Did state-owned enterprises do better during COVID-19? Evidence from a survey of company executives in China

Abstract

In a survey of 1,182 company executives in China, state-owned enterprises (SOEs) reported less business reductions under COVID-19. This paper examines if SOEs’ superior performance was resulted from government support rather than innate ability of coping with the pandemic. We construct a proxy for firm-level government support using firm's human resources (HR) action taken during the outbreak with firm's 2019 China revenue share as an instrument for the HR action variable. After controlling for the proxy for firm-level government support as well as other observed firm characteristics, we find SOEs in the sample performing significantly worse in the pandemic period.

1. Introduction

COVID-19's sudden outbreak in China, and the subsequent drastic measures taken by the Chinese government to stop its spread, substantially changed China's business environment. From January 23, 2020 (when Wuhan was locked down) to February 12, more than 200 Chinese cities (including 26 provincial capitals and sub-provincial cities) implemented strict quarantine regulations. As production and spending were frozen by the lockdown/quarantine measures, China's gross domestic product (GDP) plummeted 6.8% in the first three months of the year compared with a year earlier, its first such drop since the National Bureau of Statistics of China began publishing quarterly GDP data in 1992.

In this paper, we use data from an online survey of 1,182 company executives in China, which was conducted in April 2–9, 2020. With business operations in China severely impacted by the sudden outbreak of COVID-19, the survey data provides valuable first-hand information on how companies in China responded to the COVID-19 shock. In particular, facing the unprecedented situation, companies took actions which would be seldom observed in normal times. For example, when asked what HR (human resources) measures already taken in the first quarter, 199 survey participants (16.8%) reported “laid off workers”, 138 (11.7%) reported “cut salaries across company”, and 18 (1.5%) reported “raised salary or raised hiring”. Such seldom-observed variations in firm behavior provide a unique opportunity for the research presented in this paper.

This paper examines how state-owned enterprises (SOEs) fared under the COVID-19 shock compared with non-state-owned enterprises (NSOEs). There are extensive studies in the literature on firm behavior under public and private ownership. In a principal-agent framework, Sappington and Stiglitz (1987) modeled public and private firms as delegated production arrangements in which the government retains some authority to intervene directly, with the main difference lying in the transactions costs faced by the government when attempting to intervene in the delegated production activities. In their modeling, the government has greater ease to intervene under public ownership, but its promise not to intervene is more credible under private ownership. Applying this theory, we would expect more government assistance for SOEs than NSOEs but more active adaption of NSOEs than SOEs in the pandemic period.

Megginson and Netter (2001) provided a comprehensive survey on privatization and concluded that “(research) now supports the proposition that privately owned firms are more efficient and more profitable than otherwise-comparable state-owned firms” (p. 380). However, the empirical evidence on the performance of Chinese SOEs is mixed.1 Some studies show that productivity of China's SOEs had been improved (Cornelli & Li, 1997; Groves, Hong, McMillan, & Naughton, 1994), others show that the reform of SOEs in China was far from successful (Lin, Cai, & Li, 1998). In a study of China's privatization experience over the period 1994–1997, Sun, Tong, and Tong (2002) found an inverted U-shape relationship between state ownership and firm performance in a sample of China's listed companies with firm performance measured by the market-to-book ratio of equity. They explained this result by claiming “too much government holding of SOE shares means too much control and interference in the economic operations of SOEs; too little government holding means too little support from the government to pull the SOEs out from their difficulties” (p. 23). Viewing this empirical result in the theory of Sappington and Stiglitz (1987), the inverted U-shape relationship can be understood as resulted from the tradeoff between the beneficial effect enjoyed by SOEs in government support (which raises their financial performance) and the detrimental effect associated with the inefficiency of SOEs (which lowers their financial performance).2

Responses to the survey indicate that executives of SOEs expected better recovery of business (by end of June) and achievement of revenue target (by end of 2020). In this paper, we examine if SOEs’ superior performance was resulted from government support rather than innate ability of coping with the pandemic. Thanks to the sudden and gigantic COVID-19 shock, we observe companies taking drastic actions that reveal information about the underlying forces including government support. In this paper, we utilize such information to construct a proxy for firm-level government support. By controlling for the proxy for firm-level government support as well as other observed firm characteristics, we are able to estimate the SOE-NSOE performance difference associated with their innate ability of coping with the pandemic.

While SOEs receive more government support than NSOEs, they also have more burdens imposed by the government (Bai, Lu, & Tao, 2006; Song, Kjetil, & Zilibotti, 2011; Zhang, Zhang, & Zhao, 2002). For example, SOEs are obliged to hire excess labor (Berkowitz, Ma, & Nishioka, 2017; Chong, Guillen, & Lopez-de-Silanes, 2011; Cooper, Gong, & Yang, 2015) and are often used as instruments of macroeconomic policy and industry regulations (Bai, Li, Tao, & Wang, 2000). In previous studies, some examined firm-government relationship from the cost perspective (policy burdens on SOEs, e.g. Jian, Li, Meng, & Zhao, 2020), some others from the benefit perspective (ownership-based resources for SOEs, e.g. Ren, Manning, & Vavilov, 2019). In this paper, we do not distinguish between the cost side and the benefit side of the firm-government relationship. The proxy we construct should be interpreted as measuring the net government support effect after considering the cost related to government-imposed burdens.

The remainder of the paper is organized as follows. Section 2 describes the data. Section 3 lays out the empirical approach. Section 4 reports the results. Section 5 concludes.

2. Data description

Data for this research is drawn from an online survey of company executives in China launched at the beginning of April 2020, conducted by a research team of China Europe International Business School (CEIBS).3 The survey data captured the sharp and immediate responses of companies in China to the COVID-19 shock in the first quarter of 2020.4 As the survey was distributed mainly via CEIBS channels, 98.6% (1166) respondents are CEIBS alumni/students, among whom two-thirds are Executive MBAs (EMBAs). It is worth noting that over 97% of CEIBS EMBA students/alumni hold top/senior management positions in their companies, which ensures the reliability of the survey data.5

In this paper, we use the survey data to examine the implication of state ownership on firm performance after controlling for firm-level government support. In our sample of 1,182 companies, 113 (9.6%) are SOEs, 735 (62.2%) are Chinese private enterprises, and 290 (24.5%) are foreign/overseas-owned enterprises in China or joint ventures with more than 50% foreign/overseas ownership.6 The ownership distribution of our sample is consistent with that of the population: 1.8% of the total number of corporate enterprises in China are SOEs, 89.5% are Chinese private enterprises, and 1.2% are foreign/overseas enterprises (data in 2017, from China's National Bureau of Statistics). In terms of total current assets in 2017, the share of industrial SOEs is 3.1%, the share of Chinese industrial NSOEs is 72.9%, and the share of foreign/overseas industrial enterprises is 24.0%. SOEs remain a significant employer of workers in China. In 2017, 14.3% of urban workers were employed by SOEs, 31.4% by Chinese private enterprises, 22.0% self-employed, and 6.0% by foreign/overseas enterprises.

The survey contains three indicators on company's assessment of COVID-19's impact on their business operations: (1) Estimated reduction of business activities in China in the first quarter; (2) Expected recovery of business activities by end of June; (3) Estimated adjustment of 2020 target revenue. Table 1 provides a comparison of these three indicators between the SOE sample and the NSOE sample.

Table 1.

COVID-19's impact and company responses: SOEs vs. NSOEs.

Level	Impact on first-quarter business activities ($\mathit{FBR}$)	SOE sample	NSOE sample
	(in ascending order of business reduction)
1	Small reduction ($<$20%)	37.5% ($N$ = 42)	26.6% ($N$ = 283)
2	Medium reduction (20–39%)	25.9% ($N$ = 29)	25.1% ($N$ = 267)
3	Large reduction (40–59%)	15.2% ($N$ = 17)	18.7% ($N$ = 199)
4	Extra-large reduction (60–79%)	6.3% ($N$ = 7)	11.8% ($N$ = 125)
5	Huge reduction ($\ge$80%)	11.6% ($N$ = 13)	12.5% ($N$ = 133)

Level	Expected recovery by end of June 2020 ($\mathit{REC}$)	SOE sample	NSOE sample
	(in ascending order of expected recovery)
1	Small recovery ($<$20%)	3.5% ($N$ = 4)	6.5% ($N$ = 70)
2	Medium recovery (20–39%)	4.4% ($N$ = 5)	7.6% ($N$ = 81)
3	Large recovery (40–59%)	12.4% ($N$ = 14)	12.1% ($N$ = 129)
4	Extra-large recovery (60–79%)	15.9% ($N$ = 18)	23.4% ($N$ = 250)
5	Huge recovery ($\ge$80%)	57.5% ($N$ = 65)	43.0% ($N$ = 460)

Level	Estimated adjustment of 2020 revenue target ($\mathit{REV}$)	SOE sample	NSOE sample
	(in descending order of downward adjustment)
1	Large downward adjustment ($\ge$20%)	23.9% ($N$ = 27)	33.7% ($N$ = 360)
2	Medium downward adjustment (10–19%)	15.0% ($N$ = 17)	20.0% ($N$ = 214)
3	Small downward adjustment (3–9%)	8.0% ($N$ = 9)	8.4% ($N$ = 90)
4	Little/no adjustment ($<$3%)	46.9% ($N$ = 53)	31.1% ($N$ = 332)
5	Small upward adjustment (3–9%)	1.8% ($N$ = 2)	1.2% ($N$ = 13)
6	Medium/large upward adjustment ($\ge$10%)	4.4% ($N$ = 5)	5.1% ($N$ = 55)

Level	Salary/personnel actions already taken ($\mathit{HR}$)	SOE sample	NSOE sample
	(in descending order of harshness to employees)
1	Laid off workers	3.5% ($N$ = 4)	18.2% ($N$ = 195)
2	Cut salaries across the company	4.4% ($N$ = 5)	12.4% ($N$ = 133)
3	Cut salaries of staff only	0.0% ($N$ = 0)	0.7% ($N$ = 8)
4	Cut salaries of senior executives only	6.2% ($N$ = 7)	11.4% ($N$ = 122)
5	No salary cut	67.3% ($N$ = 76)	53.4% ($N$ = 571)
6	No layoffs	87.6% ($N$ = 99)	64.7% ($N$ = 692)
7	Raised salary or raised hiring	0.9% ($N$ = 1)	1.6% ($N$ = 17)

The top part of Table 1 displays the comparison in first-quarter business reductions (denoted by $\mathrm{FBR}$) measured in five levels from “Small reduction ($<$20%)” ($\mathrm{FBR}=1$) to “Huge reduction ($\ge$80%)” ($\mathrm{FBR}=5$). Based on a t-test, the hypothesis that “SOE sample mean ($\mathrm{FBR}$) $<$ NSOE sample mean ($\mathrm{FBR}$)” is accepted ($p$-value = 0.014). Similar results are obtained (shown in middle parts of Table 1) for expected business recovery by end of June (denoted by $\mathrm{REC}$) and estimated adjustment of 2020 revenue target (denoted by $\mathrm{REV}$). The hypothesis that “SOE sample mean ($\mathrm{REC}$) $>$ NSOE sample mean ($\mathrm{REC}$)” is accepted ($p$-value = 0.006), and the hypothesis that “SOE sample mean ($\mathrm{REV}$) $<$ NSOE sample mean ($\mathrm{REV}$)” is accepted ($p$-value = 0.004). Thus, in all three dimensions, SOEs fared better than NSOEs. The bottom part of Table 1 shows a comparison between the SOE sample and the NSOE sample in terms of the HR actions that companies had already taken in the first quarter (denoted by $\mathrm{HR}$), measured in seven levels in descending order of harshness to employees from “Laid off workers” ($\mathrm{HR}=1$) to “Raised salary or hiring” ($\mathrm{HR}=7$). In our empirical estimation, we utilize this HR data, which exhibits variations seldom observed in normal times. The survey also provides data on firm's industry (20 classified industries, 10 in manufacturing and 10 in services), firm's rating of government support to the industry where it operates, share of firm's revenue generated from China, firm size measured by number of employees, and firm's client type (selling to individuals, to firms, or to both). We use these data as control variables in our regression analysis.

3. Empirical approach

In this section, we first lay out an illustrative model. Consider company $i$ seeking profit ${\pi }_i$. We specify the following reduced-form equation for company $i$'s expected profit:

$$E({\pi }_i)=f({\text{S}}_i,{\text{G}}_i,{\text{X}}_i),$$ (1)

where we distinguish between variables of ownership-based firm behavior (${\text{S}}_i$), variables of ownership-related government policies (${\text{G}}_i$), and other firm characteristic variables (${\text{X}}_i$).

Our survey data provides two measures that correspond to expected profit $E({\pi }_i)$. The first one is “Expected recovery by end of June” (${\mathrm{REC}}_i$), and the second one is “Estimated adjustment of 2020 revenue target” (${\mathrm{REV}}_i$). The survey classifies firms into (1) Chinese state-owned or state-holding company (${\mathrm{SOE}}_i=1$); (2) Chinese private or private-holding company; (3) Wholly foreign-owned enterprise; (4) Joint venture with both Chinese and foreign share-holding; (5) Others. In our analysis, we combine all non-SOE type enterprises into one NSOE category (${\mathrm{SOE}}_i=0$).7 Ideally, we would like to estimate:

$$R_i={\mathrm{\Sigma }}_j{\beta }_j+{\alpha }_1{\mathrm{SOE}}_i+{\alpha }_g{G}_i+\sum _k({\alpha }_k{\text{X}}_{\mathrm{ki}})+{\epsilon }_i,$$ (2)

where $R_i$ (either ${\mathrm{REC}}_i$ or ${\mathrm{REV}}_i$) is the dependent variable, ${\mathrm{\Sigma }}_j{\beta }_j$ are industry fixed effects ($j=1,2,\dots ,19$), $G_i$ measures government assistance firm $i$ received or expected to receive during the pandemic period, $\alpha$ denotes coefficients, ${\text{X}}_{\mathrm{ki}}$ are all potential exogenous factors, and ${\epsilon }_i$ is an error term. By estimating Eq. (2), we would get an unbiased estimate of ${\alpha }_1$ (estimated effect specific to SOEs) with firm-level government assistance ($G_i$) controlled for.

The key to this estimation is to find measures of firm-level government assistance ($G_i$). The survey asked participants to rate Chinese government's support to their industry under COVID-19 (first quarter) on a scale from 0 (lowest support) to 10 (highest support). Based on this data, we construct variable $\mathrm{GI}$. Not surprisingly, the hypothesis that “SOE sample mean ($\mathrm{GI}$) $>$ NSOE sample mean ($\mathrm{GI}$)” is accepted in a t-test ($p$-value = 0.000). The average rating of SOEs is 6.76 as opposed to 5.62 of NSOEs. Although the survey question was about government support to the $\mathrm{industry}$, the rating from participants of the same industry varies significantly.8 Thus, we consider ${\mathrm{GI}}_i$ a variable capturing part of government assistance to the $\mathrm{firm}$. For our estimation, we specify the following regression equation:

$$R_i={\mathrm{\Sigma }}_j{\beta }_j+{\alpha }_1{\mathrm{SOE}}_i+{\alpha }_2{\mathrm{SIZ}}_i+{\alpha }_3{\mathrm{CLT}}_i+{\alpha }_4{\mathrm{GI}}_i+{\alpha }_5{\mathrm{GF}}_i+\sum _k({\alpha }_k{\text{X}}_{\mathrm{ki}})+{\epsilon }_i,$$ (3)

where ${\mathrm{SIZ}}_i$ is firm size, ${\mathrm{CLT}}_i$ is firm's client type, ${\mathrm{GF}}_i$ is firm-specific government assistance not captured by ${\mathrm{GI}}_i$ and not related to firm size/client type, and ${\text{X}}_{\mathrm{ki}}$ denotes all other firm-specific factors ($k=6,7,\dots ,K$). To get an unbiased estimate of ${\alpha }_1$, we need to meet two conditions: (i) there are measures of ${\mathrm{GF}}_i$ (firm-level government support) and ${\text{X}}_{\mathrm{ki}}$ (all other firm-specific factors); (ii) there is no correlation between the explanatory variables and the error term.

Unmeasured firm-specific government support (${\mathrm{GF}}_i$) is not observable, but certain decisions made under the sudden and massive COVID-19 shock may reveal the impact of such government support. HR decision can be thought of as the decision that packages all types of government support and associated burdens. Government assistance comes in many forms; explicit ones (such as direct subsidy and lower loan rate) and implicit ones (such as expected future support and more purchase orders); existing ones and expected ones. For example, SOEs are known to maintain workforce and act as a safety net (Berkowitz et al., 2017; Chong et al., 2011; Cooper et al., 2015). Chong et al. (2011) found that on average 78% of the SOEs downsized their labor force prior to privatization in a sample of 84 countries, implying the big role played by SOEs in supporting the government's employment goal. Thus, the firm-government relationship is a crucial driver behind personnel decisions. A more supportive HR decision signals closer ties with the state and possibly more assistance, whereas a harsher HR decision signals less net government support.

Since HR action reveals the underlying firm-government relationship, it is subject to the endogeneity issue of being possibly determined at the same time as expected performance. To mitigate the measurement errors and the potential endogeneity of $\mathrm{HR}$ as an explanatory variable, we use a two-stage estimation approach. In the first stage, we run a regression with $\mathrm{HR}$ as the dependent variable:

$${\mathrm{HR}}_i={\mathrm{\Sigma }}_j{\gamma }_j+{\delta }_s{\mathrm{SOE}}_i+{\delta }_g{\mathrm{GI}}_i+{\delta }_m{\mathrm{MCH}}_i+\sum _k({\delta }_k{\text{X}}_{\mathrm{ki}})+{\eta }_i,$$ (4)

where ${\mathrm{\Sigma }}_j{\gamma }_j$ are industry fixed effects ($j=1,2,\dots ,19$), $\delta$ denotes coefficients, and ${\eta }_i$ is an error term. ${\text{X}}_{\mathrm{ki}}$ is a set of variables that affect firm's HR decision, which includes $\mathrm{SIZ}$ (firm size), $\mathrm{CLT}$ (firm's client type), and $\mathrm{FBR}$ (the underlying factors impacting firm performance as reflected in first-quarter business reduction). Variable $\mathrm{MCH}$ is constructed from the survey question “share of company's 2019 revenue generated from business operations in China” (five levels), which we use as an instrument variable for HR action. In our survey data, $\mathrm{MCH}$ is positively correlated with $\mathrm{HR}$ (Pearson correlation coefficient = 0.062; $p$-value = 0.038), and is not correlated with $\mathrm{REC}$ (Pearson correlation coefficient = $-$0.025; $p$-value = 0.412). As the COVID-19 situation in the first quarter was much severer in China than outside China, companies with a higher share of revenue generated from China were more pressed to take quick and drastic HR measures; this explains the high correlation found in our data between the China-revenue-share variable $\mathrm{MCH}$ and the HR measure variable $\mathrm{HR}$. However, entering March, the severity of the COVID-19 situation fell in China but raised significantly outside China, and consequently the degree of recovery expected by end of June and the degree of adjustment of revenue target estimated for the year became insensitive to the share of revenue generated from China or from outside China; this explains the lack of correlation in our data between the China-revenue-share variable $\mathrm{MCH}$ and the expected recovery/revenue variable $R$. These two statistical features make the China-revenue-share variable $\mathrm{MCH}$ a valid instrument variable for our estimation.

From estimating (4), we obtain the predicted value of $\mathrm{HR}$ (denoted by $\widehat{\mathrm{HR}}$), which we use as a proxy for unmeasured firm-specific government support (${\mathrm{GF}}_i$ in Eq. (3)) in the following second-stage regression:

$$R_i={\mathrm{\Sigma }}_j{\beta }_j+{\alpha }_1{\mathrm{SOE}}_i+{\alpha }_2{\mathrm{SIZ}}_i+{\alpha }_3{\mathrm{CLT}}_i+{\alpha }_4{\mathrm{GI}}_i+{\alpha }_5{\widehat{\mathrm{HR}}}_i+{\epsilon }_i.$$ (5)

To summarize, our study adopts a two-stage estimation approach. In the first stage, we estimate Eq. (4) to get an estimate of what was behind firm's HR action during the outbreak of COVID-19, and we consider it as a proxy for the underlying firm-government relationship. In the second stage, we estimate Eq. (5) by controlling for this proxied firm-government relationship effect. With our industry-level government support measure (${\mathrm{GI}}_i$, reported by firms in the survey) capturing part of firm-specific government support effect, adding the proxy variable (${\widehat{\mathrm{HR}}}_i$) controls for part of the firm-specific government support effect not captured by ${\mathrm{GI}}_i$. In this way, we hope to obtain a less biased estimate of the coefficient on $\mathrm{SOE}$, which we interpret as associated with ownership-related firm behavior. Admittedly the estimated effect of state ownership from this approach may still be biased as there remains an omitted variable issue that some unobserved firm-level factors are not controlled for.9 In addition, the endogeneity and measurement error issue still remains as our procedure relies on the assumption that HR decisions do not determine firm's expected performance. While applying this two-stage estimation approach helps reduce the concerns of these two issues, our estimation does not solve these two issues partly due to the lack of data for constructing a better proxy for the firm-government relationship effect.

4. Estimation results

Table 2 provides a description of the variables used in our regression estimation. As the survey asks multiple-choice questions, the dependent variables ($\mathrm{REC}$, $\mathrm{REV}$ and $\mathrm{HR}$) and some explanatory variables ($\mathrm{SIZ}$, $\mathrm{GI}$, $\mathrm{MCH}$, and $\mathrm{FBR}$) are ordinal variables. We also construct several dummy variables as explanatory variables ($\mathrm{SOE}$ and $\mathrm{CLT}$).

Table 2.

Description of variables.

Variable	Description	Type and value	Mean	Stdev
$\mathit{FBR}$	Business reduction in first quarter	Ordinal variable (5 levels):	2.53	1.36
		Smallest (=1) to Largest (=5).
$\mathit{REC}$	Expected recovery by end of June	Ordinal variable (5 levels):	3.99	1.24
		Smallest (=1) to Largest (=5).
$\mathit{REV}$	Adjustment of 2020 revenue target	Ordinal variable (6 levels):	2.60	1.50
		Large downward (=1)
		to Medium/large upward (=6).
$\mathit{SOE}$	Firm ownership (SOE vs. NSOE)	Dummy variable:	0.10	0.29
		$\mathit{SOE}=1$ for SOEs;
		$\mathit{SOE}=0$ for NSOEs.
$\mathit{SIZ}$	Firm size (number of employees)	Ordinal variable (7 levels):	3.78	1.41
		Smallest (=1) to Largest (=7).
$\mathit{CLT}$	Firm's client type	Dummy variable:
		$\mathit{CLT}=1$ if B2B only;	0.55	0.50
		$\mathit{CLT}=0$ otherwise.
$\mathit{GI}$	Government support to industry	Ordinal variable (10 levels):	5.72	2.63
		Lowest (=0) to Highest (=10).
$\mathit{HR}$	HR measures taken in first quarter	Ordinal variable (7 levels):	3.60	1.43
	(in descending order of harshness)	Laid off workers (=1)
		Sizable increase in hiring (=7).
$\mathit{MCH}$	Share of 2019 revenue from China	Ordinal variable (5 levels):	2.33	1.48
		Highest (=1) to Lowest (=5).

Table 3 reports results from six regressions with expected business recovery by end of June ($\mathrm{REC}$) as the dependent variable. In the survey question, there are five choices corresponding to five equally-divided percentage ranges. Based on the answers, we construct $\mathrm{REC}$ as a five-level ordinal variable, with “$\mathrm{REC}=1$” indicating “Smallest expected recovery ($<$20%)” and “$\mathrm{REC}=5$” indicating “Largest expected recovery (80% or more)”. Since the survey was conducted on April 2–9, a company's $\mathrm{REC}$ level reflected the first-quarter impact it felt and the second-quarter situation it expected to face with regard to the COVID-19 shock.

Table 3.

Expected business recovery by end of June.

	(3.1)	(3.2)	(3.3)	(3.4)	(3.5)	(3.6)
$\mathit{SOE}$	0.284**	0.159	0.058	$-$0.046	$-$0.696***	$-$0.996***
	(0.121)	(0.123)	(0.129)	(0.129)	(0.146)	(0.174)
$\mathit{SIZ}$		0.166***	0.140***	0.141***	0.146***	0.170***
		(0.027)	(0.028)	(0.027)	(0.027)	(0.027)
$\mathit{CLT}$		0.158*	0.142*	0.116	$-$0.157*	$-$0.106
		(0.084)	(0.085)	(0.083)	(0.087)	(0.092)
$\mathit{GI}$			0.093***	0.084***	$-$0.004	$-$0.016
			(0.016)	(0.016)	(0.018)	(0.020)
$\mathit{HR}$				0.190***
				(0.028)
$\widehat{\mathit{HR}}$					1.316***	1.279***
					(0.119)	(0.124)
Constant	3.962***	3.256***	2.841***	2.075***	$-$2.125***	$-$1.943***
	(0.039)	(0.132)	(0.152)	(0.185)	(0.472)	(0.473)
Industry fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Observations ($N$)	1,088	1,088	1,012	978	931	810
R-squared ($R^2$)	0.078	0.110	0.152	0.201	0.273	0.288

Notes: The dependent variable is $\mathit{REC}$. $\widehat{\mathit{HR}}$ is the predicted value of $\mathit{HR}$ of the regressions displayed in Table 4, with (3.5) using (4.1) and (3.6) using (4.2). See Table 2 for description of variables. Numbers in parentheses are heteroscedasticity-corrected standard errors. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

Early in the data section, we performed t-tests on sample differences between SOEs and NSOEs and found that SOEs on average suffered less business reduction in the first quarter, expected better business recovery by end of June, and estimated less downward adjustment of their 2020 revenue target. In Table 3, regression (3.1) shows an estimated coefficient on $\mathrm{SOE}$ which is positive and statistically significant at the 5% level. As the regression includes industry fixed effects (20 industries), this result adds evidence supporting the observation that SOEs in our sample on average fared better than NSOEs in the pandemic time.

In regression (3.2), we add two firm characteristic variables ($\mathrm{SIZ}$ and $\mathrm{CLT}$). The results show that the estimated coefficients of the two variables have expected signs. First, the larger the firm size ($\mathrm{SIZ}$), the better the expected recovery; this is explained by the fact that larger firms tend to be more able to cope with the pandemic situation. Second, firms with other firms as clients ($\mathrm{CLT}=1$ for “B2B only”) expected better recovery than firms with (at least partly) individuals as clients ($\mathrm{CLT}=0$ for “B2C only” and “B2C&B2B”); this can be explained by the fact that the pandemic tends to make business recovery more difficult for firms serving individuals than for firms serving other firms.10 Once these firm characteristics are controlled for, we find that the estimated efficient on $\mathrm{SOE}$ becomes statistically indifferent from zero. For SOEs and NSOEs of same industry, same client type, and similar size, they expected the same level of business recovery. Thus, the higher expected recovery of SOEs shown in regression (3.1) can be explained largely by the fact that the SOEs in our sample have on average a much larger firm size ($\mathrm{SIZ}$ has mean values of 4.45 for SOEs and 3.71 for NSOEs).11

In Eq. (3.3) we include $\mathrm{GI}$ (government support to the industry reported by firm). The estimated coefficient on $\mathrm{GI}$ is positive (as expected) and is statistically significant at the 1% level. Since a firm's rating of government support to its industry was based on its own situation, inclusion of $\mathrm{GI}$ should control for part of the firm-specific government support. We find that the estimated coefficient on $\mathrm{SOE}$ remains statistically insignificant with this partial control of firm-specific government support effects.

In regression (3.4) we add HR measures taken by companies in the first quarter ($\mathrm{HR}$) as an explanatory variable. Value of $\mathrm{HR}$ is in descending order of harshness of HR measures taken: Laid off workers ($\mathrm{HR}=1$); Cut salaries across the company ($\mathrm{HR}=2$); Cut salaries of staffs only ($\mathrm{HR}=3$); Cut salaries of senior executives only ($\mathrm{HR}=4$); No salary cut ($\mathrm{HR}=5$); No layoffs ($\mathrm{HR}=6$); Raised salary or raised hiring ($\mathrm{HR}=7$).12 As shown in Table 2, there exist significant sample differences between the HR measures taken by SOEs and NSOEs. The estimated coefficient on $\mathrm{HR}$ is positive and statistically significant; firms taking harsher HR measures (lower $\mathrm{HR}$) are the ones expecting less recovery (lower $\mathrm{REC}$). Different degrees of harshness in HR measures reflected the underlying firm-specific forces including firm-government relationship. Thus inclusion of $\mathrm{HR}$ helps control these unmeasured firm-specific effects.

To mitigate the measurement errors and the potential endogeneity of $\mathrm{HR}$ as an explanatory variable, we use a two-stage estimation approach. In the first stage, we run regression (4) and the results are reported in column (4.1) of Table 4 . We find that the estimated coefficients of all the explanatory variables are statistically significant and the signs are expected.13 Regression (4.1) reports a sufficiently large $F$-statistic ($=$19.25), which suggests that $\mathrm{MCH}$ is an acceptable instrument variable (Staiger & Stock, 1997).

Table 4.

Human resources measures taken in the first quarter.

	(4.1)	(4.2)
$\mathit{SOE}$	0.595***	0.700***
	(0.114)	(0.138)
$\mathit{SIZ}$	$-$0.068**	$-$0.066**
	(0.034)	(0.039)
$\mathit{CLT}$	0.199**	0.149
	(0.097)	(0.108)
$\mathit{GI}$	0.056***	0.068***
	(0.019)	(0.021)
$\mathit{MCH}$	0.072**	0.072**
	(0.033)	(0.036)
$\mathit{FBR}$	$-$0.239***	$-$0.238***
	(0.036)	(0.039)
Constant	4.564***	4.460***
	(0.214)	(0.247)
Industry fixed effects	Yes	Yes
Observations ($N$)	988	862
F-statistic ($F$)	19.25	16.85
R-squared ($R^2$)	0.128	0.120

Notes: The dependent variable is $\mathit{HR}$. See Table 2 for description of variables. Numbers in parentheses are heteroscedasticity-corrected standard errors. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

In the second stage, we run regression (5) on expected business recovery, using the predicted value $\widehat{\mathrm{HR}}$ obtained from regression (4.1) as an explanatory variable. The results, shown in column (3.5) of Table 3, indicate a positive estimated coefficient on $\widehat{\mathrm{HR}}$ that is statistically significant at the 1% level. When $\widehat{\mathrm{HR}}$ is controlled for, the estimated coefficient on $\mathrm{SOE}$ becomes negative and statistically significant at the 1% level. Notice that the estimated coefficient on $\mathrm{GI}$ becomes statistically insignificant, suggesting that its effect has been absorbed by the estimated effect of $\widehat{\mathrm{HR}}$.14

Our main finding from the two-stage estimation is that, unlike initially appeared (in regressions (3.1) to (3.4)), SOEs had lower expected business recovery ($\mathrm{REC}$) than NSOEs once a sufficient amount of unmeasured firm-specific factors (captured by $\widehat{\mathrm{HR}}$) are controlled for. In an oversimplified scenario where government policy is the dominant underlying factor behind $\widehat{\mathrm{HR}}$, a positive estimated coefficient on $\widehat{\mathrm{HR}}$ indicates the effect of more government support to SOEs improving their expected recovery from the pandemic shock, and a negative estimated coefficient on $\mathrm{SOE}$ indicates the effect derived from SOEs’ innate behavior lowering their expected recovery from the pandemic shock. While this interpretation is admittedly an over-stretch of our estimation results, it is a useful first attempt to separate the effect of the innate behavior of SOEs from the government-support effect in empirical research that compares SOEs and NSOEs. After applying various robustness tests, we find that this result holds. Regression (3.6) shows results from dropping financial firms.15 Regressions in Table 5 show results from using $\mathrm{REV}$ (estimated adjustment of 2020 revenue target) as the dependent variable, with regression (5.6) also using only the non-financial-firm sample. In all these regressions, we find that SOEs on average had worse performance than NSOEs when both the measure of government support to industry reported by firm ($\mathrm{GI}$) and the measure of firm-level government support proxied by HR action ($\widehat{\mathrm{HR}}$) are controlled for.

Table 5.

Estimated adjustment of 2020 revenue target.

	(5.1)	(5.2)	(5.3)	(5.4)	(5.5)	(5.6)
$\mathit{SOE}$	0.387***	0.316**	0.232	0.115	$-$0.415**	$-$0.763***
	(0.142)	(0.144)	(0.152)	(0.151)	(0.172)	(0.218)
$\mathit{SIZ}$		0.090***	0.055	0.060*	0.088***	0.123***
		(0.032)	(0.034)	(0.034)	(0.033)	(0.036)
$\mathit{CLT}$		0.140	0.097	0.028	$-$0.167	$-$0.115
		(0.099)	(0.102)	(0.101)	(0.104)	(0.111)
$\mathit{GI}$			0.091***	0.080***	$-$0.001	$-$0.008
			(0.019)	(0.018)	(0.021)	(0.024)
$\mathit{HR}$				0.198***
				(0.031)
$\widehat{\mathit{HR}}$					1.216***	1.174***
					(0.138)	(0.147)
Constant	2.613***	2.204***	1.835***	1.048***	$-$2.902***	$-$2.817***
	(0.045)	(0.154)	(0.174)	(0.201)	(0.533)	(0.552)
Industry fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Observations ($N$)	1,167	1,167	1,080	1,043	986	860
R-squared ($R^2$)	0.064	0.071	0.095	0.136	0.183	0.174

Notes: The dependent variable is $\mathit{REV}$. $\widehat{\mathit{HR}}$ is the predicted value of $\mathit{HR}$ of the regressions displayed in Table 4, with (5.5) using (4.1) and (5.6) using (4.2). See Table 2 for description of the other variables. Numbers in parentheses are heteroscedasticity-corrected standard errors. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

5. Summary and conclusions

SOEs and NSOEs differ in multiple dimensions, which fall into two broad categories. The first category is firm-government relationship. In general, SOEs receive more support from the government but bear more burdens imposed by the government. The second category is firm behavior. In general, SOEs are less driven by market signals. For studies aiming at finding if SOEs and NSOEs behave differently, it is crucial to distinguish between influences from these two categories. In the literature, many studies estimated the SOE-NSOE difference using financial performance measures, and the results were largely mixed. As a company's financial performance depends both on its reaction to market signals and on its relationship with the government, it is hardly surprising to find mixed results on the SOE-NSOE difference in financial performance as the influences from the two broad categories may cancel each other, yielding a net effect that is negative, zero or positive, depending on the given scenario.

Firm-level government policy data is hard to come by. Without controlling for firm-specific government support, one cannot interpret the observed/estimated SOE-NSOE performance difference as necessarily reflecting their ownership-based behavioral difference. This paper is an attempt to tackle this issue. We use data from an online survey of 1,182 company executives in China conducted in early April 2020. It is widely known that China was most severely hit by the COVID-19 shock in the first quarter of 2020. The shock was sudden and massive, causing companies in China to take measures that would not be observed in normal times. We believe that the drastic measures taken by companies in response to the COVID-19 shock contain information on underlying firm-specific factors including the influence of government policies. Accordingly we construct a proxy for firm-level government support using firm's human resources (HR) action taken during the outbreak and using firm's 2019 China revenue share as an instrument variable for the HR action. We apply a two-stage estimation approach to first extract information from the HR action data, and then utilize it in estimating the SOE-NSOE difference associated with their ownership-based behavioral difference.

The main finding of this paper is that, despite the seemingly superior performance of SOEs over NSOEs as shown in the raw data as well as in some preliminary regressions, SOEs in our sample are found to perform significantly worse than NSOEs once a sufficient amount of firm-level differences in government support is controlled for. Specifically, after considering the effect of government support to industry (reported by firm in the survey) and the effect of unobserved firm-specific government support (proxied by firm's HR action taken at the outbreak of COVID-19), SOEs in our sample are found to perform significantly worse than NSOEs in the pandemic period. This result is found robust when the estimation is applied to the subsample of non-financial firms and when alternative firm performance measures are used.

Our study is limited in several aspects. First, with survey data, the variables are ordinal and measured with integer values; thus they are rather crude compared with continuous variables. Second, despite the care taken in constructing the proxy for unobserved firm-specific government policy factors, we cannot be certain if it captures the underlying firm-government relationship that is crucial for the identification of the ownership-based firm behavior effect; correlation between the SOE dummy variable and the proxy variable may contaminate the identification of their distinctive effects. Last but not least, limited by the small number of questions and the anonymity nature of the survey, our estimation is subject to the omitted variable issue. Given these limitations, our results should be interpreted with caution. Nevertheless, our study is useful in drawing the distinction between firm performance derived from firm-government relationship and firm performance derived from ownership-based firm behavior, which sets the right direction for future research.

Conflict of interest

The authors declare that there is no conflict of interest.