.344⁎⁎⁎	15.760⁎⁎⁎	15.296⁎⁎⁎
	(0.162)	(0.059)	(0.202)	(0.084)	(0.182)	(0.071)	(0.171)	(0.066)
Observations	1480	592	905	362	1115	446	1215	486
R-squared	0.991	0.999	0.993	0.998	0.992	0.998	0.992	0.999
Cov*Q.Dummy	YES	YES	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES	YES	YES
Quarter FE	YES	YES	YES	YES	YES	YES	YES	YES
Period	2019.Q1–2020.Q1	2019.Q1, 2020.Q1,	2019.Q1–2020.Q1	2019.Q1, 2020.Q1,	2019.Q1–2020.Q1	2019.Q1, 2020.Q1,	2019.Q1–2020.Q1	2019.Q1, 2020.Q1,

^†Robust standard errors clustered at the city level and reported in the parentheses;.

^⁎, ^⁎⁎ and.

^⁎⁎⁎denote significance at 10%, 5% and 1%.

5.3. Parallel trends

A potential concern regarding the DID estimation of the impacts of the pandemic and city lockdown is that the changes in the GRP was caused by a differential pre-existing trend. A necessary condition for satisfying our identification assumption is that the city groups separated by the severity of the pandemic or lockdown policies have similar time trends in the outcome variable. We test this assumption for the pre-treatment period. Specifically, we estimate the following:

$$Y_{ct}=\sum _{t=1}^6{\beta }_{t}Pandemi{c}_i*Quate{r}_t+\sum _{t=1}^6{\gamma }_{t}Lockdow{n}_i*Quarte{r}_t+\alpha X_{it}+{\mu }_i+{\eta }_t+{\epsilon }_{it}$$ (2)

where $Quarte{r}_t$ represents the quarterly dummies, and the fourth quarter in 2019 is the omitted category. $Lockdow{n}_i$ takes the value of 1 if the city is a lockdown city in the first quarter of 2020 and 0 otherwise. Similarly, the pandemic measure ($\text{Log}\left(Death{s}_i\right)$) take the values in the first quarter of 2020. For this test, we include data for all periods from the first quarter in 2019 to the third quarter in 2020. This also allow us to examine the marginal effects of the pandemic and intervention policies by quarter.

Fig. 4 plots the quarterly estimates for $Pandemi{c}_i$ (i.e. $\text{Log}\left(Death{s}_i\right)$) and $Lockdow{n}_i$ along with the 95% confidence intervals. The estimates show no consistent pre-treatment trend between the cities with different pandemic severity or cities with different lockdown status. Fig. 4 also suggests that city lockdown had only immediate impacts as the estimates became insignificant in the post-pandemic periods. On the other hand, the negative impact of pandemic remained significant in the second and third quarter although declined substantially. Parallel trend tests were also conducted for cities above or below the median values of chosen city characteristics. The results are provided in the Appendix. Overall, the results do not suggest violation of the parallel trend assumption.

Fig. 4.

Parallel trend tests for $\text{Log}\left(Death{s}_i\right)$ and $Lockdow{n}_i.$

5.4. City lockdown

The pandemic is largely exogenous but lockdown interventions were not randomly assigned. The validity of the identification depends on the assumption that the outcome variable is independent of the lockdown assignment, conditional on selected controls. Following Chetty et al. (2009) and La Ferrara et al. (2012), we conduct a robustness check by randomly assigning treatment (lockdown) status to cities. Specifically, we randomly draw and assign 88 cities out of the matched 176 cities as the lockdown cities. We then construct a false regressor of $Lockdow{n}_{it}^{false}$ and replace $Lockdow{n}_{it}$ in Specification (1). For this robustness check, we use data including only the first quarters of 2019 and 2020, as the baseline results using different sample periods do not differ significantly (Table 2). The necessary condition for satisfying conditional independence is that the falsified treatment regressor should have no effect on Log(GRP). We conduct the random sampling and assignment process 500 times to avoid possible impacts of incidental events.

Fig. 5 presents the distribution of the estimates of $Lockdow{n}_{it}^{false}$ and corresponding p-values controlling for COVID-19 deaths. The distribution centers around zero and most estimates are statistically insignificant. The baseline estimate (−0.028, from Column (8) in Table 2) indicated by the dashed vertical line in Fig. 3, is clearly beyond a critical value of a 1% rejection region in the placebo test. These results provide stronger support for our identification strategy.

Fig. 5.

The placebo test for the city lockdown policies.

5.5. Pandemic and lockdown spillovers

The economic performance in a city may be affected by the severity of pandemic and lockdown policies in neighboring cities, which leads to biased estimate of the causal effects. Because the pandemic spillover can occur in both ways between adjacent cities, its impact is a matter of empirical investigation. The impact of the lockdown spillover is also ambiguous. Lockdown may temporarily affect input-output linkages and restrict market access but it also helps to enhance consumer confidence and maintain normal production in neighboring cities.

Our first strategy here is to employ a spatial exclusion approach. Adjacent areas proximate to the treatment boundary are potentially most affected by the spillover effects (Ehrlich and Seidel, 2018, Kline and Enrico, 2014). We dropped all the neighboring cities of the lockdown cities and re-estimate Eq. (1). The results are presented in Columns (1) and (2) in Table 4 . The estimates for city lockdown are slightly lower (in absolute value) than those of the baseline (Columns (7)-(8) in Table 2). This may suggest a small positive spillover effect that the economy of adjacent areas benefits from a city's lockdown intervention. The estimates for the pandemic are larger than those of the baseline; however, the exclusion approach does not specifically address possible pandemic spillovers. Given the nature of the virus spread, it is possible that the pandemic is also more severe in the neighboring areas of lockdown cities. The larger estimates may be a result of the exclusion of these areas.

Table 4.

Test for spillover effects.

	(1)	(2)	(3)	(4)
Variables	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)
Log(Deaths)	−0.038⁎⁎⁎	−0.027⁎⁎⁎	−0.024⁎⁎⁎	−0.017⁎⁎⁎
	(0.005)	(0.004)	(0.004)	(0.003)
Log(Deaths_Neighbor)			−0.012⁎⁎⁎	−0.009⁎⁎⁎
			(0.003)	(0.002)
Lockdown	−0.042*	−0.030⁎⁎	−0.048⁎⁎	−0.028⁎⁎
	(0.025)	(0.015)	(0.021)	(0.013)
Lockdown_Neighbor			−0.002	−0.002
			(0.022)	(0.013)
Observations	600	240	880	352
R-squared	0.994	0.998	0.994	0.999
Period	2019.Q1–2020.Q1	2019.Q1& 2020.Q1	2019.Q1–2020.Q1	2019.Q1& 2020.Q1

^† Robust standard errors clustered at the city level and reported in the parentheses;.

^⁎,.

^⁎⁎and.

^⁎⁎⁎denote significance at 10%, 5% and 1%; all specifications control for covariates interacted with quarterly dummies, pandemic spillover, city FEs and quarter FEs.

A disadvantage of the spatial exclusion approach is that it does not differentiate and consider the nature of the pandemic and lockdown interventions of these adjacent areas. In Columns (3) and (4), we take an alternative approach by specifically controlling for the pandemic and lockdown intervention in the neighboring cities. Deaths_Neighbor is total new COVID-19 deaths in all neighboring cities. Lockdown_Neighbor takes the value of 1 if a city borders at least one lockdown city and 0 otherwise. The results on Log(Deaths_Neighbor) indicate significant spillovers of the pandemic. The economic impact of the pandemic on surrounding areas is more than half of the impact of the pandemic on the city itself. Once the pandemic spillover is controlled for, the main estimates of Log(Deaths) become slightly smaller than those of the baseline in Table 2. Specifically, a 1% increase in the number of confirmed deaths corresponds to a decline of GRP in the lockdown city and the neighboring cities by 0.017% and 0.009%, respectively (Column 4).

The results on Lockdown_Neighbor suggest no significant lockdown spillovers. For lockdown cities, our data show that the economy fell by 9.5% on average in the first quarter compared with the same period in 2019. The results in Column 4 indicate that city lockdown caused reduction in GRP by 2.8 percentage points, which explains 29% of the nominal decline, or 17.2% of the overall recession from the 5-year average growth of 6.74% prior to the pandemic. This is largely consistent with existing studies (Kong and Prinz, 2020; Forsythe et al., 2020; Lin et al., 2020; Goolsbee and Syverson, 2020). The pandemic is a common shock and has broader economic impacts whereas the interventions explain a smaller part of the impacts. Other factors, such as the decline of consumer demand due to fears of virus spread and contagion, may play more important roles in the economic slowdown.

Nevertheless, our estimate of the lockdown impact is much larger than those reported in other studies. Goolsbee and Syverson (2020) found that the shelter-in-place order explained 12% of the nominal traffic fall in the USA. Similarly, Kong and Prinz (2020) found that six NPIs can only explain 12.4% of UI claims filed. The difference may be a result of the different economic outcome variable used in the analysis. Previous studies mostly rely on indirect proxy data and may have only captured partial effects. We use GRP and the estimate therefore reflects the economy-wide impacts. A second explanation could be the stringency of intervention measures. Restrictions imposed in many developed countries are often much less stringent than the complete city lockdowns implemented in China. However, a more likely interpretation for the higher estimate of lockdown impact may be China's unique economic structure. Intra-regional and inter-regional migrant workers contribute significantly to Chinese economy. China has the world's largest migrant population (mostly migrant workers) accounting for over one sixth of the national population (Department of Floating Population, 2016). The largest seasonal migration also occurs in the first quarter of the year when migrant workers return home before the Spring Festival and then to the work place afterwards. The impact of lockdown restrictions may therefore be much greater in China than in other economies. Infrastructure investment and entrepreneurship have also been key drivers of China's economy growth. Impacts of pandemic and intervention measures on capital supply and new firm investment may also exact a heavier toll on the economy. We explore some of these mechanisms in the next section.

5.6. Impacts on labor migration, land supply, new firm entry and investment

The pandemic and city lockdown interventions can affect the economy through various microeconomic channels, such as restricting access to input and output markets (Fang et al., 2020; Bartik et al., 2020) and new infant firm entry and investment. We are particularly interested in the impacts on labor mobility, private and public investment activities, and new firm entries as these are among the most important drivers of Chinese economy. Examination on the variations of these microeconomic activities can help us understand the mechanisms through which the pandemic and city lockdown interventions affect economy. Specifically, we estimate Eq. (1) using five alternative outcome variables: 1) outgoing labor migration; 2) incoming labor migration; 3) total land supply for industrial, commercial, residential and public infrastructure uses; 4) new firm registrations; 4) new firm investment. The results are presented in Table 5 . All estimates are statistically significant and much larger in magnitude than baseline estimates of the impacts on the overall economy. In addition, the magnitude of the coefficients on city lockdown is more than four times than those of the pandemic in human mobility, new firm registration and investment regressions (Columns 1, 2, 4 and 5 in Table 5), and more than three times in the land supply regression (Column 3), indicating much larger impact of lockdown intervention on human mobility, land supply, new firm entries and investment than the pandemic. Taken together, our results suggest that the unique characteristics of Chinese economy (and perhaps also more stringent lockdown intervention) may help explain our much larger estimates of lockdown impacts than those reported in other studies.

Table 5.

The impacts of pandemic and lockdown on human mobility, land supply, new firm registration, and investment.

	(1)	(2)	(3)	(4)	(5)
Variables	Log(Moveout)	Log(Movein)	Log(Land)	Log(Firms_new)	Log(Investment)
Log(Deaths)	−0.054⁎⁎⁎	−0.037⁎⁎⁎	−0.105⁎⁎⁎	−0.032⁎⁎⁎	−0.031⁎⁎⁎
	(0.007)	(0.007)	(0.026)	(0.007)	(0.006)
Lockdown	−0.243⁎⁎⁎	−0.172⁎⁎⁎	−0.330⁎⁎	−0.139⁎⁎⁎	−0.128⁎⁎⁎
	(0.055)	(0.049)	(0.144)	(0.034)	(0.033)
Observations	26,400	26,400	1056	352	352
R-squared	0.900	0.917	0.611	0.991	0.992
City FE	YES	YES	YES	YES	YES
Time FE	YES	YES	YES	YES	YES
Period	2019.Jan 1st-2019, Mar 15th, 2020.Jan 1st-2020, Mar 15th,	2019.Jan 1st-2019, Mar 15th, 2020.Jan 1st-2020, Mar 15th,	2019. Jan-Mar, 2020. Jan.-Mar.,	2019.Q1, 2020.Q1,	2019.Q1, 2020.Q1,

^† Intercity daily data are used in the human mobility regressions, monthly data are used for land supply regression, and quarterly data are used in the new firm and investment regressions; all specifications control for covariates interacted with time dummies, pandemic spillover, city FEs and time FEs. Robust standard errors clustered at the city level and reported in the parentheses;.

*,.

^⁎⁎and.

^⁎⁎⁎denote significance at 10%, 5% and 1%.

5.7. Heterogeneity in economic slowdown and recovery

He et al. (2020) and Bonaccorsi et al. (2020) both found that the effects of city lockdown policies can vary across cities with different characteristics. Note that our heterogeneity analyses do not have causal interpretations but help us to understand the channels through which the pandemic and city lockdown affect economy. As described in Section 2.2, we examine five city characteristics: industrial structure, passenger traffic intensity, population density, internet access, and fiscal capacity. Panels A and B in Table 6 only use data for the first quarters of 2019 and 2020 and therefore examine the heterogeneous effects of the pandemic and city lockdown at the recession stage. In contrast, Panels C and D use data for the second and third quarters of 2019 and 2020 and examine recovery relative to the same periods in the previous year. Panels E and F further compare the second and third quarters of 2020 to the first quarter of 2020 and therefore reflect how cities differ in recovery from the recession. In all specifications, the pandemic measure and lockdown indicator are as defined in Specification (2). We also control for the pandemic spillover.

Table 6.

The heterogeneous impacts of the pandemic and city lockdown policies^†.

	(1) Share of 2nd industry	(2) Passenger traffic intensity	(3) Density of population	(4) Internet Access	(5)  Fiscal capacity
Recession / Panel A: Above Median
Log(Deaths)	−0.027⁎⁎⁎	−0.021⁎⁎⁎	−0.012⁎⁎⁎	−0.013⁎⁎⁎	−0.010*
	(0.005)	(0.004)	(0.004)	(0.004)	(0.006)
Lockdown	−0.043⁎⁎⁎	−0.047⁎⁎⁎	−0.021	−0.015	−0.003
	(0.016)	(0.016)	(0.014)	(0.016)	(0.011)
Recession / Panel B: Below Median
Log(Deaths)	−0.013⁎⁎⁎	−0.014⁎⁎⁎	−0.026⁎⁎⁎	−0.025⁎⁎⁎	−0.023⁎⁎⁎
	(0.004)	(0.005)	(0.005)	(0.005)	(0.004)
Lockdown	−0.020	−0.006	−0.033⁎⁎	−0.034⁎⁎	−0.056⁎⁎⁎
	(0.012)	(0.012)	(0.015)	(0.013)	(0.015)
Sample	N. Obs.: 176 / Period: 2019.Q1 & 2020.Q1
Recovery / Panel C: Above Median
Log(Deaths)	−0.012⁎⁎⁎	−0.009⁎⁎⁎	−0.006⁎⁎⁎	−0.005⁎⁎⁎	−0.003
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
Lockdown	−0.016⁎⁎	−0.018⁎⁎	−0.007	0.000	−0.005
	(0.007)	(0.007)	(0.006)	(0.006)	(0.006)
Recovery / Panel D: Below Median
Log(Deaths)	−0.004⁎⁎	−0.005⁎⁎⁎	−0.010⁎⁎⁎	−0.011⁎⁎⁎	−0.010⁎⁎⁎
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
Lockdown	−0.012⁎⁎	−0.004	−0.016⁎⁎	−0.025⁎⁎⁎	−0.023⁎⁎⁎
	(0.006)	(0.006)	(0.008)	(0.007)	(0.007)
Sample	N. Obs.: 352 / Period: 2019.Q2–2019.Q3 & 2020.Q2–2020.Q3
Recovery / Panel E: Above Median
Log(Deaths)*Post	0.040⁎⁎⁎	0.025⁎⁎	0.016*	0.020⁎⁎	0.011
	(0.006)	(0.012)	(0.009)	(0.008)	(0.010)
Lockdown*Post	0.090⁎⁎⁎	0.081⁎⁎⁎	0.044*	0.047⁎⁎	0.009
	(0.030)	(0.027)	(0.025)	(0.020)	(0.013)
Recovery / Panel F: Below Median
Log(Deaths)*Post	0.016⁎⁎	0.020⁎⁎⁎	0.037⁎⁎⁎	0.034⁎⁎⁎	0.034⁎⁎⁎
	(0.008)	(0.005)	(0.008)	(0.008)	(0.006)
Lockdown*Post	0.024	0.024	0.073⁎⁎⁎	0.056*	0.112⁎⁎⁎
	(0.019)	(0.023)	(0.027)	(0.033)	(0.032)
Sample	N. Obs.: 264 / Period: 2020.Q1-Q3

^†Robust standard errors clustered at the city level and reported in the parentheses;.

^⁎,.

^⁎⁎and.

^⁎⁎⁎denote significance at 10%, 5% and 1%.; all specifications control for covariates interacted with quarterly dummies, pandemic spillover, city FEs and quarter FEs.

Columns (1) and (3) in Panel A show that cities with higher share of secondary industry and lower population density were more affected by the pandemic and city lockdown. These results are generally consistent with national statistics. For example, in the first quarter of 2020, the national GDP declined by 6.8% whereas the secondary industry declined by 9.6%. Data from the National Statistical Bureau of China (NBSC) also indicate that the unemployment rate was higher in smaller cities (mostly low population density) than larger cities (mostly high population density) in the first quarter of 2020. Our results are broadly consistent with Dai et al. (2021b) and Pei et al. (2022). Pei et al. (2022) shows that densely populated coastal cities and cities with better-developed ICT infrastructure experienced less drop in export.8

The fear for virus contagion reduces people's willingness to travel (Goolsbee and Syverson, 2020). City lockdown policies further impose more stringent restrictions on travel. Therefore, cities depend more on passenger transportation may suffer more from the pandemic and related countermeasures. As expected, the pandemic and city lockdown policies had greater economic impacts on cities with higher passenger traffic intensity (Column (2), Panels A and B). In particular, the estimates for the lockdown in cities with low traffic intensity is not significant.

Column (4) in Panel A and B shows that cities with lower internet access suffered significantly more from lockdown interventions and the pandemic. On one hand, internet access can facilitate information dissemination and thus improve the resilience of cities under lockdown and pandemic. On the other hand, internet access also allows for opportunities for people to work from home. Our results are broadly consistent with Barrero et al. (2021) which also indicate that internet access can promote social and economic resilience during disasters like the COVID-19 pandemic.

The last column in Panel A and B has results for cities with different government fiscal capacity. Cities with greater fiscal capacity were more resilient to the shocks of the pandemic and city lockdown policies. The results also suggest that regions with weak government resources should be prioritized for fiscal transfers from the central government to mitigate the negative impacts. Bonaccorsi et al. (2020) found that mobility contraction is more severe in municipalities with stronger fiscal capacity in Italy. Our interpretation is that stronger fiscal capacity can help implement and enforce lockdown restrictions (hence greater mobility reduction). However, fiscal resources can also help to cushion the negative impacts and make cities more resilient during the pandemic and lockdown.

While many economies are still struggling in the economic shock of the pandemic and intervention measures, Chinese national statistics show that the country's economy has already started recovery with positive growth in the second to fourth quarters of 2020. Firm-level surveys also show that most small and medium-sized enterprises that have temporarily closed in the first quarter of 2020 have reopened by May (Dai et al., 2021a). Due to data limitation, we focus on the short-term recovery in the second and third quarters of 2020. We offer two perspectives: recovery relative to the same periods in 2019 and recovery from the recession in the first quarter of 2020. The results in Panels C and D indicate that those city groups hit harder by the pandemic and lockdown (i.e. cities with higher share of secondary industry, higher passenger traffic intensity, lower population density, lower internet access and lower fiscal capacity) were still lagging behind by the end of the third quarter in 2020 compared with the same periods in the previous year. However, when compared to the recession in the first quarter, the results in Panels E and F show that the same groups of cities recovered faster than those groups less affected by the pandemic and lockdown. For example, Column (1) in Panels E and F show that the cities with higher shares of secondary industry recovered twice as fast as those with lower shares secondary industry during the post-pandemic periods. Zhang et al. (2020) also find economic recovery is heterogeneous across economic sectors in China. Specifically, the employment recovery in the agri-food system is slower than that of other sectors largely due to the sluggish recovery of restaurants.

Taken together, the results suggest cities were quickly closing the gap in the economic performance induced by the pandemic and lockdown interventions. This may be explained partially by the immediate recovery interventions targeted at the most affected regions. For example, the People's Bank of China established lending facilities with a capacity of over 2 trillion RMB to fund loans for small businesses and poverty alleviation. However, we are unable to unpack the causal link due to lack of data. As we are providing the first analysis on the immediate recovery from the pandemic, the results are only suggestive. Further research and comparisons to other recovering economies would be very useful when data becomes available in the near future.

6. Conclusions

The COVID-19 crisis has had significant impacts on the economy of most countries. However, different countries experienced heterogeneous decline and recovery trajectories. Combining a city-level dataset and a microeconomic dataset on labor mobility, total land supply, and entrepreneurship, this study examines the economic impacts of the pandemic and city lockdown polices in a major developing country during both the recession and the recovery stages. Our findings complement existing studies and have important implications for policies adopted to contain the spread of COVID-19 and to reduce its impact on the economy and residents.

Our results indicate that both the pandemic and strict city lockdown implemented in China play important roles in explaining the economic recession. However, we document much larger impact of lockdown interventions than those reported for developed countries. This could be a result of differences in data used, stringency of intervention measures or economic structure. Our findings also reveal significant pandemic spillover but no lockdown spillover, which may be of interest for intervention coordination in adjacent areas and simulation of regional economic impacts.

The findings also reveal significant heterogeneity in cities’ economic responses to the pandemic and lockdown measures. The greater impacts on less developed regions characterized by lower internet access, low population density and lower fiscal capacity may enlarge the existing inequality and increase the incidence of poverty. Although these cities were quickly closing the pandemic-induced economic gap, they still hadn't fully caught up with the better performed areas. These heterogeneous effects have important implications for government to prioritize resources to mitigate possible impacts on poverty and inequality, and to facilitate economic recovery.

Perhaps the largest concern why many governments are reluctant to implement the most drastic interventions is possible impacts on the economy. Both economic and political risks make the pandemic-economy trade-off extremely challenging (Goolsbee and Syverson, 2020). Of course, the immediate impacts may depend on the structure of the economy and the long-term impacts are often obscure. There may be a wide spectrum of impact and recovery pathways. Our analysis on China provides an important case and useful contrast with experiences elsewhere. China adopted maybe the world's most stringent lockdown restrictions. The economy was hit hard by the restrictions but the impacts seemed very much short-term and the recovery was relatively quick (at least for the period under study). In Europe, the lockdowns were shorter and less severe and the economic losses in the short run may have been smaller. However, the recovery has been much more difficult as cases continued to be high. It is worth mentioning though, what we presented in the paper was evidence over a very short period of time – that is immediate impacts and immediate recovery. We are not sure how the long-term impacts would unfold. Given China's recent move to relaxing restrictions, it would be very interesting to explore the mid-to-long term effects on the economy. We hope that our study, along with other empirical evidences, would provide useful insights to the policy debate and post-pandemic assessments.

Further research is desirable to extend our investigation of economic impact of the COVID-19 pandemic and its related policies. First, our results are limited to the short-run effects of the pandemic and intervention policies. It remains unknown whether the impacts are just a one-time shock or have changed some industries permanently. Future research could provide evidence on the long-run effects on economy and its structural change. Second, China implemented more strict intervention measures compared with most other countries and achieved a rapid recovery from the recession. Studies of different intervention measures and comparisons with other recovering economies would also be promising research areas.

Contributions

All authors have materially participated in the research and article preparation. J.X.W. and C.M. designed research. J.X.W., H.X. and X.L.Z. collected data. C.M., J.X.W., H.X. and X.L.Z. performed research and analyzed data. C.M. and J.X.W. wrote the paper.

Declaration of Competing Interest

The authors declare no known interests related to their submitted manuscript.

Appendix

Fig. A1, Fig. A2

Table A1.

Balance tests.

Variables	Sample	Bias (%)	P-value
Industrial structure	Unmatched	−7.50	0.567
	Matched	−0.60	0.966
Hospital beds	Unmatched	−0.70	0.958
	Matched	−5.30	0.744
Distance to Wuhan	Unmatched	−37.30	0.005
	Matched	−4.80	0.694
FDI Intensity	Unmatched	13.90	0.313
	Matched	−5.30	0.708
Population mobility	Unmatched	53.50	0.000
	Matched	11.90	0.466

Table A2.

The main results with and without control variables.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Variables	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)	Log(GRP)
Log(Deaths)	−0.029⁎⁎⁎	−0.022⁎⁎⁎	−0.030⁎⁎⁎	−0.022⁎⁎⁎	−0.027⁎⁎⁎	−0.018⁎⁎⁎	−0.028⁎⁎⁎	−0.017⁎⁎⁎
	(0.005)	(0.004)	(0.005)	(0.004)	(0.005)	(0.003)	(0.005)	(0.003)
Lockdown	−0.046⁎⁎⁎	−0.028⁎⁎⁎	−0.048⁎⁎⁎	−0.028⁎⁎⁎	−0.039⁎⁎⁎	−0.034⁎⁎⁎	−0.053⁎⁎⁎	−0.036⁎⁎⁎
	(0.017)	(0.010)	(0.018)	(0.011)	(0.015)	(0.009)	(0.016)	(0.010)
Income level *2019.Q1			−0.029	−0.013			−0.008	0.007
			(0.019)	(0.016)			(0.013)	(0.010)
Income level *2019.Q2			−0.034				−0.029
			(0.028)				(0.021)
Income level *2019.Q3			−0.059⁎⁎				−0.069⁎⁎⁎
			(0.026)				(0.021)
Income level *2019.Q4			−0.058				−0.121⁎⁎⁎
			(0.037)				(0.039)
R&D intensity *2019.Q1			−0.003⁎⁎	−0.003⁎⁎⁎			−0.001	−0.001
			(0.001)	(0.001)			(0.002)	(0.002)
R&D intensity *2019.Q2			−0.006⁎⁎⁎				−0.003
			(0.001)				(0.002)
R&D intensity *2019.Q3			−0.004*				−0.002
			(0.002)				(0.003)
R&D intensity *2019.Q4			−0.001				0.001
			(0.003)				(0.003)
Population mobility *2019.Q1			0.000	0.000			−0.000	−0.000
			(0.000)	(0.000)			(0.000)	(0.000)
Population mobility *2019.Q2			0.000				−0.000
			(0.000)				(0.000)
Population mobility *2019.Q3			0.000				0.000
			(0.000)				(0.000)
Population mobility *2019.Q4			0.000				0.000*
			(0.000)				(0.000)
Trade openness *2019.Q1			−0.033	−0.031			−0.029	−0.028*
			(0.028)	(0.024)			(0.020)	(0.017)
Trade openness *2019.Q2			−0.022				−0.016
			(0.046)				(0.036)
Trade openness *2019.Q3			−0.022				−0.000
			(0.048)				(0.040)
Trade openness *2019.Q4			−0.056				−0.036
			(0.059)				(0.047)
Doctors *2019.Q1			0.000	−0.000			−0.000	−0.000
			(0.000)	(0.000)			(0.000)	(0.000)
Doctors *2019.Q2			−0.000				0.000
			(0.001)				(0.001)
Doctors *2019.Q3			−0.001				−0.000
			(0.001)				(0.001)
Doctors *2019.Q4			0.000				0.001
			(0.001)				(0.001)
Hospital beds *2019.Q1			−0.006	−0.010			0.006	−0.003
			(0.016)	(0.016)			(0.011)	(0.011)
Hospital beds *2019.Q2			−0.002				0.007
			(0.031)				(0.022)
Hospital beds *2019.Q3			0.017				0.019
			(0.034)				(0.025)
Hospital beds *2019.Q4			−0.014				−0.002
			(0.038)				(0.033)
Constant	15.546⁎⁎⁎	15.440⁎⁎⁎	15.940⁎⁎⁎	15.532⁎⁎⁎	15.237⁎⁎⁎	15.140⁎⁎⁎	15.692⁎⁎⁎	15.119⁎⁎⁎
	(0.022)	(0.014)	(0.207)	(0.088)	(0.020)	(0.013)	(0.162)	(0.059)
Observations	880	352	880	352	1480	592	1480	592
R-squared	0.993	0.998	0.993	0.998	0.990	0.999	0.991	0.999
Period	2019.Q1–2020.Q1	2019.Q1& 2020.Q1	2019.Q1–2020.Q1	2019.Q1& 2020.Q1	2019.Q1–2020.Q1	2019.Q1& 2020.Q1	2019.Q1–2020.Q1	2019.Q1& 2020.Q1

^† Column (1) and (2) are the results using matched control sample without controls, Column (5) and (6) are the results using unmatched full sample without controls. Column (3) and (4) correspond to column (7)-(8) in Table 2, and Column (7) and (8) correspond to column (1)-(2) in Table 3. Robust standard errors clustered at the city level and reported in the parentheses;.

^⁎,.

^⁎⁎and.

^⁎⁎⁎denote significance at 10%, 5% and 1%.

Fig. A1.

Parallel trend tests for $\text{Log}\left(Death{s}_i\right)$ by city characteristic.

Fig. A2.

Parallel trend tests for $Lockdow{n}_i$ by city characteristic.

Data availability

• Data will be made available on request.