Data cut-off: 14 October 2026
The age of a factory is written on its business licence.
This date—the founding date—is the least conspicuous column in commercial registration, and it is rarely taken out and examined on its own. When people discuss Chinese manufacturing they habitually use other kinds of figures: output value, export value, employment, the number of enterprises. These are flows or stocks, states at a particular point in time. The founding date is not. What it records is an event that happened on a certain day in the past and has not changed since. A factory founded in 1994 is thirty-two years old in 2026, no matter how many rounds of capacity expansion, product switching, production stoppage and resumption of production it has since gone through.
Line up the founding dates of 2.31 million existing manufacturing entities and the result is an age distribution curve. The question this curve answers is different from those answered by output value and by counts alike: the Chinese factories still operating today—from which years do they respectively come.
Of the 2,313,658 manufacturing entities recorded in the Tianxia Gongchang enterprise database, 106,672 were founded before 2000, accounting for 4.6%; 1,143,562 were founded after 2015, accounting for 49.4%. Half of the Chinese factories that exist today are less than twelve years old.
These entities are not all of one kind of organisation. Factories founded in the 1980s and factories founded in the 2020s display systematic differences in scale, in the degree of participation in foreign trade and in the discharge of social insurance contributions, and these differences vary monotonically with the age of the cohort. This set of comparisons is precisely the core of this report—it explains the change in the generational structure of Chinese manufacturing better than any single year's figure.
This report is in five parts. Part One deals with definitions and external anchors, including what the founding-date column itself records and the stock references provided by three National Economic Censuses. Part Two discusses the age curve since 1978 in chronological order, in eight chapters, of which Chapter 11 treats the data after 2018 separately—that segment contains a problem that must be set out clearly. Part Three compares entities of different generations in scale, foreign trade participation and employment patterns. Part Four takes the age structure down to two dimensions, geography and industry. Part Five discusses how this body of information may be used and where its limits lie.
The founding date has a property in the data that easily leads to misreading: what is counted is the entities that have "survived to today", not the entities that were "established in that year". When an enterprise is deregistered, its founding date disappears from the data along with it, though it once appeared in a particular year. Therefore, for a curve of existing entities arranged by founding year, the shape is determined by two variables at once: the number of establishments in that year, and the survival rate thereafter. These two variables cannot be separated out of this single curve.
This report does not estimate the number of enterprise establishments in historical periods, does not infer survival rates, and makes no assessment of any specific enterprise. All proprietary data in this report come from our enterprise database, extracted on 14 October 2026; the definitions and exclusion rules are consistent with "Chinese Factories in Social Insurance Insured Employee Numbers", published by the Institute in the same period, and are set out in detail in Chapter 3 of that report.
Part One: Definitions and External Anchors
1. A Stock Is Not a Flow
Before telling any story with a curve arranged by founding year, one must first make clear what it counts.
The enterprise registration database records the entities currently still on the register. A factory founded in 2005 and deregistered in 2018 is no longer in the data after 2018; it leaves behind only a founding date that has already vanished. Therefore, if the founding dates of existing entities are made into a histogram, what is obtained is not "how many factories were newly established in China in 2005" but "how many factories founded in 2005 were still on the register as of October 2026".
The difference between these two figures is the entire process of survival screening. It means the shape of the curve is determined jointly by two variables:
- how many entities were actually established in that year;
- the proportion of those entities that are deregistered, have their licence revoked or move out in each subsequent year.
These two variables cannot be separated in the data. A factory established in 2005 and deregistered in 2010 and another established in 2005 and still producing in 2026 carry the same value in the founding-year column; only because the latter of the two remains in the database does the former disappear from the statistics. The curve therefore converges naturally towards the early years—the further a year lies from today, the fewer the entities that remain and the more that have been eliminated, and the elimination is selective: those that survive are more likely to be the larger enterprises, those operating in a more standardised manner, and those more closely connected to external markets.
This regularity places a direct constraint on how this report is written. When cohorts are divided by decade of founding and their scale, foreign trade participation rate and social insurance participation rate are compared, the resulting differences contain two components at once: generational differences that already existed at the time of establishment, and survival differences left by years of subsequent screening. The two cannot be separated. In reporting such comparisons, this report describes the differences as "among existing entities, the cohort born in a given decade displays a certain characteristic" rather than "factories founded in a given decade have a certain characteristic"—the former is a fact in the data, the latter requires additional assumptions.
A change in the registration system that must be set out
The founding-date column itself underwent a change in its institutional environment around 2014, and this affects the interpretability of the curve.
Before 2014, establishing a company carried a minimum registered capital requirement. In February 2014 the State Council issued the "Reform Plan for the Registered Capital Registration System", abolishing the limits of a minimum registered capital of RMB 30,000 for a limited liability company, RMB 100,000 for a one-person limited liability company and RMB 5 million for a joint stock company, and replacing them with the subscription-based registered capital registration system, effective from 1 March 2014.
Thereafter, facilitation of registration continued to be stepped up: from 1 October 2015 the business licence, the organisation code certificate and the tax registration certificate were combined under the "three-in-one, one-licence-one-code" arrangement; from 1 October 2016 the social insurance registration certificate and the statistics registration certificate were merged in as well, under the "five-in-one" arrangement; and from 1 October 2017 the "multiple-in-one" arrangement was further introduced.
The common direction of this group of reforms was to lower the threshold and the time cost of establishing an enterprise entity. The effect on this report's data is this: establishment activity after 2014 is not a quantity directly comparable with that before 2014. The same real economic activity may, before the reform, have existed in the form of a self-employed business or an individual operation, and after the reform is more likely to be registered as a company. Part of the rise in the curve is a change in the form of registration, not an addition of production activity.
Another registration rule that requires attention
In 2025, 61,000 self-employed businesses nationwide completed registration to convert into enterprises through the tax authorities (January to October 2025, up 134.4% year on year), and by the end of 2025 the cumulative number of such "self-employed-to-enterprise" conversions nationwide was 358,000.
This registration has one key rule: the enterprise after conversion retains the founding date and the unified social credit code of the original self-employed business. That is to say, a self-employed business registered in 2010 that converts into a company in 2024 still shows in the data as founded in 2010. The effect of this rule is to make the reading for recent years too low rather than too high—entities that ought to have been added in recent years are counted in earlier years.
In using recent-year data, this report records this point as a known bias whose direction is clear.
2. The Stock Anchors Drawn by Three Economic Censuses
Our enterprise database is a part of China's manufacturing entities, not all of them. To judge how much it covers and where it is missing, an external anchor is needed. At present the most complete external anchor is the National Economic Census.
The economic census is held once every five years, and differs from the regular statistical annual reporting in that it is a full-scope inventory covering three kinds of entities—legal entities, establishments and self-employed businesses—rather than a sample survey. The results for the number of manufacturing legal entities in the three censuses are as follows.
| Indicator | End-2013 (Third Census) | End-2018 (Fourth Census) | End-2023 (Fifth Census) |
|---|---|---|---|
| Number of manufacturing legal entities | 2.252 million | 3.270 million | 4.049 million |
| Persons employed in manufacturing legal entities | — | 104.713 million | 104.815 million |
| Manufacturing self-employed businesses | — | 4.480 million | 5.123 million |
| Persons employed in manufacturing self-employed businesses | — | 16.374 million | 18.361 million |
Table compiled from the communiqués of the three National Economic Censuses. Communiqué of the Third National Economic Census, Communiqué of the Fourth National Economic Census, Communiqué of the Fifth National Economic Census
The first layer of information in these three sets of figures is that the stock is growing. Manufacturing legal entities rose over ten years from 2.252 million to 4.049 million, a cumulative increase of 1.797 million, or about 79.8%; of this, 2013 to 2018 added 1.018 million (about 45.2%) and 2018 to 2023 added 779,000 (about 23.8%). The growth rate is slowing, but the direction has not changed.
The second layer of information is the most noteworthy divergence in this entire report.
Manufacturing legal entities grew 23.8% over five years, while persons employed in manufacturing legal entities rose over the same five years only from 104.713 million to 104.815 million, an increase of 102,000, or 0.1%. Over the same period, persons employed in all legal entities in the secondary and tertiary industries rose from 383.236 million to 428.984 million, a growth of 11.9%, and the increment contributed by manufacturing was close to zero.
The wording of this comparison requires caution. "More factories, no more people" is an accurate reading, but it is not equivalent to "manufacturing is shrinking". Over the same period, persons employed in manufacturing self-employed businesses rose from 16.374 million to 18.361 million, an increase of 1.987 million, and this increment falls outside the scope of the statistics on persons employed in legal entities. At the same time, the number of persons employed in legal entities is a stock at the census reference date and does not reflect changes in output per person, in the level of automation, or in the labour intensity per unit of output value.
A safe conclusion is this: between 2018 and 2023, what increased in Chinese manufacturing was the number of entities, not the employment scale of legal entities. The 779,000 newly added manufacturing legal entities brought, on average, no observable net employment increment. This conclusion is directly related to the cohort analysis to be set out in Chapter 12 of this report—the size structure of newly added entities is entirely different from that of the entities formed at the beginning of this century.
Three cautions on definitions
In using the above data, three definitional points must be stated clearly, or the conclusions will be wrong.
First, "industrial enterprises above designated size" are not "manufacturing legal entities". The definition of industrial enterprises above designated size is industrial legal entities with annual main business revenue of RMB 20 million or more, numbering 517,300 in 2024. They are only a subset of manufacturing legal entities (4.049 million, end-2023), and they cover industry without including self-employed businesses. The two differ by nearly eight times, and any treatment that places them on the same curve is wrong.
Second, the versions of the industry classification differ across the three censuses. The Third and Fourth Censuses used the "Industrial Classification for National Economic Activities" (GB/T 4754—2011) and the Fifth used GB/T 4754—2017, with adjustments to the assignment of small classes within sections. All three use the manufacturing section, so the trends are comparable, but comparison at the level of year-by-year decimal places is not.
Third, legal entities, establishments and self-employed businesses are three different denominators. A single legal entity may have several establishments (branch plants, workshops, sales outlets), and self-employed businesses include both registered self-employed businesses and operators that are unregistered but have carried on business at a fixed location for three months or more. In citing them, this report marks the definition in each case and does not calculate across definitions.
This database compared with the census stock
The Tianxia Gongchang enterprise database records 2,313,658 manufacturing entities (for the definition, see Chapter 3 of "Chinese Factories in Social Insurance Insured Employee Numbers", published by the Institute in the same period). Against the 4.049 million manufacturing legal entities of the Fifth National Economic Census at end-2023, this database is about 57% of that figure. Against the 4.480 million to 5.123 million manufacturing self-employed businesses, this database focuses on legal entities, and its coverage of self-employed businesses is limited.
The 57% is a rough comparison, not a coverage rate. The definitions of the two are not entirely consistent: this database takes manufacturing entities registered with the commercial registration authority as the basis for selection, while the census's "legal entities" also include types of units that are not required by law to complete commercial registration; the industry classification of the two may also differ. The purpose of giving this figure is to indicate the order of magnitude of this database—it covers more than half of China's manufacturing entities, so the distribution characteristics within it are representative; at the same time it does not cover all of them, so absolute values within it cannot be used directly as national totals.
3. The Reliability of the Founding-Date Column
The analysis throughout this report rests on a single field: the founding date. Before discussing what history it records, it is necessary to make clear first what it does not record.
The founding date is the date on which a market entity obtains legal-person status or business status, and is the legal moment of birth of the entity. It is not the same thing as three other dates with which it is easily confused: the date business commenced (the time of actual production or opening), the date of moving into the locality (the time at which the entity began operating at a new address after relocating between regions), and the date of approval of the current business scope. In reality these four dates may be years or even decades apart.
For a factory that moved in from another province in 2019, if its legal-person status continued from the original address, then its founding date in the data is the year it was established at the original address—possibly 1998. Its operating history in the place it moved to is only seven years, but in the data it is twenty-seven years old. All judgements concerning moves in and moves out in Chapters 15 and 17 of this report are subject to this limitation.
The founding date becomes disconnected from industrial history in the following situations.
The first: continuation and resetting of legal-person status. When an enterprise is restructured, reorganised, split or merged, which path is taken directly changes the reading of the founding date. If the original entity is deregistered and a new entity takes over the assets and business, the founding date is reset to the year of restructuring; if the original entity's status is retained and only the shareholders and the name are changed, the founding date continues. In the restructuring of collective and state-owned enterprises in the late 1990s, both paths were widely used, and in differing proportions across regions and industries. When Chapter 18 of this report discusses the difference in age structure between Northeast China and southern Jiangsu, it is precisely this mechanism that is proposed, but this report has no annual data on restructuring paths and cannot verify it.
The second: cross-regional relocation. When an enterprise registers a relocation it normally retains the original founding date and the unified social credit code. The effect of this rule is that the destination gains a set of entities whose "age is greater than their actual operating history in the locality", while the origin loses a set. For the movement of people and production capacity between manufacturing clusters, this effect cannot be ignored. This report has no records of moves in or moves out.
The third: registration of self-employed businesses converting into enterprises. Chapter 1 of this report has explained that an enterprise after "self-employed-to-enterprise" conversion retains the founding date of the original self-employed business. This registration makes the reading for recent years too low.
The fourth: retroactive registration and legacy entities. For some entities the founding date derives from a history predating the establishment of the registration system, and its accuracy depends on the registration authority's assessment of the original documents. The 518 entities founded before 1978 in this report's data fall within this category. These 518 are very few in number and have no effect on the conclusions, but the nature of their dates differs from that of entities founded after 2010.
The fifth: the interval between the founding date and the time production actually begins. A company founded in 2023 may only be completed and put into production in 2025. This interval is especially pronounced in industries that require long construction cycles—the plant construction cycle in chemicals, metallurgy and large equipment manufacturing is often two to three years. The age structure in this report therefore mixes two points in time, "the birth of the enterprise" and "the formation of production capacity", and the interval between the two differs across industries. That capital-intensive industries are older in Chapter 16 comes partly from their smaller number of newly established entities and may also come partly from this interval—within the same observation period, their capacity formation lags behind the birth of the entities.
The above five situations cannot be distinguished in the data, and together they determine how this report should phrase its conclusions.
The founding date records the moment of birth of the legal entity, not the starting point of production and operation. Therefore:
- It is suitable for describing the composition of existing entities—the years in which the Chinese factories still operating today respectively obtained their legal-person status.
- It is not suitable for inferring the time at which production capacity was formed, nor for inferring the time at which an industry began in a given place. The latter requires the date business commenced, production records or output value data.
- When comparing age structures between regions, it must be remembered that part of the difference may come from differences in relocation and restructuring paths, and not only from differences in establishment behaviour.
The purpose of this chapter is to set an upper bound on all the comparisons that follow. Every cohort comparison and every provincial comparison in Parts Three and Four of this report should be read subject to these five limitations.
Part Two Forty Years of the Age Curve
Chapter 4. The Age Structure of This Database
The Tianxia Gongchang enterprise database contains 2,313,658 manufacturing entities, tabulated in segments by date of establishment as follows.
| Period of establishment | Entities | Share | Cumulative share |
|---|---|---|---|
| Before 1978 | 518 | 0.02% | 0.02% |
| 1978—1989 | 11,602 | 0.50% | 0.52% |
| 1990—1999 | 94,552 | 4.09% | 4.61% |
| 2000—2007 | 398,506 | 17.22% | 21.83% |
| 2008—2014 | 664,918 | 28.74% | 50.57% |
| 2015—2019 | 804,758 | 34.78% | 85.36% |
| 2020—2026 | 338,804 | 14.64% | 100.00% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing sector, tabulated in segments by date of establishment; data retrieved on 14 October 2026.
The first way to read this set of figures is the median age. Ordering all entities by date of establishment, the one falling in the exact middle was established in 2014—to be precise, the cumulative total to the end of 2013 is 1,037,019, the cumulative total to the end of 2014 is 1,170,096, and half of the total number of entities is 1,156,829. The median year of establishment falls around the eleventh month of 2014.
The median age of surviving Chinese manufacturing entities is approximately twelve years (calculated precisely at the data cut-off, 11.9 years). Half of all factories were established after 2014.
The second way to read them is the two ends. Entities established before 2000 number 106,672, or 4.61%; those established before 1990 number only 12,120, or 0.52%; those established before 1978 number only 518. At the other end, entities established after 2015 number 1,143,562, or 49.43%.
Taken together, the two ends form this structure: of the Chinese factories still operating today, one in two was established after 2015; while those established before 1990, more than thirty-six years old, number fewer than one in two hundred.
This shape is consistent with the survival logic set out earlier. Entities established before 1990 have by now passed through more than thirty years of selection; of them, those that lacked scale, that were poorly run, and that were eliminated as the industry changed have for the most part been deregistered. The 12,120 that remain are the survivors of that era.
Attention must be paid to the direction of causation here. The fact that only 12,000 entities were established before 1990 does not warrant the inference that "there were very few Chinese factories before the 1990s". There were of course large numbers of factories in China before 1990—a large part of them were state-owned and collective enterprises that withdrew from the register during the restructuring, mergers and bankruptcies from the late 1990s into the 2000s; another part was never registered in the form of a modern company; and another part exists in the data in the form of new entities created after restructuring, with the date of establishment reset to the year of restructuring. The cohort comparison discussed in Chapter 4 of this report concerns precisely this group of survivors, not all the entities that existed at the time.
Chapter 5. 1978—1991: Fourteen Years, 17,000 Entities
Listing each year before 1992 separately yields the flattest stretch of this curve.
| Year | Entities | Year | Entities |
|---|---|---|---|
| Before 1978 | 518 | 1985 | 1,077 |
| 1978 | 121 | 1986 | 848 |
| 1979 | 290 | 1987 | 978 |
| 1980 | 777 | 1988 | 1,383 |
| 1981 | 783 | 1989 | 3,423 |
| 1982 | 538 | 1990 | 3,391 |
| 1983 | 357 | 1991 | 2,466 |
| 1984 | 1,027 | — | — |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing sector, annual counts.
The fourteen years from 1978 to 1991 total 17,459, and adding the 518 from before 1978 gives 17,977. The annual readings in this stretch remained in the three digits for a long period, the lowest being 121 in 1978 and 357 in 1983.
Two features of this curve require explanation.
The first feature is that it does not reflect the actual number of factories set up at the time. From 1978 to 1991, China's township and village enterprises went through the fastest phase of development in their history, and township and individual industry appeared in large numbers in southern Jiangsu, Wenzhou and the Pearl River Delta. The great majority of these entities were not registered as "companies" in today's sense—the registration system of the time rested on several regulations that preceded the Regulations on the Administration of Registration of Enterprise Legal Persons, and township collective enterprises, affiliated enterprises and self-employed businesses were the main forms. The data in this report cover only the portion that survived thereafter and still existed as legal entities as of 2026, so the low readings in this stretch do not mean that production activity at the time was sparse.
The second feature is a step-like rise around 1989: 1,383 in 1988, 3,423 in 1989 and 3,391 in 1990, two to three times the levels of the preceding years.
The Regulations on the Administration of Registration of Enterprise Legal Persons were promulgated by State Council Decree No. 1, took effect on 1 July 1988, and repealed three earlier registration regulations at once. The jump in the 1989 reading is close in time to the entry into force of the regulations that year, but this report cannot assert causation on that basis. At least two explanations are equally compatible with the data: one is that after the registration system was unified, entities that had previously existed in other forms entered the system of legal-person registration; the other is that there was indeed an increase in newly established entities in this period. This report lacks the historical registration data needed to distinguish between the two explanations, and therefore records this jump only as a feature of the data.
From 1989 to 1991 the reading fell back from 3,423 to 2,466. This fall likewise admits several explanations, and this report makes no judgment on it.
Chapter 6. 1992—2001: The First Sustained Rise
From 1992 the curve entered a sustained rise lasting ten years.
| Year | Entities | Year | Entities |
|---|---|---|---|
| 1992 | 5,296 | 1997 | 12,207 |
| 1993 | 7,705 | 1998 | 16,519 |
| 1994 | 7,690 | 1999 | 19,962 |
| 1995 | 9,374 | 2000 | 29,989 |
| 1996 | 9,942 | 2001 | 37,136 |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing sector, annual counts.
Over the ten years the reading rose from 5,296 to 37,136, a six-fold increase. This stretch is the only one in all the data of this report that approximates exponential growth. Two points between 1992 and 2001 deserve to be marked out: first, the 1992 reading exceeded five thousand for the first time, more than four times the level of the 1980s; second, the 2000 reading approached thirty thousand for the first time, more than three times that of 1995.
Within this stretch of data, the starting point in 1992 has a clear policy background. In the spring of 1992, Deng Xiaoping made his inspection tour of the South and delivered his talks there; in the same year, China's reform and opening entered a new round of expansion. Five years later, in 1997, the 15th National Congress of the Party established the basic economic system under which public ownership is the mainstay and diverse forms of ownership develop side by side, and made clear that the non-public sector of the economy is an important component of the socialist market economy. Both of these events were favourable in direction to the registration of private entities.
It should likewise be noted that the data in this report are insufficient to quantify the effects of these policies. The curve rises during this period, but the reasons for the rise include three sources: improvements in the registration system, changes in statistical definitions, and real economic activity. This report presents only the curve itself.
What this 1990s stretch has left behind, ten years on, is a specific group of entities. Chapters 9 and 10 will show that among surviving entities, those established in the 1990s are markedly larger, and have markedly higher foreign trade participation rates and social insurance enrolment rates, than later cohorts. They are the oldest and also the most stable layer of Chinese manufacturing today.
Chapter 7. 2002—2008: Seven Years of Slowing Growth
In December 2001 China formally joined the World Trade Organization. Over the following seven years, the annual readings in the data of this report are as follows.
| Year | Entities | Year-on-year | Year | Entities | Year-on-year |
|---|---|---|---|---|---|
| 2002 | 44,918 | +21.0% | 2006 | 61,813 | +15.3% |
| 2003 | 53,737 | +19.6% | 2007 | 61,128 | −1.1% |
| 2004 | 56,143 | +4.5% | 2008 | 58,160 | −4.9% |
| 2005 | 53,642 | −4.5% | — | — | — |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing sector, annual counts; year-on-year changes calculated by the Institute from the available data.
These seven years provide a reading that differs from intuition. Joining the WTO is usually described as a key juncture in the expansion of Chinese manufacturing, yet in the data of this report, over the six years from 2002 to 2007 the annual reading rose from 44,918 to 61,128, a cumulative increase of 36.1%, or about 5.3% a year on average; in 2008 it fell back to 58,160. This pace is lower than that of the preceding ten years (a six-fold increase over 1992—2001) and also lower than that of the following ten years.
There are two explanations for this phenomenon, and they cannot be distinguished on the data of this report.
The first explanation is that what the data in this report reflect is the outcome of survival, not the number established in a given year. Factories established between 2002 and 2008 have by 2026 passed through eighteen to twenty-four years of selection. This period was precisely the phase in which competition in Chinese manufacturing was fiercest, capacity expanded fastest, and subsequent elimination was most concentrated—the contraction of external demand after the 2008 international financial crisis and the campaign against overcapacity around 2015 both left their mark on this group of entities.
The second explanation is that industrial expansion in this period was realised mainly in the form of capacity expansion by existing entities, rather than in the form of the number of newly established entities. After joining the WTO, the rapid growth of export orders led existing coastal factories to accelerate capacity expansion, add production lines and extend working hours; at the same time, labour from the interior concentrated in the coastal regions rather than setting up new factories locally. This explanation is consistent with the low establishment counts in the central and western provinces in the late 2000s seen in the data, but this report has no year-by-year data on capacity and employment with which to verify it.
Both explanations point to the same conclusion: the number of newly established entities is not a suitable indicator of manufacturing expansion. 2002 to 2008 was one of the fastest periods of growth in Chinese manufacturing, and at the same time a period in which the growth rate of the number of newly established entities was relatively low. Using the number of entities as a proxy variable for industrial scale yields a wrong reading for this stretch of time.
Chapter 8. 2009—2014: The Rise After the Crisis
The 2009 reading was 72,928, an increase of 25.4% over 2008. It then rose for five consecutive years: 91,208 in 2010, 96,370 in 2011, 95,210 in 2012, 117,965 in 2013 and 133,077 in 2014.
| Year | Entities | Year-on-year |
|---|---|---|
| 2009 | 72,928 | +25.4% |
| 2010 | 91,208 | +25.1% |
| 2011 | 96,370 | +5.7% |
| 2012 | 95,210 | −1.2% |
| 2013 | 117,965 | +23.9% |
| 2014 | 133,077 | +12.8% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing sector, annual counts.
The jump in 2009 coincides in time with the implementation of the RMB 4 trillion investment programme, but here too this report cannot assert causation on that basis. The rise in this period has three possible sources: an increase in the entities actually newly established, the advance of registration facilitation, and an improvement in survival rates. The two consecutive years of high growth in 2013 and 2014, on the other hand, are directly adjacent to the subscription-based registered capital registration system that took effect on 1 March 2014.
The growth rates of 23.9% in 2013 and 12.8% in 2014 deserve to be placed side by side. The subscription-based registered capital registration system took effect on 1 March 2014, so its effect on the full-year data should appear mainly in 2014, yet the growth rate in 2013 is higher than that in 2014. This shows that the causes of the growth lie in more than one factor. Other measures to facilitate enterprise registration were also introduced one after another around 2013, and this report does not have sufficient information to separate these sources.
In the cohort analysis of this report this stretch corresponds to the 2008—2014 group, 664,918 in total, or 28.74% of all surviving entities. Their ages range from twelve to eighteen years, and they are the second-largest group among surviving Chinese factories.
The Current Profile of This Generation of Entities
The 664,918 entities established between 2008 and 2014 account for 28.74% of all surviving manufacturing entities and form the second-largest group. Their present characteristics fall between those of the two generations.
The median number of insured employees is 0, and the 90th percentile is 29; the zero-insurance rate is 51.6%; insured employees total 13,278,943, or 25.8% of all insured employees in manufacturing; 1,323 entities have more than 1,000 insured employees, 0.2% of this cohort; the foreign trade participation rate is 16.9%; 11.3% have a registration status other than active.
Placing this set of figures alongside the two adjacent cohorts:
| Indicator | 1990—1999 | 2008—2014 | 2015—2019 |
|---|---|---|---|
| Number of entities | 94,552 | 664,918 | 804,758 |
| 90th percentile of insured employees | 130 | 29 | 15 |
| Zero-insurance rate | 30.4% | 51.6% | 61.9% |
| Foreign trade participation rate | 35.2% | 16.9% | 9.6% |
| Share with 1,000 or more insured employees | 1.3% | 0.2% | 0.1% |
| Non-active rate | 5.9% | 11.3% | 14.9% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing sector, by period of establishment.
The values of the five indicators in the table—the 90th percentile, the zero-insurance rate, the foreign trade participation rate, the share of entities with 1,000 or more insured employees, and the non-active rate—all fall between those of the other two cohorts for this cohort, and which end they are closer to is not consistent: the foreign trade participation rate of 16.9% sits about midway between 35.2% and 9.6%, whereas the share with 1,000 or more insured employees, 0.2%, and the non-active rate, 11.3%, are closer to the post-2015 cohort.
The significance of this set of readings is that it shows that the generational difference described in this report is not a single abrupt change but a continuous movement. Factories established between 2008 and 2014 stand between the older and the newer generation on every dimension, and no dimension shows a discontinuity. Any approach that divides Chinese manufacturing into two classes, "old plants" and "new plants", loses resolution over this stretch.
Chapter 9. 2015—2017: The Peak of the Curve
The 2015 reading was 133,204, 2016 was 157,308 and 2017 was 234,699. 2017 is the highest point in all the data of this report.
| Year | Entities | Year-on-year |
|---|---|---|
| 2015 | 133,204 | +0.1% |
| 2016 | 157,308 | +18.1% |
| 2017 | 234,699 | +49.2% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing sector, annual counts.
The year-on-year growth rate in 2017 was 49.2%. This growth rate ranks second among the years since 2000; the only one higher is 50.23% in 2000 (rising from 19,962 to 29,989). The manufacturing entities established in that year and still on the register today number 234,699, or 10.15% of all surviving entities—one in ten surviving Chinese factories was established in 2017.
This stretch is dense with policy junctures. The subscription-based registered capital registration system took effect on 1 March 2014; the "three-in-one" business licensing reform and "one licence, one code" on 1 October 2015; the "five-in-one" reform on 1 October 2016; and the "multiple-in-one" reform on 1 October 2017. Within four years, the material, procedural and time costs of establishing and registering an enterprise entity fell continuously.
The 2017 figure can be checked against official statistics, which is possible for only a few of the years covered in this report.
Data published by the State Administration for Industry and Commerce show that 518,000 manufacturing enterprises were newly registered nationwide in 2017, a year-on-year increase of 16.3%. State Administration for Industry and Commerce, 2017 annual data In the preceding year, 446,000 manufacturing enterprises were newly registered nationwide in 2016, a year-on-year increase of 16.9%. Ibid.
Placing the two sets of figures side by side:
| Year | Newly registered manufacturing enterprises nationwide (official) | Surviving entities established that year in this database | Ratio |
|---|---|---|---|
| 2016 | 446,000 | 157,308 | 35.3% |
| 2017 | 518,000 | 234,699 | 45.3% |
Table compiled by the Institute by comparing public data with the data of this database.
What these two ratios mean is that of the manufacturing enterprises newly registered in 2016, 35.3% still appear in the Tianxia Gongchang enterprise database as of 2026; the corresponding proportion for 2017 is 45.3%. This proportion reflects two processes at once—the deregistration and elimination of the following nine to ten years, and the completeness of this database's coverage of surviving entities. This report cannot separate the two, and therefore this ratio cannot be read as a survival rate, nor as a coverage rate. Its value lies in providing an order-of-magnitude reference: the coverage by this database of entities established in 2016 and 2017 lies within the upper half of the same order of magnitude.
One difference in definition should be noted: the official "newly registered manufacturing enterprises" are classified by the type of enterprise registration, whereas "manufacturing entities" in this database are classified by Section C of the Industrial Classification for National Economic Activities and have undergone industry classification; the two do not correspond exactly. This difference affects the specific value of the ratio but does not change the order of magnitude.
The peak in 2017 itself also requires explanation. Nationwide, newly registered manufacturing enterprises grew 16.3% year on year that year, whereas the 2017 reading in this database grew 49.2% year on year, a three-fold difference between the two. This gap shows that a considerable part of the jump in the 2017 reading in this database came not from an increase in the actual number established, but from a change in the form of registration—the subscription-based registered capital system and the "multiple-in-one" reform led a group of entities that had originally existed as self-employed businesses or in other forms to be registered as company-form enterprises around 2017. In the data these entities appear as additions in 2017, while in reality they had long existed.
Chapter 10. The Number of Industrial Enterprises Above Designated Size: First Falling, Then Rising
Beyond the data of this database there is another independent stock series, one whose time span is much longer than that of the economic census and which can be consulted for every year: the number of industrial enterprise units above designated size.
The definition of industrial enterprises above designated size is industrial legal entities whose annual revenue from principal business reaches a certain threshold. That threshold changed once, in 2011—raised from RMB 5 million to RMB 20 million. Under the current definition, industrial enterprises above designated size are industrial legal entities with annual revenue from principal business of RMB 20 million or more; the threshold was raised from RMB 5 million in 2011. National Bureau of Statistics, China Statistical Yearbook, Table 13-3
| Year | Number of enterprise units | Year | Number of enterprise units |
|---|---|---|---|
| 1998 | 165,100 | 2015 | 383,100 |
| 2005 | 271,800 | 2016 | 378,600 |
| 2008 | 426,100 | 2017 | 372,700 |
| 2009 | 434,400 | 2018 | 375,000 |
| 2010 | 452,900 | 2019 | 377,800 |
| 2011 | 325,600 | 2020 | 399,400 |
| 2012 | 343,800 | 2021 | 441,500 |
| 2013 | 369,800 | 2022 | 472,000 |
| 2014 | 377,900 | 2023 | 493,200 |
| — | — | 2024 | 517,300 |
Table from the National Bureau of Statistics, China Statistical Yearbook, Table 13-3, "Main Indicators of Industrial Enterprises Above Designated Size", selected years. The precise figure for 2025 was not available to this report.
Two points about this series must be set out.
The first is the cliff in 2011, and it is not business failures. The reading fell from 452,900 in 2010 to 325,600 in 2011, a decrease of 127,300, or 28.1%. The reason for the decrease is not that 127,000 enterprises disappeared, but that after the threshold was raised, enterprises with annual revenue from principal business between RMB 5 million and RMB 20 million left the statistical scope as a whole. The period around 2011 in this series is not directly comparable, and any year-on-year calculation that crosses 2011 is wrong.
The second is the net decrease for three consecutive years from 2015 to 2017. 383,100, 378,600, 372,700—this is the only interval in this series, after the definition was fixed in 2011, in which the count fell for three consecutive years, a cumulative decrease of 10,400. The period around 2015 was precisely when the campaign against overcapacity was being advanced. The official figure recorded in Chapter 18 of this report, that ferrous metal smelting and rolling processing fell 42.0% over five years, is one industry-side aspect of this phase.
Over the following seven years the reading rose continuously: 375,000 in 2018, 377,800 in 2019, 399,400 in 2020, 441,500 in 2021, 472,000 in 2022, 493,200 in 2023 and 517,300 in 2024. Over the seven years the cumulative increase was 142,300, or 37.9%.
This series and the data of the first nine chapters of this report form a complementary pair of evidence.
The curve in this database, tabulated by year of establishment, describes the age composition of surviving entities; the number of industrial enterprises above designated size describes the number of entities above designated size on the register in each year. The two differ in definition—the former covers manufacturing entities across all categories (including those below designated size and self-employed businesses), while the latter covers only legal entities in industry that reach the revenue threshold—but they point in the same direction.
To state that direction clearly: in 2024 there were 517,300 industrial enterprises above designated size, and at the end of 2023 there were 4,049,000 manufacturing legal entities (see Chapter 2); the former equals 12.8% of the latter. That is to say, close to nine tenths of manufacturing legal entities do not appear in the statistics above designated size. This is consistent with the census data cited in Chapter 2—manufacturing legal entities had 104,815,000 persons employed, while the average number of employed persons in manufacturing above designated size was 69,455,300, and the difference of 35,359,700 all falls in enterprises below designated size and self-employed businesses.
In other words: the number of manufacturing entities in China is made up for the most part of enterprises below designated size and self-employed businesses, while nearly two thirds of manufacturing employment is in enterprises above designated size. The centre of gravity of numbers and the centre of gravity of employment fall in two different parts.
This structure provides an answer to the question raised earlier in this chapter. Between 2018 and 2023 manufacturing legal entities increased by 23.8% while persons employed increased by only 0.1%; this combination of "more plants, no more people" is not contradictory—the great majority of the newly added entities fall below designated size, and their average employment is far smaller than that of enterprises above designated size, so growth in the number of entities does not bring a proportional growth in employment.
It should be noted that the year-on-year changes in the number of industrial enterprises above designated size are likewise affected by adjustments to the statistical scope. The roster of enterprises above designated size changes each year as enterprises cross the threshold or exit, so the entity counts of two adjacent years do not correspond to exactly the same set of enterprises. National Bureau of Statistics This series can therefore be read for trend, but precise growth rates should not be calculated year by year.
Chapter 11. After 2018: This Segment Cannot Be Read as a Decline
The preceding eight chapters unfold in chronological order and met with no obstacle; here one appears. This section must first deal with a data problem before discussing substance, otherwise all the subsequent conclusions about recent years will be unreliable.
Look at the data first.
| Year | Entities | Year-on-year |
|---|---|---|
| 2017 | 234,699 | +49.2% |
| 2018 | 151,642 | −35.4% |
| 2019 | 127,905 | −15.7% |
| 2020 | 105,282 | −17.7% |
| 2021 | 84,728 | −19.5% |
| 2022 | 55,609 | −34.4% |
| 2023 | 43,539 | −21.7% |
| 2024 | 28,757 | −33.9% |
| 2025 | 16,246 | −43.5% |
| 2026 | 4,643 | — |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing sector, annual counts.
From 234,699 in 2017 to 28,757 in 2024, the figure fell 87.7% over seven years, and it fell in every one of those years, with not a single year of recovery. The 4,643 for 2026 correspond to an incomplete year (data cut off on 14 October) and are not comparable.
If this reading were true, it would describe an extremely serious matter: the number of newly established manufacturing entities in China falling to one eighth of its former level within seven years. This conclusion is in direct conflict with the economic census data cited in Chapter 2 of this report.
Where the Conflict Lies
The Fifth National Economic Census shows that at the end of 2023 there were 4,049,000 manufacturing legal entities nationwide; the Fourth National Economic Census shows 3,270,000 at the end of 2018. Over the five years the stock grew 23.8%, an increase of 779,000.
The stock is growing while the new-establishment reading tabulated by year of establishment is collapsing; these two things cannot both hold. The stock equals the cumulative number established over the years minus the cumulative number that exited over the years. If only twenty or thirty thousand manufacturing entities were newly established each year from 2022 to 2024, while the stock was still growing net each year, then the number of exits would have to be a large negative number—which does not hold arithmetically.
Three Independent Pieces of Evidence
The first piece of evidence comes from within this database. The same annual curve plotted separately on three different samples gives the following results.
| Year | Whole database | Manufacturing | Non-manufacturing |
|---|---|---|---|
| 2017 | 348,624 | 234,700 | 113,924 |
| 2018 | 252,096 | 151,644 | 100,452 |
| 2019 | 232,375 | 127,906 | 104,469 |
| 2020 | 214,888 | 105,288 | 109,600 |
| 2021 | 159,076 | 84,729 | 74,347 |
| 2022 | 116,496 | 55,609 | 60,887 |
| 2023 | 102,649 | 43,540 | 59,109 |
| 2024 | 68,846 | 28,757 | 40,089 |
| Decline, 2017→2024 | −80.3% | −87.7% | −64.8% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database by year of establishment; the three samples are, respectively, all self-collected entities, entities in manufacturing Section C, and entities in sections other than manufacturing. This table is presented in order to show the shape of the curve; none of the three samples excludes anomalous values for the number of insured employees, so the figures in the manufacturing column differ from those in the other chapters of this article (which exclude 45 anomalous records and use a denominator of 2,313,658) by one to several entities. The difference is negligible, but the definitions must be distinguished.
The three samples decline by different amounts (manufacturing the most, 87.7%, and non-manufacturing the least, 64.8%), and the shapes of the curves are not entirely the same: manufacturing declines year by year from 2018, whereas non-manufacturing rises for two consecutive years in 2019 and 2020 (from 100,452 to 109,600) and only turns downward from 2021. But the position of the turning point is the same in all three samples—each reaches its high point around 2017; and the net direction from 2017 to 2024 is the same, a large downward movement in all of them. An industrial phenomenon that occurred only in manufacturing would not make the non-manufacturing sections also trace out a downward curve of the same order of magnitude.
The second piece of evidence comes from registration rules. The "individual-to-enterprise conversion" rule explained in Chapter 1—under which a converted enterprise retains the date of establishment of the original self-employed business—makes the readings for recent years lower than they would otherwise be. The direction in which this rule operates runs counter to the hypothesis that "there really are fewer entities in recent years", so it cannot explain the decline.
The third piece of evidence is corroborative: the nationwide counts of newly established market entities published annually by the State Administration for Market Regulation have remained in the same order of magnitude since 2018. Manufacturing’s share of newly established enterprises has not been published year by year since 2018, however, and this report could not obtain an official series for year-by-year comparison, so this item stands only as corroboration and does not by itself support the conclusion. The share of manufacturing in newly established enterprises has not been published year by year since 2018, so a direct year-by-year comparison cannot be made, but the stock data are already sufficient to settle the question.
A Quantifiable Test
Using the ratio obtained in Chapter 9, an extrapolation test can be performed on this reading.
In 2017, the surviving entities established in that year in this database numbered 234,699, corresponding to the 518,000 newly registered manufacturing enterprises officially published, a ratio of 45.3%. Assuming that this ratio still holds in 2024—that is, that this database's coverage completeness for recent years is comparable to that of 2017—then the 28,757 entities in the database in 2024 correspond to about 63,500 newly registered manufacturing enterprises.
It must be emphasised that this calculation is a hypothetical extrapolation, not an official figure. It has only one use: to test internal consistency. On this extrapolation, the number of newly registered manufacturing enterprises in 2024 is about 87.7% lower than in 2017. Over the same period the stock of manufacturing legal entities grew 23.8%. A system in which only sixty thousand are added each year while the stock is growing would require the number of exits to be negative over the long term in order to be sustained—which does not hold.
Therefore, the assumption about the ratio does not hold. What is more likely to be the case is that this database's coverage completeness for entities established after 2018 is markedly lower than its coverage completeness for earlier years.
Why Coverage Completeness Varies with Vintage
This phenomenon has technical causes, and it is a common failing of enterprise databases.
First, the construction of a stock database is usually based mainly on an inventory of the stock at a particular point in time. When a group of entities is collected, what is captured is the enterprises still on the register at that moment together with their historical records. The records of an enterprise established in 2013 and still on the register in 2026 will be captured in multiple crawls; whereas an enterprise established in 2024 can appear only in crawls made after 2024. If the peak period of entity collection was around 2020, then the probability that entities established after that were captured is naturally lower.
Second, newly established entities have a lower completeness of information and are therefore more likely to be filtered out at the quality-screening stage. A newly established entity often has no social insurance contribution records, no customs records, no certification information and no official website, and the fields available for it when a "factory database" is being built are few. If the collection rules impose requirements on field completeness, newly established entities are the first to be affected.
Third, there is a time lag in the availability of corporate annual report data. Most of the fields used in this report come from annual reports, and annual reports are submitted before 30 June of the following year. An entity established in 2025 submits its first annual report in 2026; an entity established in 2026 has no annual report data at all. This makes the data added for the most recent two years very thin in themselves.
The combined effect of the three causes is this: in a curve tabulated by year of establishment, the right-hand end (recent years) is necessarily below the true value, and the further to the right, the lower it is.
Rules for Reading This Segment
On the basis of the above, this report applies the following treatment to the readings after 2018.
The decline in this stretch is not taken as a conclusion. This report does not assert that "newly established manufacturing entities in China fell sharply after 2017", because the data in this report are insufficient to support that assertion, and the stock data contradict it.
This stretch is marked as an interval pending verification. Once this database completes the filling-in of its coverage of recent entities, or once year-by-year official figures on newly established manufacturing enterprises are obtained, the readings in this stretch should be recalculated. This report sets out the conditions required for recalculation: year-by-year official figures on newly established manufacturing enterprises after 2018, or a dedicated collection and verification exercise for entities established in recent years.
Groups after 2015 are handled with caution in the cohort analysis. In the cohort comparisons by vintage to be presented in Chapter 12, the samples for the 2015—2019 and 2020—2026 groups are less complete than those for the earlier groups. This means that part of the characteristics shown by these two groups (smaller scale, lower foreign trade rate, higher zero-insurance rate) may come from sample selection rather than from the properties of the cohorts themselves—if coverage completeness screens newly established entities in a way related to scale, then the entities included are more likely to be the larger ones, which would bias the readings towards overstating the scale of these two groups. Biases in both directions are present at the same time, and this report cannot determine the net effect.
The existence of this section is itself a conclusion of this report: when the shape of a data curve changes abruptly at its right-hand end, the first thing to be examined is the data themselves, not the world that the data describe. In the draft stage this report wrote this decline as an industrial phenomenon; it was the stock data from the economic census and the shape of the whole-database comparison curve that overturned that treatment. This process of revision is recorded here so that users may judge the reliability of the remaining parts of this report.
Part Three: Generational Differences
12. Profiles of Seven Cohorts
The preceding nine sections discussed the curve itself. This section turns to what lies behind each point on the curve: entities established in different periods display systematic differences in size, foreign-trade participation and social insurance enrolment.
| Period of establishment | Number of entities | Zero-insurance rate | Median insured headcount | 90th percentile | Foreign-trade rate | 1,000 or more | Non-active rate | Total insured headcount |
|---|---|---|---|---|---|---|---|---|
| Before 1978 | 518 | 32.4% | 10 | 333 | 29.0% | 16 | 6.9% | 70,967 |
| 1978–1989 | 11,602 | 28.9% | 5 | 139 | 26.0% | 157 | 6.4% | 1,054,754 |
| 1990–1999 | 94,552 | 30.5% | 5 | 130 | 35.2% | 1,252 | 5.9% | 7,823,233 |
| 2000–2007 | 398,506 | 38.5% | 3 | 69 | 28.5% | 2,072 | 7.8% | 17,028,335 |
| 2008–2014 | 664,918 | 51.8% | 0 | 29 | 16.9% | 1,323 | 11.3% | 13,278,943 |
| 2015–2019 | 804,758 | 62.3% | 0 | 15 | 9.6% | 728 | 14.9% | 8,128,558 |
| 2020–2026 | 338,804 | 62.2% | 0 | 19 | 8.1% | 348 | 9.3% | 4,121,506 |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing scope, grouped by period of establishment. Median insured headcount and the 90th percentile are percentiles of insured headcount; the "non-active rate" is the share of entities whose operating status is marked as non-active. For the data sources and the exclusion rules, see Chapter 3 of Social Insurance Headcount in China's Factories, published by the Institute in the same period.
This table can be read in four ways.
The first reading is that the median insured headcount falls as the cohort becomes more recent. The cohort established before 1978 has a median insured headcount of 10; the mid-value of both the 1978–1989 and the 1990–1999 groups is 5; 2000–2007 stands at 3; and all three groups established in 2008 or later have a median of 0.
The 90th percentile points in the same direction and by a larger margin: 333 before 1978, 130 in the 1990s, 69 in the 2000s, 29 in 2008–2014 and 15 in 2015–2019. From the 1990s to the late 2010s, the employment scale at the same percentile fell by close to ninety per cent.
The zero-insurance column in this table differs slightly in definition from the corresponding column in the size-distribution table in Chapter 13: this table counts entities that did not disclose an insured headcount together with those reporting zero, whereas the Chapter 13 table counts only those reporting zero. The third definition used in this batch — the denominator being entities that did disclose a headcount — appears in Chapter 3 of Social Insurance Headcount in China’s Factories. The differences do not affect the ordering conclusions of this chapter.
The second reading is that the zero-insurance rate rises as the cohort becomes more recent. Entities established in 1978–1989 have a zero-insurance rate of 28.9%; the 1990s, 30.5%; 2000–2007, 38.5%; 2008–2014, 51.8%; and both groups established after 2015, around 62%.
Does this gradient survive once the size factor is removed? This article has not performed that cross-tabulation. Numerically, however, the rise in the zero-insurance rate and the fall in median size occur together, and the two may be two faces of one thing: the smaller the entity, the more likely it is not to use forms of employment that require insurance enrolment.
The third reading is that the foreign-trade participation rate forms an inverted V. The cohort from before 1978 has a foreign-trade rate of 29.0%, 1978–1989 has 26.0%, 1990–1999 rises to 35.2%, and thereafter it falls level by level: 28.5% in 2000–2007, 16.9% in 2008–2014, 9.6% in 2015–2019 and 8.1% in 2020–2026.
The cohort established in the 1990s is the generation with the highest foreign-trade participation rate. This accords with the historical period in which that group of entities was formed — the 1990s were precisely the period when China's export-oriented manufacturing reached scale, and the foreign-trading part of this generation has maintained export business ever since. Entities established after 2015 have a foreign-trade rate of less than 10%.
It should be noted that Chapter 11 of this article has already explained that the sample completeness for the two groups established after 2015 is low. The foreign-trade rate readings for these two groups may therefore be too low: if the collection process under-covers entities with export records, the actual foreign-trade rate would be higher than the figures in the table. The direction of this bias is clear; its magnitude is unknown.
The fourth reading is that the largest factories are concentrated in the middle cohorts. Entities with an insured headcount of 1,000 or more number 5,896 in total, of which 2,072 (35.1%) were established in 2000–2007, 1,323 (22.4%) in 2008–2014 and 1,252 (21.2%) in 1990–1999. The three groups together account for 4,647, or 78.8%.
All entities established before 2007 — the four groups before 1978, 1978–1989, 1990–1999 and 2000–2007 — number 505,178 in total, or 21.83% of all extant manufacturing entities; their insured headcount totals 25,977,289, or 50.4% of the total manufacturing insured headcount.
This is the most important set of figures in this article: factories more than nineteen years old account for less than a quarter of the extant population, yet carry half of manufacturing employment.
The Inverted U of the Non-Active Rate
One column of the table in Chapter 12 has not yet been discussed: the non-active rate. It records the share, in each cohort, of entities whose business registration status no longer falls into the "active" or "in operation" category.
| Period of establishment | Non-active rate |
|---|---|
| Before 1978 | 6.9% |
| 1978–1989 | 6.4% |
| 1990–1999 | 5.9% |
| 2000–2007 | 7.8% |
| 2008–2014 | 11.3% |
| 2015–2019 | 14.9% |
| 2020–2026 | 9.3% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing scope, by period of establishment.
The shape of this column is an inverted U, with the peak falling on the cohort established in 2015–2019 rather than on the oldest cohort. It requires explanation, because this shape runs counter to the intuition that the older an entity is, the more likely it is to have exited.
The explanation is made up of three factors acting together.
The first factor: the earlier a cohort was established, the more rounds of screening it has already passed. For a factory established in 1995 to appear in today's database, the precondition is that it was not deregistered in the thirty years that followed. It has already passed the screen. A factory established in 2016 has passed less than ten years of screening. The non-active rate for earlier cohorts is therefore necessarily low — not because that generation of entities was healthier, but because the unhealthy ones have already been removed. This is the second appearance of survivorship bias in the same field.
The second factor: exposure time. It takes time for an entity to "become non-active". In the cohort established after 2020, the shortest-lived members were established less than a year ago, and most of them have not yet had time to complete the process from operating difficulty to deregistration. The non-active rate for the most recent group is therefore necessarily low — this is the same direction as the first factor but a different mechanism: the former is about survivors being retained, the latter about failures not yet having left the field.
The third factor: the middle cohorts satisfy both the condition of having been exposed long enough and that of not yet having been screened out sufficiently. The cohorts established in 2008–2014 and 2015–2019 have existed for seven to eighteen years, which is ample time to complete the exit process; at the same time they have not yet undergone the multiple rounds of elimination that the 1990s cohort has faced over as long as thirty years. The intersection of the two conditions brings the non-active rate to its peak in this range.
Taken together, the existence of the inverted U does not mean that entities established in 2015–2019 are of the worst quality. What it shows is that the non-active rate column is driven simultaneously by three time-related mechanisms, and those three mechanisms do not point in the same direction.
The column is therefore usable in one direction only: set it against other indicators within the same cohort, rather than comparing its level across cohorts. The conclusion in Chapter 12 that entities established before 2007 carry half of manufacturing employment is unaffected — it rests on insured headcount, not on operating status.
A Necessary Note on Causation
All the comparisons in the preceding passage belong to "cohort differences among extant entities", not to "inherent attributes of factories established in a given period". As Chapter 1 has already explained, differences between cohorts contain three components at once:
Generational establishment differences. Factories established in the 1990s were, at the moment of their establishment, in an industrial environment oriented towards exports and competing by means of scale, and were therefore more likely to be large and export-oriented. This is a difference that existed at the time.
Survival screening differences. Factories established in the 1990s have by now been screened for thirty years. The great majority of those that were too small, poorly run or eliminated by industrial change have been deregistered. What remains are survivors. Factories established in 2015 have been screened for only ten years, and smaller entities are still on the register.
Sample completeness differences. Chapter 11 explains that this database's coverage completeness for recently established entities is low, and this difference affects every reading for the two groups established after 2015.
The three components cannot be separated in this article's data. The statement that "the older the factory, the larger it is" is therefore an accurate description, but its cause is not that "factories grow larger as they age"; it is that "the factories that have survived to today and have been collected in this database are more likely to be those that were larger at the time".
This distinction has a practical effect on use. In judging a specific enterprise, greater age is a positive signal, but what it means is that this enterprise has already passed many years of screening, not that this enterprise will grow larger year by year.
13. The Older the Factory, the Larger: The Shape of the Size Distribution
The percentiles in Chapter 12 give the differences in size, but a percentile indicates only a position on the distribution, not how many points the distribution has. Splitting insured headcount into seven bands and crossing them with period of establishment reveals changes in the shape of the distribution itself.
The table below gives, for each cohort, the share of the entities in that cohort falling into each band.
| Period of establishment | Number of entities | 0 persons | 1–9 persons | 10–49 persons | 50–99 persons | 100–299 persons | 300–999 persons | 1,000 persons or more |
|---|---|---|---|---|---|---|---|---|
| Before 1978 | 518 | 32.4% | 17.4% | 21.6% | 8.5% | 9.1% | 7.9% | 3.1% |
| 1978–1989 | 11,602 | 28.9% | 28.9% | 22.6% | 7.0% | 7.4% | 3.8% | 1.4% |
| 1990–1999 | 94,552 | 30.4% | 29.3% | 21.4% | 6.7% | 7.3% | 3.5% | 1.3% |
| 2000–2007 | 398,506 | 38.4% | 28.7% | 20.0% | 5.5% | 5.0% | 1.9% | 0.5% |
| 2008–2014 | 664,918 | 51.6% | 26.4% | 15.4% | 3.1% | 2.4% | 0.7% | 0.2% |
| 2015–2019 | 804,758 | 61.9% | 23.8% | 10.6% | 1.7% | 1.1% | 0.3% | 0.1% |
| 2020–2026 | 338,804 | 58.0% | 22.1% | 11.8% | 2.1% | 1.4% | 0.4% | 0.1% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing scope, cross-tabulating period of establishment with insured-headcount band. Each row sums to 100% (individual rows may differ by 0.1 percentage point owing to rounding).
The most important thing to read in this table is not the level of any one band but the fact that the whole distribution is shifting its position.
From the 1978–1989 cohort to the 2015–2019 cohort, the four bands with a size of 10 or more — 10–49 persons, 50–99 persons, 100–299 persons and 300–999 persons — fall from 22.6%, 7.0%, 7.4% and 3.8% of the cohort to 10.6%, 1.7%, 1.1% and 0.3% respectively. The four bands shrink at the same time, and the extent of the shrinkage increases with the band: the 10–49 band falls by 53.1%, the 50–99 band by 75.7%, the 100–299 band by 85.1% and the 300–999 band by 92.1%.
By contrast, the 0-person band rises from 28.9% to 61.9%. The 1–9 band changes little, falling from 28.9% to 23.8%.
This is not a thinning of the right tail but a compression of the entire distribution towards the left. If only large entities were declining, the middle bands ought to be relatively stable; in fact every band above one person is shrinking, and the share lost goes entirely into the "0 persons" band.
The 1,000-or-more band changes most sharply: from 1.4% in the 1978–1989 cohort to 0.1% in the 2015–2019 cohort, a difference of 14 times. The 3.1% of the cohort from before 1978 is the highest of all cohorts — of that group's 518 entities, 16 have an insured headcount of 1,000 or more.
Placed side by side, the three comparisons give three expressions of the same conclusion: in numbers, the 94,552 entities of the 1990s cohort produced 1,252 factories with 1,000 or more insured employees (1.3% of that cohort), whereas the 804,758 entities of the 2015–2019 cohort produced only 728 (0.1%); in proportion, the latter is one-thirteenth of the former; in distribution, the centre of gravity of the latter's entire curve is pressed down into the smallest band.
As for causes, the three points set out at the end of Chapter 11 apply here as well — generational establishment differences, survival screening differences and sample completeness differences cannot be separated. What this table can establish is the shape; what it cannot establish is the cause.
14. The Inverted V of Foreign-Trade Participation
The cohort difference in the foreign-trade rate is the only one of this article's six indicators that does not change monotonically.
| Period of establishment | Foreign-trade rate |
|---|---|
| Before 1978 | 29.0% |
| 1978–1989 | 26.0% |
| 1990–1999 | 35.2% |
| 2000–2007 | 28.5% |
| 2008–2014 | 16.9% |
| 2015–2019 | 9.6% |
| 2020–2026 | 8.1% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing scope, by period of establishment. For the meaning and definition of the foreign-trade flag, see Chapter 16 of Social Insurance Headcount in China's Factories, published by the Institute in the same period.
The curve forms a peak in the 1990s with a value of 35.2%, and declines level by level thereafter. This indicates that foreign-trade participation in Chinese manufacturing is a characteristic belonging to a particular generation, not one that diffuses evenly over time.
The explanation for this shape has three layers.
The first layer is the industrial environment of the period. Manufacturing entities established in the 1990s grew up during the fastest period of export-oriented industrialisation. By the time China joined the World Trade Organization in 2001, most of them had been operating for five to ten years and had the production capacity, quality systems and foreign-trade channels needed to take on overseas orders. Entities established after that point face a market with more intense competition and higher barriers.
The second layer is the direction in which survival screening operates. Chapter 16 of Social Insurance Headcount in China's Factories shows that the insured headcount of foreign-trading entities accounts for 69.96% of manufacturing in total, making it the most discriminating of all fields. This means that foreign-trading entities are systematically larger than entities that do not trade abroad. On that basis, thirty years of screening causes the share of foreign-trading entities among those that survive to fall more slowly — they are less easily eliminated. The 35.2% foreign-trade rate of the 1990s cohort therefore contains both "a higher share of foreign-trading enterprises at the time" and "a higher survival rate among foreign-trading enterprises".
The third layer is sample completeness. Chapter 10 explains that this database's coverage completeness for entities established after 2015 is low. If the collection process is correlated with export records, the readings for the two recent groups will be too low.
The three layers cannot be separated. This article therefore reports only the inverted V itself and does not attribute it to any single cause.
From the standpoint of use, the shape has one direct implication: in seeking suppliers with export capability, period of establishment is a valid sorting dimension. The foreign-trade participation rate of entities established in the 1990s is three point seven times that of entities established after 2015. The magnitude of this difference is sufficient for initial screening, and comes ahead of dimensions such as province and industry.
Part Four: Geography and Industry
15. Regional Age Structure: From 69.4% in Beijing to 7.9% in Guizhou
Put the same question along the geographic dimension, and the differences in the answers are larger than along the period dimension.
The table below is ranked in descending order of "the share of entities established before 2008" and gives "the share of entities established after 2015" alongside, the two indicators forming a pair.
| Provincial-level unit | Number of entities | Before 2008 | After 2015 |
|---|---|---|---|
| Beijing | 5,494 | 69.4% | 8.0% |
| Shanghai | 27,819 | 56.1% | 18.7% |
| Jiangsu Province | 273,689 | 33.3% | 37.7% |
| Liaoning Province | 57,525 | 32.7% | 38.2% |
| Tianjin | 23,856 | 28.9% | 40.2% |
| Zhejiang Province | 368,769 | 27.7% | 38.7% |
| Fujian Province | 79,329 | 25.4% | 45.5% |
| Jilin Province | 20,814 | 23.5% | 41.2% |
| Shanxi Province | 22,981 | 23.1% | 48.7% |
| Sichuan Province | 67,349 | 22.7% | 50.0% |
| Inner Mongolia Autonomous Region | 12,456 | 22.2% | 43.5% |
| Heilongjiang Province | 22,841 | 20.9% | 48.8% |
| Shaanxi Province | 24,437 | 20.4% | 52.7% |
| Hubei Province | 42,527 | 19.6% | 47.8% |
| Shandong Province | 202,862 | 19.0% | 56.4% |
| Guangdong Province | 438,592 | 16.6% | 53.5% |
| Anhui Province | 80,031 | 16.0% | 53.1% |
| Chongqing | 45,247 | 16.0% | 54.2% |
| Xinjiang Uygur Autonomous Region | 14,254 | 14.9% | 52.3% |
| Henan Province | 100,649 | 13.9% | 64.0% |
| Hebei Province | 197,039 | 12.8% | 63.6% |
| Jiangxi Province | 48,472 | 12.8% | 61.7% |
| Guizhou Province | 26,331 | 7.9% | 60.6% |
| Xizang Autonomous Region | 3,062 | 4.2% | 73.4% |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing scope, by province; selected entries. Data for provinces not listed can be found in search results from this database.
The gap between the two extremes is substantial. In Beijing, entities established before 2008 account for 69.4% and those established after 2015 for only 8.0%; the corresponding figures for the Xizang Autonomous Region are 4.2% and 73.4%. The two indicators are almost exact mirror images of each other, which indicates that 2008 is an effective dividing line.
Three Recognisable Patterns
The first category is the old industrial regions and the directly administered municipalities. In all seven places — Beijing, Shanghai, Jiangsu, Liaoning, Tianjin, Zhejiang and Fujian — the share established before 2008 is above 25%, with Jilin following at 23.5%. Four of the top six (Beijing, Shanghai, Tianjin and Liaoning) are old industrial bases or directly administered municipalities; Jiangsu and Zhejiang are the two provinces where the restructuring of township and village enterprises was most concentrated in the 1990s; and Fujian is one of the early coastal open provinces.
What these areas share is that manufacturing entities formed early and that the relative pace of subsequent new additions has been below the national average. Their manufacturing age structure is "an old base plus a small amount of new entry".
The second category is the relatively young among the large coastal manufacturing provinces. Guangdong Province has 16.6% established before 2008 and 53.5% after 2015; Shandong Province, 19.0% and 56.4%; Zhejiang Province, 27.7% and 38.7%. The three provinces are three of those with the largest numbers of manufacturing entities, but their age structures differ — Guangdong and Shandong are younger, Zhejiang older. This accords with the difference in the forms of industrial organisation of the two areas: Zhejiang's block economy is dominated by specialised small enterprises that persist over the long term, whereas the manufacturing entities of Guangdong and Shandong include more recently established supporting enterprises.
The third category is the populous provinces of central and western China. Hebei (12.8% and 63.6%), Henan (13.9% and 64.0%), Jiangxi (12.8% and 61.7%) and Guizhou (7.9% and 60.6%) all have a pre-2008 share below 14% and an after-2015 share above 60%. The manufacturing age structure of these four provinces is one in which "a new generation of entities predominates".
The Beijing-Tianjin-Hebei Contrast
Hebei Province and Beijing are adjacent in space yet at the two extremes of age structure. Of Beijing's 5,494 entities, 69.4% were established before 2008; of Hebei's 197,039 entities, only 12.8% were established before 2008. The number of entities is 35.9 times that of Beijing, while the number of entities established before 2008 is 6.6 times that of Beijing.
Converting at the ratios in the table, Beijing has roughly 3,800 manufacturing entities established before 2008 and Hebei roughly 25,200; the two together total about 29,000, of which about 86.9% are in Hebei.
This pattern accords with the direction of the Beijing-Tianjin-Hebei coordinated development strategy, in which Beijing relieves itself of functions non-essential to the capital and manufacturing shifts to the surrounding area. But the data in this article are a stock snapshot and cannot determine whether these old entities have "always been in Hebei" or "moved in from Beijing but retained their original date of establishment". Relocating entities usually retain their original date of establishment and their unified social credit code, so the age structure of the destination area is pulled older by the incoming enterprises. This article has no in-migration or out-migration records and cannot distinguish between the two.
Regional Shifts in the Census Data
The interprovincial differences in age structure accord with the direction of the regional shifts recorded in the official census.
The fifth National Economic Census (at year-end 2023, released on 26 December 2024) gives the changes in the shares of legal entities and persons employed across the four regions of east, central, west and northeast, with the fourth National Economic Census at year-end 2018 as the base for comparison.
| Region | Change in share of legal entities | Change in share of persons employed |
|---|---|---|
| East | −2.0 percentage points | −1.8 percentage points |
| Central | +1.4 percentage points | +1.1 percentage points |
| West | +0.6 percentage points | +0.8 percentage points |
| Northeast | broadly unchanged | −0.2 percentage points |
Table compiled by the Institute by comparing the communiqués of the fifth and fourth National Economic Censuses. Communiqué of the Fifth National Economic Census (No. 7)
Definitional caveat (which must be written into any cross-census comparison): the communiqué of the fourth National Economic Census uses a three-way classification of east, central and west and does not list the northeast separately; the fifth changes to a four-way classification. The row for "the change in the northeast's share" therefore cannot be aligned directly between the two definitions; it is listed in this table because the communiqué itself gives that statement, not because the two values can be subtracted directly.
This set of data and the interprovincial age structure in Chapter 15 of this article corroborate each other: the direction of the shift of legal entities and persons employed towards the central and western regions accords with the age-structure pattern in which "provinces such as Hebei, Henan, Jiangxi and Guizhou are dominated by a new generation of entities". Wherever new entities appear, the age structure is young there.
The direction of the shift is the same, but the magnitudes differ. On the census basis, the rise in the central region's share over five years is 1.4 percentage points, which is a slow movement; in this article's data, by contrast, both Hebei and Henan have more than 63% of their entities established after 2015, nearly eight times Beijing's 8.0%. The source of this contrast is the inertia of the stock — moving the structure of the stock takes decades, whereas the distribution of new additions can change very quickly. Age structure measures the accumulated result of flows, while the regional share measures the stock; the two give readings of different magnitude for the same phenomenon.
The Limits of Explanation for Age Structure
It must be stated once more that differences in provincial age structure contain at once "regional differences in establishment behaviour at the time", "regional differences in survival rates thereafter" and the "differences in coverage completeness" explained in Chapter 11 of this article. The three cannot be separated.
One concrete example is the three northeastern provinces. Liaoning has 32.7% established before 2008, Jilin 23.5% and Heilongjiang 20.9%. The three provinces rank in the top tier, the middle tier and the middle tier nationwide respectively. Looking only at the "share of old entities", the manufacturing of the northeast is not older than that of other old industrial regions. This reading is not entirely consistent with the conventional impression of "an ageing manufacturing sector in the northeast"; the reason may be that this article's data cover only entities still on the register — during the large-scale restructuring of state-owned enterprises in the northeast from the late 1990s through the 2000s, an extremely large number of entities left the register, and the age structure of the remaining entities was thereby reshaped. The "old" in the data is the old age of the survivors, not the old age of all historical entities.
16. Industry Age Structure: Licences and Equipment Determine Age
Differences in age across industries are more regular than across provinces.
| Industry division | Number of entities | Before 2000 | Before 2010 | After 2015 | Median insured headcount |
|---|---|---|---|---|---|
| Manufacture of tobacco products | 436 | 22.9% | 58.3% | 23.2% | 105 |
| Manufacture of medicines | 15,418 | 14.9% | 48.4% | 34.4% | 19 |
| Printing and reproduction of recording media | 47,133 | 14.0% | 52.7% | 28.8% | 3 |
| Manufacture of railway, ship, aerospace and other transport equipment | 18,505 | 9.2% | 40.3% | 38.3% | 2 |
| Manufacture of instruments and meters | 22,405 | 8.9% | 36.8% | 40.5% | 5 |
| Manufacture of raw chemical materials and chemical products | 71,015 | 8.3% | 40.3% | 39.8% | 4 |
| Processing of petroleum, coal and other fuels | 16,435 | 7.7% | 39.7% | 42.1% | 0 |
| Manufacture of electrical machinery and equipment | 112,150 | 6.8% | 34.5% | 41.1% | 3 |
| Manufacture of automobiles | 61,355 | 6.3% | 33.3% | 43.2% | 5 |
| Smelting and rolling of ferrous metals | 15,190 | 6.2% | 38.9% | 40.7% | 0 |
| Smelting and rolling of non-ferrous metals | 18,760 | 5.8% | 36.8% | 41.7% | 3 |
| Manufacture of general-purpose machinery | 198,918 | 5.1% | 31.4% | 44.5% | 1 |
| Manufacture of rubber and plastic products | 157,660 | 4.7% | 29.5% | 47.7% | 0 |
| Manufacture of special-purpose machinery | 136,718 | 4.7% | 29.5% | 47.0% | 2 |
| Manufacture of metal products | 229,611 | 3.7% | 25.6% | 51.0% | 0 |
| Manufacture of computers, communication and other electronic equipment | 94,954 | 3.7% | 25.7% | 47.8% | 6 |
| Manufacture of textile garments and apparel | 91,904 | 3.4% | 21.8% | 54.8% | 0 |
| Manufacture of non-metallic mineral products | 188,318 | 3.3% | 24.1% | 49.8% | 0 |
| Other manufacturing | 107,152 | 1.9% | 16.4% | 59.0% | 0 |
| Manufacture of furniture | 88,159 | 1.8% | 15.0% | 64.1% | 0 |
| Processing of timber and manufacture of wood, bamboo, rattan, palm and grass products | 96,048 | 1.4% | 15.6% | 64.5% | 0 |
| Comprehensive utilisation of waste resources | 27,234 | 0.8% | 10.0% | 78.8% | 0 |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing scope, by industry division. The column for median insured headcount is taken from Chapter 15 of Social Insurance Headcount in China's Factories, published by the Institute in the same period.
Ordered by the share established before 2000, the industries fall into two clear groups.
In the leading group, entry is constrained by licences, permits or very large fixed assets. The 22.9% of the tobacco products industry comes from the monopoly system; the 14.9% of medicines manufacturing comes from drug manufacturing permits and GMP certification; the 14.0% of printing and reproduction of recording media comes from permits to engage in publication printing; the 9.2% of railway, ship, aerospace and other transport equipment manufacturing comes from the fixed investment required at the final assembly stage; and the 8.3% of raw chemical materials and chemical products manufacturing comes from safety and environmental approvals. What these industries share is that entry requires prior qualification, and obtaining qualification takes time and capital, so that once an entity has entered it tends to remain active over the long term.
In the trailing group, entry requires nothing more than a single machine. Comprehensive utilisation of waste resources stands at 0.8%, processing of timber and manufacture of wood, bamboo, rattan, palm and grass products at 1.4%, manufacture of furniture at 1.8% and other manufacturing at 1.9%. What these industries share is that one production step corresponds to one machine, and an entity can be a combination of three or five people. Neither entry nor exit carries institutional cost, so the roster of entities is updated quickly — old entities are deregistered and new ones established without interruption, and the average age of extant entities is therefore low.
The zero-insurance rate and age structure are highly correlated in this table. The leading group (tobacco 28.2%, medicines 22.4%, railway, ship and aerospace 44.3%, raw chemical materials 34.8%) has low zero-insurance rates; the trailing group (timber processing 82.5%, waste resources 77.4%, furniture 68.5%) has high zero-insurance rates. The same holds for median insured headcount.
This correlation suggests one possible explanatory path: an industry's entry barrier determines the minimum viable scale of an entity, the minimum viable scale determines the form of employment, and the form of employment determines how social insurance enrolment is fulfilled. At a timber-processing operation of three or five people, work is more likely to be performed in the form of family members, temporary helpers or part-time workers, and these forms are handled differently from full-time employees in social insurance registration. At a medicines plant with a thousand workers, employment is necessarily full-time and requires registration.
This path is one explanation supported by this article's data, but it is not the only explanation. The same correlation could also arise from a mechanism in the reporting stage — namely that enterprises in low-barrier industries are small and therefore more likely to under-report in their annual reports. This article has no data capable of distinguishing between the two.
17. The Age Profile of Industrial Clusters
The provincial scale conceals the differences inside industrial clusters. The actual unit of manufacturing agglomeration is the county and the county-level city, and pushing the age structure down to this level yields a set of portraits more detailed than interprovincial comparison.
This section selects the 25 county-level statistical units with the largest number of manufacturing entities (Dongguan and Zhongshan, the two prefecture-level cities without districts, are counted at city level in the data), ranked in descending order of "the share of entities established before 2000".
| County or district | Number of entities | Before 2000 | Before 2010 | After 2015 | Total insured headcount |
|---|---|---|---|---|---|
| Xiaoshan District (Hangzhou) | 9,688 | 12.4% | 58.5% | 17.7% | 197,146 |
| Wujin District (Changzhou) | 16,944 | 12.2% | 40.9% | 39.5% | 295,376 |
| Huishan District (Wuxi) | 9,933 | 11.3% | 47.1% | 32.5% | 138,173 |
| Wenling (Taizhou) | 11,442 | 10.4% | 33.7% | 37.7% | 127,878 |
| Jiangyin (Wuxi) | 12,376 | 10.0% | 44.5% | 40.0% | 283,872 |
| Ruian (Wenzhou) | 9,928 | 8.7% | 25.2% | 41.9% | 134,317 |
| Jinjiang (Quanzhou) | 10,256 | 8.5% | 28.2% | 45.8% | 150,880 |
| Cixi (Ningbo) | 19,403 | 8.5% | 35.8% | 33.3% | 251,338 |
| Yuhuan (Taizhou) | 9,659 | 8.4% | 36.9% | 39.4% | 118,210 |
| Yuyao (Ningbo) | 13,478 | 7.1% | 40.8% | 28.7% | 170,510 |
| Chao'an District (Chaozhou) | 12,378 | 6.5% | 23.5% | 57.3% | 51,761 |
| Longwan District (Wenzhou) | 9,827 | 5.8% | 26.2% | 42.3% | 127,867 |
| Shunde District (Foshan) | 23,523 | 4.4% | 28.2% | 48.0% | 588,945 |
| Wujiang District (Suzhou) | 13,249 | 4.4% | 40.9% | 28.5% | 304,682 |
| Kunshan (Suzhou) | 12,281 | 4.4% | 36.2% | 37.3% | 604,106 |
| Haining (Jiaxing) | 10,493 | 3.9% | 36.4% | 28.8% | 139,675 |
| Longgang District (Shenzhen) | 12,524 | 3.7% | 30.4% | 30.7% | 418,259 |
| Nanhai District (Foshan) | 29,769 | 3.3% | 26.0% | 52.7% | 406,057 |
| Zhuji (Shaoxing) | 11,490 | 2.8% | 22.3% | 49.5% | 82,177 |
| Tongxiang (Jiaxing) | 10,072 | 2.6% | 32.5% | 40.4% | 119,679 |
| Yongkang (Jinhua) | 11,537 | 2.3% | 20.3% | 48.8% | 88,888 |
| Zhongshan | 38,282 | 2.3% | 18.7% | 49.6% | 775,407 |
| Bao'an District (Shenzhen) | 20,160 | 2.2% | 26.7% | 30.9% | 694,512 |
| Dongguan | 89,229 | 1.5% | 14.5% | 62.2% | 2,594,630 |
| Bazhou (Langfang) | 16,129 | 0.5% | 9.6% | 79.0% | 27,131 |
Table compiled by the Institute from the Tianxia Gongchang enterprise database, manufacturing scope, counted by district or county, taking the top 25 by number of entities.
Yangtze River Delta Clusters Are Older Than Pearl River Delta Clusters as a Whole
Ranked by the share established before 2000, 10 of the top 13 belong to the Yangtze River Delta (Xiaoshan, Wujin, Huishan, Wenling, Jiangyin, Ruian, Cixi, Yuhuan, Yuyao and Longwan); the three exceptions are Jinjiang in Fujian, Chao'an District in Guangdong and Shunde District in Guangdong.
Looked at in reverse, the four core Pearl River Delta units — Nanhai District at 3.3%, Shunde District at 4.4%, Zhongshan at 2.3% and Dongguan at 1.5% — have pre-2000 shares only one-third to one-eighth of those of the main Yangtze River Delta clusters.
The magnitude of this difference is large enough that sampling fluctuation cannot explain it. Three explanations may be consistent with the data.
First, the period of formation differs. For the township and village enterprises of southern Jiangsu and the individual and private economy of Wenzhou, the peak of entity formation fell in the mid-to-late 1980s and the 1990s. The early stage of manufacturing in the Pearl River Delta likewise included Hong Kong-funded "three-plus-one" (san lai yi bu) processing arrangements in the 1980s, but a very large part of that cohort consisted of processing units registered in a non-corporate form, which do not enter today's statistics on corporate entities; the subsequent proportions of deregistration, conversion of ownership and relocation were also higher.
Second, the rate of new additions differs. Dongguan's manufacturing entities total 89,229, the largest of the 25 units in this section. Such a large base means that Dongguan has been adding entities continuously for forty years. The more entities are added, the more heavily the share of early entities in the stock is diluted. By comparison, Cixi has 19,403 entities and Yuyao has 13,478 — far fewer in absolute terms — so early entities carry a stronger presence there.
Third, the industrial form differs. The traditional clusters of the Yangtze River Delta (hardware, valves, bearings, eyewear, small home appliances, textiles) consist mainly of small and medium-sized enterprises founded by local people, and once an entity has been formed it tends to remain active over the long term. In the electronics and home appliance industries of the Pearl River Delta, contract manufacturing relationships and brands change frequently, and the lifespan of an entity is correspondingly shorter.
The three factors point in the same direction, and this article cannot determine their respective weights.
The Award Years of Cluster Designations
Most of the clusters in the table in this section have officially recognised designations and award years, which can serve as an external reference for their period of formation.
Wenzhou was awarded the designation "China's Shoe Capital" on 25 September 2001 by joint document of the China National Light Industry Council and the China Leather Industry Association; Wenzhou Municipal Bureau of Economy and Information Technology Jinjiang was awarded the same designation in 2001 at the opening ceremony of the third Shoe Fair; Jinjiang Municipal People's Government Yongkang's hardware industry designation was upgraded in 2012 on reassessment from "Home of Hardware" to "China's Hardware Capital". China Hardware Products Association
The award years and the age structures in the table in this section do not always move in the same direction, and this is worth recording. Wenzhou and Jinjiang were both awarded in 2001, and the counties and districts to which they belong (Ruian, Longwan and Yongjia belong to Wenzhou; Jinjiang belongs to Quanzhou) do indeed occupy positions in the table with a relatively high share of early entities; but Yongkang was upgraded to "Hardware Capital" only in 2012, and only 2.3% of its entities were established before 2000, making it the youngest age structure in this section. This indicates that there is no stable correspondence between the year a designation is awarded and the age structure of a cluster — a designation reflects the recognition by an industry organisation of the industrial scale at a given point in time, whereas the age structure reflects the combined result of the two curves of subsequent establishment and exit.
Two Comparisons That Can Be Examined Separately
Cixi and Dongguan. Cixi has 19,403 entities, 8.5% established before 2000 and 33.3% after 2015, with a total insured headcount of 251,338, or 12.95 persons per entity on average. Dongguan has 89,229 entities, 1.5% established before 2000 and 62.2% after 2015, with a total insured headcount of 2,594,630, or 29.08 persons per entity on average.
Dongguan's number of manufacturing entities is 4.6 times that of Cixi, its insured headcount is 10.3 times, and its average size is 2.25 times. Both cities are concentrations of the home appliance and electronics industries, but Dongguan's individual entities are distinctly larger and its age structure is distinctly younger.
Bazhou and Kunshan. These two units are already compared in Chapter 17 of Social Insurance Headcount in China's Factories, published by the Institute in the same period; the age dimension is added here. Bazhou has 16,129 entities, only 0.5% established before 2000 and 79.0% after 2015, the youngest age structure among all 25 units. Kunshan has 12,281 entities, 4.4% established before 2000 and 37.3% after 2015.
In Bazhou, 79.0% of entities were established after 2015, yet its total insured headcount is only 27,131, or 1.68 persons per entity. This is an extreme form, within this section, of "very new entities, very small individual size", and it corresponds to the large number of newly established micro-scale processing sites in the metal rolling and metal furniture industries. Kunshan's age structure sits in the younger part of the middle range, but its average of 49.19 persons per entity is 29 times that of Bazhou.
The two comparisons show that age structure and size structure are two independent dimensions that do not necessarily move in the same direction. Bazhou's entities are the youngest and its size the smallest; Dongguan's are the second youngest and its size in the middle-to-upper range; and the units with the oldest entities, such as Xiaoshan, Wujin and Huishan, sit in the middle range of size.
A Note on the Selection Criterion in This Section
The top 25 by number of entities selected in this section were screened by the single, reproducible criterion of "number of manufacturing entities". This criterion omits some clusters that have distinctive industrial features but a smaller number of entities. Among Shanghai's districts, for example, Chongming District, Pudong New Area and Minhang District have pre-2000 establishment shares of 31.0%, 27.1% and 24.4% respectively, all higher than any unit in the table in this section, yet their numbers of entities have not entered the top 25 (1,132, 3,018 and 1,608 respectively). Nationwide, the county-level units with the highest pre-2000 establishment shares are concentrated in Shanghai's municipal districts, not in the large manufacturing counties in the table in this section.
This addition clarifies the scope within which this section's conclusion holds: the statement presented in the table, that "Yangtze River Delta clusters are older than Pearl River Delta clusters", is a comparison that holds between county-level units with a large number of manufacturing entities, that is, a high degree of industrial agglomeration, and is not a general rule for county-level units nationwide. Shanghai's districts, with few manufacturing entities but an extremely old structure, are another form — the same category as the case described in Chapter 15, in which Beijing has only 5,494 manufacturing entities yet a pre-2008 share of 69.4%: land and labour costs exclude small and micro entities, leaving a small number of entities that have persisted for a long time.
18. Old Industrial Bases: the Northeast Is Not in the Oldest Column
By common impression, manufacturing in the three northeastern provinces should be among the oldest in the country by age structure. The data in this article show a more complicated picture.
Shares of entities established before 2008: Liaoning Province 32.7%, Jilin Province 23.5%, Heilongjiang Province 20.9%. Among provincial-level units nationwide, Liaoning ranks fourth (the top three being Beijing 69.4%, Shanghai 56.1% and Jiangsu 33.3%), while Jilin and Heilongjiang each sit in the middle of the range.
All three provinces rank lower than the position that the impression of "an ageing Northeast manufacturing base" would lead one to expect. Jilin's 23.5% ranks 8th nationwide and Heilongjiang's 20.9% ranks 14th — below Shanxi (23.1%), Sichuan (22.7%), Ningxia (22.4%), Inner Mongolia (22.2%) and Gansu (21.6%).
One Possible Explanation: The Age of the Survivors
Chapter 1 of this article has already explained that counting by year of establishment counts entities still on the register, not all the entities established in a given year. This property has its largest effect on the readings for the Northeast.
From the late 1990s through the 2000s, the three northeastern provinces underwent large-scale reform of state-owned enterprises and adjustment of the industrial structure. In the course of this, a great many manufacturing entities were deregistered, had their licences revoked, were merged or went bankrupt, and thus left the registration database. The dates of establishment of these entities had originally been concentrated in the 1950s to the 1980s — they should have made the Northeast's age structure the oldest column in the country.
After these exits occurred, the entities that remained consisted of two parts: one part comprises entities that continued to remain active after restructuring, whose dates of establishment were mostly reset to the year of restructuring (the late 1990s to the 2000s); the other part comprises entities established after this period. Both parts fall after the 1990s, so the age structure of the Northeast's extant entities appears younger than its industrial history.
In other words: the Northeast's manufacturing "history" is long, but the "age" of the Northeast's extant manufacturing entities is not. The two statements are not contradictory; they describe different objects.
This article does not have deregistration data by year for manufacturing entities in the three northeastern provinces, so the explanation above is an inference consistent with the data, not a verified conclusion. Verifying it would require year-by-year exit data for manufacturing entities in the three northeastern provinces since the late 1990s.
One Case of Entity Exit with an Official Attribution
The exit of entities because of industrial adjustment is rarely quantified directly in public materials. The ferrous metal smelting and rolling industry is an exception.
In its series of reports on the fourth National Economic Census, the Department of Industrial Statistics of the National Bureau of Statistics gives the following: at the end of 2018 the industry had 22,000 enterprises, 42.0% fewer than at the end of 2013, an average annual decline of 10.3%; the number of persons employed fell by 47.5%, an average annual decline of 12.1%. The report's attributed cause reads: "In recent years, as the various policy measures of supply-side structural reform have continued to advance, capacity reduction in the steel industry has achieved notable results." Total Industrial Economic Output Grows Steadily, Employment Declines — Sixth Report in the Series on the Fourth National Economic Census
This is a rare official statement in public materials that links industrial policy directly to a decline in the number of entities in a given industry. Its use must be bounded: it applies only to the single industry division of ferrous metal smelting and rolling, and cannot be extrapolated into "the number of manufacturing entities fell because of capacity reduction" — over the same period the total number of manufacturing legal entities rose from 2,252,000 to 3,270,000, an increase of 45.2% (see Chapter 2 of this article). What capacity reduction cut back was the industry structure, not the manufacturing total.
Its significance for this section is that it provides a quantified sample: in an industry that underwent mandatory capacity exit, 42.0% of entities disappeared from the register within five years. The adjustment that the three northeastern provinces underwent from the late 1990s to the 2000s exceeded this one in both scope and intensity, and its effect in reshaping the region's age structure through entity exit can be understood by reference to the order of magnitude of this sample.
A Contrasting Observation
Jiangsu, Zhejiang and Shanghai provide the contrast to the Northeast. These three regions likewise went through the restructuring of collective enterprises in the 1990s, yet the age structure of their extant entities ranks in the oldest column nationwide (Shanghai 56.1%, Jiangsu 33.3%, Zhejiang 27.7%).
A possible source of the difference is that the mode of restructuring differed. The restructuring of township and village enterprises in southern Jiangsu took "outright sale" as its main form, and in many places the original enterprise's entity status was retained and taken over by the operator, so the date of establishment continued; in the reform of state-owned enterprises in the Northeast, the proportion of bankruptcy and deregistration was higher. The two modes have entirely different effects on the field of date of establishment.
This contrast likewise requires exit data for verification; this article can only list this phenomenon.
How to Use This Section
For users, the practical meaning of this section is one operating rule: do not use age structure to infer the length of a region's industrial history.
A region's extant-entity age structure reflects that "at some point in the past forty years, this region produced a concentrated batch of enterprises that still remain active today". It is affected by both establishment behaviour and exit behaviour, and the effect of exit behaviour is especially large in regions that underwent drastic industrial adjustment.
To judge a region's industrial history requires historical materials, not current enterprise registration data. What the data in this article can do is describe what remains today, not what once existed.
Part Five: Use and Limits
19. The Timing of Regional Policy and the Age Pattern
A natural hypothesis is that regions covered early by opening policies formed manufacturing entities early and should therefore have older age structures, while regions covered later should have younger ones. The data in this report support this hypothesis only in part.
The time sequence of regional policy since 1978 is clear.
| Time | Event |
|---|---|
| 26 August 1980 | The 15th session of the Standing Committee of the Fifth National People's Congress approved the "Regulations on Special Economic Zones in Guangdong Province", establishing special economic zones in Shenzhen, Zhuhai, Shantou and Xiamen |
| May 1984 | The Central Committee of the Communist Party of China and the State Council approved and transmitted the "Summary of the Symposium on Certain Coastal Cities", further opening the 14 coastal port cities of Tianjin, Shanghai, Dalian, Qinhuangdao, Yantai, Qingdao, Lianyungang, Nantong, Ningbo, Wenzhou, Fuzhou, Guangzhou, Zhanjiang and Beihai |
| February 1985 | Decision to establish coastal economic open zones in the Yangtze River Delta, the Pearl River Delta and the Xiamen-Zhangzhou-Quanzhou triangle in southern Fujian, forming a graded pattern of "special economic zones—coastal open cities—coastal economic open zones" |
| 13 April 1988 | The first session of the Seventh National People's Congress adopted the "Resolution on Establishing the Hainan Special Economic Zone", making Hainan the largest special economic zone in the country |
| 18 April 1990 | The central authorities announced agreement to accelerate the development of the Pudong area of Shanghai |
| 26 October 2000 | The "Notice of the State Council on Several Policy Measures for Implementing the Development of the West" (Guofa [2000] No. 33) |
| October 2003 | The Central Committee of the Communist Party of China and the State Council issued the "Several Opinions on Implementing the Strategy of Revitalising Old Industrial Bases Such as the Northeast" (Zhongfa [2003] No. 11) |
| 5 March 2004 | The Government Work Report explicitly proposed "promoting the rise of the central region" for the first time; on 15 April 2006 the "Several Opinions of the Central Committee of the Communist Party of China and the State Council on Promoting the Rise of the Central Region" (Zhongfa [2006] No. 10) was issued |
Table compiled from public policy documents and authoritative historical materials; the specific sources are given at the end of this report.
If the timing of policy determined age structure, then Guangdong, Fujian, Zhejiang, Jiangsu and Shanghai, the earliest to open, should stand at the front of the age sequence. The actual ranking is: Shanghai 56.1%, Jiangsu 33.3%, Zhejiang 27.7%, Fujian 25.4%, Guangdong 16.6% (all being the share of entities established before 2008).
Shanghai, Jiangsu, Zhejiang and Fujian do indeed rank near the front of the age sequence, consistent with the hypothesis. Guangdong is the clear exception—it is the province in which the first special economic zones were set up in 1980, yet it is one of the regions with a younger age structure among all the provincial-level units in this report, with a pre-2008 share of 16.6%, lower than Shandong's 19.0% and only slightly above Hebei's 12.8%.
This exception has an explanation. Guangdong's manufacturing entities total 438,592, the largest number of any province. While it was the first to form a manufacturing agglomeration in the 1980s, it was also the region that added the most entities over the following forty years. Early opening determined that a region formed a manufacturing agglomeration first; what determines today's age structure is the speed of addition and the speed of elimination over the three-plus decades since. Guangdong is high on both, so its age structure reflects "many old entities, and even more new ones"; Zhejiang's speed of addition is relatively low, so while it has retained just as many early entities, its age structure appears older.
By the same logic, the regions covered after 2000 by the policies for developing the West and for the rise of the central region generally have younger age structures (Xizang 4.2%, Guizhou 7.9%, Henan 13.9%, Hebei 12.8%, Jiangxi 12.8%), which is consistent with the timing of the policies. But this consistency can equally be explained by industrial relocation under the National Bureau of Statistics definition—as manufacturing moved inland, the registration of newly established entities moved inland with it.
The data in this report cannot distinguish "policy caused a region to form manufacturing earlier" from "a region was covered by policy earlier because it met the conditions". The two coincide in time but differ in causation.
20. How to Use Age Information in Making Judgements
Age is the field in business registration that is easiest to obtain and least used. This section sets out three uses of it in practical judgement, and two uses it should not be put to.
Use 1: as a ranking dimension for supplier screening
Chapter 14 shows that entities established in the 1990s have a foreign-trade participation rate 3.7 times that of entities established after 2015. Chapter 12 shows that the 505,178 entities established before 2007 account for 21.83% of the surviving total while carrying 50.4% of manufacturing employment.
These two facts together show that when what is needed is production capacity and foreign-trade capability rather than coverage, year of establishment is an effective dimension for initial screening. Restricting the candidate pool to entities established before 2008 reduces the number of candidates to one-fifth, and what is retained is a subset with several times the employment intensity.
The concrete operating rule for this use is that year of establishment should be used together with industry and region, not on its own. Only when years of establishment are compared within the same industry and the same region does the difference carry meaning; in comparisons across industries or across regions, age differences are masked by structural differences between industries and regions. Within a single industry, age differences are easier to interpret than in comparisons across industries or regions: on the “established before 2000” measure, the tobacco products industry and the wood, bamboo, rattan, palm and grass products industry differ by 21.5 percentage points, a difference that reflects their different entry thresholds.
Use 2: as an observation window on regional industrial vitality
The median age of a region's manufacturing entities is one indicator for judging whether that region's industry is in a phase of expansion or of consolidation.
In the data in this report, the regions with a pre-2008 share above 25% (Beijing, Shanghai, Jiangsu, Liaoning, Tianjin, Zhejiang, Fujian) have a manufacturing entity structure dominated by existing stock; the regions with a pre-2008 share below 14% (Hebei, Henan, Jiangxi, Guizhou, Xizang) have a structure dominated by additions.
The two types of region face different industrial problems: for the former the policy focus is on transforming the existing stock and upgrading capacity, while for the latter it is on the quality of newly added entities and the survival of newly founded enterprises. Measuring both types of region with the same set of investment-attraction indicators produces misleading conclusions—assessing the former by the number of new entities, or the latter by the output value of the existing stock, will mismatch in either case.
This use must be applied together with the note in Chapter 11 of this report: recent readings are affected by completeness of collection, and the right-hand end of the curve in this report cannot be used directly to judge "whether additions are slowing".
Use 3: as a supporting signal in company due diligence
At the level of an individual enterprise, the date of establishment provides two kinds of information.
The first is the length of time it has passed through selection. A factory established in 1995 has already passed thirty years of market selection; one established in 2024 has not. This is a positive signal in assessing supply-chain stability, especially in scenarios that require long-term supply commitments.
The second is its correspondence with the industrial cycle. The growth period of entities established between 2008 and 2014 covered the RMB 4 trillion investment programme and the capacity expansion, and they subsequently went through the capacity governance around 2015; the growth period of entities established between 2015 and 2017 corresponds to the market clearing after supply-side structural reform. Enterprises from different cycles will have different characteristics in the age of their equipment, their capacity utilisation rates and their customer structure.
Two uses it should not be put to
First, age cannot be used to judge enterprise size. The data in Chapter 13 show that the share of factories with more than a thousand persons in the 1990s cohort (1.32%) is 14.7 times that in the 2015–2019 cohort (0.090%). This is a statistical fact at the cohort level and does not apply to individuals. A factory established in 2018 may well have a thousand people, and a factory established in 1995 may well have only five. Age is a probabilistic signal, not a size reading.
Second, age cannot be used to infer the number of establishments in past years. This is the problem set out in Chapter 1 of this report, and it is the most common misuse of this kind of data. The age distribution of surviving entities is the result of establishment behaviour and survival behaviour superimposed. Reading the table in Chapter 4 of this report as "China established 235,000 new factories in 2017 and only 29,000 in 2024" is a fundamental misreading of the data, an error comparable to inferring the number of births in past years from the age distribution of the population alive today—the latter can be corrected with a life table, while for the former no usable survival table exists.
Moving from "the distribution of years of establishment" to "the number of new establishments in each past year" requires a survival-rate function in between. There are no official year-by-year data on the survival rate of Chinese manufacturing enterprises; the estimates used in academic research vary considerably, and back-cast estimates for early years have limited reliability. This report therefore does not perform this conversion, nor does it adopt any externally estimated survival rate as a correction.
21. Questions This Report Does Not Resolve
First: the survival rate cannot be separated out of the data
This is the most fundamental of all the limitations of this report, and the problem already raised in Chapter 1.
The age distribution of surviving entities is the product of "the number of establishments in each past year" and "the survival rate in each past year". All the cohort differences given in this report—older cohorts being larger, having a higher foreign-trade rate and a higher insurance enrolment rate—contain both generational establishment differences and survival-selection differences at the same time, and the two cannot be separated.
Separating the two requires year-by-year numbers of new establishments and year-by-year numbers of exits, both published on the manufacturing definition. What can currently be obtained is stock census data (once every five years) and the number of newly established enterprises for some years (the manufacturing definition has not been published year by year since 2018). As for exit data, this report has not been able to obtain year-by-year manufacturing-definition data on enterprise deregistration.
In the absence of these two sets of data, the cohort analysis in this report can only make comparisons "among surviving entities", not "among entities established in a given year". This boundary must be maintained when the data in this report are used.
Second: readings after 2018 are affected by completeness of collection and cannot be used directly
Chapter 11 has used three sets of evidence to show that the decline at the right-hand end of the curve is mainly a problem of data collection rather than an industrial phenomenon, and has listed that stretch as a segment pending verification.
It should be added that this report has not been able to quantify the size of this effect. Two pieces of work could be done: first, obtain the official year-by-year number of newly established manufacturing enterprises after 2018, in order to calculate the ratio point by point; second, carry out a dedicated collection verification for entities established in recent years in this database, sampling several regions and comparing the difference set between this database's collection list and the register held by the registration authority.
Neither piece of work has been done in this report. The data for the former are currently unavailable (the authorities have not published the manufacturing definition year by year since 2018), and the latter lies beyond the data scope of this report.
Third: the effect of enterprise relocation in and out on regional age structure
When enterprises relocate across regions they usually retain their original date of establishment and unified social credit code. This means that a region's age structure is pulled older by inbound enterprises, while the age structure of the place they leave changes accordingly.
This report has no records of enterprise relocation in or out, so the comparison of the age structures of provincial-level units in Chapter 15 includes both "locally established" and "relocated in" origins and cannot distinguish between them. This matters most for regions with active internal migration, such as Beijing-Tianjin-Hebei, the Yangtze River Delta and the Pearl River Delta.
Fourth: the scale and direction of the conversion of individual businesses into enterprises
Chapter 1 explained that an enterprise converted from an individual business retains the date of establishment of the original individual industrial and commercial household, so this registration counts entities added in recent years under earlier years, making recent readings too low.
This report has obtained two aggregate figures: 61,000 households nationwide from January to October 2025 (the basis on which the tax authorities handle them), and a cumulative 358,000 households as of the end of 2025 (the basis on which the market regulation authorities register them). These two definitions cannot be subtracted from each other, and both are data for all industries, with no breakdown for manufacturing.
This report can therefore only state the direction in which this rule operates, and cannot estimate the size of its specific effect on the data in this database.
Fifth: the causes of cohort characteristics cannot be distinguished
Chapters 12 to 16 present a large number of differences between cohorts: the zero-insurance rate rises as cohorts become newer, the foreign-trade rate takes an inverted V shape, and the median size falls as cohorts become newer. These differences are data facts.
But as to their causes, this report offers several possibilities rather than a definite attribution. Taking the cohort gradient of the zero-insurance rate as an example, at least three explanatory paths are consistent with the data: an entity-size effect (small entities do not use forms of employment that require enrolment), an industrial-environment effect (the industrial organisation of recent years is better suited to flexible employment), and a reporting-behaviour effect (the annual reports of small entities are more likely to be filled in casually as zero). All three produce the same reading on the data in this report.
Distinguishing them requires enterprise-level employment data and field surveys. The study cited in Chapter 7 of "Social Insurance Headcount in China's Factories", published by the Institute in the same period, gives the magnitude of the reporting-behaviour effect, but its sample is private enterprises in six provinces and municipalities, not a manufacturing sample stratified by year of establishment.
Conclusion
The date of establishment of a factory is the only piece of information on its business licence that will never change again from the day it is set up.
Registered capital may rise or fall, the business scope may widen or change, the shareholders may be replaced, the address may move, and the contact number may become a dead line. The date of establishment will not. It is like a stratigraphic marker, fixing the birth of an entity in a particular month of a particular year; whatever this factory goes through afterwards, that marker stays where it is.
For this reason, lining up the dates of establishment of 2.31 million surviving manufacturing entities yields not just a curve but a roster of ages. What this roster records is the composition of Chinese manufacturing today—which vintages of factory are still here, and which are gone.
This composition has three core features.
The first is that it is young. The median year of establishment is 2014 and the median age is about 11.9 years; entities established after 2015 account for 49.43%, while those established before 2000 account for only 4.61%. More than half of the Chinese factories still operating today appeared in the last ten years.
The second is that it is steep. Factories more than nineteen years old account for 21.83% of the surviving total while carrying 50.4% of manufacturing employment; those less than nineteen years old account for 78.17% and carry 49.6%. The pyramid of numbers and the pyramid of capacity slope in opposite directions, and by a wide margin.
The third is that its right-hand end cannot be seen clearly. Chapter 11 of this report overturned the judgement in the draft that "newly established entities declined sharply after 2017". The stock of manufacturing legal entities rose from 3.270 million to 4.049 million between 2018 and 2023, yet persons employed increased by only 102,000. The stock is growing, employment is not growing, and the reading of new additions counted by year of establishment is collapsing—of the three, the last is a data problem, while the first two are industrial facts.
Taken together, these three yield a judgement about the generational structure of manufacturing: Chinese manufacturing has entered a phase in which "the number of entities keeps rising, the individual entity keeps shrinking, and total employment no longer grows". The census data for 2018 to 2023 show this directly—manufacturing legal entities increased by 779,000, while persons employed in legal entities increased by 102,000, an average of 0.13 persons per additional entity.
This structure is neither a problem nor an achievement; it is a fact that needs to be described correctly. Its meaning depends on what is being asked: if the question is capacity, then the fact that newly added entities contribute almost no employment can only be assessed together with year-by-year data on automation and output value; if the question is industrial organisation, then the growth in the number of entities and the shrinking of the individual entity point to a more dispersed manufacturing structure with a longer tail.
What this report can establish is this last point: under existing conditions of public data, any judgement about whether Chinese manufacturing is "gaining factories or losing factories" must first state whether it counts entities, employment or output value. Over this period the three numbers point in three different directions.
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 the data defects have been explained in the corresponding chapters of the text. All binned and grouped statistics in the text were calculated directly by the Institute on those definitions, without interpolation or extrapolation. Content concerning the national economic census and policy documents corresponds to the published originals, with dates taken as the date of issue or the date of effect of the document.
Principal references:
- Tianxia Gongchang industrial platform — Chinese factory database and industrial chain data
- Main Data Communiqué of the Third National Economic Census, the National Bureau of Statistics
- Communiqué of the Fourth National Economic Census, the National Bureau of Statistics
- Communiqué of the Fifth National Economic Census (No. 2): Basic Information on Units, the National Bureau of Statistics
- 2017 national data on newly registered manufacturing enterprises, Workers' Daily
- 2016 national data on newly registered manufacturing enterprises, Guangming Daily
- Historical materials on the decision to establish the special economic zones, the Institute of Party History and Literature of the Central Committee of the Communist Party of China
- Historical materials on the decision to develop and open up Pudong, the Institute of Party History and Literature of the Central Committee of the Communist Party of China
- Historical materials on the strategy of revitalising old industrial bases such as the Northeast, the Institute of Party History and Literature of the Central Committee of the Communist Party of China
- Chronicle of events in promoting the rise of the central region, People's Daily Online
- Award of the "China's Shoe Capital" designation to Wenzhou, the Wenzhou Municipal Bureau of Economy and Information Technology
- Award of the "China's Shoe Capital" designation to Jinjiang, the Jinjiang Municipal People's Government
- Reassessment and upgrading of the "China's Hardware Capital" designation for Yongkang, the China Hardware Products Association
- Communiqué of the Fifth National Economic Census (No. 7): Units and Persons Employed by Region, the National Bureau of Statistics
- Historical materials on the 1984 Symposium on Certain Coastal Cities and the decision on coastal economic open zones, China Economic Net (source noted as: the Ministry of Commerce)