Hiring and Downsizing Read One Floor Below Employment Statistics — The Direction Indicated by Si-Gun-Gu and Industry-Level Administrative Data

In July 2026, the number of employed persons stood at 29,136,000, an increase of 108,000 from a year earlier. This is the opening line of the 'July 2026 Employment Trends' published by the National Data Agency on August 12. Since the expansion returned to the 100,000 range for the first time in four months, it is not a bad scorecard at first glance.

However, there are questions that this single line cannot answer. Where in which region, industry, and workplace did those 108,000 jobs increase? Are they production workers at a semiconductor supplier in southern Gyeonggi Province, kitchen staff at a restaurant in downtown Busan, or care jobs scattered across elderly care facilities nationwide? Employment statistics count 'working people,' but they do not count 'companies that increased their headcount.'

There is data available that can bridge this gap. It is the 'National Pension Enrolled Workplace Details' released monthly by the National Pension Service via the Public Data Portal. Each workplace is recorded as a single row, which includes the legal-dong-level address, industry code, number of subscribers, number of newly acquired subscribers, number of lost subscribers, and the current month's billed amount. The key point is that this is administrative record-keeping reported directly by companies for insurance premium assessment, rather than a survey.

The Layer That Employment Statistics Cannot See

Employment trends are derived from a sample survey. Surveyors visit sampled households and ask whether anyone worked for even an hour for income during the previous week. While optimal for viewing national and city/provincial trends, the sample becomes too thin once descending below the si-gun-gu (municipal) level, making monthly publication impossible. In fact, si-gun-gu employment indicators are published biannually through a separate 'Regional Employment Survey.' The National Data Agency conducted the survey for the first half of 2026 from April 21 to May 6 targeting approximately 232,000 households nationwide, and released the main si-gun-gu employment indicators only on August 20.

In contrast, National Pension workplace data is updated monthly. Furthermore, the unit of observation is not 'people' but 'workplaces.' The amount of information generated by this difference is larger than expected. Even if the employment count in a certain region remains flat, if an exchange took place wherein 300 workplaces in industry A reduced headcount and 250 workplaces in industry B increased headcount, the regional industrial structure is already changing. Aggregate statistics offset this turnover and report 'no change.'

Indeed, such turnover is verified by statistics. According to the 'Wage Worker Job Trends for the First Quarter of 2026 (based on February)' published by the National Data Agency on August 24, out of a total of 20,828,000 wage-earning jobs, 'continuous jobs' where the same workers were employed as a year prior accounted for 74.0%, or 15,416,000 jobs. The remaining 26% consisted of positions where workers changed (replacement jobs of 3,186,000, 15.3%), newly created positions (new jobs of 2,226,000, 10.7%), or vanished positions (dissolved jobs of 1,934,000). While the net increase was 292,000, this means that over 4 million jobs were created and eliminated behind the scenes.

Data: What Is Included and What Is Omitted

To use this data for articles or research, one must first read its specifications accurately. The conditions specified on the Public Data Portal are as follows:

First, the coverage does not include all workplaces. It only includes corporate workplaces with 3 or more subscribers and individual businesses with 10 or more subscribers as of July 2025 and thereafter. Corporations with 1 to 2 employees and small-scale individual businesses are excluded from the outset. Therefore, discussing 'alley commercial district employment' using this data is inappropriate.

Second, there is a time lag. The 'data generation year-month' reflects reports filed up to the qualification deadline, which is the 15th of the month following the month in which the event occurred. An employee hired in June is captured in the June data if reported by July 15, but if reported thereafter, it is pushed to the following month. Operationally, a delay of around a month and a half must be assumed.

Third, the 'number of newly acquired subscribers' and 'number of lost subscribers' do not precisely match hires and retirements. Newly acquired subscribers include resumed payments, while lost subscribers include payment exemptions. The NPS specifies that because this item is compared with the previous month's billing targets, it may differ from the actual number of acquisitions, and if an acquisition does not occur on the first day of the month and is not billed in the current month, it is reflected in the following month's acquisition count. In other words, trends such as a 3-month moving average are more reliable than the absolute values of monthly figures.

Fourth, the amount items are not income indicators. The current month's billed amount applies an upper limit on the standard monthly income under the Enforcement Decree of the National Pension Act. The upper limit was 6.37 million won from July 2025 to June 2026, and 6.59 million won from July 2026 to June 2027. The higher the concentration of high-income earners at a workplace, the greater the discrepancy between actual salaries and billed amounts. Estimating regional average wages using this item will inevitably result in a downward bias.

Fifth, and most importantly, there is a constraint. National Pension workplace subscribers are principally targeted at individuals aged 18 to 60 and under. Workers aged 60 and over and subscribers to other public pensions, such as the Government Employees Pension or Private School Personnel Pension, differ from the population of this statistic. However, in the July 2026 employment trends, employed persons aged 60 and older increased by 231,000 from a year earlier, far exceeding the total increase (108,000). This means that the age group essentially driving current Korean employment growth is not sufficiently reflected in this data.

In summary, this data is not a substitute for 'aggregate employment.' It should be understood as workplace-level employment signals based on social insurance subscriptions, centered on those aged 18–59. Looking at employment types, it does not capture regular workers only. If subject to the national pension, fixed-term and part-time workers are also included; conversely, if exempted or granted payment exceptions, they are omitted even if they are actually working. Calling it a 'regular hiring indicator' is inaccurate.

July 2026: Overlapping Three Administrative Statistics

To verify the directionality of the National Pension data, it is safest to place administrative statistics of similar character side by side: employment insurance subscribers, establishment labor survey workers, and wage worker job trends.

According to the 'Labor Market Trends Based on July 2026 Employment Administrative Statistics' published by the Ministry of Employment and Labor on August 10, regular employment insurance subscribers stood at 15,877,000, an increase of 277,000 (1.8%) compared to the same month last year. The expansion maintaining the upper 200,000 range marks the seventh consecutive month. The expansion, which was 183,000 last December, rose to 264,000 this January, and has remained stable within a narrow band: 270,000 in March, 271,000 in April, 270,000 in May, 266,000 in June, and 277,000 in July.

The general direction is the same in the 'July 2026 Survey on Labor Conditions at Establishments' published by the Ministry of Employment and Labor on August 27. As of the last business day of July, the number of workers at establishments with 1 worker or more was 20,718,000, an increase of 226,000 (1.1%) from the same month last year. However, examining the details reveals a different texture. While regular workers increased by 78,000 (0.5%), temporary and daily workers increased by 147,000 (7.6%), with temporary and daily workers accounting to 65.0% of the total increase. Other workers increased by 1,000. By scale, establishments with fewer than 300 workers increased by 164,000, and those with 300 or more increased by 62,000. Accessions during July rose to 1,143,000, and separations increased as well.

The magnitude of wage worker job increases also forms a recovery curve. Having dropped to 15,000 in the first quarter of 2025, the increase expanded through 111,000 in the second quarter, 139,000 in the third quarter, and 221,000 in the fourth quarter, reaching 292,000 in the first quarter of 2026. This is the largest increase in two years since the first quarter of 2024 (314,000).

The fact that the three indicators coincide in aggregate direction is meaningful. Rather than the monthly fluctuations of the sample-based employment trends (an increase of 74,000 in April, a decrease of 40,000 in May, an increase of 63,000 in June, and an increase of 108,000 in July), administrative record-based indicators draw a much more gradual and consistent curve. The same attitude is required when viewing National Pension workplace data. Direction and persistence are more informative than a single month of sharp spikes or drops.

Same Manufacturing Industry, Why Exactly Opposite Answers Arise

However, when descending to the industry level, answers diverge among statistics. This point is a pitfall that must be known when handling National Pension data.

Based on employment insurance, July manufacturing subscribers fell by 3,000, marking a 14-month consecutive decline. In the same month's Survey on Labor Conditions at Establishments, manufacturing workers increased by 14,000, marking a 7-month consecutive rise. The directions are exact opposites. Construction is similar. Employment insurance subscribers fell by the 7,000 range (7,200 according to detailed press release figures), marking a 36-month consecutive decline, whereas workers in the Survey on Labor Conditions at Establishments increased for the second consecutive month. However, the Ministry of Employment and Labor stated that given the preceding 23 consecutive months of decline and recent increases hovering around the 3,000 and 7,000 levels, it is premature to judge this as a recovery trend.

Neither of the two is wrong; their populations differ. Employment insurance counts regular subscribers only, omitting labor providers, daily workers, self-employed persons, and uninsured employed persons. The Survey on Labor Conditions at Establishments is an establishment sample survey that includes other workers, and the Ministry of Employment and Labor previously explained during its April 2026 results announcement that reflecting the latest population data from the 2024 Business Labor Status Survey and applying the 11th Korean Standard Industrial Classification could upwardly adjust previously underestimated increases in manufacturing and other sectors.

National Pension workplace data also holds another population. Therefore, one must not immediately write "manufacturing employment is declining" just because manufacturing subscribers in a specific region decreased in this data. The accurate description is "workplace subscribers of National Pension-applicable workplaces in the corresponding region have decreased." The moment population notation is omitted in administrative data journalism, the article becomes an incorrect sentence.

Industries: Services Pulled, Manufacturing and Construction Suppressed

Even considering population differences, industry-level structures are relatively clear. In July employment insurance subscribers, the service sector grew by 285,000 (2.6%) to 11,139,000, practically pulling the entire increase single-handedly. In detail, health and welfare saw the largest increase at 112,000, followed by accommodation and food services (57,000), business service activities (26,000), professional, scientific, and technical services (23,000), and education services (19,000). In the Survey on Labor Conditions at Establishments, wholesale and retail trade declined for 28 consecutive months, which the Ministry of Employment and Labor explained reflected the impact of large supermarket closures and shutdowns.

The drop in construction employment insurance subscribers has narrowed for five straight months, from 9,100 in March to 8,800 in April, 8,400 in May, 8,100 in June, and 7,200 in July. While there is room to interpret this as a bottoming-out phase, the decline has not stopped.

There is one more interpretive trap in manufacturing. July manufacturing employment insurance subscribers decreased by 3,000, among which Employment Permit System (E-9) foreign subscribers increased by 13,000. Excluding foreign subscribers, the decline widens to 15,000. The Ministry of Employment and Labor explained that 90.0% of Employment Permit System foreigners are concentrated in manufacturing, requiring caution when interpreting subscriber trends in that industry. Because National Pension workplace data likewise does not separate foreign workers into a distinct category, the same trap applies identically.

For future directions, the 'Job Outlook for Major Industries in the Second Half of 2026' released on August 19 by the Korea Employment Information Service and the Korea Institute for Advancement of Technology serves as reference. In this outlook based on employment insurance insured worker data, among nine core manufacturing industries, semiconductors were projected to increase by 5.1% (8,000 people) and shipbuilding by 2.7% (3,000 people) year-on-year. Conversely, textiles were projected to decline by 3.5% (5,000 people), while petroleum and chemicals (-0.6%), automobiles (-0.5%), steel (-0.3%), metal fabrication (-0.3%), and electronics and displays (-0.1%) were classified within the range of maintaining the previous year's levels. These are forecasts at best and not finalized performances.

However, overestimating semiconductor's employment contribution is problematic. As of the first half of 2026, semiconductor industry workers numbered 157,000, accounting for about 1.0% of all workers. In the wage worker job trends for the first quarter, semiconductor jobs increased by only 5,000 from a year earlier. Among manufacturing sub-sectors, the industry with the largest increase was other food products (6,000 jobs), with semiconductors ranking second. Compared to the scale of the export boom, the spillover into employment is limited. This is a structural characteristic of capital-intensive industries, and the high semiconductor unfilled vacancy rate of 14.8% suggests that a separate labor bottleneck exists.

Regions: Where Are Jobs Increasing?

The strengths of National Pension workplace data are most clearly revealed when reading signals at the si-gun-gu level. However, regional statistics published as of late August 2026 have disparate base periods, requiring caution in comparison.

Based on the Survey on Labor Conditions at Establishments, the number of workers by city and province in January 2026 increased by 40,000 in Seoul and 25,000 in Gyeongnam, with growth rates high in order of Sejong (2.8%), Ulsan (2.3%), and Gyeongnam (2.2%). The fact that Gyeongnam and Ulsan—regions dense with shipbuilding industries—ranked together at the top aligns with the direction of shipbuilding employment forecasts. As of May 2026, domestic shipyard order backlogs stood at 38.5 million CGT, securing more than three years' worth of work. More recent worker data by city/province and si-gun-gu is contained in the Regional Survey on Labor Conditions at Establishments for the first half of 2026 (as of April), published concurrently on August 27.

Industrial concentration is revealed in the Regional Employment Survey. As of the first half of 2025, regions with the highest proportion of mining and manufacturing workers were Geoje-si, Gyeongnam (38.4%) among provincial regions, and Dong-gu, Ulsan (43.4%) among special and metropolitan city regions. In such places, a single fluctuation in core industry conditions shakes the entire regional employment. If the number of newly acquired subscribers in a specific si-gun-gu has exceeded the number of lost subscribers for months in National Pension workplace data, it can be examined as an early signal that orders in the region's core industry are translating into actual workforce deployment.

Counter-signals are also verified. In the Regional Employment Survey for the second half of 2025, employment in ward (gu) districts of seven special and metropolitan cities stood at 11,589,000, a decrease of 40,000 from the same period last year, and the employment rate dropped by 0.2 percentage points to 58.8%. Employment in county (gun) districts of nine provinces similarly fell by 11,000 to 2,104,000. This signifies that a structure of employment draining simultaneously from metropolitan downtowns and rural county districts is persisting.

Look at Turnover Rates, Not Net Increases

The most common mistake when handling this data is looking solely at month-on-month increases or decreases in the number of subscribers. Doing so eliminates the reason for providing the number of newly acquired subscribers and lost subscribers separately.

Even with the same net increase of 100 people, the contents are entirely different. If there are 120 new acquisitions and 20 losses, it is a gradual expansion. If there are 800 new acquisitions and 700 losses, it is a high-turnover state where labor is churning massively. While the net increase is the same in both cases, the character of the labor market differs. However, the fact that the turnover rate is high does not suffice to judge employment quality. Seasonal industries, expansion of hiring in fast-growing sectors, and workplace restructuring also create high turnover. Turnover rates become meaningful only when read alongside indicators such as wages, employment types, tenure, and voluntary versus involuntary turnover proportions.

In fact, those supplementary indicators sent recent signals. In the July Survey on Labor Conditions at Establishments, 65.0% of the worker increase came from temporary and daily workers. This composition was invisible from the single figure of a 226,000 net increase. By creating turnover rate indicators dividing (newly acquired subscribers + lost subscribers) by the number of subscribers by region and industry, clues to this compositional shift can also be grasped in National Pension data.

Age-specific indicators show another layer. While employed persons aged 60 and older increased by 231,000 in July 2026, employed persons in their 20s decreased by 204,000. Employed persons aged 15–29 stood at 3,441,000, dropping by 191,000 to continue a downward trend for 45 consecutive months, and the youth employment rate fell by 0.2 percentage points to 44.2% for 27 consecutive months. The youth unemployment rate rose by 1.3 percentage points to 6.8%, which the National Data Agency stated is the largest increase in 5 years and 6 months since January 2021 (1.8 percentage points). In employment insurance as well, subscribers aged 29 and under fell by 57,000, extending a consecutive decline for 47 months since September 2022.

In summary, while totals are growing, that growth is concentrated among those aged 60 and older and temporary/daily workers, and within the 18–59 age bracket targeted by National Pension workplace subscribers, the youth segment continues to recede. This is why National Pension workplace data is likely to present a more conservative picture than aggregate statistics. If the two data sets diverge, it is likely a population difference rather than an error.

What Can and Cannot Be Done With This Data

Three things can be done. First, identifying cells where headcount increased and cells where it decreased within the si-gun-gu × industry grid at monthly renewal cycles. Second, separating new acquisitions and losses to distinguish between expansion, contraction, and high-turnover phases. Third, viewing the number of workplaces and subscribers together to distinguish whether an increase is due to 'more workplaces increasing' or 'existing workplaces increasing headcount.' The latter is a much stronger signal.

What cannot be done is also clear. Movements of employment for those 60 and older, the self-employed, special-type labor workers, corporate entities with fewer than 3 employees, and individual businesses with fewer than 10 employees are not sufficiently captured. Because addresses reflect the registered addresses of National Pension-applicable workplaces, they may not match companies' actual work locations or production bases. In particular, variances in regional attribution occur depending on whether head offices, branches, and factories are applied separately and how workplaces are merged or closed. Wage levels are also not subject to estimation due to standard monthly income upper limit constraints. Above all, this data is public data opening material, not nationally approved statistics. Aggregation results must not be presented as definitive statistics.

The realistic utilization method is triangular verification. It is the sequence of narrowing down candidate regions and industries with National Pension workplace data, cross-checking directions with employment insurance subscriber trends and the Survey on Labor Conditions at Establishments, and confirming through regional employment surveys and field reporting. Administrative data is a tool to refine questions, not a tool to substitute for conclusions.

Key points to watch in South Korea's employment market for the second half of 2026 condense into three: whether the narrowing decline in construction leads to an actual rebound, whether the export booms in semiconductors and shipbuilding spread to hiring at the supplier level, and at what point youth employment—declining for 45 months—stops falling. The fact that the Ministry of Employment and Labor's First Half 2026 Occupational Survey on Labor Conditions at Establishments showed planned recruitment openings for the second and third quarters at 460,000, down 1.8% from the same period last year, demands cautious answers to all three questions. And the answers to those three questions will appear first across the grid of si-gun-gu districts and industries, rather than national aggregates. National Pension workplace data provides that grid every month.

Data Transparency Notice

This article was written based on materials published as of August 28, 2026. The sources and characteristics of the figures are as follows:

① Confirmed Official Statistics and Publication Dates. National Data Agency, 'July 2026 Employment Trends' (published Aug 12) — employed persons, employment rate, age-specific changes, youth unemployment rate. National Data Agency, 'Q1 2026 (based on February) Wage Worker Job Trends' (published Aug 24) — total wage-earning jobs, continuous, replacement, new, and dissolved jobs, semiconductor job changes. Ministry of Employment and Labor, 'Labor Market Trends Based on July 2026 Employment Administrative Statistics' (published Aug 10) — regular employment insurance subscribers, industry and age-specific changes, Employment Permit System foreigners. Ministry of Employment and Labor, 'July 2026 Survey on Labor Conditions at Establishments and First Half 2026 Regional Survey on Labor Conditions at Establishments' (published Aug 27) — number of workers, changes by employment status and scale, manufacturing, wholesale/retail, and construction trends, accessions. Ministry of Employment and Labor, 'First Half 2026 Occupational Survey on Labor Conditions at Establishments' (published June 30) — planned recruitment openings. National Data Agency, 'Regional Employment Survey' (first and second halves of 2025) — si-gun-gu employment rates and industry proportions. Korea Employment Information Service & Korea Institute for Advancement of Technology, 'Job Outlook for Major Industries in Second Half of 2026' (published Aug 19) — industry-specific forecasts, shipbuilding order backlogs, semiconductor worker scale and unfilled vacancy rate.

② Dataset Specifications. Coverage of the National Pension Service 'National Pension Enrolled Workplace Details' (corporate workplaces with 3 or more subscribers, individual businesses with 10 or more subscribers since July 2025), definition of data generation year-month, definition of newly acquired and lost subscribers, current month billed amount, and standard monthly income upper limit quote the NPS's official explanations disclosed on the Public Data Portal. Descriptions regarding workplace subscribers' age requirements and separate application are based on NPS institutional guidance and workplace practical guides.

③ What This Article Does Not Include. Results directly aggregating the raw data (CSV) of National Pension Enrolled Workplace Details are not included. Si-gun-gu × industry-level subscriber counts and new acquisition/loss figures require downloading and independently computing large-scale raw files distributed monthly, and those computational results were not independently verified at the time of writing this article. Therefore, specific si-gun-gu and industry change figures are not presented. All regional and industry descriptions in the text are based on published government statistics, covering only reading methods and interpretive cautions regarding National Pension data.

④ Forecast and Interpretive Descriptions. Employment change figures by industry for the second half, such as semiconductors, shipbuilding, and textiles, are Korea Employment Information Service forecasts and not finalized performances. Judgments on whether narrowing construction declines will lead to a rebound, methods of constructing turnover rate indicators, and comprehensive explanations of why manufacturing figures between employment insurance and the establishment labor survey diverge are the reporter's analytical interpretations based on published materials.

⑤ Cautions When Comparing. Employment trends (household sample survey), employment insurance (administrative records), Survey on Labor Conditions at Establishments (establishment sample survey), wage worker job trends (combined administrative data), and National Pension workplace details (administrative records) have different survey targets, populations, and definitions. As verified in the text, even the same manufacturing sector can show opposite directions of change depending on statistics. Figures from different statistics must not be summed into a single time series or directly compared.