Constructing alternative unemployment statistics in China

Abstract

This paper presents alternative unemployment statistics for China, encompassing the extended unemployment rate (U5), which incorporates marginally attached workers, and the nonemployment index (NEI) proposed in recent literature (Hornstein et al. in Econ Q 100(1):1–21). The analysis spans from 1988 to 2021, merging recent data (2012–2018 and 2018–2021) with early estimates (1988–2009). Our findings indicate that both standard and alternative unemployment statistics reveal similar trends in China: a rise in the labor force participation rate and a decline in the unemployment rate since 2012, marking a reversal from the trend before 2009. Additionally, unemployment measures in China align more closely with those of affluent countries from 2007 to 2018 and with less affluent countries from 2018 onwards. Moreover, U5 demonstrates enhanced predictive capabilities for wage dynamics in China’s labor market during the recent 2012–2018 period. Lastly, we delve into the discussion of errors in labor market status data.

1 Introduction

Unemployment statistics are vital economic indicators that garner close attention. However, they have consistently sparked controversy in China.¹ The primary focus is on the limitations of the official register-based unemployment rate published by China’s National Bureau of Statistics (NBS), which only includes individuals registered with public employment agencies as unemployed. Since the registration rate is not one hundred percent,² this statistic is considered inadequate for accurately reflecting the labor market’s actual condition.³

Several early initiatives have aimed to enhance unemployment statistics in China. These include adjusting the official register-based unemployment rate upward by incorporating supplementary data on the register rate, or recalculating the unemployment rate using representative surveys (Giles et al. 2005; Knight and Xue 2006; Feng et al. 2017).⁴ Furthermore, the NBS began to emulate the prevailing international standard practice by publishing a survey-based unemployment rate monthly for urban areas, utilizing their monthly Urban Household Survey (UHS) data, commencing in 2018.

While prior efforts have made progress in aligning China’s unemployment statistics with standard definitions and practices, they fail to offer alternative statistics for much of the period post-2010, leaving our understanding reliant on unreliable official data. Moreover, recent advancements have adhered to the standard definition of unemployment, overlooking marginally attached workers and individuals conventionally categorized as “out of the labor force.” However, emerging evidence from other countries suggests that those “out of the labor force” may be significant, as they often have a substantial chance of transitioning to employment. Accounting for the employability of these individuals could lead to a more accurate measurement of total labor resource utilization and the health of the labor market (Jones et al. 2003; Schweitzer 2003; Hornstein et al. 2014). Nonetheless, the significance of marginally attached workers and those “out of the labor force” in shaping labor statistics in China remains poorly understood.

Hence, this paper seeks to enhance China’s unemployment statistics by leveraging the comprehensive labor market data available in the recent China Family Panel Studies (CFPS) spanning 2012–2018. These datasets offer crucial insights into marginally attached workers and individuals categorized as “out of the labor force.”

Firstly, we construct standard unemployment and labor force participation statistics, comparing them with NBS’s official statistics, data from other countries during the same period, and historical statistics from early periods in China. Our analysis reveals an upward trend in labor force participation and a declining trend in the standard unemployment rate from 2012 to 2018, marking a reversal from the previous trend observed between 1988 and 2009. China’s labor force participation rate and its growth during this period notably exceed those of other countries. Notably, China’s unemployment rate has remained lower than the global average since 2014, with a more significant decline compared to other countries. Moreover, the trajectory of China’s labor statistics aligns more closely with that of developed nations from 2007 to 2018, but exhibits greater similarity to less affluent nations since 2018.

Secondly, we utilize data on marginally attached workers to develop extended unemployment statistics comparable to the U5 rate in the USA. Additionally, we construct the nonemployment index (NEI) proposed by Hornstein et al. (2014), which incorporates individuals “out of the labor force” (O) into unemployment calculations, assigning a weight based on O individuals’ probability of transitioning to employment relative to standard unemployed individuals. This marks the inaugural effort to compile such statistics for China. Our analysis reveals that the U5 rate exceeds the standard unemployment rate (U3) in China by 1–4 percentage points while maintaining the trend observed from 2012 to 2018. Both the NEI and standard unemployment statistics indicate similar trends in labor force participation. Moreover, the U5 rate demonstrates superior predictive capability for wage and GDP dynamics, particularly wage growth, in China. Despite the exploratory nature of this study and existing data limitations, we believe the findings are informative and offer encouragement for future research.

The remainder of the paper proceeds as follows. Section 2 reviews the developments and limitations of China’s unemployment statistics. Section 3 introduces the data. Section 4 explains the statistics to be constructed. Section 5 reports the results. Section 6 discusses. Section 7 concludes.

2 Overview

2.1 Development in unemployment statistical work in China

In China, employment was formerly centrally planned by the government.⁵ In rural areas, laborers were organized into production teams, while in urban areas, all workers were employed by state-owned enterprises (SOEs). This system persisted until the mid-1990s, during which SOEs faced strict limitations on dismissing redundant workers. Only a few unemployed individuals were officially registered by the government, if any, and received unemployment stipends. Thus, during this period, unemployment statistics primarily served the purpose of administering unemployment welfare and only accounted for registered unemployed workers. However, starting in 1995, the government could no longer sustain meaningful employment and decent wages for all SOE workers. Under budgetary pressure, authorities gradually permitted SOEs to lay off workers. Between 1995 and 2001, approximately 34 million SOE workers were laid off (Giles et al. 2006).

During this period, the official statistics bureau continued the practice of only counting unemployed workers registered with the government. However, this approach had a limitation: not all unemployed individuals were registered. The under-coverage stemmed from two primary reasons. Firstly, some individuals lacked the necessary identification papers or employment history qualifications required for registration. Secondly, others chose not to register due to either the low unemployment benefits or the social stigma associated with being labeled as unemployed. These issues have been extensively discussed in the literature (Giles et al. 2005; Feng et al. 2017).

Due to this limitation, the official register-based unemployment rate remains consistently low, offering limited insight into China’s evolving economy (Cai et al. 2013; Feng et al. 2017). Consequently, the literature has explored various methods to enhance the accuracy of official unemployment statistics. Some scholars debate whether redundant and laid-off SOE workers should be included in unemployment figures (Solinger 2001, 2002). Others advocate for the incorporation of additional data, such as nationally representative labor surveys, to compute the unemployment rate based on the international standard definition (Giles et al. 2005; Knight and Xue 2006; Feng et al. 2017). Previous research has yielded more standardized estimates of the unemployment rate from the late 1980s to 2009.

However, there is a lack of alternative unemployment estimates for the period after 2010, leaving the official register-based unemployment rate as the sole available statistic. This limitation may impede our comprehension of the contemporary economy. Since 2012, the government has implemented significant labor policies. Notably, the Ministry of Human Resources and Social Security of the People’s Republic of China launched the Plans to Promote Employment (2011–2015) as a key policy initiative. These plans outlined various strategies to boost employment, including fostering entrepreneurship, investing in marketable human capital, and enhancing labor market infrastructure, institutions, and services.⁶ These policies are expected to induce changes in the labor market. However, the persistently stagnant official register-based unemployment rates and the absence of alternative statistics would hinder our ability to comprehend the shifts occurring in the labor market during this period.

Hence, the primary objective of this paper is to offer alternative unemployment statistics for the period after 2010. The initial alternative entails computing the unemployment rate based on the standard definition of unemployment. This standardized unemployment rate would enhance the inclusion of unemployed workers, thereby rendering it more informative than the official register-based unemployment statistic.

Since 2018, the official statistics bureau has finally commenced publishing unemployment rates (survey-based unemployment rate) derived from representative surveys that adhere to the international standard definition of unemployment. Although this marks a welcome improvement, it is essential to acknowledge the limitations inherent in the standard unemployment definition, which may constrain its informativeness. Therefore, it may be beneficial to further explore alternative statistics. Here, we outline the primary endeavors undertaken to explore alternative labor statistics in both the literature and practice.

2.2 Explorations for alternative labor statistics

The standard international practice involves querying respondents about their search activity during a specific reference period to ascertain unemployment. Debates frequently arise concerning the definition of search activity and the selection of the reference period. Firstly, what constitutes search activity? While the common practice is to consider only “positive search” activities, such as responding to job advertisements by submitting resumes, “passive search” activities, such as browsing job advertisements or even “waiting,” could also result in employment (Hall 1983; Coles and Smith 1998; Riddell 2000). Consequently, excluding “passive search” from the definition of job search, if job search is defined as an activity potentially leading to a job offer, may overlook significant labor market activities.

Secondly, what should be the duration of the reference period? A common reference period is one month. However, there is debate regarding the justification for this one-month duration. Studies have revealed that individuals’ labor force status can change rapidly within weeks (Jones and Riddell 1999). Moreover, individuals on the cusp of the four-week reference period may exhibit similar transition probabilities to employment as the unemployed, suggesting comparable search behaviors (Brandolini et al. 2006). Additionally, empirical evidence indicates substantial heterogeneity in labor market dynamics within the “nonemployed” category, which encompasses individuals not currently working, including both the conventionally defined “unemployed” and those classified as “out of the labor force” (Clark and Summers 1982; Ellwood 1982; Flinn and Heckman 1983; Tano 1991; Gönül 1992). Neglecting individuals categorized as “out of the labor force” may therefore overlook crucial labor market insights.

Researchers and statistical bureaus are actively exploring alternative unemployment statistics. For instance, the US Bureau of Labor Statistics (BLS) has consistently reported the extended unemployment rate (U5) alongside the standard unemployment rate (U3). The U5 rate broadens the definition of unemployment to include marginally attached individuals (those discouraged or for other reasons). Some scholars suggest estimating individuals’ unobserved labor market attachment by assessing their likelihood of transitioning to employment (Brandolini et al. 2006; Jones and Riddell 1999, 2006, 2019). In line with this approach, Hornstein et al. (2014) develop a nonemployment index (NEI), wherein individuals categorized as “out of the labor force” (O) are weighted based on their probability of transitioning to employment relative to that of standard unemployed individuals (U). This adjustment for attachment status within the O group bears resemblance to quality adjustments made for employment, wherein relative wages serve as proxies for relative labor efficiency.⁷ The NEI has the advantage of providing information about the level of labor market attachment compared to traditional unemployment measures, in addition to indicating the size of the nonemployed population relative to the total population. Therefore, this metric offers insights into the extent of idleness within the population and may offer a more comprehensive assessment of overall labor resource utilization and labor market health (Jones et al. 2003; Schweitzer 2003; Hornstein et al. 2014). Currently, the Federal Reserve Bank of Richmond publishes the NEI for the USA on a monthly basis.

2.3 Directions for improvement in China’s unemployment statistics

At present, there are no alternative unemployment statistics accessible for China, potentially impeding our comprehension of the country’s labor market dynamics. For instance, marginally attached workers might engage with the labor market, offering insights into market tightness and structure. Similarly, individuals categorized as “out of the labor force” could signify untapped productive potential based on their willingness to participate in the workforce. Developing and juxtaposing alternative unemployment metrics to encompass various categories of “non-conventional” unemployed individuals would enhance our insights into the labor market and the broader economy.

Constructing alternative unemployment measures, however, necessitates detailed information regarding the reasons people are classified as “out of the labor force.” In this study, we delve into a recent dataset that provides an opportunity to formulate and compare alternative unemployment and nonemployment statistics for China.

3 Data

In this study, we primarily utilize recent data from the China Family Panel Studies (CFPS), spanning the period 2012–2018. The CFPS is a biannual representative household survey initiated in 2010 by the Institute of Social Science Studies at Peking University (also known as Beijing University). It comprises panel data and employs a nationally representative stratified design. The CFPS samples encompass 25 provinces, representing 94.5% of China’s population.⁸ Each wave of the survey incorporates new households into the sample,⁹ and the dataset includes cross-sectional and panel weights to ensure the sample accurately reflects the population it represents.¹⁰

We concentrate on the urban samples since the urban areas encompass the primary labor market. Rural areas primarily engage in agricultural work, subject to distinct market mechanisms. This selection criterion is common in the literature. We limit the sample to individuals aged 16–60 to accurately represent the working-age population, as 16 is the legal minimum working age and 60 is the upper limit for workforce participation.¹¹ Alternative sample selection criteria, such as excluding women aged above 55 and/or individuals at the exact retirement age, were considered. Our findings remained consistent regardless of these variations in sample selection.

Fig. 1

Labor force status questions in CFPS. Sources: CFPS 2012–2018

The CFPS surveys feature a comprehensive array of questions designed to ascertain the labor force status of the sample (Fig. 1). The initial query is: “Have you worked for at least one hour last week?” In accordance with standard practice, CFPS categorizes respondents who answer “Yes” to this question as “employed” (E). If the response is “No,” but the respondent affirms “Yes” to any of the subsequent follow-up questions¹² (“Do you have a job, but are currently on temporary vacation, sick leave, or other leave, or undergoing on-the-job training?”/“Will you return to the original job position within a certain period or within six months?”/“Are you self-employed, with your business currently in an off-season, but set to resume later?”/“Is your agricultural work (including crop cultivation, orchard management, harvesting, forestry, fish farming, fishing, livestock raising, or selling agricultural products in markets, etc.) experiencing an off-season?”), the survey also classifies the respondent as “employed.”

If a respondent answers “No” to all of the preceding questions,¹³ the subsequent inquiry is: “Did you look for a job in the past month?” (Search, or S), followed by: “Can you start working in two weeks if you receive a job offer?” (Ready, or R). According to convention, CFPS categorizes a respondent who responds “No” to either question as “out of the labor force” (O), and a respondent who answers “Yes” to both questions as unemployed (U).

For respondents who answer “No” to both the S and R questions (NS-NR), CFPS records their response regarding the reason for unemployment.¹⁴ Respondents are prompted to select from eight options: “Do not need/want to work” (NoNeed/Want, or NN/W), “Do housework” (Homemaker, or H), “Retired” (Retiree, or RT), “Too old and physically feeble (primarily referring to elderly farmers)” (Old/PhysicallyFeeble, or OPF), “No working capacity due to disability/illness” (Disabled/Ill, or DI), “Cannot find a suitable job” (Discouraged, or DG), “Attending school or training” (Student/Trainee, or S/T), and “Other” (Other, or OT). The distinction between RT and OPF primarily concerns occupation and is associated with the diversity of social welfare systems: the latter predominantly pertains to farmers excluded from mainstream social welfare systems.

Table 1 presents the summary statistics of the percentage of the working-age population in each labor force status. It reveals that approximately 80% of the working-age population is classified as “employed.” Only 2% of the working-age population falls into the “search and ready to work” (S-R) status, representing the conventional “unemployment” category. Around 20% of the working-age population is categorized as conventional O status. Within the O category, there is a small subset (less than 0.4%) of “search but not ready to work” (S-NR) individuals, a larger group (4–6%) of “not search but ready to work” (NS-R) individuals, and the largest group (10–14%) of “not search and not ready to work” (NS-NR) individuals, the majority of whom are homemakers, disabled/ill, old/physically feeble, or retirees.¹⁵

Table 1 Percent of the working-age population in each labor market status

In addition, we occasionally utilize various supplementary data sources to construct series for comparative analysis. These additional sources encompass the official Chinese statistical yearbooks, data from the World Bank, and information from labor statistics bureaus in the USA. Below, we will provide detailed explanations of these data sources when necessary.

4 Methodology

4.1 Correcting measurement errors

Note that errors in the labor force states reported by CFPS interviewees may exist. It is reasonable to suspect that some individuals classified as unemployed or out of the labor force might falsely claim to be employed to evade the potential social stigma linked with unemployment. To explore this potential issue, we employ the approach outlined by Feng and Hu (2013) to rectify misreporting errors. This method involves estimating the following misreporting matrix

$$\begin{aligned} M_{S_{t}|S_{t}^{*}}\equiv [\Pr \left( S_{t}=i|S_{t}^{*}=j\right) ]_{i,j}, \end{aligned}$$

(1)

where $S_{t}$ denotes the self-reported labor force status at time t, $S_{t}^{*}$ denotes the true (latent) labor force status, $i,j\in \left\{ E=1,U=2,O=3\right\}$ denotes the ith row and jth column. Feng and Hu (2013) show how to estimate $M_{S_{t}|S_{t}^{*}}$ under the assumption of Markov transitions and regularity conditions using panel data.¹⁶ Then we can correct for misreporting error as:

$$\begin{aligned} \left[ \begin{array}{c} \Pr \left( S_{t}^{*}=E\right) \\ \Pr \left( S_{t}^{*}=U\right) \\ \Pr \left( S_{t}^{*}=O\right) \end{array}\right] =M_{S_{t}|S_{t}^{*}}^{-1}\times \left[ \begin{array}{c} \Pr \left( S_{t}=E\right) \\ \Pr \left( S_{t}=U\right) \\ \Pr \left( S_{t}=O\right) \end{array}\right] . \end{aligned}$$

(2)

We may need to adjust for reporting errors within the O subgroups. While theoretically, we could extend the methodology by subdividing the O group, practical constraints arise due to data limitations. The small sample size of O subgroups results in a matrix with numerous rows and columns of zero, rendering it non-invertible and precluding the use of the aforementioned methodology. Consequently, we pragmatically assume that the misreporting probabilities of all O subgroups are identical to those of the overall O group and proceed accordingly. That is, we assume that, for the misreporting matrix similar to Eq. (1), $\Pr \left( S_{t}=i|S_{t}^{*}=O_{k}\right) =\Pr \left( S_{t}=i|S_{t}^{*}=O\right)$ and $\Pr \left( S_{t}=O_{k}|S_{t}^{*}=j\right) =\Pr \left( S_{t}=O|S_{t}^{*}=j\right)$ for $i,j=\{E,U\}$, then correct misreporting errors using the counterpart to Eq. (2).

4.2 Statistics to be constructed

4.2.1 Standard unemployment rate and labor force participation rate (LFPR)

We begin by constructing two primary statistics: the standard unemployment rate (u) and the labor force participation rate (LFPR). Following the conventional definition, in CFPS, the standard unemployment rate is calculated as $U/(U+E)$, while the standard LFPR is computed as $(U+E)/(U+E+O)$.¹⁷

The official unemployment statistics released by the NBS employ the same formula, although there are disparities in the coverage of U. In the register-based unemployment rate, U encompasses solely individuals who register at public employment agencies, disregarding additional unemployed workers who are unregistered. Conversely, in the survey-based unemployment rate introduced since 2018, the scope of U was established through representative surveys and adheres to the standard definition.

Regarding LFPR, the NBS does not furnish the official version of this statistic.¹⁸ Therefore, we derive it here solely for comparison purposes, utilizing officially published information: (register-based “official”) LFPR = number of register-based unemployed individuals (aged 16 to retirement age) / register-based unemployment rate / N, where N represents the count of urban residents aged 15–59 (estimated using statistics on the overall population distribution across 5-year age brackets and the urban population size¹⁹).²⁰

From 2018 onwards, we can also calculate the survey-based “official” LFPR using the formula $\left( \frac{E}{1-u}\right) /N$, where we utilize the survey-based official unemployment rate for u and the total number of the working population for E.²¹ These data are sourced from the official statistics yearbooks published by the NBS.

4.2.2 Extended unemployment and labor force participation statistics

Using data from CFPS regarding the reasons for “not working,” we define two groups: the “discouraged” (DG) group and the “other reason” (OT) group. By incorporating DG and OT into unemployment, we formulate China’s variation of U5 as $U5=(U+\textrm{DG}+\textrm{OT})/(U+\textrm{DG}+\textrm{OT}+E)$. Consequently, we derive the extended LFPR (“LFPR5”) as $\textrm{LFPR}5=(U+\textrm{DG}+\textrm{OT}+E)/(U+E+O)$.²²

4.2.3 Nonemployment statistics

Hornstein et al. (2014) propose the following nonemployment index (NEI):

$$\begin{aligned} \textrm{NEI}=\Pr \left( S=U\right) +\frac{\bar{p_{O,E}}}{\bar{p_{U,E}}}\Pr \left( S=O\right) \end{aligned}$$

(3)

where $\Pr \left( S=U\right)$ ($\Pr \left( S=O\right)$) denotes the fraction of the population that is in status U (O), and $\bar{p_{U,E}}$ ($\bar{p_{O,E}}$) denotes the average transition probability from state U (O) to E in a reference period. Hornstein et al. (2014) argue that averages are utilized to mitigate cyclical fluctuations in relative transition probabilities, thereby accentuating shifts in the size of the nonemployed population.

We construct the NEI using CFPS following Eq. (3). Additionally, we can generate two statistics, consistent with the NEI approach, which incorporate weighted O into unemployment: the O-weighted unemployment rate and the LFPR that incorporates weighted O into unemployment.

We can demonstrate the construction of the “official” version of these statistics using data officially published by the NBS for comparative analysis. Specifically, we can estimate $\Pr (S=U)=E\times u/(1-u)/N$ and $\Pr (S=O)=1-E/(1-u)/N$, where u, E and N represent the unemployment rate, the number of employed individuals, and the total working-age population, respectively.²³ We use the official survey-based unemployment rate for u, and E and N are the same as in Sect. 4.2.1.²⁴ We utilize these estimates to generate the “official” statistics, incorporating the weighted O, which encompasses the NEI, as well as the O-weighted unemployment rate and LFPR, where the weighted O is integrated with U.

To construct the NEI for China, we need to ascertain the transition probabilities $p_{U,E}$ and $p_{O,E}$. However, such data are scarce in China. An exception is Feng and Guo (2021), who estimate these probabilities using UHS data from 2003–2006, focusing on a sample of males aged 25–54 with local urban hukou (household registration) and not in the government sector. While we could estimate these probabilities for a two-year period using CFPS, our unemployment statistics are based on a one-month frame, making it technically inappropriate to directly use our estimates for NEI construction. Therefore, we opt to utilize the estimates from Feng and Guo (2021), who indicate that for the 2003–2006 period, $p_{O,E}$ is 0.3 times $p_{U,E}$ for the one-month frame. Hence, throughout our analysis, we consider 1 O as 0.3 U in constructing the NEI for China. That is, $\bar{p_{O,E}}/\bar{p_{U,E}}=0.3$ in Eq. (3). This practice is in line with Hornstein et al. (2014) who construct their NEI for the USA using the sample average $\bar{p_{O,E}}/\bar{p_{U,E}}$.

4.2.4 Comparing statistics

We begin by comparing the magnitudes of alternative statistics for the same year to depict the labor market’s state from varied perspectives. It is important to note that, according to its definition, the standard unemployment rate u can be expressed as $\frac{\Pr (S=U)}{1-\Pr \left( S=O\right) }$. Consequently, NEI can be reformulated as $\textrm{NEI}=u+\left( \frac{\bar{p_{OE}}}{\bar{p_{UE}}}-u\right) \Pr \left( S=O\right)$. Thus, whether NEI exceeds or falls below u hinges on the magnitude of the relative transition probability $\frac{\bar{p_{OE}}}{\bar{p_{UE}}}$: if the transition probability of O relative to U is high, NEI will consistently surpass u.

When comparing the O-weighted and alternative unemployment rates, it is important to recognize that, by design, the O-weighted unemployment rate will exceed the standard unemployment rate. Whether the O-weighted unemployment rate surpasses or falls short of the extended unemployment rate, which incorporates some O (considered equivalent to U), hinges on the composition of O and the relative transition probability to employment.

We proceed by comparing the trends indicated by alternative statistics and assessing their magnitudes relative to those of other countries. This trend analysis aims to validate the reliability of our understanding derived from alternative statistics. Additionally, the cross-country comparison is intended to position China within the global labor market landscape. Such an exercise facilitates a deeper comprehension of China’s labor market dynamics compared to its historical performance and that of other nations.

Finally, we conduct a comparative analysis of the statistics concerning their correlation with significant economic indicators. One such indicator is wages, where the correlation between unemployment (nonemployment) statistics and wages provides insights into labor market tightness and mechanisms. Another crucial indicator is GDP, where the correlation elucidates the efficiency of labor resource utilization in production. To gauge the degree of correlation, which reflects the informational significance of a statistic, we assess the goodness of fit of each statistic in predicting the desired economic outcomes. Specifically, we regress the outcome variable against each statistic, incorporating a constant and a linear trend, and compare the predicted outcomes with the actual ones to discern which statistic better identifies periods of accelerated growth in the outcome variable.

5 Results

5.1 Standard unemployment statistics in China

Table 2(A1) presents the standard unemployment rate (u) and labor force participation rate (LFPR) statistics derived from the CFPS samples. We utilize the cross section weights provided by CFPS to account for the sampling design. The findings indicate a notable increase in the labor force participation rate, climbing from 79.9 to 85%, while concurrently witnessing a decrease in the unemployment rate, declining from 3.5 to 1.8% over the period spanning 2012 to 2018 in China.

How do these findings align with the official statistics published by the NBS? Table 2(B1) presents the NBS register-based unemployment rate for the same period. The data reveals a marginal decrease in the NBS register-based unemployment rate from 4.1 to 3.8%. Notably, these figures exhibit minimal variation compared to those obtained from CFPS data (Fig. 2A1). This echoes prior observations in the literature, indicating that the NBS register-based unemployment rate displays limited movement in response to economic cycles.

Table 2(B1) presents the register-based “official” LFPR (labor force participation rate), derived from NBS-published data as elucidated in Sect. 4.2.1. It marginally increases from 44.8 to 45.4% during 2012–2018. Similar to the official register-based unemployment rate, these figures exhibit minimal deviation (Fig. 2B1).

Table 2 Unemployment rate, nonemployment rate, and labor force participation rate in urban China vs. USA, 2012–2018 (in percent)

In the literature, it is widely acknowledged that the NBS register-based unemployment rate in China tends to underestimate actual unemployment levels. Therefore, it is surprising to find that the unemployment rate derived from CFPS data is even lower than the NBS register-based unemployment rate. This discrepancy could stem from two potential factors. First, CFPS might not adequately represent the population, thus impacting its suitability for analyzing unemployment in China. Alternatively, discrepancies could arise from errors in the labor force status reported by CFPS respondents. Given CFPS’s widespread usage and acceptance of its representativeness, we concentrate on the latter explanation: reporting errors. We present the estimated misreporting probabilities matrix in Table 4 in the Appendix, following the methodology outlined in Sect. 4.1. To account for attrition, CFPS panel weights are applied, and the results represent averages over the sample period.

The findings reveal a noteworthy trend: 13% of individuals categorized as “out of the labor force” erroneously report themselves as employed, while a substantial 75.9% of those classified as unemployed similarly assert employment. This discrepancy sheds light on the seemingly low unemployment rate observed in CFPS. Table 2(A2) presents the misreporting error-corrected data from CFPS. Correcting for misreporting significantly elevates the CFPS unemployment rate prior to 2014 and slightly elevates it post-2014. Misreporting leads to an underestimation of the standard unemployment rate by 1.5–6.7 percentage points in CFPS. Conversely, corrected LFPR figures remain comparable to the uncorrected data, primarily due to the relatively limited misreporting of “out of the labor force” individuals as employed. The corrected results indicate a decline in the unemployment rate from 10.2 to 3.3% and an increase in LFPR from 79.6 to 86% during the 2012–2018 period. These trends are depicted in Fig. 2A1, B1, respectively.²⁵

Since 2018, the NBS initiated monthly publications of the survey-based unemployment rate. How does this rate compare to estimates from CFPS? Table 2(B2) reveals that the average NBS survey-based unemployment rate for 2018 stands at 4.9%, exceeding the CFPS error-corrected unemployment rate for the same year. One potential explanation is that the official survey-based unemployment rate relies on the Urban Household Survey (UHS) of the NBS, which lacks a clearly defined reference week for labor force status in a given month (Feng et al. 2017). Consequently, it may encompass more individuals in the unemployment category than permitted by CFPS’s narrower criteria, which entail a “four-week reference period for job search and two-week reference period for readiness to work.” This disparity is depicted in Fig. 2A1.

Table 2(B2) presents the survey-based “official” LFPR for 2018, calculated using the NBS survey-based unemployment rate and other NBS-published data, as detailed in Sect. 4.2.1. The estimate stands at 82.6%, which closely aligns with our CFPS estimate of 86%. This result is depicted in Fig. 2B1.

Fig. 2

Statistics of labor resource utilization, urban China vs. USA and world, 2012–2018. Sources: CFPS 2012–2018 and statistics published by NBS for Panel (A1) and (B1); data published by the Federal Reserve Bank of Richmond for Panel (A2) and (B2); World Bank for Panel (A3) and (B3). Notes U3: standard definition of unemployment. U5: U3 plus marginally attached (discouraged and other reasons). U_wO: U3 plus weighted O. NEI: nonemployment rate. U5_1: U5 plus S-NR (search but not ready). U_reg_official: official register-based unemployment rate. U_survey_official: official survey-based unemployment rate. Misreporting errors in CFPS are corrected. “highinc”: high-income countries. “uppmid”: upper-middle-income countries. “lowmid”: lower-middle-income countries. “lowinc”: low-income countries

5.2 Extended unemployment statistics in China

Table 2(A2) presents the extended unemployment rate, accounting for the DG and OT groups after correcting for reporting errors. In 2012, the extended unemployment rate exceeds the standard unemployment rate by over 4 percentage points, while for subsequent years, it remains 1–1.5 percentage points higher than the standard unemployment rate (Fig. 2A1).

Note that the extended unemployment rate for 2018 stands at 4.5%, aligning more closely with the average official survey-based unemployment rate published by the NBS. Table 2(A2) presents an additional iteration of the extended unemployment rate, encompassing not only the DG and OT groups but also the S-NR (“search but not ready”) group. Remarkably, this version of the extended unemployment rate matches precisely with the average official survey-based unemployment rate for 2018. This observation implies that the NBS official survey-based unemployment rate is better aligned with the extended unemployment rate than the standard unemployment rate.

Table 2(A2) also presents the estimated LFPR using the extended definition of unemployment, which encompasses DG, OT, and S-NR. The extension results in LFPR estimates surpassing 80% throughout all the studied years, altering the LFPR trend somewhat. While the standard definition of unemployment indicates an increase in LFPR during the period 2012–2018, the extended definition suggests a decline initially before eventually rising (Fig. 2B1).

The extended unemployment rate indicates that disregarding marginally attached workers would underestimate the unemployment issue in China. However, it still overlooks other categories of individuals classified as “out of the labor force.” Hereafter, we will devise a nonemployment statistic that incorporates all forms of individuals categorized as “out of the labor force” in a weighted fashion.

5.3 Constructing the nonemployment statistics in China

Table 2(A3) and (B3) presents, while Fig. 2A1, B1 illustrates, three statistics incorporating weighted O into unemployment. The initial statistic is the unemployment rate inclusive of weighted O. Figure 2A1 indicates that the O-weighted unemployment rate surpasses the two alternative rates but shares the same qualitative trend as the standard and extended U5-like unemployment rates, all depicting a decrease from 2012 to 2018.

The second statistic is the labor force participation rate incorporating weighted O. Integrating weighted O boosts the labor force participation rate by 4.2–6.1 percentage points compared to the standard definition of unemployment, while maintaining an upward trend throughout the studied period (Fig. 2B1).

The third statistic is the NEI, which gauges the proportion of the labor force equivalent to standard unemployed job seekers, illustrating the extent of idleness within the population (Hornstein et al. 2014). The findings indicate that approximately 14.3% of China’s working-age population was idle in 2012. However, this nonemployment rate gradually decreased to 7% by 2018, with an annual decline of $-$ 1.2 percentage points during this period. This trend aligns with the decrease in the standard unemployment rate and the increase in the standard labor participation rate (Fig. 2A1, B1).

Table 2(B3) indicates that in 2018, the “official” O-weighted unemployment rate stands at 10.6%, marking a difference of 2.8 percentage points compared to our CFPS-based estimate. Similarly, the “official” O-weighted labor force participation rate is reported as 87.8%, reflecting a 2.4 percentage point downward variance from our CFPS-based estimate. Additionally, the “official” nonemployment rate is recorded at 9.3%, surpassing our CFPS-based estimate by 2.3 percentage points.

To assess the sensitivity of NEI to the estimate of $p_{O,E}/p_{U,E}$, we also present NEI under varying values of $p_{O,E}/p_{U,E}$. By design, the nonemployment rate that assigns weight to O relative to U falls between the simple U/population ratio (equivalent to weighting 1 $O = 0 U$) and the ratio of the sum of U and O to the population, ($U+O$)/Population (equivalent to weighting 1 $O = 1 U$). The closer the transition probability to employment for O is to U, the more the nonemployment rate will resemble the ($U+O$)/Population ratio. Table 2(A3) presents NEI under weights 1 $O = 0 U$ and 1 $O = 1 U$, respectively, representing the lower and upper bounds of NEI. Although the magnitudes of NEI differ, the implied trends are similar.

6 Discussion

6.1 Historical comparison

How do our estimates compare with previous estimates in China? Figure 3 illustrates our estimates of the standard unemployment rate and LFPR alongside those calculated by Feng et al. (2017) for the period 1988–2009 using UHS data. The findings indicate that, although the standard unemployment rate experiences a slight increase between 2009 and 2012, it subsequently decreases to a level comparable to that of the 1990s before the peak observed around 2000.²⁶ Our error-corrected estimate for 2012 (10.2%) aligns with (though exceeds) the average unemployment rate (9.5%) recorded for the period 2002–2009, whereas our error-corrected estimate for 2018 (3.3%) resembles (but is lower than) the average unemployment rate (3.7%) for the period 1988–1995. The average annual change in the 2012–2018 period is $-$ 1.2 percentage points, contrasting with the 0.9 percentage point change observed during the 1995–2002 period.

In terms of the standard LFPR, our error-corrected LFPR for 2012 (79.6%) stands approximately 5 percentage points higher than the average (74.2%) observed during the 2002–2009 period. Similarly, our error-corrected LFPR for 2018 (86%) surpasses the average (82.2%) recorded during 1988–1995 by about 5 percentage points. The average annual change over the 2012–2018 period amounts to 1.1 percentage points, contrasting with a decrease of $-$ 0.8 percentage point observed during 1995–2002. Overall, both the standard unemployment rate and LFPR using the standard definition of unemployment in China exhibit a reversal of the trend seen in the earlier period from 1988 to 2009.

Similarly to the standard unemployment rate and LFPR, both the extended unemployment rate and the NEI exhibit a decline during the 2012–2018 period, while the LFPR under the extended definition of unemployment demonstrates a rise. Additionally, we extend the NEI series back to 1988 utilizing the labor force status estimates provided by Feng et al. (2017), with a weighting of $\bar{p_{O,E}}/\bar{p_{U,E}}=0.3$. This series is depicted in Fig. 3A, revealing that the NEI consistently surpasses the standard unemployment rate.²⁷ Like the standard unemployment rate, the NEI remains elevated post-2000 and has experienced a decline since 2012.

Fig. 3

Long-term trend of labor utilization in China. Sources: CFPS 2012–2018 and UHS 1988–2009 (estimates reported by Feng et al. 2017). US series is from Federal Reserve Bank of Richmond; Series for the rest of the world are from the World Bank. Notes For details of the UHS data, see Feng et al. (2017). LFPR: labor force participation rate. “highinc”: high-income countries. “uppmid”: upper-middle-income countries. “lowmid”: lower-middle-income countries. “lowinc”: low-income countries

6.2 Cross-country comparison

6.2.1 Standard labor statistics

How do China’s labor statistics compare with those of other countries? We begin by comparing China’s standard labor statistics with those of the USA, which not only boasts the largest economy globally but also possesses arguably the most advanced labor statistics practice. Table 2(C) presents the annual average of the US’ standard unemployment rate (U3) and LFPR using the standard unemployment definition for the same period, allowing for direct comparison. These data points are depicted in Fig. 2A2, B2, respectively. Our estimation of the standard unemployment rate surpasses the US’ U3 rate by approximately 2 percentage points in both 2012 and 2014 but declines to half to one percentage point below it in 2016 and 2018. The LFPR with the standard unemployment definition in the USA hovers around 63% between 2012 and 2018, significantly lower than that in China during the same period, and experiences a marginal decrease of approximately 1 percentage point.

Figure 2A3, B3 also depicts the standard unemployment rate and LFPR of the rest of the world for comparison.²⁸ The data reveal that China’s standard unemployment rates in 2012 and 2014 surpass not only those (U3 rates) of the USA but also those of other countries. However, in 2016 and 2018, China’s standard unemployment rate has decreased below that of other countries. Regarding LFPR with the standard unemployment definition, China’s LFPR during this period significantly exceeded not only those of the USA but also those of other countries.

Figure 3 depicts the U3 series and the LFPR series under the standard unemployment definition for the USA and the rest of the world from 1988 onwards for comparison. Notably, China’s standard unemployment rate surpassed the U3 rate of the USA between 1997 and 2008. However, in 2009, the standard unemployment rate in the USA exceeded that of China. From 2009 to 2014, the standard unemployment rate in the USA remained below that of China, only to be surpassed again by China’s rate after 2016. Comparatively, China exhibited the highest unemployment rate (standard definition) globally from around 2000 until 2014, but since 2016, China’s unemployment rate has fallen below that of other countries. Regarding LFPR, China consistently maintained a higher LFPR than both the USA and the rest of the world throughout the period spanning 1988 to 2018.

What is noteworthy is the similarity between China’s standard labor statistics and those in high-income countries. Since 2009, the standard unemployment rates in China, the USA, and high-income countries have exhibited a decline from their peaks, a pattern not observed in other countries.²⁹ Additionally, it is worth noting that the standard LFPR has increased in both China and high-income countries since 2009, a trend not observed in many other countries, including some high-income nations like the USA, where the LFPR has instead declined.³⁰

Figure 4 extends the series to include the most recent period from 2018 onward, utilizing official data published by the NBS. It plots the official survey-based unemployment rates from the NBS and the “official” LFPR we derive using NBS-provided information as outlined in Sect. 4.2.1, for the period after 2018, alongside corresponding statistics for other countries. A notable observation compared to Fig. 3 is a shift in the relationship between China and high-income countries: the resemblance between China’s standard labor statistics and those of high-income countries diminishes, while the resemblance with statistics from low to upper-middle-income countries increases.³¹

Fig. 4

Recent trends in standard labor statistics in China vs. world. Sources: A Unemployment series for China: survey-based unemployment rate published by NBS; unemployment series for other countries: World Bank. B LFPR series for China: constructed using information published by NBS (see Sect. 4.2.1); LFPR series for other countries: World Bank. Notes LFPR: labor force participation rate. “highinc”: high-income countries. “uppmid”: upper-middle-income countries. “lowmid”: lower-middle-income countries. “lowinc”: low-income countries

6.2.2 Extended labor statistics

For the extended labor statistics, we primarily compare China with the USA, where such statistics are available. Table 2(C) presents the US U5 rates for comparison, with these figures illustrated in Fig. 2A2. Similar to the standard unemployment rate, China’s extended unemployment is higher than the corresponding U5 rate in the USA in 2012 and 2014, but lower than the US U5 rate in 2016 and 2018. Additionally, as previously discussed in Sect. 5.2, our findings indicate that China’s official survey-based unemployment rate, as published by the NBS, closely aligns with the rate we calculate using CFPS under the extended definition of unemployment, which includes discouraged and “other reason” jobless individuals (equivalent to the definition for the US U5 rates). This suggests that China’s official NBS survey-based unemployment rate is more comparable to the US U5 unemployment rate than to the standard U3 unemployment rate.

Table 2C presents and Fig. 2B2 depicts the extended LFPR in the US for comparison. Both under the standard and the extended definition of unemployment, the LFPR in the USA remains approximately 63% between 2012 and 2018, significantly lower than that in China during this period. Furthermore, the LFPR in the USA has slightly decreased by about 1 percentage point in this period, regardless of the definition of unemployment.

Figure 2A2 also includes the nonemployment rate in the USA for comparison.³² Before 2018, the nonemployment rate (representing the extent of idleness of the working-age labor) in China exceeds that in the USA. However, since then, the nonemployment rate in China has decreased below that in the USA, with the gap narrowing from 4.3 percentage points in 2012 to $-$ 0.7 percentage points in 2018.

Figure 3B also includes the NEI in the USA for comparison, spanning from 1994 onward. In the USA, the NEI series consistently remains above the U3 series. It is worth noting that both the standard unemployment rate and NEI in China surpass the U3 and NEI in the USA from 1994 to 2014.

6.3 Comparing alternative un/nonemployment statistics: the predictive power

6.3.1 Predicting wage and GDP

How effective are alternative unemployment (nonemployment) statistics in capturing labor market tightness? To assess this, we examine the predictive power of each statistic in forecasting wage dynamics, considering that a tight labor market typically correlates with accelerated wage growth. To illustrate, we evaluate the goodness of fit of each unemployment statistic in predicting wage fluctuations. This involves employing each unemployment metric to forecast wage levels, utilizing coefficients derived from a regression model that includes a constant, time variables, and the respective unemployment indicator. Subsequently, we compare the projected wage values with actual wage data sourced from the NBS.³³ The wage data represent real average wages, with 2012 prices serving as the base year.

We initially contrast the statistics derived from the CFPS dataset for China spanning 2012–2018. Figure 5A illustrates the actual wage data during this timeframe, alongside the predicted wages represented by different lines. It is evident that during the 2012–2018 period, the U5 rate exhibits superior predictive accuracy compared to both the U3 rate and the NEI in forecasting wage dynamics. Moreover, the U3 rate demonstrates marginally higher predictive capability than the NEI.³⁴

A parallel trend emerges when examining the growth rate of real average wage as the dependent variable. As depicted in Fig. 5B, the U5 rate exhibits superior predictive capability for wage growth compared to both the U3 and the NEI during the 2012–2018 period. Notably, the advantage of the U5 rate is particularly pronounced in this context.

Fig. 5

Goodness of fit for unemployment statistics predicting wage in China. Sources: Wage data: NBS; unemployment statistics: U3, U5, and NEI are calculated using CFPS (2012–2018), and u_survey_official and NEI_official are published by or calculated using information published by NBS (2018–2021). Notes “wagepc”: Real average wage; the base year is 2012. “pred_x”: predicted wage using coefficients from a regression of wage on a constant, time, and the respective unemployment statistic denoted by “x” in “_x”

Similarly, Fig. 6A, B depicts the comparative performance of alternative unemployment and nonemployment statistics in predicting GDP in China.³⁵ Notably, for both GDP levels and growth rates, the U3, U5, and NEI exhibit nearly identical predictive efficacy during the 2012–2018 period, with differences being negligible.

Fig. 6

Goodness of fit for unemployment statistics predicting GDP in China. Sources: GDP data: NBS; unemployment statistics: U3, U5, and NEI are calculated using CFPS (2012–2018), and u_survey_official and NEI_official are published by or calculated using information published by NBS (2018–2021). Notes “gdppcindex”: real GDP per capita index; the base year is 2012. “pred_x”: predicted GDP using coefficients from a regression of GDP on a constant, time, and the respective unemployment statistic denoted by “x” in “_x”

To ensure the robustness of our findings by increasing the sample size, we extend the analysis to the provincial level using CFPS data for the years 2012–2018. Specifically, we compute the unemployment and nonemployment statistics for each province based on CFPS data, gather provincial average wage and GDP per capita (in real terms) data from the NBS, and subsequently conduct regressions of provincial wage and GDP (both in level and growth rate) on the relevant unemployment or nonemployment statistic, alongside province and year dummy variables. We then compare the goodness of fit of each regression. The results remain consistent: U5 demonstrates superior predictive capability for both wage and GDP, particularly in forecasting wage growth.³⁶

6.3.2 Additional comments

The findings suggest that, in China, the U5 rate outperforms the U3 rate in reflecting wage dynamics within the labor market. Conversely, the U5 rate exhibits no greater predictive power than the U3 rate concerning GDP dynamics in China. This divergence could stem from the fact that the relationship between unemployment and GDP primarily mirrors the utilization of the country’s labor resources. In contrast, the correlation between unemployment and wage encompasses the relative supply and demand of labor in the market. Consequently, the U5 rate may be considered a superior gauge of labor market tightness in China, while remaining an equally effective indicator as U3 for GDP growth.³⁷

To expand this comparison to the present period, we utilize the unemployment and nonemployment statistics published by the NBS or constructed using NBS data for the span 2018–2021. As the NBS does not offer adequate information for constructing a counterpart of the U5 rate, we primarily juxtapose the official survey-based unemployment rate from the NBS with the “official” NEI crafted using NBS data, as detailed in Sect. 4.2.3. However, two points merit attention in this comparison.

Firstly, if we consider the official NBS survey-based unemployment rate as a closer approximation of the U5 rate, as suggested by previous finding (e.g., Section 5.2), then the outcomes for the 2018–2021 period may be viewed as a demonstration of the predictive efficacy of U5 during this timeframe. Alternatively, it underscores, at the very least, the utility of the official NBS survey-based unemployment rate as a predictor for wage or GDP.

Secondly, the “official” NEI comprises diverse NBS datasets, which may lack strict comparability in population coverage, such as age range. Consequently, we anticipate a notable measurement error in the “official” NEI, potentially resulting in a weak correlation between the “official” NEI and wage or GDP data. Hence, the effectiveness of the “official” NEI in predicting wage or GDP for this periodmight underestimate the general utility of the NEI.

That being said, it is worth noting that in Fig. 5A, the official survey-based unemployment rate outperforms the “official” NEI in predicting wage for the 2018–2021 period,³⁸ and Additionally, Fig. 5B demonstrates even more prominently that the official survey-based unemployment rate is superior in predicting wage growth compared to the “official” NEI in the same period. This trend is similarly evident in Fig. 6A for predicting GDP levels and in Fig. 6B for predicting GDP growth: the official survey-based unemployment rate consistently outperforms the “official” NEI in both scenarios.

While the “official” NEI may not fully capture its predictive utility for essential economic indicators, it is noteworthy that the NEI constructed using CFPS does not demonstrate a significant advantage over the standard unemployment rate. As depicted in Fig. 5 and 6 for the 2018–2021 period, the NEI appears to have similar predictive capability for wage or GDP compared to U3.

Although the NEI does not consistently demonstrate a significant advantage over alternative unemployment statistics in predicting either wage or GDP for China, it occasionally exhibits superior predictive power over alternatives for other countries, particularly the USA, during specific episodes. For instance, Fig. 7 in the Appendix presents the actual and predicted GDP data for the USA from 1994 to 2022,³⁹ Notably, while the NEI marginally outperforms U3 and U5 in predicting GDP for most of this period, it notably excels in forecasting GDP for the second quarter of 2022. During this period, the onset of COVID-19 led to a significant decline in GDP and a substantial outflow from the labor force in the USA. This exceptional performance is attributed to the closer attachment of individuals out of the labor force to the labor market than conventionally believed. Consequently, this outcome suggests that the NEI is particularly valuable when individuals “out of the labor force” are unusually connected to the labor market and may require policy intervention.

Furthermore, it is essential to acknowledge that the utility of the NEI may be constrained by the utilization of a constant $\bar{p_{O,E}}/\bar{p_{U,E}}$. Allowing this relative transition probability to vary temporally in NEI’s construction could augment the additional information conveyed by this statistic. Another avenue to enrich the information content of the NEI involves weighting distinct subsets of the “out of the labor force” demographic based on their group-specific relative transition probabilities, rather than applying a uniform weight across all individuals in this category. However, this approach necessitates more comprehensive data from labor statistics bureaus, including regular updates on the transition probabilities of individuals in various labor force statuses. We identify this as a prospective direction for future research.

Finally, it is worth noting that alternative labor market statistics, such as the U5 rate and the NEI, hold significant potential for informing economic policies. For instance, given that the U5 rate provides a more comprehensive measure of labor market tightness, policies aimed at improving job search and matching processes can benefit from incorporating U5 alongside the standard unemployment rate. Similarly, with adequate data, we could develop two versions of NEI: one with a fixed transition probability weight and another with a contemporary transition probability weight. Comparing the “fixed weight” NEI with the standard unemployment statistic can offer insights into changes in the size of the “out of the labor force” population, while comparing the “variant weight” NEI with the “fixed weight” NEI can shed light on changes in the transition probability of the population. Substantial discrepancies between these statistics could indicate significant increases in the number of individuals “out of the labor force” who remain closely tied to the labor market, signaling labor market challenges necessitating policy interventions. Hence, alternative labor market statistics have the potential to substantially enhance policy formulation processes.

6.4 Implications

One notable finding is the striking similarity observed between the trends in China’s standard labor statistics and those of high-income countries since 2009, as discussed in Sect. 6.2.1. This resemblance may stem from the strong economic ties between China and high-income countries. The rationale is straightforward: China’s economy relies heavily on external demand, with high-income countries being its primary trading partners. Therefore, as these affluent economies rebounded from the 2008 recession, leading to improvements in their labor markets, the demand for Chinese goods and services also rebounded. Consequently, China experienced a recovery in its economic activities, including its labor market. This finding underscores the deep interdependence between China and the developed world.

The same logic can elucidate the developments post-2018. During this period, the USA rallied affluent Western countries to initiate economic and technological decoupling efforts against China, commencing with the trade war launched by the USA in 2018. Consequently, this action weakened China’s ties with high-income countries while bolstering its connections with other nations as China sought alternatives to markets in affluent countries. This shift elucidates the observed change in China’s relationship with high-income countries since 2018: regarding standard labor statistics, China appears to align more closely with less affluent countries and less so with wealthy nations.

Another significant finding concerns the trajectory of labor statistics in China. While prior research had documented trends up to 2009, our study extends this analysis by examining alternative statistics before 2009 and the evolving trends in standard and alternative statistics post-2010. From the late 1990s to 2009, China consistently exhibited higher levels of unemployment and nonemployment compared to other countries, irrespective of the measures used. This phenomenon can be attributed, in part, to the lingering effects of the SOE reform of the preceding decade, as documented by Feng et al. (2017).⁴⁰ The global recession in 2008 further exacerbated the unemployment issue in China. Consequently, during this period, China grappled with a more severe unemployment and nonemployment challenge relative to other countries.

The emerging trend indicates a decline in China’s unemployment and nonemployment levels after 2012, with a notable decrease observed post-2014. Particularly noteworthy is that China’s unemployment and nonemployment rates have consistently fallen below those of other countries since 2014. This holds true for both the standard unemployment measures and the alternative extended unemployment and nonemployment measures.

The recent trend suggests a potential recovery in China’s labor market, possibly signaling an improvement from past reforms. This could be attributed to contemporary policies aimed at employment promotion. The Plans to Promote Employment (2011–2015) issued by the Ministry of Human Resources and Social Security of the People’s Republic of China outlined initiatives to foster entrepreneurship, invest in marketable human capital, and enhance labor market infrastructure and services. These efforts likely contributed to a resilient labor market. For instance, China’s investment in information infrastructure since 2015 has spurred rapid growth in the digital economy (China Academy of Information and Communications Technology 2022).⁴¹ In theory, the advancement of the digital economy should enhance job search and matching in the labor market while also generating more employment opportunities, thus mitigating unemployment. As depicted in Fig. 8b in the Appendix, provinces experiencing higher levels of digital economy development exhibit a significant correlation with lower unemployment rates during this period.⁴² It would be intriguing to delve deeper into this relationship and its implications.⁴³ Additionally, exploring further causes of recent labor market changes and assessing the relative contribution of different causal factors are areas we intend to investigate in future research.

Section 5 demonstrates that the official register-based unemployment rate remains unchanged after 2010, whereas all alternative measures indicate an improvement in the labor market. Moreover, Figure A.8 in the Online Appendix illustrates that the official register-based unemployment rate exhibits limited predictive power for wage or GDP growth. This disparity underscores the inadequacy of the official register-based unemployment rate in comprehending changes in the labor market. The limitation arises from its sole consideration of registered unemployment, overlooking more informative changes in unregistered unemployment. Alternative measures address this limitation by expanding the coverage of unemployment to encompass unregistered unemployment, thereby furnishing more comprehensive insights than the official register-based unemployment rate.

In contrast, Sect. 5 demonstrates that alternative measures indicate similar trends, implying that minor deviations from the standard definition of unemployment do not substantially distort our comprehension of labor market changes over extended periods. Therefore, the findings provide reassurance regarding the robustness of the labor market trend in China. Similarly, this resilience is observed in other countries, such as the USA, highlighting that the standard definition of unemployment encapsulates the most crucial aspect of long-term change.

Still, non-conventional unemployed workers may exhibit different trends compared to standard unemployed workers. Consequently, alternative measures of unemployment and nonemployment may depict distinct trends and carry different implications. For instance, a significant rise in the number of nonemployed individuals engaged in passive job searches may warrant policy intervention, with the NEI being more suitable for signaling this need, as it provides insights into the size and engagement of the nonemployed population. Therefore, it is advisable to assess the robustness of long-term trends by comparing alternative statistics.

Furthermore, the comparison in Sect. 6.3 indicates that one statistic does not universally suffice for predicting important economic outcomes. Different economic indicators require different un/nonemployment statistics as some are more informative than others depending on the economic context. This necessity arises from the varying levels of attachment among different “nonemployed” individuals, which can impact different economic activities across diverse economic cycles. For instance, marginally attached workers can offer valuable insights into labor market tightness and subsequent wage dynamics. Neglecting this information may diminish the predictive power of unemployment statistics regarding deviations in wage growth from the trend. Incorporating these workers into alternative measures is thus essential for more effectively monitoring this aspect of the economy. Additionally, other non-conventional unemployed workers may offer further insights into various economic aspects. Therefore, it is worthwhile to explore additional alternative measures that capture diverse correlations between the labor market and the economy from various perspectives.

7 Conclusion

Totally disregarding individuals “out of the labor force” (O) in labor statistics construction might overlook crucial insights into the labor market. Various alternative statistics have been proposed, such as the U5 rate, which incorporates discouraged workers, and the nonemployment index (NEI), encompassing all O individuals with weights based on their probability of transitioning to employment relative to standard unemployed individuals. However, such statistics have been unavailable in China. Here, we present such statistics for China, utilizing the nationally representative household survey dataset CFPS from 2012 to 2018, supplemented by data published by China’s official statistics bureau NBS from 2018 onwards. This facilitates temporal and cross-country comparisons, enabling an evaluation of alternative labor statistics in terms of their efficacy in capturing market tightness and measuring resource utilization.

Firstly, demonstrate that alternative labor statistics suggest a similar trend in labor market activity among the population in recent years: a decline in unemployment coupled with an increase in labor force participation, marking a reversal from the trend observed before 2009. Furthermore, a cross-country comparison of standard labor statistics positions China in closer alignment with wealthy nations during the period from 2007 to 2018, whereas since 2018, it aligns more closely with less affluent nations.

Secondly, when comparing the predictive performance of the standard unemployment rate (U3), the U5 rate, and the NEI index for wage and GDP dynamics, we observe that the U5 rate outperforms the other two measures. However, the NEI does not demonstrate a significant improvement over the U3 rate. We discuss the limitations inherent in the current method of constructing the NEI, providing insights for potential enhancements to this statistical measure in future research.

We consider this paper as an initial endeavor to enhance the labor market statistics database for China, making it more comparable internationally. Future tasks, such as publishing these alternative labor statistics using the more extensive and current official UHS monthly data from NBS, are left for future research. This step is essential for gaining a more accurate and real-time understanding of labor market activities and their consequences.

Appendix

See Tables 3, 4, 5 and Figs. 7, 8.

Table 3 Percent of the demographic subpopulations in each labor market status

Table 4 Misreporting probabilities $\Pr \left( \textrm{LFS}=i|\textrm{LFS}^{*}=j\right)$ (in percent)

Table 5 Coefficient of correlation between China’s standard labor statistics and other countries

Fig. 7

Goodness of fit for unemployment statistics predicting GDP in the USA Sources: GDP data: Federal Reserve Bank of St. Louis; unemployment statistics (U3, U5 and NEI): Federal Reserve Bank of Richmond. Notes “gdppcindex”: real GDP per capita index. The base period is the fourth quarter of 2012. “pred_x”: predicted GDP using coefficients from a regression of GDP on a constant, time, and the respective unemployment statistic denoted by “x” in “_x”

Fig. 8

Relationship between unemployment rate and digital economy development in China. Sources: Unemployment rates (U3) are calculated based on CFPS 2012–2018. Digital economy index is calculated using the method of Huang et al. (2019), Zhao et al. (2020), Bai and Zhang (2021), Wang et al. (2021), Chen et al. (2022), and China Academy of Information and Communications Technology (2022) and data from various issues of China Statistics Yearbook. Data are on province level