Data cut-off: 14 October 2026
How many manufacturing entities there are in China is a question with a pile of answers computed on different definitions. Placing the three columns of registered capital, contributed capital and the number of insured employees side by side makes the answer clearer, and also harder to give.
Of the 2,313,700 manufacturing entities recorded in our enterprise database, 1,223,400 entered zero in the "number of insured employees" field of their corporate annual reports, or 52.9%. Within the same group of entities, 43,300 have registered capital above RMB 100 million, while only 5,896 have more than 1,000 people enrolled in social insurance. These two figures differ by more than seven times, and they point to two aspects of the same thing: how much money a company says it has, and how many people a company actually employs.
This report uses the single field of the number of insured employees to describe afresh the size distribution of China's manufacturing entities. It is chosen because its source differs from that of the other two fields, not because it is more accurate. Registered capital is a figure that shareholders subscribe at the time of establishment and, under the subscription system, does not need to be actually paid in; contributed capital is the money shareholders have actually put in, one step closer to day-to-day operations; the number of insured employees, by contrast, is the number for whom the enterprise pays social insurance as disclosed to the market regulatory authorities in its annual report, and behind it lie monthly contribution records of real money. Placed together, the three figures form a chain running from "intention" to "fact".
The practical use of this chain for manufacturing enterprises is screening. A buyer needs to judge whether a supplier is genuinely producing, a financial institution needs to judge the actual operating scale of an enterprise, and an investment promotion department needs to judge the real volume of manufacturing in a region; all of them need an indicator that does not depend on the enterprise's own statements. The number of insured employees is the closest of the currently publicly available data to meeting this requirement, but it has clear boundaries, and there are ways in which it can be misused. This report sets out both the boundaries and the method of use.
This report does not estimate the operating revenue of any enterprise, does not derive a national figure for total manufacturing employment, and does not assess the social insurance compliance status of any specific enterprise. All proprietary data appearing in the text come from our enterprise database; the date of data extraction is 14 October 2026, and the definitions and exclusion rules are set out in full in Chapter 3.
Part One Definitions: What This Field Records
1. An Overlooked Indicator
On public enterprise information lookup pages, registered capital always occupies the most prominent position. It is a line of bold figures that a reader remembers at a glance. The number of insured employees, by contrast, usually hides further down: some platforms place it in the "business registration information" table, and some show it only on the detail page of the annual report. Both are a line of figures, yet one is taken as the first impression of an enterprise's strength while the other is cited by almost no one.
The two are entirely different in nature. Registered capital is the amount of contribution that shareholders undertake to subscribe at the time of the company's establishment. After the subscription-based registered capital registration system came into effect in 2014, this figure no longer needs to be actually paid in at establishment, nor does it require capital verification. A company with RMB 10 million in registered capital may have not a single yuan in its accounts. Contributed capital records the portion shareholders have actually paid, and is closer to the real investment than registered capital, but it is a point-in-time figure reflecting the completion of historical contributions; it has no necessary connection with how many people the company currently employs or how many production lines it runs.
The number of insured employees is a different matter. Under Article 57 of the Social Insurance Law of the People's Republic of China, an employer shall complete social insurance registration within thirty days of its establishment; Article 58 provides that an employer shall, within thirty days of the date on which it begins employing workers, apply to complete social insurance registration for its employees. the Social Insurance Law of the People's Republic of China The combined effect of these two articles is that as long as a company employs workers in accordance with the law, it leaves a record at the social insurance administration corresponding to the number of workers it employs. Once the corporate annual report discloses this record, it becomes a figure that outsiders can see.
The difference among the three indicators can be distinguished by the questions each is able to answer.
| Indicator | Data source | Questions it can answer | Questions it cannot answer directly |
|---|---|---|---|
| Registered capital | Subscription undertaken at the time of establishment registration | How much shareholders are willing to vouch for this company | Whether the money has actually been paid in, how many people it actually employs |
| Contributed capital | The extent to which shareholders' contributions have been completed | How much money shareholders have already put in | Whether it is still producing now, how many people it employs |
| Number of insured employees | The number of insured employees disclosed in the annual report | For how many people this entity actually pays social insurance | Labour dispatch and outsourced personnel, part-time employment, operating revenue |
Table compiled by the Institute from the registration or disclosure basis of each of the three indicators.
A manufacturing enterprise that treats registered capital alone as a size indicator will run into a very typical kind of misjudgement. In our enterprise database, 43,300 manufacturing entities have registered capital above RMB 100 million, and of these, 3,869 disclose a number of insured employees of zero in their annual reports, or 8.9%. This does not mean that these 3,869 are all shell companies; each of the following situations produces the same reading: the group headquarters or investment platform itself does not directly employ workers; the entity has only just completed establishment and has not yet begun hiring; the annual report was filed by an agency but the social insurance data was not disclosed at the same time. In any case, however, the 8.9% share shows that having registered capital above RMB 100 million is by itself not sufficient to prove that a company is organising production.
Conversely, entities with modest registered capital but a substantial number of insured employees are usually the real production segment. In our enterprise database, manufacturing entities with registered capital below RMB 1 million and more than 100 insured employees form an important class of sample for identifying "small but real" factories in an industrial cluster. Such entities are not dominant in number, but they constitute the actual production capacity of county-level industrial clusters.
There is also a technical reason why the number of insured employees is overlooked: it is not a "good" field for direct horizontal comparison. The same factory fluctuates from year to year; two factories of the same size may differ several-fold because of differing levels of automation; the same headcount may correspond to very different operating revenue on the coast and in the interior. These properties make it hard to use as a ranking indicator, but better suited to use as a screening indicator—what screening requires is not precise ordering but the exclusion of objects that plainly do not belong to a given class.
2. The Three Roles of the Number of Insured Employees
To make this field usable, it is first necessary to set out clearly what several things it carries at once. These three things have different boundaries, and using them mixed together leads to erroneous conclusions.
The first role is that of a proxy variable for employment scale.
The number of insured employees records the number of people the employer has brought into social insurance registration. For a manufacturing enterprise, the relationship between this number and the number actually on the job depends on the employment structure: employees who have formally signed a labour contract and whose social insurance is paid by the unit itself are counted; personnel dispatched in through a labour dispatch company have their social insurance paid by the dispatching unit and are not counted in the employing unit's number of insured employees; processes that are wholly outsourced likewise do not enter the plant's roster; part-time employment, workers who continue working beyond the statutory retirement age, and student interns each have separate enrolment rules, and in most cases do not enter the "number of insured employees under urban employee basic old-age pension insurance" field.
The more accurate reading of the number of insured employees is therefore "the number of people directly employed by the unit and lawfully enrolled in social insurance", not "the number of people working in this factory". A factory of 100 people whose workforce includes 40 dispatched workers may have a number of insured employees of 60. This gap is not rare in manufacturing, especially in assembly, packaging and seasonal processing steps where employment fluctuates widely.
The second role is that of a compliance signal, but it is valid in only one direction.
Article 4 of the Social Insurance Law establishes the general contribution obligation of employers within China, Article 57 requires social insurance registration within thirty days, Article 84 sets a penalty for failure to complete social insurance registration, and Article 86 sets, for failure to pay in full and on time, a late payment penalty of 0.05% per day together with a fine of not less than one time and not more than three times the amount. the Social Insurance Law of the People's Republic of China In 2019, the Comprehensive Plan for Reducing Social Insurance Contribution Rates lowered the employer contribution rate for urban employee basic old-age pension insurance to 16%, and at the same time specified that "outstanding historical contributions of enterprises shall not be collected in a concentrated clearing campaign on one's own initiative". the Comprehensive Plan for Reducing Social Insurance Contribution Rates, Guobanfa [2019] No. 13 (国办发〔2019〕13号) Article 19 of the Supreme People's Court's Interpretation (II) on Several Issues Concerning the Application of Law in the Trial of Labour Dispute Cases, effective from 1 September 2025, provides that where an employer and a worker agree, or the worker undertakes, that social insurance premiums need not be paid, the people's court shall hold that the agreement or the undertaking is void. Interpretation (II) on Several Issues Concerning the Application of Law in the Trial of Labour Dispute Cases, Fashi [2025] No. 12 (法释〔2025〕12号)
Together, these provisions make the number of insured employees a figure with legal consequences. But its signal runs in one direction only: a high number of insured employees shows that the entity is at least complete in its social insurance registration; a number of insured employees of zero does not, however, directly imply that the entity employs no workers. Annual reports are filed by the enterprises themselves, the grasp of the filing definition varies, and not disclosed and disclosed as zero are two different things in terms of data structure—a point developed in the next chapter. To read "zero insurance" directly as "no workers" is to treat a weak signal as a strong one.
The third role is that of the degree of cost exposure.
The social insurance cost of manufacturing consists of two parts: the contribution base and the contribution rate. The contribution base is determined by the employee's own wage and has upper and lower limits. Both the upper and the lower limit are assessed on the basis of the province's "average wage of urban unit employees on a full-calibre basis", and the lower limit is 60% of that average wage. the Comprehensive Plan for Reducing Social Insurance Contribution Rates, Guobanfa [2019] No. 13 Because average wages differ across provinces, a worker at the same wage level corresponds to a different minimum contribution amount in different provinces.
For the 2025 scheme year, the monthly contribution base for employee social insurance in Shanghai had a lower limit of RMB 7,460 and an upper limit of RMB 37,302; Shanghai Municipal Human Resources and Social Security Bureau Beijing had a lower limit of RMB 7,162 and an upper limit of RMB 35,811; Beijing Municipal Human Resources and Social Security Bureau Guangdong Province had two tiers, with a lower limit of RMB 5,510 for Guangzhou and for units directly under the province, a lower limit of RMB 4,775 for other areas, and an upper limit of RMB 27,549. Human Resources and Social Security Department of Guangdong Province Under the same system, the lowest-tier contribution base differs by about 1.56 times between provinces.
Among the five social insurance schemes, work injury insurance and maternity insurance are paid by the employer alone, with employees contributing nothing; basic old-age pension insurance, basic medical insurance and unemployment insurance are paid jointly by the employer and the employee. the Social Insurance Law of the People's Republic of China Work injury insurance additionally applies differentiated rates according to the degree of industry risk, and takes the unit's total employee wage bill as its base. ibid. This means that the larger the employment scale, the higher the social insurance cost, and that this cost grows approximately linearly with headcount rather than being diluted as scale expands.
Once the three roles are separated, this report's way of using the number of insured employees is settled: use it to look at distributions, not to produce rankings; use it for initial screening, not to draw conclusions; use it to compare entities within the same region and the same industry, not to compare magnitudes directly across regions or industries.
3. The Boundaries of This Sample and Three Defects
The proprietary data used in this report come from the Tianxia Gongchang enterprise database, which is built on public business registration information and information disclosed in corporate annual reports. Its reference point is 14 October 2026, and the search index contains 3,837,462 entity records in total.
Of these, 3,711,619 are business registration entities collected by the Institute itself, and a further 125,843 are entities added externally by company name. Externally added entities do not go through the collection process for the number of insured employees field when they are entered, and that field is uniformly set to zero, so that mixing them into the statistics would artificially inflate the count of "zero-insurance" entities. All analysis in this report takes the 3,711,619 independently collected entities as its scope, with externally added entities excluded entirely.
The definition of manufacturing follows Section C of the Industrial Classification for National Economic Activities (GB/T 4754), comprising 31 divisions, from the processing of agricultural and sideline food products to the repair of metal products, machinery and equipment. Filtered on this definition, the database contains 2,313,703 manufacturing entities.
This figure requires one correction.
An Outlier That Must Be Written Into the Main Text
The number of insured employees field contains records that plainly do not belong to employment data. Across the whole database, 74 records have values above 100,000 in this field; the largest single record is 621,000,048, and others include 411,023,198, 130,000,000 and 88,855,000. The form of these numbers resembles administrative division codes or registration serial numbers rather than the number of employees any factory could have. Even the world's largest employers by headcount have total workforces on the order of a million.
Leaving the 74 records in the total produces a distortion of orders of magnitude. Before exclusion, the total number of insured employees across the database is 1,817,080,415; after excluding these 74 records it is 78,401,686. The two differ by a factor of 23, and only the latter falls within a range that can be checked against public statistics.
This report therefore handles the matter as follows: under the manufacturing definition, 45 records with a number of insured employees greater than 100,000 are excluded, giving an effective denominator of 2,313,658. Both figures, before and after exclusion, are written out so that they can be checked.
This one treatment illustrates a general rule for using an enterprise database: totals are more fragile than distributions. Binned statistics are unaffected, because the outliers fall only in the highest band; once sums are computed or means are taken, individual records can change the conclusion. All subsequent analysis in this report that involves the total number of insured employees is based on the data after exclusion.
Not Disclosed and Disclosed as Zero Are Two Different Things
Among the 2,313,658 manufacturing entities in the effective denominator, 19,171 have a missing value in the number of insured employees field, or 0.83%; 1,223,420 disclose a number of insured employees of zero in their annual reports, or 52.89%.
The two groups of records are entirely different in data provenance. In the first case the enterprise did not disclose this item at all, which is missing information; in the second the enterprise disclosed a zero, which is information that exists. Merging the two into "entities without social insurance" would inflate the total by 1.6% and, more importantly, would conflate two situations entirely different in nature.
In calculating the zero-insurance rate, this report uses "entities that disclosed a number of insured employees" as the denominator, that is, 2,313,658 minus 19,171, which equals 2,294,487. On this definition, the zero-insurance rate among manufacturing entities is 53.3%. Throughout the text, every occurrence of the term "zero-insurance rate" refers to this definition, and every occurrence of "number of entities" or "effective denominator" refers to 2,313,658.
The Enterprise Database Is Not the Complete Register of Registered Entities
Our enterprise database was built for factory retrieval and is not a mirror of the complete register of market entities. Besides manufacturing, the database also contains such sections as animal husbandry, wholesale, retail, agriculture, and science and technology promotion and application services. Counted by industry name, animal husbandry ranks first in the database with 329,128 entities, and the manufacture of metal products second with 229,616.
This point determines how this report must be phrased: it cannot say broadly that "China has 2,310,000 factories"; it can only say that "our enterprise database records 2,313,700 manufacturing entities". A difference in collection definition stands between the two, and this report does not derive a national total.
The Time Lag in Annual Report Data
A corporate annual report discloses information for the preceding business year. The current number of insured employees for a given enterprise in the database therefore in fact reflects the situation in the year covered by its most recent annual report. For enterprises whose operating conditions change between two annual reports, the reading lags. None of the analysis in this report distinguishes the year to which an annual report belongs; this is a limitation arising from data availability and should be taken into account when interpreting data on entities newly established in the past two years.
4. Why Manufacturing Entities Are Comparable
The comparability of the number of insured employees field is not uniform across industries; manufacturing is one of the sections where comparability is relatively good, for three reasons.
First, in manufacturing there is a relatively stable correspondence between employment and physical output. How many machines an injection moulding workshop can run and how many machine operators and inspectors it needs are determined by the number of machines and the number of shifts, and the empirical range within the industry is fairly narrow. By contrast, turnover in wholesale and retail can be very large with very few workers, and in consulting services there is almost no fixed technical coefficient between output and headcount. Employment in manufacturing is closer to a proxy variable for production capacity.
Second, the industry attribution of manufacturing entities is relatively clear. The registered business scope of a company may span several fields, but its actual production activity usually falls within one particular division. When this report groups statistics by industry division, the employment structure of each group shows differences consistent with its process characteristics, a point that will be developed in Chapter 15.
Third, the obligation to provide the five social insurance schemes is not waived for any industry in manufacturing. Article 4 of the Social Insurance Law establishes the general contribution obligation of employers within China; manufacturing falls into no exempt category, and the five social insurance schemes are all legally mandatory for manufacturing plants. the Social Insurance Law of the People's Republic of China Differences in the number of insured employees within manufacturing therefore reflect differences in employment scale and in the fulfilment of enrolment, not differences in the scope of obligation.
It should be stated in advance that this report does not put the number of insured employees to three uses. First, it is not used to derive operating revenue. Two factories with the same 50 workers may differ several-fold in revenue because of different product unit prices and levels of automation. Second, it is not used to derive production capacity. Capacity is determined by equipment, not by headcount. Third, it is not used to assess the compliance status of an individual enterprise. Zero insurance has multiple causes; this report describes distributions only and does not name or assess any individual enterprise.
Part Two Size Distribution
5. The Size Pyramid of 2.31 Million Manufacturing Entities
Grouping the 2,313,658 manufacturing entities into bands by number of insured employees produces a shape that is an extremely steep pyramid.
| Number of insured employees | Number of entities | Share | Cumulative share |
|---|---|---|---|
| 0 (disclosed as zero) | 1,223,420 | 52.89% | 52.89% |
| 1—9 | 587,342 | 25.39% | 78.28% |
| 10—49 | 330,768 | 14.30% | 92.57% |
| 50—99 | 70,080 | 3.03% | 95.60% |
| 100—299 | 57,096 | 2.47% | 98.07% |
| 300—999 | 19,885 | 0.86% | 98.93% |
| 1,000—4,999 | 5,289 | 0.229% | 99.16% |
| 5,000 and above | 607 | 0.026% | 99.18% |
| Not disclosed | 19,171 | 0.83% | 100.00% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, banded by the number of insured employees field; date of data extraction 14 October 2026; 45 outliers excluded.
The most direct information this table conveys is this: manufacturing entities with fewer than 50 insured employees account for 92.57% of all entities. Entities with more than 1,000 insured employees number 5,896 in total, or 0.255%. Of these, the 607 with more than 5,000 insured employees account for 0.026% (two point six per ten thousand).
The steepness of the pyramid can be read from another angle as well. Combining the 0 and the 1 to 9 bands gives 1,810,762 entities, or 78.28%. In other words, four out of every five manufacturing entities have fewer than 10 insured employees. The annual-report insured-employee counts of the entities in this band are either zero or a single digit; among them are a large number of entities that do not in fact organise production, and also a large number of genuine micro-scale processing points—one machine and two or three people in household-workshop production, which is common in industries such as hardware, wood processing and garment processing.
The quantiles give a more precise shape. Within the valid denominator, the 50th percentile of the number of insured employees is 0, the 75th percentile is 7, the 90th percentile is 29.02, the 95th percentile is 69.39, the 99th percentile is 334.09, the 99.9th percentile is 2,005.08, and the 99.99th percentile is 9,322.16.
In terms of cumulative shares, of the 2,313,658 entities, those with no more than 29 insured employees account for about nine tenths, those with no more than 69 for about ninety-five per cent, and those with no more than 334 for about ninety-nine per cent. The size distribution of manufacturing entities is not normal, and it has no concentrated peak of a "typical size".
This shape is consistent with everyday observation of the industrial clusters of Guangdong, Zhejiang and Jiangsu. In a county-level hardware industrial cluster, a few firms have more than a thousand people, several dozen firms have from a few dozen to a few hundred, and the rest are thousands of small processing points. The distribution of the number of insured employees quantifies this structure.
It should be noted that this table reflects the distribution of the number of entities, not the distribution of production capacity. Small and micro entities, which account for 92.57% by number, account for a far lower share of the total number of insured employees. The 5,896 entities with more than 1,000 insured employees (0.255% of entities) account for 17,327,500 insured employees in total, or 33.6% of the 51.506 million total insured employees; of these, the 607 with more than 5,000 account for 7,301,554 insured employees in total, or 14.2%. The pyramid of numbers and the pyramid of production capacity slope in opposite directions.
6. A Median of Zero: The Composition of the 1.22 Million Zero-Insurance Entities
The median number of insured employees of manufacturing entities is zero. The meaning of this reading needs to be unpacked.
An annual report disclosing a number of insured employees of zero may correspond to any of the following situations, and these cannot be distinguished in the data.
Situation one: the enterprise genuinely did not employ workers requiring insurance in the year concerned. This includes entities that have been established but have not yet begun operating, entities that have ceased production but have not yet completed deregistration, and entities that hold only factory buildings or land without organising production.
Situation two: all of the enterprise's employment is arranged through forms that do not enter the unit's own insurance register. Three forms are common in manufacturing: employment through a labour dispatch company, with social insurance paid by the dispatching unit; the outsourcing of an entire process to another entity, whose employees do not enter the register of the party placing the order; and the use of persons beyond the statutory retirement age, part-time workers and student interns, whose insurance rules differ from those of serving employees and who in most cases do not appear in the column "number of insured employees under urban employee basic old-age pension insurance".
Situation three: differences at the reporting stage. Corporate annual reports are filed by the enterprises themselves, and some entities file through bookkeeping agencies. Understanding of, and habits in completing, the "number of insured employees" column differ, and a minority of entities enter zero in that column rather than the actual number.
The three situations appear in the data as the same zero; this report cannot distinguish them at the individual level, and can only offer some clues about them in aggregate. The clues come from two comparisons.
The first comparison groups by registered capital. If zero insurance were caused mainly by "not actually operating", the zero-insurance rate should fall markedly as registered capital rises—entities with high registered capital are more likely to be enterprises in actual operation. The actual data support this judgement, and the gradient is very clean:
| Registered capital | Number of entities | Zero insured | Zero-insurance rate |
|---|---|---|---|
| Below RMB 1 million | 1,143,152 | 830,726 | 72.7% |
| RMB 1—5 million | 536,158 | 224,475 | 41.9% |
| RMB 5—10 million | 217,729 | 75,601 | 34.7% |
| RMB 10—50 million | 314,010 | 78,759 | 25.1% |
| RMB 50—100 million | 59,291 | 9,990 | 16.8% |
| Above RMB 100 million | 43,318 | 3,869 | 8.9% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing definition, cross-tabulated.
The zero-insurance rate falls monotonically from 72.7% to 8.9%, with no reversal in any of the six bands. This monotonicity indicates that zero insurance is highly correlated with the actual operating status of the entity, rather than being randomly distributed reporting noise.
The second comparison groups by year of establishment. The preliminary reading is that the earlier an entity was established, the lower its zero-insurance rate. This is consistent with the plain judgement that entities which survive are more likely to be actually operating. A full cohort comparison by year of establishment involves confounding with the completeness of data collection, and this report does not adopt it; the relevant analysis appears in "The Age of Chinese Factories", published by the Institute at the same time.
Taken together, the two comparisons provide a central estimate for zero insurance: it mainly reflects the entity's actual operating status and employment form, rather than being purely a question of compliance. This estimate, however, carries errors in two directions that must be noted.
One direction is understatement. Factories that make heavy use of dispatched workers and outsourced processes have more people actually on site than the number of insured employees. The number of insured employees of such factories is not zero, but the reading is low.
The other direction is overstatement. Reading a number of insured employees of zero directly as an "empty shell" would misjudge as non-existent the processing points described in situation two that genuinely organise production. In county-level industrial clusters, a processing point where five or six people run a few machines and take outside orders from large plants may well have no social insurance registration at all. In the count of entities it is a zero; in production capacity it is not.
For buyers and for investment promotion departments, this implies a concrete operating rule: zero insurance is a signal requiring further verification, not a conclusion that can be used directly. The way to verify is not to keep querying data but to return to the production site—to look at the equipment, the electricity meter and the logistics documents. The role of the data is to narrow the objects requiring verification from 2.31 million to 1.22 million; the remaining judgement must be made outside the data.
7. Where This Column in the Annual Report Comes From
The previous two chapters repeatedly use the phrase "the number of insured employees disclosed in the annual report" without explaining the origin of this column. The history of this column is shorter than most people assume, and the way it is produced directly determines the nature of the data.
2014: The Annual Report System Is Established, but It Contains No Social Insurance
The Interim Regulations on Enterprise Information Disclosure were promulgated by State Council Order No. 654 and took effect on 1 October 2014. "the Interim Regulations on Enterprise Information Disclosure" Article 8 provides: "An enterprise shall, between 1 January and 30 June each year, submit the annual report of the previous year to the administration for industry and commerce through the enterprise credit information publicity system, and disclose it to the public." Enterprises registered in the year of establishment submit and disclose from the following year. ibid.
Article 9 lists the seven items the annual report shall contain: mailing address and contact details; information on the state of existence such as commencement of business, suspension of business and liquidation; information on enterprises invested in and established, and on the purchase of equity; the subscribed and actually contributed capital contributions of shareholders or promoters, the time of contribution and the form of contribution; information on equity changes; information on the enterprise's website and online stores; and information on the enterprise's number of persons engaged, total assets, total liabilities, guarantees provided to outside parties, total owners' equity, total operating revenue, revenue from principal operations, total profit, net profit and total taxes paid. ibid.
Among the seven items there is no social insurance insured-employee count. Article 9 contains a further paragraph distinguishing the disclosure obligations: items one to six shall be disclosed to the public, while item seven is left to the enterprise to choose whether to disclose to the public. In other words, in the 2014 annual report system the only column relating to employment was "number of persons engaged", and it fell among the items the enterprise could decide on its own whether to disclose. ibid.
2016: Social Insurance Items Are Added to the Annual Report
The number of insured employees entered the annual report on the basis of the "Notice on Issues Concerning the Addition of Social Insurance and Statistical Items to Enterprise Annual Reports", issued jointly by the State Administration for Industry and Commerce, the Ministry of Human Resources and Social Security and the National Bureau of Statistics on 17 November 2016, document number Gongshang Qijianzi [2016] No. 226 (工商企监字〔2016〕226 号). Its higher-level basis is the "Notice of the General Office of the State Council on Accelerating the Reform of the Registration System of the "five-in-one" business licensing reform and "one licence, one code"", a reform that abolished the periodic verification and renewal of social insurance registration certificates and statistical registration certificates and replaced them with enterprises filing annual reports themselves and disclosing them to the public. "Notice on Issues Concerning the Addition of Social Insurance and Statistical Items to Enterprise Annual Reports"
The notice added five social insurance items: "type of insurance scheme, number of insured employees of the unit, contribution base of the unit, amount actually paid in the current period, and cumulative amount in arrears of the unit". Of these, the first and second items shall be disclosed to the public, while the third to the fifth are left to the enterprise to choose whether to disclose to the public. ibid.
This provision has two direct consequences for the data used in this report.
First, the "number of insured employees of the unit" is a mandatory disclosure item. Enterprises may choose not to disclose the contribution base, the amount actually paid and the cumulative amount in arrears, but they cannot do so for the number of insured employees and the type of insurance scheme. This explains why this report can obtain the number of insured employees for 99.17% of manufacturing entities but cannot obtain their contribution amounts—the latter is a voluntary disclosure item with a far lower coverage.
Second, this field has existed only since the annual report for the 2016 reporting year. Enterprises filing supplementary annual reports for 2015 and earlier years file according to the content prescribed by the original regulations, without the social insurance items. Modifications to the system had to ensure that from 1 January 2017 enterprises filed annual reports under the adjusted items. ibid. The number of insured employees used in this report therefore has an upper time limit of the 2016 reporting year; entities established before that time that had ceased operating and no longer filed annual reports are blank in this field.
The notice also specifies a data transfer mechanism: the departments of industry and commerce and of market regulation must, within one month after the annual report work for each year ends, transfer enterprise annual report information in bulk to the human resources and social security departments and the statistics departments, and regions with the necessary capacity may transfer monthly. ibid.
Definition: A Point-in-Time Figure at the End of December, Excluding Retired Personnel
The "number of insured employees of the unit" has an official definition. According to annual report completion guidance issued by local market regulation departments, this column refers to the number of employees participating in social insurance at the end of December of the year concerned, excluding the number of retired personnel; the contribution base of the unit refers to the sum of the contribution bases on which the unit paid social insurance contributions for its employees from January to December of the year, including retroactive contribution bases and the differences arising from adjustments to contribution bases; the amount actually paid in the current period refers to the social insurance contributions paid on time by the unit from January to December of the year, excluding arrears made up and prepayments spanning years or quarters; and the cumulative amount in arrears of the unit refers to the principal amount of all social insurance contributions in arrears as at the end of December of the year. Annual report completion guidance of Sishui County, Jining, Shandong Province
This is a set of point-in-time data, not an annual average, nor a sum of monthly figures across the year. It is therefore sensitive to the employment situation at the year end: a factory that employed 200 people through the year and cut back to 80 by the year end because of falling orders records a number of insured employees of 80.
The authorities also explicitly distinguish the two concepts of "number of insured employees" and "number of persons engaged". In answering questions on annual reports, a market regulation department wrote: "Must the 'number of insured employees' and the 'persons engaged' in an enterprise's annual report be kept consistent? The 'number of insured employees' refers to the number of employees of the unit who pay social insurance, and its concept differs from that of 'persons engaged'." Annual report guidance of Jiangcheng District, Yangjiang, Guangdong Province This distinction is the official basis for the statement in Chapter 6 of this report that the number of insured employees is not the number of people at the plant.
"Not Disclosed", "Withheld from Public Disclosure" and "Entering 0" Are Three Different Things
These three expressions are often conflated in practice, but their legal character is entirely different.
Not disclosed means the enterprise failed to submit its annual report within the prescribed period, or failed to disclose the relevant information within the period ordered. This is an unlawful act, and the consequence is inclusion in the list of enterprises with abnormal operations; after the 2024 amendment it is further provided that where an enterprise has been included in the list of enterprises with abnormal operations for failing to submit annual reports as prescribed for two consecutive years without rectification, and cannot be contacted at its registered domicile or place of business, the market regulation department at or above the county level shall revoke its business licence. "the Interim Regulations on Enterprise Information Disclosure" (current text), "the Interim Regulations on Enterprise Information Disclosure" (text promulgated in 2014)
Withheld from public disclosure means the enterprise has filed truthfully in the system but has checked the columns within the optional range as not displayed to the public. This is lawful, and on the public-facing side one sees the words "the enterprise has chosen not to disclose" rather than a blank.
Entering 0 means the enterprise truthfully reported that the actual number of insured employees at the end of the year was zero. Local market regulation department completion guidance states this explicitly: "Number of persons insured under each scheme: the number of employees participating in social insurance at the end of the reporting period (excluding retired personnel) ... if social insurance has not yet been paid, enter '0' in each place." Annual report guidance of Yongdeng County, Lanzhou, Annual report guidance of Xinhui District, Jiangmen
This sentence is key to understanding the core data of this report. Entering 0 is a form of completion explicitly permitted by official guidance, and it corresponds to the factual state of "social insurance not paid". As to whether the reason for not paying social insurance is that no employees were hired, or that the employment form does not enter the unit's register, or that contributions should have been paid but were not, the guidance does not distinguish and the annual report system does not collect this. The three types of cause listed in Chapter 6 of this report have their origin here—the data field by design has only one zero, and cannot carry three meanings.
A Study That Matched Annual Reports Against Actual Insurance Participation, Enterprise by Enterprise
Among the three types of cause above, "differences at the reporting stage" had long been treated only as a possibility. A study published in 2026 has turned it into a fact with an associated figure.
Researchers at the Guanghua School of Management of Peking University matched the micro-survey data of the Peking University Enterprise Survey of Innovation and Entrepreneurship (ESIEC) for 2023, enterprise by enterprise, against the 2022 annual reports of the surveyed enterprises. ESIEC 2023 completed interviews with 6,110 enterprises in total, of which 2,687 were successfully matched to the number of employees under old-age pension insurance disclosed in their 2022 annual reports. On this basis the researchers compared the social insurance headcounts reported by the same enterprise through the two channels. "China Journal of Economics and Management", 2026, Volume 5, Issue 1
The results are as follows. Among enterprises whose social insurance headcount in the survey data was greater than 0, 54% still disclosed a social insurance headcount of 0 in their annual reports. The paper's wording is: "This indicates that enterprises have a tendency to enter the value '0' when completing annual report information." By way of contrast, among enterprises whose social insurance headcount in the survey data was 0, 94% also disclosed 0 in their annual reports. ibid.
Among the samples where both the annual report and the survey social insurance headcounts were greater than 0, the annual report figure was smaller than the survey figure in 64.2% of cases, and the two were equal in 21.5%. The paper concludes that enterprises tend to under-report their actual number of social insurance contributors in their annual reports. ibid.
The paper offers four explanations for the reporting tendency: first, the probability that an enterprise is penalised for disclosing false annual report information is relatively low, so enterprises do not much care about the veracity of that information; second, enterprises may be unwilling to inform the government or their peers of their true operating situation; third, enterprises are unlikely to over-report the social insurance headcount in their annual reports, since this could lead the tax authorities or the human resources and social security departments to demand make-up contributions; fourth, enterprises often regard filing an annual report each year as an additional burden and delegate it to an ordinary employee, who may enter 0 if unaware of the enterprise's true situation. The fourth explanation also accounts for why larger enterprises are in fact more likely to under-report—the larger the enterprise, the lower the probability that an employee picked up casually to complete the annual report happens to know the true situation. ibid.
The researchers also carried out third-party cross-validation. Comparing "estimates from the survey data", "aggregates of the annual report data" and "data disclosed by provincial government departments", they found that the publicly available data for Henan Province explicitly state their source as "annual reports" and that their values are close to the annual report aggregates, while in Guangdong and Zhejiang, where the gap between the annual reports and the survey is large, the two provinces' publicly available figures for social insurance contributions of serving enterprise employees are closer to the estimates from the survey data. The paper therefore warns data users: "if the data source is enterprise annual reports, there will be substantial bias." ibid.
This result has three direct consequences for the treatment adopted in this report, which must be stated clearly.
First, the 52.89% zero-value share reported in Chapter 6 of this report cannot be read as "52.89% of manufacturing entities do not pay social insurance for their employees". It reads as "52.89% of manufacturing entities disclosed a number of insured employees of zero in their most recent annual report". According to the study above, the gap between these two statements may be considerable.
Second, the banded distribution in this report remains valid, but its meaning requires adjustment. A reading greater than 0 still points to employment that genuinely exists—an enterprise that enters 200 in its annual report has a high probability of actually employing around 200 people or more, because the tendency to under-report biases annual report readings downwards. A reading of zero, by contrast, provides no directional information: it may correspond to genuinely not being insured, or to being insured in fact while not reporting truthfully.
Third, the proposition advanced in Chapter 3 of this report, that "totals are more fragile than distributions", needs one further clause here: in a field subject to systematic under-reporting, totals are not only fragile but also directional. According to the study above, the direction is downward.
The sample boundary of the study must be stated as well: ESIEC 2023 is a random-sample survey of 107 districts and counties in six provinces and municipalities (Liaoning, Henan, Zhejiang, Shanghai, Gansu and Guangdong), with a matched sample of 2,687 enterprises, predominantly private enterprises. The two proportions of 54% and 64.2% reflect the situation of that sample and cannot be directly generalised as the prevailing level for manufacturing nationwide. This report cites these two figures in order to show that the magnitude of reporting bias may be substantial, not in order to provide a nationwide correction coefficient.
Another Type of Systematic Risk Already on Record
There is another mechanism at the reporting stage that distorts the reading, and it is already on public record.
In an interactive question-and-answer exchange, the market regulation department of Shenzhen replied to a member of the public: "Upon verification, the social insurance information in the enterprise annual report is pushed by the social insurance department, and does not require the enterprise to complete it." The enterprise that raised the question reported that the company actually had 1 person insured, but the annual report preview page showed that "the number of insured employees is 0, and the contribution amount is also 0", and the system provided no entry point for completion, leaving only the choice of disclosing or not disclosing. Interactive question and answer, Shenzhen Municipal Market Regulation Bureau
Several days later the same department answered a similar question in consistent terms: "The social insurance information in the enterprise annual report is obtained from the social insurance department, and does not require the enterprise to complete it." That question reported that in the previous year employees had actually contributed to social insurance, yet the exported annual report showed the social insurance headcount and other information all as 0. Interactive question and answer, Shenzhen Municipal Market Regulation Bureau
The value of these two records is that they show that in at least some regions the "number of insured employees" is no longer completed by the enterprise itself but pushed to the annual report system by the social insurance department. The direction of this change is towards higher data quality—data provided by the department that holds the contribution records is more reliable than self-reporting by enterprises. At the same time, however, it introduces a new source of distortion: problems at the system interface stage can make the reading inconsistent with actual insurance participation, and enterprises can neither see a route to correction nor judge the matter.
This report cannot estimate the share of all entities accounted for by such cases. They are listed here to suggest a rule of judgement: when an enterprise shows zero insurance while other evidence indicates that it is producing normally, one should not first assume that the enterprise is concealing something, but should consider whether there is a problem in the data chain. The two records from Shenzhen demonstrate that this possibility genuinely exists.
About Nine Per Cent of Entities Have No Annual Report Data for the Year
The annual report system itself has boundaries of coverage.
Enterprises that do not file their annual reports on time are included in the list of enterprises with abnormal operations. As at October 2024, 11.388 million entities nationwide were on the list of enterprises with abnormal operations. Press conference of the State Administration for Market Regulation In terms of the filing rate, the national enterprise annual report filing rate for the 2022 reporting year was 91.33%, and since the regulations took effect in 2014 it has remained above 91% for six consecutive years. The State Council website
From this one can infer a structural gap in the data used here: each year roughly 8% to 9% of existing enterprises have no annual report data for that year, including the number of social insurance insured employees. In the statistics of this report these entities appear as missing field values (0.83% under the manufacturing definition) rather than as zero. The two are treated separately in this report.
It is worth noting that the missing-field rate in this report's sample (0.83%) is markedly lower than the national non-filing rate (about 8.7%). A possible reason for the difference is that the collection process of the Tianxia Gongchang enterprise database itself imposes certain requirements on data completeness, so that entities that do not file annual reports are filtered out at the collection stage. This means that this report's denominator is already a relatively clean subset, and its zero-insurance rate reading may be lower than the actual situation across all registered entities.
How External Research Treats "All-Zero" Reports
A final piece of corroborating evidence worth recording comes from an academic survey. In its 2023 sampling, the Peking University Enterprise Survey of Innovation and Entrepreneurship listed "all entries in the sample's annual reports across years are 0" as the first exclusion rule, and this rule accounted for 68% of all excluded samples. "China Journal of Economics and Management", 2024, Volume 3, Issue 2
That study addresses the problem of identifying whether an entity is actually operating, and its approach is to judge records in which every column of the annual report is zero as very probably not operating, and to remove them from the sample. Two-thirds of the exclusions arise from this single rule, which shows that "all-zero annual reports" form a category of considerable size that has already been widely recognised in research.
This report does not adopt that exclusion method, because what this report seeks to describe is precisely the distribution of this category of entities, rather than to exclude them and then analyse. The external practice nevertheless supports the judgement in Chapter 6 of this report: linking "all-zero annual reports" to "not actually operating" is not an inference of this report but a rule already in use in research practice.
8. Four Figures from Four Sources
The previous chapter showed that the number of insured employees in enterprise annual reports is subject to systematic under-reporting. That being so, a natural question arises: what is the true situation?
There are official and academic answers to this question, but they measure different things. Placing them side by side is more illuminating than giving only one of them.
This Report's Definition: 52.89%
Of the 2,313,658 manufacturing entities in the Tianxia Gongchang enterprise database, 1,223,420 disclosed a number of insured employees of zero in their most recent annual report, or 52.89%. This is an entity-based definition—the unit of statistics is one enterprise, and the answer is "zero or not zero".
NPC Law Enforcement Inspection: A Relatively Large Gap of "Should Be Insured but Is Not" Among Small, Medium and Micro Enterprises
On 5 November 2024, Wang Dongming, Vice-Chairman of the Standing Committee of the National People's Congress, delivered the "Report of the Law Enforcement Inspection Team of the Standing Committee of the National People's Congress on the Inspection of the Implementation of the Social Insurance Law of the People's Republic of China" at the twelfth session of the Standing Committee of the 14th National People's Congress. The text of the report states: "The law enforcement inspection found that state-owned enterprises, foreign-funded enterprises and large private enterprises are basically able to enrol their employees in the various insurance schemes in accordance with the law, whereas the insurance participation rate of small, medium and micro enterprises is still not high enough, and the gap of 'should be insured but is not' is relatively large." National People's Congress
The same report gives another figure: "According to a survey by the China Association of Social Security, the compliance rate of employee old-age pension insurance in China (the ratio of the number of contributors to the number of insured persons) fell from 85.2% in 2011 to 80.8% in 2022." The report explains this as follows: "in recent years, under the impact of the triple pressures of contracting demand, supply-side shocks and weakening expectations, some small, medium and micro enterprises face operating difficulties, and lapses in insurance and arrears of contributions have become relatively common." ibid.
The coverage of this law enforcement inspection is worth recording, in order to judge how representative its conclusions are: the inspection teams were headed by Wang Dongming, Xiao Jie and Zheng Jianbang; from June to July 2024 they went in three groups to six provinces (autonomous regions and municipalities)—Heilongjiang, Shanghai, Jiangxi, Guangdong, Sichuan and Ningxia—went into 16 cities and prefectures, conducted on-site inspections of 46 enterprises and 11 rural communities, and organised 13 discussion sessions; the standing committees of the people's congresses of a further six provinces (municipalities)—Liaoning, Zhejiang, Anhui, Henan, Chongqing and Qinghai—were also commissioned to inspect their own administrative areas, and the Chinese Academy of Social Sciences and the China Association of Social Security were commissioned to carry out third-party assessments. ibid.
CASS Survey: Basic Old-Age Pension Insurance Participation Rate of 92.0%
In 2024 the World Social Security Research Centre of the Chinese Academy of Social Sciences carried out the "Survey on Social Insurance Participation Status", covering 31 provinces (municipalities and autonomous regions) excluding Hong Kong, Macao and Taiwan; based on the district and county population structure of the Seventh National Population Census, it adopted multistage probability sampling proportionate to population size, with a target population of employed persons aged 20 to 60 working in urban areas, and collected 15,080 valid questionnaires. World Social Security Research Centre, Chinese Academy of Social Sciences
The survey results are: the basic old-age pension insurance participation rate among all respondents was 92.0%, of which urban employee basic old-age pension insurance 61.9% and urban and rural resident basic old-age pension insurance 30.1%; the work injury insurance participation rate was 45.7% and the unemployment insurance participation rate 43.1%; the supplementary old-age pension insurance participation rate was 7.4%. 8% of all respondents were in a lapsed state of "should be insured but is not". ibid.
The same NPC report provides two further figures relating to cost: "at present the contributions for the five insurance schemes account for around 35% of employee wages"; and "paying contributions on the minimum contribution base (60% of the average social wage) is widespread. A contribution base that does not reflect reality both affects the sustainability of the social insurance fund and will drag down future pension levels." National People's Congress
The Four Figures Cannot Be Added or Subtracted
Placed together, the four figures are as follows:
| Source | Unit of statistics | Definition | Reading |
|---|---|---|---|
| Our enterprise database | Manufacturing entities | Whether the number of insured employees disclosed in the annual report is zero | 52.89% zero |
| Peking University ESIEC matching study | Enterprises (six provinces and municipalities, 2,687 matched samples) | Among enterprises that actually have social insurance in the survey, the proportion entering zero in the annual report | 54% |
| NPC law enforcement inspection report | Enterprises (qualitative, 12 provinces, autonomous regions and municipalities) | The compliance of small, medium and micro enterprises with insurance participation | A relatively large gap of "should be insured but is not" |
| CASS survey | Individuals (31 provinces, 15,080 questionnaires) | Participation rate of surveyed employed persons in the five social insurance schemes | Old-age pension 92.0%, work injury 45.7%, unemployment 43.1% |
Table compiled by the Institute from the publicly stated definitions of each source.
These four figures cannot validate one another, because they measure different things. The 52.89% in this report is an entity-based figure—an enterprise with one insured employee and an enterprise with one thousand insured employees both count as "non-zero" in this statistic. The participation rate in the CASS survey is a person-based figure—counted by head, a thousand-worker plant contributes a thousand times as much to numerator and denominator as that one-person enterprise does. The 54% in the Peking University study is a ratio between two channels for the same set of enterprises, measuring reporting bias rather than the level of insurance participation.
The only thing that can be said directly about the three is their relationship of magnitude. Under the entity-based definition, about half of manufacturing entities show non-participation in their annual reports; under the person-based definition, old-age pension insurance coverage exceeds ninety per cent, while coverage of work injury and unemployment insurance is below half. The two readings are not contradictory: employment in manufacturing is highly concentrated in a small number of entities (Chapter 15 of this report shows that seven high-threshold industries account for 18.3% of entities and 44.1% of employment), so "most entities are not insured" and "most individuals are insured" can hold at the same time.
The implications of this structure for use are clear. Ask "how many factories in Chinese manufacturing do not pay social insurance for their employees", and the answer depends on whether the question is about the number of enterprises or the number of people, and the two answers can differ enormously. The former is around half; the latter, on the work injury and unemployment definitions of the CASS survey, lies at roughly between a half and sixty per cent—the two are close, but for entirely different reasons: the former is high because the number of small and micro enterprises is large, while the latter is high because coverage of work injury and unemployment insurance at the individual level is itself low.
This chapter sets out these four figures in order to make the point that before using any figure relating to insurance participation, one should first establish whether its unit of statistics is the enterprise or the individual, not in order to give a "correct answer". This is the point at which this kind of data is most easily misused and most easily causes serious misleading.
9. Our Database Against National Definitions
The previous two chapters discussed the internal consistency of a single body of data. This chapter brings in external definitions to answer a question deferred until now: how large a share of the national manufacturing total do the 51.506 million insured persons in the Tianxia Gongchang enterprise database represent.
Two Official Definitions That Must Be Kept Separate
The Ministry of Human Resources and Social Security publishes a "Statistical Communiqué on the Development of Human Resources and Social Security" annually. At the end of 2025, the number of persons participating in urban employee basic old-age pension insurance nationwide was 546.72 million, an increase of 12.19 million over the end of the previous year; of these, insured employees numbered 394.57 million and insured retired personnel 152.15 million. At the end of the same year, the number of persons participating in unemployment insurance nationwide was 249.18 million and the number participating in work injury insurance 305.07 million. Statistical communiqué of the Ministry of Human Resources and Social Security
There is a definition here that is misused every year. The "number of insured persons" is the sum of insured employees and insured retired personnel, and does not equal the scale of the serving workforce. At the end of 2025, the 152.15 million retired personnel accounted for 27.8% of the number of insured persons, and the ratio of retired persons to insured employees was 38.6%. In other words, reading "number of persons participating in urban employee basic old-age pension insurance" directly as "how many serving employees are paying social insurance" overstates the figure by roughly thirty-eight per cent at a stroke.
The "number of insured employees" used in this report comes from the "number of insured employees of the unit" in enterprise annual reports, whose official definition explicitly excludes retired personnel (see Chapter 7). The reading in this report is therefore a serving-employee definition, and is not the same thing as the "number of insured persons" in the communiqué, which has to be broken down.
The Denominator for Manufacturing
There are two usable official definitions for the scale of persons employed in manufacturing.
The Fifth National Economic Census (end of 2023, released on 26 December 2024) shows 4.049 million manufacturing legal entities and 104.815 million persons employed in manufacturing legal entities. Communiqué of the Fifth National Economic Census (No. 2) It should be noted that the same communiqué, when expressed under the secondary industry definition, gives 4.048 million and 104.814 million; the difference arises from the treatment of "repair of metal products, machinery and equipment" under the secondary industry definition. Both are the text of the communiqué; this report adopts the set under the section definition.
The other definition comes from industrial enterprises above designated size. In 2024 the number of industrial enterprises above designated size nationwide was 517,303 and their average number of employed persons was 76.901 million. For the manufacturing part, the average number of employed persons in manufacturing above designated size in 2023 was 69.4553 million.
The difference between these two figures is worth recording separately: the average number of employed persons in manufacturing above designated size is 69.4553 million, while persons employed in manufacturing legal entities total 104.815 million, a difference of 35.3597 million, or 33.7% of the latter. This employment lies entirely in below-designated-size enterprises and self-employed businesses and falls outside the scope of above-designated-size statistics. Any account that uses the "number of employed persons in industrial enterprises above designated size" to represent employment in Chinese manufacturing will omit more than one third.
Two Comparisons
Comparison one: the number of entities.
The Tianxia Gongchang enterprise database contains 2,313,658 manufacturing entities. Manufacturing legal entities under the Fifth National Economic Census definition number 4.049 million (end of 2023). Our database is equivalent to 57.1% of the latter.
Two points need to be made about this ratio. First, the collection basis of our database is manufacturing entities registered with the industry and commerce authorities, whereas the census legal-entity definition also includes types of unit that are not required by law to carry out industry and commerce registration, so the two do not coincide exactly. Second, the reference point of our database is October 2026 while the census data are as at the end of 2023, more than two years apart, during which period there were both new registrations and deregistrations. The 57.1% is therefore a reference of magnitude, not a precise coverage rate.
Comparison two: employment.
The number of insured employees of manufacturing entities in the Tianxia Gongchang enterprise database totals 51,506,296, that is, 51.506 million. Persons employed in manufacturing legal entities number 104.815 million. The former is equivalent to 49.1% of the latter.
This ratio embodies two differences at once, which must be separated.
The first difference is coverage: our database contains only about 57.1% of manufacturing legal entities, and the employment of the entities not contained is not part of the 51.506 million.
The second difference is definition: the number of insured employees and persons employed are two different quantities. Persons employed includes all personnel in post; the number of insured employees includes only persons brought into social insurance registration. The gap between the two arises from dispatched and outsourced personnel not entering the unit's register, from part-time and over-age personnel being subject to different rules, and from a portion of employment that should be insured but is not.
Scaling out the first difference proportionally allows a rough conversion: if the entities contained in our database have the same size structure as the national total, then the 2,313,658 entities correspond to about 104.815 million × 57.1% ≈ 59.85 million persons employed. On this basis, the ratio of the number of insured employees of our database's entities to their number of persons employed is about 86.1%.
This 86.1% is a hypothetical projection, not a measured value. Its validity depends on the assumption that the entities contained in our database have the same size structure as the national total—whereas Chapter 3 of this report has already shown that the Tianxia Gongchang enterprise database is not a mirror of the full register of market entities, and that its collection process may involve systematic omissions of newly established entities and small and micro entities; if so, the average size of the entities in our database would be higher than the national average, and the actual ratio should be lower than 86.1%. The purpose of setting out this conversion is to show how much of the 49.1% ratio may come from each of its two sources, not to produce a comparable figure.
The Conclusion of This Chapter
Taken together, the three comparisons yield one judgement: the Tianxia Gongchang enterprise database covers more than half of China's manufacturing legal entities, and its distributional features are representative; but its absolute values cannot be used directly as national totals.
Specifically:
- Using the distribution of our database (the proportions in each band, the relative position of industries, the ranking of provinces, the differences between cohorts) for analysis is feasible, and the conclusions of this report as a whole rest on this level.
- Using the absolute values of our database (2.31 million entities, 51.506 million people) to extrapolate national totals is not feasible. It would require at least two corrections: one being extrapolation according to coverage, the other being the conversion of numbers of insured employees into numbers of persons employed.
- All proportions given in this report should, when cited, retain the qualifier "the manufacturing definition of the Tianxia Gongchang enterprise database".
This chapter also answers a question a reader might raise: why, when this report repeatedly emphasises definitions, does it not deliver a national total. The reason lies in the fact that where the coverage rate is unknown, a ratio obtained by dividing two figures has no explanatory power, not in the data being insufficient—and the coverage rate is precisely what this report cannot derive from its own data. This chapter uses figures from yearbooks and the census to provide a reference of magnitude; the function of that reference is to let one know at what level our database's readings sit, not to serve as a usable conversion coefficient.
10. The 5,896 Factories with More Than a Thousand Insured Employees
Manufacturing entities with more than 1,000 insured employees number 5,896 in total. The number itself is small, 0.255% of all entities, but together they account for 17.3275 million insured employees, 33.6% of the total number of insured employees in manufacturing. To understand the organisational structure of Chinese manufacturing, this group of entities cannot be bypassed.
They are highly concentrated geographically. Counted by province, Guangdong has 1,150, Jiangsu 790, Zhejiang 481 and Shandong 464; the four provinces together account for 2,885, or 48.9% of the national total. These four provinces are also all of the four provinces with the largest total numbers of manufacturing entities—Guangdong 438,592, Zhejiang 368,769, Jiangsu 273,689 and Shandong 202,862. The ratio between the number of entities and the number of above-1,000 factories is not uniform, however: in Guangdong there is one above-1,000 factory for every 381 manufacturing entities, in Zhejiang one for every 767, in Jiangsu one for every 346 and in Shandong one for every 437. Zhejiang ranks second nationally in the number of entities, yet its number of above-1,000 factories ranks behind Guangdong and Jiangsu.
In cross-industry comparison, what is more illuminating is the share of above-1,000 factories among the entities of the industry concerned, which may be called the industry's "large-plant density".
| Industry division | Number of entities | Above 1,000 | Large-plant density |
|---|---|---|---|
| Manufacture of tobacco products | 436 | 60 | 13.76% |
| Manufacture of medicines | 15,418 | 252 | 1.634% |
| Smelting and pressing of ferrous metals | 15,190 | 240 | 1.580% |
| Manufacture of railway, ship, aerospace and other transport equipment | 18,505 | 251 | 1.356% |
| Manufacture of computers, communication and other electronic equipment | 94,954 | 1,153 | 1.214% |
| Smelting and pressing of non-ferrous metals | 18,760 | 208 | 1.109% |
| Manufacture of automobiles | 61,355 | 598 | 0.975% |
| Processing of petroleum, coal and other fuels | 16,435 | 152 | 0.925% |
| Manufacture of chemical fibres | 6,573 | 53 | 0.806% |
| Manufacture of raw chemical materials and chemical products | 71,015 | 423 | 0.596% |
| Manufacture of electrical machinery and apparatus | 112,150 | 566 | 0.505% |
| Manufacture of special-purpose machinery | 136,718 | 218 | 0.159% |
| Manufacture of general-purpose machinery | 198,918 | 228 | 0.115% |
| Processing of food from agricultural products | 84,935 | 93 | 0.109% |
| Manufacture of non-metallic mineral products | 188,318 | 136 | 0.072% |
| Manufacture of furniture | 88,159 | 64 | 0.073% |
| Manufacture of metal products | 229,611 | 146 | 0.064% |
| Comprehensive utilisation of waste resources | 27,234 | 10 | 0.037% |
| Processing of timber, and manufacture of wood, bamboo, rattan, palm and straw products | 96,048 | 12 | 0.013% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, cross-tabulated by industry division; density is "number of entities above 1,000 ÷ number of entities in the industry".
The industries at the top share one explanation: barriers of entry and capital. Of the 436 entities in the manufacture of tobacco products, 60 are above 1,000, a density of 13.76%; this figure reflects the administrative entry regime for tobacco manufacturing, not the outcome of market competition. Of the 15,418 entities in the manufacture of medicines, 252 are above 1,000, a density of 1.63%, which is directly related to the licensing of pharmaceutical production, GMP certification and investment in cleanrooms. The densities for ferrous metal smelting and non-ferrous metal smelting are 1.58% and 1.11% respectively, corresponding to equipment such as blast furnaces and electrolytic cells that cannot be operated at small scale. The 1.36% for the manufacture of railway, ship, aerospace and other transport equipment corresponds to the fixed costs of the final assembly stage.
The industries at the bottom also share an explanation: processes that can be divided, equipment that can be small. The density for the processing of timber and the manufacture of wood, bamboo, rattan, palm and straw products is 0.013%, for the manufacture of metal products 0.064%, and for the comprehensive utilisation of waste resources 0.037%. What these industries have in common is that one process can correspond to one machine and one person, or to a small team of a few people, and the scale of production can expand or contract over a wide range, so that the form of industrial organisation is inherently dispersed.
One counter-intuitive result is the manufacture of computers, communication and other electronic equipment. Its number of entities reaches 94,954 and it has 1,153 above-1,000 factories, both among the highest, with a density of 1.21%. Electronics manufacturing includes both complete-unit and module plants with more than a thousand people and a large number of small supporting plants of a dozen or so people; both ends are heavy. Its number of insured employees totals 7,379,001, the highest of the 31 divisions, 67.5% higher than the 4,405,413 of the second-placed manufacture of automobiles.
11. The Triangle of Registered Capital, Contributed Capital and the Number of Insured Employees
Placed side by side, the three fields reveal things that are not obvious when they are viewed separately.
| Registered capital | Number of entities | Insured 0 | Zero-insurance rate | Contributed 0 | Contributed equal to registered capital | Total insured employees |
|---|---|---|---|---|---|---|
| Below RMB 1 million | 1,143,152 | 830,726 | 72.7% | 836,122 | — | 4,014,170 |
| RMB 1—5 million | 536,158 | 224,475 | 41.9% | 164,589 | — | 4,949,422 |
| RMB 5—10 million | 217,729 | 75,601 | 34.7% | 42,160 | — | 3,973,391 |
| RMB 10—50 million | 314,010 | 78,759 | 25.1% | 37,392 | — | 12,475,581 |
| RMB 50—100 million | 59,291 | 9,990 | 16.8% | 4,100 | — | 6,081,133 |
| Above RMB 100 million | 43,318 | 3,869 | 8.9% | 1,073 | — | 20,012,599 |
| Total | 2,313,658 | 1,223,420 | 52.9% | 1,085,436 | 1,221,503 | 51,506,296 |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing definition, cross-tabulated. Owing to the structure of the cross-tabulation, the "contributed equal to registered capital" column is not broken down by band; the total is 1,221,503 of all manufacturing entities, or 52.80%.
Begin with a set of overall figures. Among all manufacturing entities, 1,085,436 have zero contributed capital, or 46.91%; 1,221,503 have contributed capital exactly equal to registered capital, or 52.80%; and 339,670 have registered capital itself of zero, or 14.68%.
The share of 52.80% for "contributed equal to registered capital" is high enough to require explanation. One possibility is that many small and micro entities paid up in full at the time of establishment, or under the subscription system paid up within the period stipulated in their articles of association and did not change thereafter, so that the two figures remain equal over time. Another possibility is that the contributed capital of some entities is not separately disclosed and has been entered in the data as the same value as registered capital. This report cannot distinguish these two situations at the individual level, and therefore when citing this figure treats it only as a description of a data characteristic, not as evidence that "the shareholders have actually contributed capital".
The relationship among the three columns can be seen from two extremes.
The first extreme is "much money, few people". Among the 43,318 entities with registered capital above RMB 100 million, 3,869 still have a number of insured employees of zero, and 1,073 have zero contributed capital. The zero-insurance rate in this band is 8.9%, the lowest of all bands, but 8.9% corresponds to the absolute number of 3,869—close to four thousand entities with registered capital above RMB 100 million disclosed no insured personnel at all in their most recent annual report. A considerable part of them are holding platforms, investment vehicles or project companies that do not organise production themselves; another part are newly built projects not yet in production. Whichever category they fall into, they cannot be counted as "factories" in calculating production capacity.
The second extreme is "not few people, not much money". Among the 1,143,152 entities with registered capital below RMB 1 million, a considerable number have more than 100 insured employees, and 394 have more than 1,000. An entity with registered capital below RMB 1 million and a thousand or more insured employees usually corresponds to a processing plant created by the restructuring of a former collective enterprise in earlier times, with a simple shareholder structure but complete equipment and personnel, or to a production unit established with very low registered capital while the group in practice provides the factory building and equipment. Such entities genuinely exist in industrial clusters and are often among the best supplier candidates—they have production capacity and workers, but the "first impression" from public information is poor.
The relationship between the number of insured employees and registered capital, at the aggregate level, yields a conclusion opposite to that at the level of the number of entities. By number of entities, the below-RMB-1-million band accounts for 49.42%, close to half; by number of insured employees, this band contributes only 4,014,170, or 7.79%. The 43,318 entities above RMB 100 million (1.87%) contribute 20,012,599, or 38.85%. Describing Chinese manufacturing by the number of entities and describing Chinese manufacturing by the number of employed persons yield two different Chinas.
12. The Long Tail of the Distribution and the Failure of the Mean
The number of insured employees of manufacturing entities totals 51,506,296; divided by the valid denominator of 2,313,658, the arithmetic mean is 22.26. This figure has almost no explanatory power.
The reason is the shape of the distribution. The median number of insured employees is 0, the 75th percentile is 7 and the 90th percentile is 29. The arithmetic mean of 22.26 falls between the 75th and the 90th percentile; that is, the mean is larger than at least three quarters of manufacturing entities. It is not a figure that can represent a "typical factory".
The thickness of the long tail can be seen from several quantiles: the 99th percentile is 334, the 99.9th percentile is 2,005 and the 99.99th percentile is 9,322. From the 99th to the 99.99th percentile, the proportion of entities differs by a factor of 100, and the number of insured employees differs by a factor of 27.9. The right tail of the distribution is very long and shows no sign of converging.
There are three direct implications for users.
First, do not use the mean number of insured employees to estimate the size of an individual factory. For a manufacturing entity drawn at random, the most likely count is zero or a single digit, whereas the mean gives an impression of 22 that is too high.
Second, do not use the mean number of insured employees multiplied by the number of entities to estimate a region's total employment. In the Tianxia Gongchang enterprise database, Guangdong has 438,592 manufacturing entities with a total of 10,882,275 insured employees, a mean of 24.81; the Tibet Autonomous Region has 3,062 entities with a total of 15,363 insured employees, a mean of 5.02. The two means differ by nearly a factor of five, and this difference arises both from differences in factory size and from differences in the share of "zero-insurance and small and micro entities" in the entity structures of the two places—the zero-insurance rate is 87.7% in Tibet and 42.2% in Guangdong. The mean mixes the two factors together and they cannot be separated.
Third, when comparing different regions or different industries, one should compare distributions rather than means. Distributions of the same scale may share a mean, yet one consists of "mostly small and medium plants plus a few large plants" while the other is "a few very large plants plus a large number of empty shells", and their significance for procurement and for investment promotion is entirely different. Subsequent chapters of this report all adopt banded distributions and comparison of quantiles, and do not use the mean as a principal indicator.
Part Three: Space and Industry
Chapter 13: Guangdong, Zhejiang and Jiangsu — Three Pillars
The three provinces with the largest number of manufacturing entities are Guangdong, Zhejiang and Jiangsu, with 438,592, 368,769 and 273,689 entities respectively, 1,081,050 in total, or 46.72% of all manufacturing entities nationwide. The three provinces together have 23,995,926 insured employees, 46.59% of 51,506,296. The share of entities and the share of employment are almost equal.
The differences within the three provinces are more interesting than their combined figures.
| Province | Entities | Total insured | Average insured | Zero-insurance rate | Median | 90th percentile | 1,000 or more |
|---|---|---|---|---|---|---|---|
| Guangdong Province | 438,592 | 10,882,275 | 24.81 | 42.2% | 2 | 33 | 1,150 |
| Jiangsu Province | 273,689 | 7,226,802 | 26.40 | 40.2% | 2 | 40 | 790 |
| Zhejiang Province | 368,769 | 5,886,849 | 15.96 | 45.6% | 1 | 26 | 481 |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, by province under the manufacturing definition. The column "Average insured" is the total divided by the number of entities; it is used only for relative comparison within this table and is not used to estimate the total volume of regional employment, for the reasons given in Chapter 12.
Zhejiang has 34.7% more entities than Jiangsu, yet 18.5% fewer insured employees. Average insured employees are 15.96 against 26.40, a difference of 65%. The 90th percentile is 26 against 40, and the number of factories with 1,000 or more employees is 481 against 790. These three indicators point in the same direction: the structure of manufacturing entities in Zhejiang is more skewed towards small units than that in Jiangsu.
Guangdong leads in both directions. It has 18.9% more entities than Zhejiang and 84.9% more insured employees; its average of 24.81 insured employees lies between Jiangsu and Zhejiang; and its 1,150 factories with 1,000 or more employees are close to the sum of Jiangsu and Zhejiang together (1,271).
The combination of these three figures is consistent with the organisational form of manufacturing in the three provinces. Zhejiang's industrial organisation is known for the "block economy", in which one town concentrates on a single category of product, and within the town the number of enterprises is large, the individual plant is small and the division of specialised labour is fine. In the data, this structure appears as a large number of entities, a low number of insured employees per entity and few factories with more than 1,000 employees. In Jiangsu's industrial organisation, foreign-invested and large private enterprises carry a greater weight, and individual entities are larger. Guangdong has both—the Pearl River Delta contains tens of thousands of small and medium-sized component suppliers as well as very large finished-product and contract-manufacturing enterprises.
The ordering by zero-insurance rate is Jiangsu 40.2%, Guangdong 42.2%, Zhejiang 45.6%, which does not correspond exactly to the ordering by average insured employees. Jiangsu has both the lowest zero-insurance rate and the highest average number of insured employees, which indicates that its entity structure is also lower along the dimension of "the proportion of micro and small entities". Zhejiang is skewed towards small units on both indicators.
It should be noted that all comparisons in this section are made outside a single institutional environment. The lower limits of the social insurance contribution base are not the same in the three provinces. For the 2025 contribution year, the lower limit of the contribution base for urban employee basic old-age pension insurance in Guangdong Province was RMB 5,510 for Guangzhou and for units directly under the province, and RMB 4,775 for other areas, Human Resources and Social Security Department of Guangdong Province while the specific standards of the other two provinces must be sought in the announcements of their respective human resources and social security departments. Differences in the contribution base affect enterprises' social insurance costs, but they do not affect the field of insured employees itself—the number of insured employees records a number of people, not an amount of money. The provincial comparison in this report uses the headcount definition and is therefore unaffected by differences in the base.
Chapter 14: Shanghai's 20.8% and Tibet's 87.7%
The zero-insurance rate by province runs from 20.8% to 87.7%, a difference of 4.2 times. This range is more informative than the range of average insured employees, because it is not affected by any single very large entity.
The five provincial-level units with the lowest zero-insurance rate are Shanghai 20.8%, Jiangsu Province 40.2%, Guangdong Province 42.2%, Beijing 42.6% and Zhejiang Province 45.6% (Fujian Province 45.7% follows closely). The five highest are the Tibet Autonomous Region 87.7%, Guizhou Province 79.8%, Heilongjiang Province 74.4%, Hebei Province 74.2% and Jilin Province 73.5% (Henan Province 73.4% follows closely).
Shanghai's readings require separate explanation. It has only 27,819 manufacturing entities, 6.3% of Guangdong's, but its zero-insurance rate is only half that of Guangdong. The median number of insured employees is 6 and the 90th percentile is 84, both far above the national level (national median 0, 90th percentile 29). The structure corresponding to these readings is this: land and labour costs in Shanghai exclude small processing sites, and the manufacturing entities that remain are mainly the production bases of above-designated-size enterprises, foreign-invested enterprises and headquarters-type entities. Its number of manufacturing entities is small, but each one is more likely to be an enterprise genuinely organising production.
Beijing's figures run to the extreme in the same direction. It has only 5,494 manufacturing entities, among the fewest of all provincial-level units, but its 90th percentile of insured employees is 143 and its average insured employees are 93.82, both the highest nationwide. The contrast with the Tibet Autonomous Region is the sharpest: 3,062 entities, a 90th percentile of only 2 insured employees, an average of 5.02 insured employees and a zero-insurance rate of 87.7%. That is to say, nine tenths of Tibet's manufacturing entities disclosed no insured personnel at all in their most recent annual report, and even where personnel were disclosed the number was very small.
Hebei Province presents another notable pattern. It has 197,039 manufacturing entities, the fifth largest number nationwide, but its zero-insurance rate is 74.2%, the fourth highest nationwide, and its 90th percentile of insured employees is only 10. A large number of entities, small individual plants and a low insurance rate appear all at once. Within the Beijing-Tianjin-Hebei area, Beijing and Hebei Province are almost exactly opposite on these three indicators: Beijing has 5,494 entities, a zero-insurance rate of 42.6% and a 90th percentile of 143; Hebei Province has 197,039 entities, a zero-insurance rate of 74.2% and a 90th percentile of 10. The two are adjacent in space and sit at two different tiers of industrial structure.
The explanation of this set of differences requires caution. The level of the zero-insurance rate is affected simultaneously by three factors: the proportion of micro and small entities and of entities not actually operating within the region, the forms of employment within the region (the extent to which dispatch and outsourcing are used), and differences in the annual report filing process. The data in this report cannot separate these three at the provincial level. A safe reading is this: the inter-provincial differences in the zero-insurance rate do exist and are very large in magnitude, but what they measure is entity structure, not the level of social insurance compliance of a given province.
From the standpoint of use, this set of differences yields a concrete rule: when comparing enterprises across provinces, one cannot directly compare the absolute number of insured employees, nor even directly compare "whether zero-insured". The same factory of 20 people is a relatively small entity in Zhejiang Province, while in the Tibet Autonomous Region it already belongs to the larger class. Comparison should be made within a province and within an industry.
| Provincial-level unit | Entities | Zero-insurance rate | Median | 90th percentile | 1,000 or more |
|---|---|---|---|---|---|
| Shanghai | 27,819 | 20.8% | 6 | 84 | 136 |
| Jiangsu Province | 273,689 | 40.2% | 2 | 40 | 790 |
| Guangdong Province | 438,592 | 42.2% | 2 | 33 | 1,150 |
| Beijing | 5,494 | 42.6% | 3 | 143 | 78 |
| Zhejiang Province | 368,769 | 45.6% | 1 | 26 | 481 |
| Fujian Province | 79,329 | 45.7% | 1 | 35 | 252 |
| Tianjin | 23,856 | 50.0% | 1 | 43 | 80 |
| Liaoning Province | 57,525 | 56.4% | 0 | 29 | 154 |
| Hubei Province | 42,527 | 58.3% | 0 | 46 | 210 |
| Sichuan Province | 67,349 | 59.9% | 0 | 39 | 204 |
| Anhui Province | 80,031 | 61.6% | 0 | 32 | 218 |
| Shandong Province | 202,862 | 62.0% | 0 | 23 | 464 |
| Jiangxi Province | 48,472 | 65.7% | 0 | 31 | 135 |
| Henan Province | 100,649 | 73.4% | 0 | 14 | 203 |
| Jilin Province | 20,814 | 73.5% | 0 | 19 | 54 |
| Hebei Province | 197,039 | 74.2% | 0 | 10 | 213 |
| Heilongjiang Province | 22,841 | 74.4% | 0 | 13 | 61 |
| Guizhou Province | 26,331 | 79.8% | 0 | 10 | 57 |
| Tibet Autonomous Region | 3,062 | 87.7% | 0 | 2 | 1 |
The table is an excerpt, ordered by zero-insurance rate in ascending order. Data for all provincial-level units can be found in the search results of this database.
Chapter 15: The Employment Structure of the 31 Manufacturing Divisions
Dividing manufacturing into 31 divisions, the employment structure differs greatly from one to another. The table below is ordered by number of entities in descending order and also gives the zero-insurance rate, the total number of insured employees and the percentiles.
| Industry division | Entities | Zero-insurance rate | Total insured | Median | 90th percentile |
|---|---|---|---|---|---|
| Manufacture of metal products | 229,611 | 56.7% | 2,438,849 | 0 | 19 |
| Manufacture of general-purpose machinery | 198,918 | 47.0% | 3,103,621 | 1 | 28 |
| Manufacture of non-metallic mineral products | 188,318 | 63.1% | 2,168,087 | 0 | 20 |
| Manufacture of rubber and plastic products | 157,660 | 52.6% | 2,073,237 | 0 | 20 |
| Manufacture of special-purpose machinery | 136,718 | 42.3% | 2,932,628 | 2 | 38 |
| Manufacture of electrical machinery and apparatus | 112,150 | 33.4% | 4,400,586 | 3 | 56 |
| Manufacture of textiles | 108,008 | 56.0% | 1,582,501 | 0 | 22 |
| Other manufacturing | 107,152 | 62.9% | 971,980 | 0 | 16 |
| Manufacture of wood and products of wood, bamboo, rattan, palm and grass | 96,048 | 82.5% | 284,302 | 0 | 4 |
| Manufacture of computers, communication equipment and other electronic equipment | 94,954 | 28.2% | 7,379,001 | 6 | 112 |
| Manufacture of wearing apparel and accessories | 91,904 | 62.8% | 1,056,359 | 0 | 17 |
| Manufacture of furniture | 88,159 | 68.5% | 761,373 | 0 | 11 |
| Processing of food from agricultural products | 84,935 | 70.6% | 1,099,495 | 0 | 19 |
| Manufacture of raw chemical materials and chemical products | 71,015 | 34.8% | 3,323,612 | 4 | 86 |
| Manufacture of foods | 66,202 | 61.6% | 1,256,587 | 0 | 28 |
| Manufacture of articles for culture, education, arts and crafts, sports and entertainment activities | 61,721 | 56.4% | 846,290 | 0 | 20 |
| Manufacture of automobiles | 61,355 | 34.9% | 4,405,413 | 5 | 121 |
| Manufacture of paper and paper products | 60,862 | 51.3% | 798,934 | 0 | 19 |
| Manufacture of leather, fur, feather and related products and footwear | 56,870 | 60.6% | 817,881 | 0 | 17 |
| Printing and reproduction of recording media | 47,133 | 35.4% | 723,212 | 3 | 29 |
| Manufacture of wine, beverages and refined tea | 41,741 | 75.8% | 740,691 | 0 | 12 |
| Comprehensive utilization of waste resources | 27,234 | 77.4% | 148,874 | 0 | 7 |
| Manufacture of instruments and meters | 22,405 | 29.1% | 792,324 | 5 | 64 |
| Smelting and rolling of non-ferrous metals | 18,760 | 41.0% | 1,218,115 | 3 | 90 |
| Manufacture of railway, ship, aerospace and other transport equipment | 18,505 | 44.3% | 1,649,658 | 2 | 76 |
| Processing of petroleum, coal and other fuels | 16,435 | 51.9% | 699,363 | 0 | 20 |
| Manufacture of medicines | 15,418 | 22.4% | 1,692,766 | 19 | 237 |
| Smelting and rolling of ferrous metals | 15,190 | 54.6% | 1,236,635 | 0 | 45 |
| Repair of metal products, machinery and equipment | 11,268 | 63.5% | 399,434 | 0 | 20 |
| Manufacture of chemical fibres | 6,573 | 46.9% | 288,791 | 1 | 60 |
| Manufacture of tobacco | 436 | 28.2% | 215,697 | 105 | 1,223 |
Table compiled by the Institute from this database, by industry division under the manufacturing definition.
The two ends of the zero-insurance rate are 82.5% for the manufacture of wood and products of wood, bamboo, rattan, palm and grass and 22.4% for the manufacture of medicines, a difference of 3.7 times. The two ends of the median are 0 for the manufacture of wood and several other industries and 105 for the manufacture of tobacco.
Taken together, the 31 divisions can be roughly divided into three groups.
Group one: industries with a median of 0 and a zero-insurance rate above 55%. These comprise the manufacture of metal products, the manufacture of non-metallic mineral products, the manufacture of textiles, the manufacture of wood and products of wood, bamboo, rattan, palm and grass, the manufacture of wearing apparel and accessories, the manufacture of furniture, the processing of food from agricultural products, the manufacture of foods, the manufacture of articles for culture, education, arts and crafts, sports and entertainment activities, the manufacture of leather, fur, feather and related products and footwear, the manufacture of wine, beverages and refined tea, the comprehensive utilization of waste resources, other manufacturing, and the repair of metal products, machinery and equipment. This group contains 14 divisions in total, with 1,259,171 entities, 54.4% of all manufacturing entities, and 14,572,703 insured employees, accounting for 28.3%. Their common feature is that the process can be broken down into very small units and a product can be completed by one person or a few people, so that the size distribution of entities is extremely dispersed.
Within this group, the 90th percentile for the manufacture of wood and products of wood, bamboo, rattan, palm and grass is only 4, the lowest of the 31 divisions. That is to say, among the 96,048 entities in this industry, nine tenths have fewer than 4 insured employees.
Group two: industries with a zero-insurance rate between 33% and 55% and a median of 0 to 3. These comprise the manufacture of general-purpose machinery, the manufacture of rubber and plastic products, the manufacture of chemical fibres, the processing of petroleum, coal and other fuels, the manufacture of paper and paper products, and the smelting and rolling of ferrous metals. This group consists mainly of process manufacturing and equipment manufacturing, and the size distribution of entities shows a clear bimodal pattern—there are both small and medium-sized plants of a few dozen people and large plants of over a thousand.
Group three: industries with a zero-insurance rate below 36% and a median of 3 or more. These comprise the manufacture of medicines (22.4%, median 19), the manufacture of computers, communication equipment and other electronic equipment (28.2%, median 6), the manufacture of instruments and meters (29.1%, median 5), the manufacture of electrical machinery and apparatus (33.4%, median 3), the manufacture of raw chemical materials and chemical products (34.8%, median 4), the manufacture of automobiles (34.9%, median 5), and printing and reproduction of recording media (35.4%, median 3). This group contains 7 divisions in total, with 424,430 entities, 18.3% of manufacturing entities, but 22,716,914 insured employees, 44.1% of the total number of insured employees in manufacturing.
These seven industries show the same feature in the data: the higher the threshold, the fewer the entities, the more employment per entity, and the lower the zero-insurance rate. The sources of the threshold differ—for medicines it is licensing and GMP, for electronics it is equipment and clean rooms, for automobiles it is supplier certification by the original equipment manufacturer, for raw chemical materials it is safety and environmental approval, and for instruments and meters and electrical machinery it is accumulated technology. But they all produce the same result: they exclude the "one machine, one person" processing site.
Group three accounts for 18.3% of entities but 44.1% of employment; this contrast is one of the most important structural facts in this report as a whole. Employment in Chinese manufacturing is highly concentrated in a small number of high-threshold industries, and the number of entities in those industries is not large.
Chapter 16: The 15.9% and 70% of Foreign-Trade Factories
The data in this section come from the foreign-trade flag field in the database, which indicates whether an entity has records related to import and export operations.
| Category | Entities | Share | Insured > 0 | Zero-insurance rate | Median | 90th percentile | Total insured | Share of total insured |
|---|---|---|---|---|---|---|---|---|
| Foreign trade | 366,768 | 15.86% | 309,672 | 15.6% | 17 | 180.74 | 36,035,445 | 69.96% |
| Non-foreign-trade | 1,946,890 | 84.14% | 761,395 | 60.9% | 0 | 13.87 | 15,470,851 | 30.04% |
Table compiled by the Institute from this database, by foreign-trade flag under the manufacturing definition.
The magnitude of this contrast exceeds that of any other in this report. Foreign-trade entities make up 15.86% of manufacturing entities, yet 69.96% of the total number of insured employees in manufacturing. The zero-insurance rate is 15.6% against 60.9%, a difference of 3.9 times. The median is 17 against 0. The 90th percentile is 180.74 against 13.87, a difference of 13 times.
The explanation of this gap has three layers; they cannot be fully separated in the data, but each layer has independent grounds.
The first layer is selection by scale. Factories able to take on export orders have usually already crossed a certain size threshold. Before establishing a supply relationship, overseas buyers generally conduct a factory audit covering production capacity, the quality management system, social responsibility and employment compliance. A processing site with only three or five people would find it hard to pass such an audit, and hard to bear the costs of customs declaration, foreign exchange settlement and certification that exporting requires. The flag "foreign trade" is therefore itself the result of a filter: while marking trading capability, it also marks scale.
The second layer is linked compliance. The factory inspection stage of export orders, particularly orders for customers in Europe and North America, generally includes a check on social insurance contributions. This is not a requirement of Chinese law but a supply chain compliance requirement of the buyer. Foreign-trade factories therefore fulfil social insurance enrolment to a greater extent than domestic-sales factories, for independent commercial reasons. In the data this appears as follows: the zero-insurance rate of foreign-trade entities is nearly three-quarters lower than that of non-foreign-trade entities.
The third layer is statistical definition. Export-related business records themselves have to be handled under a regular market entity identity, which filters out some production units operating in a personal or non-formal form. The effect of this layer is difficult to quantify, but its direction is the same as that of the first layer.
From the standpoint of use, the foreign-trade flag is the field with the greatest discriminating power in this database. It assigns the "non-foreign-trade" flag to 84.14% of entities and retains only 15.86% as foreign-trade entities; and that 15.86% covers nearly seven tenths of manufacturing employment. For situations in which candidate suppliers have to be narrowed from the million scale to the ten-thousand scale within a short time, this field has greater filtering power than registered capital, greater than paid-in capital, and greater than any single industry classification.
It should be noted that the foreign-trade flag does not distinguish the proportion of exports. A factory that sells 95% of its products domestically and has only a small volume of export records and a factory that exports 95% of its products are identical on this one field. To differentiate further, other information such as customs credit rating and export destination country is needed, and this lies outside the scope of the data in this report.
Chapter 17: County Level — A Contrast Between Two Kinds of Cluster
The unit at which manufacturing actually clusters is usually the county and the district rather than the province. In this database, 98.4% of manufacturing entities carry a district-or-county field, and tabulation by district or county can reconstruct the actual shape of industrial clusters.
| District or county | Entities | Zero-insurance rate | Total insured | 1,000 or more |
|---|---|---|---|---|
| Dongguan | 89,229 | 37.6% | 2,594,630 | 257 |
| Zhongshan | 38,282 | 34.0% | 775,407 | 81 |
| Nanhai District (Foshan) | 29,769 | 37.0% | 406,057 | 22 |
| Shunde District (Foshan) | 23,523 | 30.9% | 588,945 | 53 |
| Bao'an District (Shenzhen) | 20,160 | 34.9% | 694,512 | 77 |
| Cixi (Ningbo) | 19,403 | 64.9% | 251,338 | 32 |
| Wujin District (Changzhou) | 16,944 | 38.9% | 295,376 | 38 |
| Bazhou (Langfang) | 16,129 | 88.4% | 27,131 | 0 |
| Yuyao (Ningbo) | 13,478 | 64.5% | 170,510 | 15 |
| Wujiang District (Suzhou) | 13,249 | 48.8% | 304,682 | 29 |
| Longgang District (Shenzhen) | 12,524 | 38.7% | 418,259 | 37 |
| Chao'an District (Chaozhou) | 12,378 | 66.4% | 51,761 | 2 |
| Jiangyin (Wuxi) | 12,376 | 33.7% | 283,872 | 23 |
| Kunshan (Suzhou) | 12,281 | 23.5% | 604,106 | 81 |
| Yongkang (Jinhua) | 11,537 | 61.0% | 88,888 | 10 |
| Zhuji (Shaoxing) | 11,490 | 61.5% | 82,177 | 7 |
| Wenling (Taizhou) | 11,442 | 41.9% | 127,878 | 11 |
| Haining (Jiaxing) | 10,493 | 48.2% | 139,675 | 6 |
| Jinjiang (Quanzhou) | 10,256 | 51.8% | 150,880 | 20 |
| Rui'an (Wenzhou) | 9,928 | 40.3% | 134,317 | 5 |
| Yuhuan (Taizhou) | 9,659 | 44.4% | 118,210 | 4 |
Table compiled by the Institute from this database, by district or county under the manufacturing definition, taking the top 21 by number of entities.
Two comparisons in this table deserve separate examination.
The first is Kunshan against Bazhou. Kunshan has 12,281 manufacturing entities, a zero-insurance rate of 23.5%, a total of 604,106 insured employees and 81 factories with 1,000 or more employees. Bazhou has 16,129 manufacturing entities, 31% more than Kunshan, but a zero-insurance rate of 88.4% and a total of 27,131 insured employees, only 4.5% of Kunshan's, and no factory with 1,000 or more employees.
More entities but less employment: these two county-level units represent two entirely different forms of industrial organisation. Manufacturing in Kunshan is dominated by export-oriented electronics and precision manufacturing, with a relatively small number of entities, large individual entities and a high degree of employment compliance—12,281 entities account for 604,000 insured employees, an average of 49.2 each. Manufacturing in Bazhou is dominated by metal rolling and metal furniture, with a large number of entities, extremely small individual entities and a low insurance ratio—16,129 entities together account for only 27,000 insured employees, an average of 1.68 each, and not one factory with 1,000 or more employees.
The second is Cixi and Yuyao against Shunde District and Nanhai District. Both are concentrations of the home appliance and hardware industries: Cixi has 19,403 entities, a zero-insurance rate of 64.9% and an average of 12.95 insured employees; Yuyao has 13,478 entities, 64.5% and an average of 12.65; while Shunde District has 23,523 entities, a zero-insurance rate of 30.9% and an average of 25.04 insured employees, and Nanhai District has 29,769 entities, 37.0% and an average of 13.64.
The zero-insurance rate in these two county-level cities in Zhejiang Province is clearly higher than in the two districts of Foshan in Guangdong Province. This difference is consistent in direction with the provincial comparison in Chapter 13 (Zhejiang 45.6% against Guangdong 42.2%), but its magnitude is greater at the county level. A possible explanation is the form of industrial organisation: the appliance supply chain in Cixi and Yuyao rests at its base on large numbers of household-workshop-style component processing sites, from which finished-product plants purchase; the appliance industry in Shunde and Nanhai, by contrast, consists mainly of finished-product plants and above-designated-size suppliers. This explanation is consistent with the data in this report, but this report has no transaction data between enterprises and cannot verify it further, so it can only be listed as a judgement awaiting testing.
From the standpoint of use, differences at the county level have more operational value than those at the provincial level. When a buyer looks for suppliers in a particular category, it usually already knows which counties that category is concentrated in; what then needs to be judged is the structure within that county. For the same "home appliance industry cluster", screening suppliers in Cixi and screening suppliers in Shunde present different candidate pools: Cixi's candidate pool is larger and contains more micro processing sites, requiring more on-site verification work; Shunde's candidate pool is smaller and individual plants are larger, but the number to choose from is correspondingly smaller.
The data in this chapter also point to a common error in county-level investment promotion and industrial research: using the number of entities to represent the scale of an industry. Ranked by number of entities, Bazhou stands eighth nationwide, above Kunshan; ranked by number of insured employees, Bazhou's scale is less than one-twentieth of Kunshan's. The "industry scale" indicated by the two indicators differs twentyfold, and they come from the same body of data.
Part Four Institutions and Compliance
18. 1999: Private Factories Become Contributing Units
The earliest systematic basis for the social insurance obligations of manufacturing entities in China is the Interim Regulations on the Collection and Payment of Social Insurance Premiums, promulgated on 22 January 1999 by State Council Order No. 259 (the current text was amended in accordance with the Decision of the State Council on Amending Certain Administrative Regulations of 24 March 2019). Interim Regulations on the Collection and Payment of Social Insurance Premiums
Article 3, paragraph 1, of those Regulations provides that the scope of collection of basic old-age pension insurance premiums shall be: "State-owned enterprises, urban collective enterprises, foreign-invested enterprises, urban private enterprises and other urban enterprises and their employees, and public institutions managed as enterprises and their employees." Ibid. This sentence expressly includes "urban private enterprises" within the scope. As far as manufacturing is concerned, this means that from 1999 onward private manufacturing plants were already within the scope of legally designated contributing units; the subsequent change was not a transition from absence to presence, but a gradual improvement in collection and administration capacity.
Article 3, paragraphs 2 and 3, of the same Regulations respectively provide for the scope of collection of basic medical insurance premiums and unemployment insurance premiums, in wording identical to that for old-age pension insurance. Article 4 establishes the mandatory obligation to contribute: "Contributing units and contributing individuals shall pay social insurance premiums in full and on time. The social insurance premiums collected shall be placed in the social insurance fund, shall be used exclusively for their designated purpose, and shall not be misappropriated by any unit or individual." Ibid.
Article 7 provides that "a contributing unit must apply to the local social insurance administration agency for social insurance registration and participate in social insurance"; Article 8 provides that "when an enterprise completes its registration, it shall complete social insurance registration at the same time". Ibid. These two articles link commercial registration and social insurance registration procedurally, and they are the institutional starting point that allows the number of insured employees field in today's enterprise database to correspond one-to-one with commercial-registration entities.
An easily overlooked detail is the range of insurance schemes covered. Article 2, paragraph 1, of the Regulations provides: "These Regulations shall apply to the collection and payment of basic old-age pension insurance premiums, basic medical insurance premiums and unemployment insurance premiums (hereinafter collectively referred to as social insurance premiums)." Ibid. In other words, the 1999 Regulations themselves covered only three insurance schemes. The collection of work injury insurance premiums and maternity insurance premiums was left by Article 29 to the decision of provincial-level people's governments: "The people's governments of provinces, autonomous regions and municipalities directly under the Central Government may, in light of local circumstances, decide that these Regulations shall apply to the collection and payment of work injury insurance premiums and maternity insurance premiums within their respective administrative regions." Ibid. The five social insurance schemes were not written into law as a whole until the Social Insurance Law of 2010.
The consequences of underpayment were already provided for in this version of the Regulations. Article 13: "Where a contributing unit fails to pay, or to withhold and pay on behalf of others, social insurance premiums in accordance with the provisions, the administrative department for labour security or the tax authority shall order it to pay within a specified period; where it still fails to pay after the expiry of that period, it shall, in addition to paying the amount in arrears, pay a late payment penalty of 2‰ per day from the date the arrears arose." Ibid. Article 23 provides that where a unit fails to complete social insurance registration, registration of change or deregistration, or fails to declare the amount payable, it shall be ordered to rectify within a specified period; where the circumstances are serious, a fine of not less than RMB 1,000 and not more than RMB 5,000 may be imposed on the directly responsible person in charge and other directly responsible persons; where the circumstances are especially serious, a fine of not less than RMB 5,000 and not more than RMB 10,000 may be imposed. Ibid.
The penalty provisions in this version had two features: the penalty was directed at individuals, and the monetary range did not vary with the scale of the arrears. An enterprise with tens of millions of RMB in social insurance arrears and an enterprise with tens of thousands of RMB in arrears faced the same range of personal fines under this article. That structure changed after 2011.
19. 2011: Registration Obligations and Penalties Established by the Social Insurance Law
The Social Insurance Law of the People's Republic of China was adopted on 28 October 2010 and came into force on 1 July 2011; the current text was amended in accordance with the amendment decision of 29 December 2018. Social Insurance Law of the People's Republic of China
Article 2 of the Law writes all five insurance schemes into law at once: "The State shall establish social insurance systems such as basic old-age pension insurance, basic medical insurance, work injury insurance, unemployment insurance and maternity insurance." Article 4 establishes a universal obligation to contribute: "Employers and individuals within the territory of the People's Republic of China shall pay social insurance premiums in accordance with the law." Ibid. Compared with the 1999 Regulations, there is no enumeration of industries here, nor is any room left for provincial governments to decide whether the law applies.
For the data in this report, the two most important articles are Article 57 and Article 58.
Article 57, paragraph 1: "An employer shall, within thirty days of its establishment, apply to the local social insurance administration agency for social insurance registration on the strength of its business licence, registration certificate or unit seal." Article 57, paragraph 3, provides: "The departments for market regulation, the departments for civil affairs and the organs for the administration of institutional establishment shall promptly notify the social insurance administration agencies of the establishment and termination of employers." Ibid.
Article 58, paragraph 1: "An employer shall, within thirty days of the date on which it begins employing a worker, apply to the social insurance administration agency for social insurance registration on that worker's behalf. Where social insurance registration has not been completed, the social insurance administration agency shall determine the social insurance premiums that the employer shall pay." Paragraph 3 provides: "The State shall establish a unified national individual social security number. The individual social security number shall be the citizen's identity number." Ibid.
Article 58, paragraph 3, is the key to understanding the data in this report. The fact that the individual social security number is the same as the citizen's identity number means that every insured person corresponds to a unique, non-duplicable identifier, and the number of people for whom an employer has completed insurance registration is a verifiable headcount within the system. That number subsequently enters the corporate annual report and is disclosed to the public. The credibility of the number of insured employees field used in this report ultimately derives from this article.
As to penalties, Article 84 provides that where an employer fails to complete social insurance registration, it shall be ordered to rectify within a specified period, and where it fails to rectify after the expiry of that period, "the employer shall be fined not less than one time and not more than three times the amount of social insurance premiums payable, and the directly responsible person in charge and other directly responsible persons shall be fined not less than RMB 500 and not more than RMB 3,000". Article 86 provides that where social insurance premiums are not paid in full and on time, "a late payment penalty of five ten-thousandths per day shall be added from the date the arrears arose; where payment is still not made after the expiry of the specified period, the relevant administrative department shall impose a fine of not less than one time and not more than three times the amount in arrears". Ibid.
Compared with the 1999 Regulations, there are two substantive changes here. First, the target of the penalty extends from the individual to the unit, and the fine is tied to the amount payable or in arrears rather than falling within a fixed range. Second, the standard for the late payment penalty changes from "2‰ per day" to "five ten-thousandths per day". 2‰ equals twenty ten-thousandths, and five ten-thousandths equals 0.5‰; the latter standard is one quarter of the former. The two rates must not be used interchangeably in calculating the cost of arrears; after 1 July 2011, Article 86 of the Social Insurance Law should govern.
Article 95 contains a further provision directed at the population that is the source of the data in this report: "Rural residents who come to the cities to work shall participate in social insurance in accordance with the provisions of this Law." Ibid. Manufacturing is one of the main fields covered by this article. Before 2011, localities differed in how they handled insurance participation by rural migrant workers; after this article, the obligation to participate no longer varies with household registration. Thereafter, the number of insured employees of a manufacturing entity should in theory cover all of its workers, irrespective of type of work or household registration.
A distance remains between the institution and the data. The law sets out obligations; the data record the results of compliance. Chapter 3 of this report has already shown that the zero-insurance rate among manufacturing entities is 53.3%. The gap between this figure and the mandatory character of the statutory provisions does not originate in the law, but in differences of recording, enforcement and the actual condition of enterprises. The three zero-insurance situations listed in Chapter 6 are the concrete form this gap takes in the data.
20. 2019: A Two-Step Move on Contribution Reduction and the Collection System
Social insurance policy in 2019 consisted of two moves in opposite directions: reducing the contribution rate on one hand, and changing the collecting authority on the other. The two moves were placed in the same document and were mutually conditioning.
The document is the Comprehensive Plan for Reducing Social Insurance Contribution Rates, finalised by the General Office of the State Council on 1 April 2019 and issued on 4 April 2019, document number Guobanfa [2019] No. 13 (国办发〔2019〕13号). Comprehensive Plan for Reducing Social Insurance Contribution Rates
On the reduction of contributions, Article 1 provides: "From 1 May 2019, the employer contribution rate for urban employee basic old-age pension insurance (including basic old-age pension insurance for enterprises and for government agencies and public institutions) shall be reduced. Where the employer pension insurance contribution rate of a province, autonomous region, municipality directly under the Central Government or the Xinjiang Production and Construction Corps is higher than 16%, it may be reduced to 16%; where it is currently lower than 16%, transitional measures shall be studied and proposed." Ibid. At a subsequent policy briefing, the Ministry of Human Resources and Social Security explained that before the reduction the employer pension insurance contribution rate in most provinces was 20% or 19%, and was 14% in Guangdong and Zhejiang; across the three schemes combined, the reduction would lighten the social insurance contribution burden on enterprises by more than RMB 300 billion over the whole of 2019. Social Insurance Contribution Reduction to Lighten the Burden on Enterprises by More Than RMB 300 Billion
On the contribution base, Article 3 changed the basis of calculation: "Each province shall determine the upper and lower limits of the individual social insurance contribution base using the full-calibre average wage of urban unit employed persons, weighted from the average wage of employed persons in urban non-private units and the average wage of employed persons in urban private units in that province, and shall reasonably lower the social insurance contribution base of some insured persons and enterprises." Comprehensive Plan for Reducing Social Insurance Contribution Rates
This change has an indirect bearing on the data in this report. Because the basis of calculation moved from considering only the average wage of urban non-private units to weighting private units in as well, the weighted average wage is lower than under the previous basis, and the lower limit of the contribution base is correspondingly lowered. The amount contributed falls, but the number of insured employees field is unaffected—it records a headcount, not an amount.
On the collection system, the wording of Article 6 is the core of this adjustment: "Contributions to enterprise employee basic old-age pension insurance and other schemes for enterprise employees shall in principle continue to be collected under the current collection system for the time being, so as to keep the method of contribution stable, and shall be transferred one province at a time as each province is ready; the responsibility for collecting and administering social insurance premiums for government agencies and public institutions and for urban and rural residents shall be transferred on schedule. ... Enterprises' historical arrears shall be handled properly; in the course of reforming the collection system, no locality may on its own initiative carry out a concentrated clearing of enterprises' historical arrears, and no practice that increases the actual contribution burden on micro and small enterprises may be adopted, so as to avoid causing difficulties in their production and operation." Ibid.
This article needs to be read against the background of the previous year. In July 2018, the General Office of the Communist Party of China Central Committee and the General Office of the State Council issued the Plan for Reforming the System for the Collection and Administration of National and Local Taxes, which proposed that from 1 January 2019 all social insurance premiums be collected uniformly by the tax authorities. That plan was not published in full, and its content can at present be recounted only on the basis of reports from the tax system and authoritative media; in using it, this report marks its source as reporting rather than the original document text. That arrangement gave rise at the time to widespread discussion of whether historical arrears would be cleared in a concentrated way, and it was subsequently adjusted in the 2019 Comprehensive Plan to a province-by-province approach.
The actual implementation came later than originally planned. At a press conference on 20 January 2021, the State Taxation Administration explained that "since November 2020, social insurance premiums have been collected in full by the tax authorities", and disclosed that in 2020 tax authorities nationwide collected RMB 3.8 trillion in social insurance premiums. State Taxation Administration press conference
From the institution to the data, the change brought about by this adjustment is collection capacity. Social insurance registration is handled by the social insurance administration agencies, but once the declaration and collection of premiums moved into the tax system, the conditions existed for cross-checking an enterprise's declared total wage bill, its individual income tax data and its social insurance contribution base. This change does not alter the definition of the number of insured employees, but it does alter how difficult it is for an enterprise to under-report the number of its employees.
Inter-Provincial Differences in the Contribution Base
The contribution-reduction policy retained inter-provincial differences in its implementation. The upper and lower limits of the contribution base are determined on the basis of each province's full-calibre average wage of urban unit employed persons, with the lower limit set at 60% of that average wage.
For the 2025 scheme year, Shanghai's monthly contribution base had a lower limit of RMB 7,460 and an upper limit of RMB 37,302, with a 2024 full-calibre average wage of urban unit employed persons of RMB 12,434 per month; Shanghai Municipal Human Resources and Social Security Bureau Beijing's lower limit was RMB 7,162 and its upper limit RMB 35,811; Beijing Municipal Human Resources and Social Security Bureau Guangdong applied two tiers, with a lower limit of RMB 5,510 for Guangzhou and provincial-level units and RMB 4,775 for other areas, an upper limit of RMB 27,549, and a pension calculation base of RMB 9,493 per month. Department of Human Resources and Social Security of Guangdong Province
Calculated at the lowest tier, the monthly contribution base for a worker contributing at the lower limit is 56.2% higher in Shanghai than in areas of Guangdong outside the designated tier. An enterprise with a factory of equal scale in each of the two places does not bear the same social insurance cost. This explains why the inter-provincial differences in the zero-insurance rate discussed in Chapter 14 of this report cannot be read directly as differences in levels of compliance—the cost of compliance itself differs between regions.
It should be noted that this report was unable to obtain the original official announcement of the Department of Human Resources and Social Security of Zhejiang Province on the upper and lower limits of the contribution base for the 2025 scheme year, and therefore does not list figures for that province.
21. 2025: Agreements Not to Pay Social Insurance Are Invalid
In 2025, the most important change on the social insurance question came from the judicial system rather than the administrative authorities.
The Interpretation (II) of the Supreme People's Court on Several Issues Concerning the Application of Law in the Trial of Labour Dispute Cases was adopted at the 1,942nd meeting of the Judicial Committee of the Supreme People's Court on 17 February 2025, promulgated on 31 July 2025, and came into force on 1 September 2025, document number Fa Shi [2025] No. 12 (法释〔2025〕12号), and comprises twenty-one articles in all. Interpretation (II) on Several Issues Concerning the Application of Law in the Trial of Labour Dispute Cases
Article 19 is the article most directly concerned with employment relations in manufacturing.
Paragraph 1: "Where an employer and a worker agree, or a worker undertakes to the employer, that social insurance premiums need not be paid, the people's court shall hold that agreement or undertaking invalid. Where an employer fails to pay social insurance premiums in accordance with the law and the worker, in accordance with the provisions of Article 38, paragraph 1, item 3 of the Labour Contract Law, requests termination of the labour contract and payment of economic compensation by the employer, the people's court shall uphold the request in accordance with the law."
Paragraph 2: "Where the circumstances in the preceding paragraph obtain and the employer, after making supplementary payment of social insurance premiums in accordance with the law, requests that the worker return the social insurance premium compensation already paid, the people's court shall uphold the request in accordance with the law." Ibid.
Taken together, these two paragraphs change the consequences of a common arrangement in manufacturing. The practice that previously existed in some factories was for the enterprise and the employee to agree not to pay social insurance and for the sum that should have been contributed to be paid directly to the employee as a cash subsidy. After Article 19 came into force, such an agreement is held invalid in litigation; where the employee terminates the labour contract on that basis and claims economic compensation, the court upholds the claim; and after the enterprise makes supplementary payment of social insurance, it may require the employee to return the compensation cash previously received.
Article 21 provides for entry into force and transitional linkage: "This Interpretation shall come into force on 1 September 2025. Article 32, paragraph 1, of the Interpretation (I) of the Supreme People's Court on Several Issues Concerning the Application of Law in the Trial of Labour Dispute Cases (Fa Shi [2020] No. 26; 法释〔2020〕26号) is repealed at the same time. Where judicial interpretations previously issued by the Supreme People's Court are inconsistent with this Interpretation, this Interpretation shall prevail." Ibid.
The Data in This Report Cannot Yet Reflect This Change
This point needs to be stated clearly, otherwise it will be misread.
The number of insured employees used in this report comes from corporate annual reports, and a corporate annual report discloses information for the preceding business year. The reference time point of the Fifth National Economic Census is 2023, and the filing period for the 2024 annual reports ran from 1 January to 30 June 2025. That is to say, the current number of insured employees of the entities in the database reflects, in the great majority of cases, the situation in the 2024 business year and earlier, whereas Article 19 came into force on 1 September 2025. All of the data in this report predate the entry into force of that article, and therefore this report does not include, and cannot be used to assess, the effects of the article after it came into force.
The implication for how the data should be read is this: the zero-insurance rate of 53.3% given in this report is the stock position as it stood before September 2025. Observing changes after the article came into force requires waiting until the 2025 annual reports have been filed before 30 June 2026 and the data have been updated. Two indicators can then be compared: whether the zero-insurance rate among manufacturing entities has fallen, and whether the number of entities in the 1 to 9 employee band has changed—if the earlier practice (substituting cash subsidies for social insurance) is widely corrected, those first affected will be the micro and small entities that previously did not complete social insurance registration.
Part Five Use and Boundaries
22. Three Layers of Sieves
The distributions established in the preceding eighteen chapters can be reduced to one operational question: given a specific procurement requirement, how can a limited verification cost be used to identify, from among 2,310,000 manufacturing entities, the few dozen that are worth visiting in person.
The method set out in this chapter consists of three layers of screening. The order of the three layers cannot be reversed, because the cost of each layer differs—the first two are completed on the data and cost close to nothing; the third requires manual verification and is expensive.
Layer One: Operating Status and the Foreign-Trade Flag
The starting point is all 2,313,658 manufacturing entities. Adding the "active" status condition leaves 2,048,859, a retention of 88.6%. This layer mainly filters out entities that have been deregistered, whose licences have been revoked or that have suspended operations; they remain in the enterprise registration database but no longer have the capacity to supply.
Adding the "foreign-trade" flag leaves 364,247, a retention of 17.8%.
This layer is the single most effective screening step in the whole procedure. It compresses the candidate pool from 2,050,000 to 360,000, a compression ratio close to 5.6 to 1, and what is retained is a subset with a markedly higher rate of employment compliance—Chapter 16 of this report shows that the zero-insurance rate is 15.6% for foreign-trade entities and 60.9% for non-foreign-trade entities.
It should be noted that the foreign-trade flag is not universally applicable to the requirement. For a buyer that supplies only the domestic market, adding this condition would exclude a large number of qualified suppliers. In that case this layer should be skipped and the second layer entered directly, at the cost of a larger candidate pool and a heavier follow-up verification workload.
Layer Two: The Insured Employees Threshold
Adding "number of insured employees ≥ 50" leaves 104,231, a retention of 28.6%; adding "number of insured employees ≥ 100" leaves 62,418, a retention of 59.9%.
For comparison, filtering by active status and "number of insured employees ≥ 100" without the foreign-trade flag yields 81,883, or 3.54% of all manufacturing entities. Filtering by "number of insured employees ≥ 300" alone yields 25,781, or 1.11%; filtering by "number of insured employees ≥ 1,000" alone yields 5,896, or 0.25%.
| Filter layer | Entities | Share of all manufacturing | Retention at this layer |
|---|---|---|---|
| Starting point: all manufacturing entities | 2,313,658 | 100.00% | — |
| ① Add "active" status | 2,048,859 | 88.55% | 88.6% |
| ② Add "foreign-trade" flag | 364,247 | 15.74% | 17.8% |
| ③ Add "number of insured employees ≥ 50" | 104,231 | 4.51% | 28.6% |
| ④ Add "number of insured employees ≥ 100" | 62,418 | 2.70% | 59.9% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database; measured values from layer-by-layer screening, with the conditions of the layers applied cumulatively.
The choice of threshold should be adjusted by industry. The data in Chapter 15 show that size distributions differ greatly across industries: in the processing of timber and the manufacture of wood, bamboo, rattan, palm and straw products, the 90th percentile is only 4 persons, so using 50 persons as the threshold would exclude almost every entity in that industry; in the manufacture of tobacco products, the median is 105 persons, so using 50 persons as the threshold is equivalent to applying no screening at all. The operational rule is: set the threshold by industry so that it falls between the 90th and 95th percentiles of that industry. In this way the compression ratio of each layer is roughly comparable, and no industry is screened out wholesale.
Layer Three: Consistency Checks Across Three Definitions
Once the first two layers are complete, the candidate pool is usually on the order of tens of thousands. The third layer is the verification of an individual entity, and what it checks is whether three definitions are consistent with one another.
The first check is the relationship between registered capital and contributed capital. Entities where the two are equal account for 52.80% of all manufacturing entities (Chapter 11); this proportion is not in itself anomalous; but when an entity has registered capital of several tens of millions of RMB, a number of insured employees in the tens, and contributed capital of zero, the actual state of the funds is worth confirming further.
The second check is the relationship between the number of insured employees and the date of establishment. An entity established recently whose number of insured employees already reaches several hundred usually corresponds to newly built capacity or to the relocation of an existing entity; an entity established twenty years ago whose number of insured employees is only in single digits usually corresponds to an entity that has retained its business licence while actual production has ceased, or to production carried on in a form that does not enter the entity's own roster.
The third check is the match between the number of insured employees and the business scope. Where an entity's business scope covers complete production processes while its number of insured employees has long been zero, its actual state needs to be confirmed on site; this is especially common in county-level industrial clusters.
From Data to the Site
Once the three layers have been applied, the remaining work is not in the data. Confirming manufacturing capability requires looking at equipment, at production lines, at inventory, at electricity bills and at the flow of people. The three zero-insurance situations listed in Chapter 6 of this report cannot be distinguished in the data; on site they can.
This method should therefore be understood as "using data to reduce the amount of on-site verification", not as "using data to replace on-site verification". The former is what data can do; the latter lies beyond the boundary of any business-registration or annual-report data.
23. Five Common Misuses
This chapter lists the five types of error most likely to arise in using the number of insured employees, each of which corresponds to a data fact already given above.
Misuse one: multiplying the average number of insured employees by the number of entities to estimate the total manufacturing employment of a region or of the whole country.
The number of insured employees of manufacturing entities totals 51,506,296 persons, the effective denominator is 2,313,658 entities, and the arithmetic mean is 22.26 persons. But the median is 0, the 75th percentile is 7 persons, and the mean falls between the 75th and the 90th percentile (Chapter 12). In a distribution whose median is zero and whose right tail extends beyond 9,000 persons, the mean is not a valid summary statistic. A result obtained by multiplying the mean by the number of entities will vary sharply according to whether the sample contains a small number of very large entities.
Misuse two: equating "a number of insured employees of zero" with "a shell company" or "no workers".
Zero insurance has three mutually distinct causes: the entity genuinely employs no workers who require insurance; the work is carried out in the form of labour dispatch, outsourcing or re-employment beyond statutory retirement age; and differences at the reporting stage (Chapter 6). The three causes appear as the same value in the data. In a county-level industrial cluster, a processing workshop in which five or six people operate a few machines and take on orders placed out by larger factories may perfectly well be registered as having zero social insurance, yet it genuinely exists in terms of capacity. Reading zero insurance as "non-existence" would cause such entities to be ignored as a whole.
Misuse three: comparing the absolute number of insured employees directly across provinces and across industries.
The zero-insurance rate ranges from 20.8% to 87.7% across provinces, a factor of 4.2 (Chapter 14); median employment ranges from 0 to 105 persons across industries (Chapter 15). The same factory of 20 persons counts as a relatively small entity in Zhejiang and already belongs to a larger class in Xizang; it falls in the top 10% in the timber-processing industry and in the bottom 10% in the tobacco products industry. Absolute values are comparable only under the conditions of the same province, the same industry and the same point in time.
Misuse four: treating the number of insured employees as an indicator of production capacity.
Capacity is determined by equipment, not by headcount. The output of a highly automated factory of 50 persons may exceed that of a semi-automated factory of 200 persons. In the data of Chapter 15, the median number of insured employees in the manufacture of computers, communications and other electronic equipment is 6 persons and the 90th percentile is 112 persons; in the manufacture of metal products the median is 0 and the 90th percentile is 19. The output value corresponding to a unit of employment differs greatly between these two industries, and the number of insured employees cannot be converted into capacity across industries.
Misuse five: using the number of insured employees to assess the compliance status of an individual enterprise.
This report describes distributions only; it does not assess individuals. The zero-insurance rate of 53.3% is an aggregate characteristic, but no state of any particular enterprise can be inferred from an aggregate characteristic. To make a compliance judgement about a single enterprise it is necessary to examine its social insurance contribution records, its labour contracts and its wage payment vouchers, and these lie outside the scope of public data. Applying a statistical distribution to the assessment of an individual is a use that lies beyond the data boundary of this report.
24. Questions This Report Does Not Resolve
The value of a report written from enterprise database data depends not only on which conclusions it offers, but also on how clearly it states which conclusions cannot be offered. This chapter lists four groups of questions that this report has not resolved, together with the reasons why they remain unresolved.
Group one: the precision of conversion to national aggregates.
Chapter 9 has already used data from the communiqué of the Ministry of Human Resources and Social Security and from the communiqué of the Fifth National Economic Census to give an order-of-magnitude comparison between the Institute's readings and national aggregates: the manufacturing entities in this database are equivalent to 57.1% of manufacturing enterprise legal entities, and the number of insured employees in this database is equivalent to 49.1% of the persons employed in manufacturing enterprise legal entities.
These two ratios are order-of-magnitude references, not conversion coefficients. Turning them into a usable conversion would require two things that this report does not have. First, a measured value for the coverage of this database's collection—which depends on the collection rules, since the Tianxia Gongchang enterprise database was built for factory retrieval and is not a mirror of the complete register of market entities (Chapter 3), and its coverage has not been systematically measured. Second, the conversion relationship between "number of insured employees" and "persons employed" for entities in this database—the 86.1% given in Chapter 9 is a hypothetical estimate whose validity rests on the untested premise that the entities collected in this database are consistent with the national structure in terms of size.
Until these two things have been supplied, all proportions in this report retain the qualifier "manufacturing definition of the Tianxia Gongchang enterprise database".
Group two: the share accounted for by each of the three causes of zero insurance.
Chapter 6 lists three possible causes of zero insurance—no actual employment, forms of employment that do not enter the entity's own roster, and reporting differences—and states that this report cannot distinguish among them at the individual level. This limits the strength of several conclusions in this report.
For example, the statement "the zero-insurance rate of manufacturing entities is 53.3%" is accurate, but it cannot be rewritten as "53.3% of manufacturing enterprises did not pay social insurance for their employees". The latter formulation presupposes that all zero insurance belongs to the second cause, and the data of this report do not support that presupposition. Distinguishing the three causes would require obtaining, at the enterprise level, social insurance contribution details, filings of labour contracts and wage payment records, and these lie outside the scope of public data.
The indirect method that can be adopted is sample-based field survey. The study cited in Chapter 7 does precisely this—matching field survey data against annual reports entity by entity, and arriving at the finding that 54% of enterprises "had social insurance in fact but recorded zero in their annual reports". The sample of that study was 2,687 matched enterprises in six provinces and municipalities; this report was unable to obtain a comparable study covering manufacturing, so the specific shares of the three causes within manufacturing remain unknown.
Group three: the extent to which recent data are affected by the progress of data collection.
The number of insured employees used in this report comes from corporate annual reports. A corporate annual report discloses information for the preceding business year, so the data in the database lag (Chapter 3). More importantly, the degree to which the Tianxia Gongchang enterprise database has completed collection for recently established entities may differ from the degree of completion for entities established earlier.
This effect becomes visible when entities are grouped by year of establishment: if recently established entities are underrepresented in the database, then in cohort comparisons by vintage (for example, "entities established after 2015 have a higher zero-insurance rate than entities established before 2000") the difference between the two groups will contain two sources at once: a genuine generational difference, and a difference in the completeness of collection.
This report does not use cohort comparisons by vintage in its main text, precisely in order to avoid this confounding. The related analysis is placed in "The Age of China's Factories", published by the Institute at the same time, where the question of data boundaries is treated separately.
Group four: the definition and threshold of the foreign-trade flag.
The foreign-trade flag used in Chapter 16 of this report is a binary field within the database; it distinguishes neither the share of exports nor the destination of exports. A factory that sells 95% of its products domestically and a factory that exports 95% of its products take the same value on this field.
The conclusion given in Chapter 16—that "foreign-trade entities account for 69.96% of employment"—rests on a binary flag. If the flag were subdivided into several grades (for example, by export share), the conclusion might change—a factory with an export share of 5% and a factory with an export share of 80% do not necessarily have the same distribution of size and of compliance. This report does not have such subdivided data.
What the four groups of questions above have in common is this: their common nature is an insufficiency of data dimensions rather than an insufficiency of data volume. Adding samples cannot solve them; what is needed are new fields or new methods of collection. This is also the general boundary encountered in using public enterprise data—the value of public data lies in its coverage, and its limitation lies in its depth.
Conclusion
The field of the number of insured employees deserves a report of its own because it records a different kind of fact.
Registered capital records intention; contributed capital records the fulfilment of a commitment; the number of insured employees records a behaviour that actually occurs every month. Each of the three kinds of fact has its own uses, but when an analyst has only the first two, he will systematically see a group of entities that "look large in scale"; once the third is added, part of that group displays a different appearance. The 3,869 entities in Chapter 11—those with registered capital above RMB 100 million whose annual reports disclose a number of insured employees of zero—are precisely what this is about.
This report revised one of its own conclusions in the course of writing. The landing point of the first draft was that "more than half of manufacturing entities did not pay social insurance for their employees". A study comparing enterprises one by one, introduced in Chapter 7, overturned that formulation: among enterprises confirmed by field survey to have social insurance, 54% recorded zero in their annual reports. This means that a considerable part of the 52.89% reading in this report came from reporting behaviour, not from insurance behaviour.
The revised conclusion is more useful than the one it replaced, because it shifts attention from "how many factories are non-compliant" to "what this figure actually records". What this report ultimately offers is three structural facts about the data themselves.
The first is the dual meaning of the figure 52.89%. In our enterprise database it marks the share of entities whose annual reports disclose a number of insured employees of zero; in reality it contains three situations at once: genuinely not insured, forms of employment that do not enter the entity's own roster, and insured but under-reported. An entity that "has no employees" in public data does not necessarily lack production capacity, nor does it necessarily lack social insurance—this must be taken as a premise in using this field, not as an exception.
The second is 15.86% and 69.96%. Foreign-trade entities account for less than one sixth of manufacturing entities, yet they account for nearly 70% of manufacturing employment. The gap between these two figures is the screening condition with the highest discriminating power in this report as a whole, and also the one relatively least affected by reporting error—the foreign-trade flag comes from import and export operation records and is not entered in the corporate annual report.
The third is 18.3% and 44.1%. Seven high-threshold industries account for 18.3% of manufacturing entities and 44.1% of manufacturing employment. Employment in Chinese manufacturing is more concentrated than its number of entities suggests; where attention should be placed depends on whether what is needed is coverage or capacity.
These three facts point to the same methodological judgement: every field in public enterprise data is accurate in one direction and distorted in another. The key to using them lies in knowing which kind of fact each field records, within what range it is reliable, and what one obtains after crossing its boundary—not in finding a single "best" field.
Chinese manufacturing has 2,310,000 registered entities. This figure by itself explains nothing. What is revealing is its shape—how many of them are genuinely organising production, how many people work along this chain, and how differently these questions are answered in different counties and in different industries.
Data Sources and Principal References
The proprietary data used in this report come from the Tianxia Gongchang enterprise database, with a reference point of 14 October 2026. The definitions of the sample, the rules for excluding outliers and three data defects have been set out in full in Chapter 3. All binned and grouped statistics in the text were calculated directly by the Institute on those definitions, without interpolation or estimation. Content concerning legislation corresponds to currently effective texts, and quotations of provisions follow the original text.
Principal references:
- Tianxia Gongchang industrial platform — Chinese factory database and industrial chain data
- Social Insurance Law of the People's Republic of China (2018 amendment), the Chinese government website
- Interim Regulations on the Collection and Payment of Social Insurance Premiums (State Council Order No. 259), the Chinese government website, State Council Gazette
- Notice of the General Office of the State Council on Issuing the Comprehensive Plan for Reducing Social Insurance Contribution Rates (Guobanfa [2019] No. 13; 国办发〔2019〕13号), the Chinese government website
- Interpretation (II) of the Supreme People's Court on Several Issues Concerning the Application of Law in the Trial of Labour Dispute Cases (Fa Shi [2025] No. 12; 法释〔2025〕12号), the Supreme People's Court
- Social Insurance Contribution Reduction to Lighten the Burden on Enterprises by More Than RMB 300 Billion, the Chinese government website, transcript of a policy briefing
- Press conference of the State Taxation Administration on the collection of social insurance premiums, reposted on the website of the National Development and Reform Commission
- Announcement of Shanghai Municipality on the Upper and Lower Limits of the Social Insurance Contribution Base for the 2025 Scheme Year, Shanghai Municipal Human Resources and Social Security Bureau
- Notice of Beijing Municipality on the Upper and Lower Limits of the Social Insurance Contribution Wage Base for the 2025 Scheme Year (Jing Ren She Fa [2025] No. 11; 京人社发〔2025〕11号), Beijing Municipal Human Resources and Social Security Bureau
- Notice of Guangdong Province on the Upper and Lower Limits of the Contribution Base for Employee Basic Old-Age Pension Insurance in 2025, Department of Human Resources and Social Security of Guangdong Province
- The Actual Enrolment of Employees of Chinese Private Enterprises in Old-Age Pension Insurance: Evidence from Online and Offline Micro-Surveys, the Journal of Economics and Management (Jingji Guanli Xuekan), 2026, Volume 5, Issue 1
- Report of the Law Enforcement Inspection Team of the Standing Committee of the National People's Congress on the Inspection of the Implementation of the Social Insurance Law of the People's Republic of China, the National People's Congress
- Survey of Social Insurance Enrolment Status, Center for International Social Security Studies, Chinese Academy of Social Sciences
- Notice on Issues Concerning the Addition of Social Insurance and Statistical Items to Corporate Annual Reports (Gong Shang Qi Jian Zi [2016] No. 226; 工商企监字〔2016〕226号), reposted by the Changsha Municipal Administration for Market Regulation
- Interim Regulations on Enterprise Information Disclosure (text issued in 2014; State Council Order No. 654), the Chinese government website
- Interim Regulations on Enterprise Information Disclosure (currently effective text), the National Database of Administrative Regulations
- Communiqué of the Fifth National Economic Census (No. 2): Basic Information on Units, the National Bureau of Statistics
- Statistical communiqué of the Ministry of Human Resources and Social Security, the Ministry of Human Resources and Social Security