## wpiea2022032-print-pdf

## Source details

**Canonical URL:** [wpiea2022032-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022032-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022032-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022032-print-pdf.pdf.json)

---

### Declining business dynamism — six stylized facts
- Fact 1: The revenue and asset share of young Chinese firms has declined over time.
  - The share of firms under 10 years old fell from around 70 percent in 2003-04 to around 30 percent in 2017-18.
  - Similar trend when using total assets as firm size and for service sector firms (post-2013).
  - Baseline sample imposes a RMB 5 million revenue threshold in all years; expanding to all manufacturing firms increases sample from around 200,000 firms to 700,000 firms but still shows low revenue share for young firms.
- Fact 2: Life-cycle growth of young Chinese firms has declined over time.
  - 3-5 year old firms have 3-year growth rates around 17 percentage points higher than firms in the 16+ age group.
  - Revenue growth of firms under the age of 10 (relative to 16+) is substantially smaller in 2011-2018 than in 2003-2010.
  - Life-cycle growth is estimated via ∆k y_ist = α + Σ_{a=1}^5 β_a I(age group=a) + FE_st + ε_ist with k = 3 and age groups 1-2, 3-5, 6-10, 11-15 and 16+ (omitted).
- Fact 3: Productivity growth and innovation over the life-cycle has weakened.
  - Flatter life-cycle growth could reflect (a) increasing distortions preventing growth, or (b) firms investing less in R&D, process efficiency, quality improvements, or other intangible inputs.
- Fact 4: Younger and smaller firms are more capital constrained than older and larger counterparts.
  - Large financial frictions estimated by Bai et al. (2018) have remained an important constraint for young firms over the last decade.
- Fact 5: The responsiveness of capital growth to the average product of capital has declined over time.
  - A decline in the capital responsiveness elasticity could reflect worsening frictions to capital reallocation.
  - Consistent with observed decline in dispersion of capital growth across firms.
- Fact 6: SOEs have lower revenue productivity (revenue over inputs) and capital productivity (revenue over capital) than private firms.
  - Productivity gaps documented prior to 2013 and remain through 2018 for a larger representative sample.
  - Productivity gaps are also large in the service sector (data available post-2013).

### Role of SOEs and cross-sectional evidence
- SOE sector context:
  - More than 80 percent of SOEs were closed, privatized, or merged between 1998 and 2007 (Hsieh and Song, 2015).
  - SOE share of industrial assets remained at almost 40% in 2019.
- Empirical association:
  - In the cross-section of provinces, business dynamism tends to be weaker where SOEs account for a larger share of assets.
  - Measures of dynamism used: life-cycle growth of firms and responsiveness of capital growth to marginal product of capital.
  - Suggestive evidence that SOE reform could boost productivity directly via resource reallocation and indirectly by stimulating business dynamism.

### Aggregate TFP and data source (Orbis)
- Data and coverage:
  - Firm-level data from Bureau Van Dijk’s Orbis database for 2003 to 2018.
  - Cleaned manufacturing database contains on average 221,180 firms per year with revenues, capital and costs.
  - Service sector coverage available from 2013 to 2018.
  - For baseline results impose RMB 5 million revenue threshold in all years (because of minimum annual revenue threshold up to 2012 in Orbis).
  - Identify around 3,200 SOEs every year on average in manufacturing and mining sectors; for 2013-2018 identify 9,200 SOEs every year on average in services sectors.
- Constructing aggregate manufacturing TFP:
  - Aggregate TFP is a gross-output weighted average of industry-level TFP based on a Cobb-Douglas industry production function.
  - Revenues used as output, tangible fixed assets as capital, and industry cost shares identify labor and intermediate inputs from sum of cost of goods sold plus non-operating costs.
  - Variables deflated using 2-digit industry deflators for revenues and intermediate inputs; sectoral TFP estimates constructed at the same level as industry-specific deflators.
  - Aggregate manufacturing revenues from Orbis are 78 percent of official aggregates on average for above-scale manufacturing.
- TFP trend:
  - Manufacturing TFP measure points to a slowdown in productivity growth post-GFC.
  - Sampling changes post-2013 increase volatility of TFP series, but TFP growth slowed down considerably post-GFC relative to the previous decade.

### Productivity measurement and life-cycle productivity findings
- Measurement framework:
  - TFPR as Q_si/I_si and TFPQ as Q_si^σ/(I_si) where Q_si is firm revenues and I_si are firm inputs with I_si = (K_α_si s wL_1−α_s si )_γ_s X_1−γ_s si. α_s and γ_s are industry cost shares, and we set σ= 4.
- Empirical findings:
  - No evidence that TFPR varies over the life-cycle, nor that it changes between the 2000s and the 2010s.
  - TFPQ of young firms relative to older firms declined substantially in the 2010s.
  - In the 2000s TFPQ growth declined monotonically with age; in the 2010s TFPQ growth is U-shaped with 6-10 year old firms experiencing lower TFPQ growth than 16+ firms.
  - Data on intangible fixed assets for a subset of firms show young firms investing relatively less in process efficiency and quality improvements in the 2010s than in the previous decade.

### Regional SOE intensity and its associations with dynamism
- Motivation and facts:
  - SOE share of industrial assets fell from over 54 percent in 2003 to below 43 percent in 2008, then plateaued with SOEs still accounting for 39 percent of assets in 2019.
- Life-cycle growth and SOE intensity (province-level evidence):
  - Regression: lc_pt = α + β SOEI_pt + FE_p + FE_t + ν_pt where lc_pt is 3-year average growth of young firms relative to old firms and SOEI_pt is SOE intensity (asset share).
  - Being in a province with 10 percentage point higher SOE share of assets reduces 3-year revenue and input growth by 3.5 percent for young firms, relative to an average 3-year growth rate of 17.5 percent.
  - Province-level results: higher SOE intensity associated with weaker revenue growth, capital growth, and TFPQ growth for young firms. Industry-level regressions do not find an association.
  - Interpretation: negative spillovers of local SOE intensity on private firms may reflect local regulatory barriers or political economy mechanisms limiting young firm investment or expansion.
- Capital reallocation and SOE intensity:
  - No significant link in 2003-2007; in 2014-2018 provinces with high SOE intensity have significantly lower capital responsiveness.
  - Suggestive mechanism: local banking systems in high-SOE provinces may be less efficient at allocating capital to productive private firms because lending incentives favor SOEs, worsening allocation among private firms.

### Key numerical results and regression estimates (selected)
- Dispersion in total asset growth: 0.126 (2003-2007) → 0.083 (2013-2018).
- Responsiveness of capital growth to lagged ARPK: 0.096 (2004-2007) → 0.059 (2016-2018).
- Province-level life-cycle regressions (selected coefficients):
  - 3-year revenue growth: -0.00349** (standard error 0.00147)
  - 3-year asset growth: -0.00291** (standard error 0.00136)
  - 3-year TFPQ growth: -0.00120 (standard error 0.000867)
  - 3-year intangibles’ growth: -0.00503 (standard error 0.0110)
  - Observations: 362, 363, 362, 210 (columns 1–4)
  - R-squared: 0.268, 0.196, 0.254, 0.233
- Province-level capital responsiveness (selected):
  - Years 2003-2007 (col 1): coefficient on log(Average Product of Capital) = 0.114*** (standard error 0.000923); interaction log(ARPK) x High State Presence = 0.00171 (standard error 0.00188)
  - Years 2014-2018 (col 2): coefficient on log(Average Product of Capital) = 0.0766*** (standard error 0.00161); interaction log(ARPK) x High State Presence = -0.0325*** (standard error 0.00395)
  - Observations: 601,764 (2003-2007), 186,697 (2014-2018)
- Orbis cleaned manufacturing database — Table 3 (observations and averages, selected years):
  - 2003: Observations 86,968; Revenues 91,636,403; Costs 85,481,774; Total Assets 94,817,986; Tangible Fixed Assets 36,034,866
  - 2010: Observations 262,169; Revenues 153,797,953; Costs 142,712,045; Total Assets 125,190,272; Tangible Fixed Assets 40,524,759
  - 2015: Observations 98,698; Revenues 425,288,257; Costs 412,465,565; Total Assets 545,848,812; Tangible Fixed Assets 161,837,954
  - 2018: Observations 280,115; Revenues 329,358,590; Costs 321,326,740; Total Assets 548,890,078; Tangible Fixed Assets 137,220,768

### Data, measurement and construction notes (selected)
- Revenue inclusion thresholds in raw Orbis:
  - 2003 to 2010: firms with more than 5 million RMB in revenues.
  - 2011 and 2012: firms with revenues over a 20 million RMB threshold.
  - 2013 to 2018: no minimum revenue threshold.
- Authors restrict sample to firms with at least 5 million RMB in revenues in all years; robustness checks include threshold of 20 million RMB.
- Capital variables: tangible fixed assets, fixed assets, and total assets; imputation procedures used in specified years (impute tangible fixed assets from fixed assets in 2003, fixed assets from tangible fixed assets in 2010; impute both from total assets in 2014, 2017 and 2018).
- Costs constructed as cost of goods solds + other operating costs; revenues measured as total operating revenues.
- Data cleaning rules: drop firms with missing or zero assets, revenues or costs; drop plants with year-to-year changes by a factor of 20 or more; trim the 0.5% tails.

### Conclusion and policy implications
- Summary of documented trends:
  - After strong productivity growth in the 2000s, China’s productivity growth stagnated in recent years.
  - Six empirical facts document a decline in business dynamism linked to weaker life-cycle growth of young firms, lower young-firm productivity and innovation, capital constraints for young/small firms, weaker capital reallocation to high-ARPK firms, and persistent SOE–private productivity gaps.
  - Higher regional SOE intensity is associated with weaker business dynamism and lower capital responsiveness in recent years.
- Policy recommendations emphasized in the text:
  - Undertake pro-market reforms to boost productivity growth.
  - SOE reform could boost productivity growth directly through better resource reallocation and indirectly by stimulating business dynamism.

*Source: wpiea2022032-print-pdf — China’s Declining Business Dynamism — Working Paper No. WP/22/32*

### 2007. Our findings suggest that the large financial frictions estimated by Bai et al. (2018)

### wpiea2022032-print-pdf - 2007. Our findings suggest that the large financial frictions estimated by Bai et al. (2018)

### Declining business dynamism — six stylized facts
- Fact 1: The revenue and asset share of young Chinese firms has declined over time.
  - The share of firms under 10 years old fell from around 70 percent in 2003-04 to around 30 percent in 2017-18.
  - Similar trend when using total assets as firm size and for service sector firms (post-2013).
  - Baseline sample imposes a RMB 5 million revenue threshold in all years; expanding to all manufacturing firms increases sample from around 200,000 firms to 700,000 firms but still shows low revenue share for young firms.
- Fact 2: Life-cycle growth of young Chinese firms has declined over time.
  - 3-5 year old firms have 3-year growth rates around 17 percentage points higher than firms in the 16+ age group.
  - Revenue growth of firms under the age of 10 (relative to 16+) is substantially smaller in 2011-2018 than in 2003-2010.
  - Results become more pronounced when controlling for initial firm size.
  - Life-cycle growth is estimated via ∆k y_ist = α + Σ_{a=1}^5 β_a I(age group=a) + FE_st + ε_ist with k = 3 and age groups 1-2, 3-5, 6-10, 11-15 and 16+ (omitted).
- Fact 3: Productivity growth and innovation over the life-cycle has weakened.
  - Flatter life-cycle growth could reflect (a) increasing distortions preventing growth, or (b) firms investing less in R&D, process efficiency, quality improvements, or other intangible inputs.
- Fact 4: Younger and smaller firms are more capital constrained than older and larger counterparts.
  - Large financial frictions estimated by Bai et al. (2018) have remained an important constraint for young firms over the last decade.
- Fact 5: The responsiveness of capital growth to the average product of capital has declined over time.
  - Estimated by regressing capital growth on the average product of capital by sub-period.
  - A decline in the capital responsiveness elasticity could reflect worsening frictions to capital reallocation.
  - Consistent with observed decline in dispersion of capital growth across firms.
- Fact 6: SOEs have lower revenue productivity (revenue over inputs) and capital productivity (revenue over capital) than private firms.
  - Productivity gaps documented prior to 2013 and remain through 2018 for a larger representative sample.
  - Productivity gaps are also large in the service sector (data available post-2013).

### Role of SOEs and cross-sectional evidence
- SOE sector context:
  - More than 80 percent of SOEs were closed, privatized, or merged between 1998 and 2007 (Hsieh and Song, 2015).
  - SOE share of industrial assets remained at almost 40% in 2019.
- Empirical association:
  - In the cross-section of provinces, business dynamism tends to be weaker where SOEs account for a larger share of assets.
  - Measures of dynamism used: life-cycle growth of firms and responsiveness of capital growth to marginal product of capital.
  - Suggestive evidence that SOE reform could boost productivity directly via resource reallocation and indirectly by stimulating business dynamism.

### Aggregate TFP and data source (Orbis)
- Data and coverage:
  - Firm-level data from Bureau Van Dijk’s Orbis database for 2003 to 2018.
  - Cleaned manufacturing database contains on average 221,180 firms per year with revenues, capital and costs.
  - Service sector coverage available from 2013 to 2018.
  - For baseline results impose RMB 5 million revenue threshold in all years (because of minimum annual revenue threshold up to 2012 in Orbis).
  - Identify around 3,200 SOEs every year on average in manufacturing and mining sectors; for 2013-2018 identify 9,200 SOEs every year on average in services sectors.
- Constructing aggregate manufacturing TFP:
  - Aggregate TFP is a gross-output weighted average of industry-level TFP based on a Cobb-Douglas industry production function.
  - Revenues used as output, tangible fixed assets as capital, and industry cost shares identify labor and intermediate inputs from sum of cost of goods sold plus non-operating costs.
  - Variables deflated using 2-digit industry deflators for revenues and intermediate inputs; sectoral TFP estimates constructed at the same level as industry-specific deflators.
  - Aggregate manufacturing revenues from Orbis are 78 percent of official aggregates on average for above-scale manufacturing.
- TFP trend:
  - Manufacturing TFP measure points to a slowdown in productivity growth post-GFC (global financial crisis).
  - Sampling changes post-2013 increase volatility of TFP series, but TFP growth slowed down considerably post-GFC relative to the previous decade.

### Interpretation and connections to literature
- Possible drivers of declining dynamism:
  - Demographic factor: an aging population may be important (analogous to U.S. evidence: Pugsley and Şahin, 2018).
  - Post-reform transition dynamics: one-off events such as WTO accession enabled private firm growth in early 2000s; rapid private entry and SOE reform raised productivity and the bar for future entrants.
  - Declining life-cycle growth may reflect less investment in intangibles or rising distortions.
- Relation to broader literature:
  - Other studies show entry of new firms accounts for a substantial share of productivity growth (e.g., Garcia-Macia et al., 2019 find entry accounts for around a quarter of U.S. productivity growth between 1983 and 2013).
  - Evidence of declining firm responsiveness to shocks in the U.S. contributing to slow productivity growth (Decker et al., 2020).
  - Large state presence may be an additional factor influencing business dynamism in some countries.

*Source: wpiea2022032-print-pdf - 2007. Our findings suggest that the large financial frictions estimated by Bai et al. (2018) — https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022032-print-pdf.pdf*

### 2012.  This can introduce a selection bias when we consider revenue growth for firms who cross the 20

### wpiea2022032-print-pdf - 2012.  This can introduce a selection bias when we consider revenue growth for firms who cross the 20

### Productivity measurement and life-cycle productivity findings
- Measurement framework:
  - TFPR asQ_si/I_si and TFPQ asQ_σ σ−1_si/I_si where Q_si is firm revenues and I_si are firm inputs with I_si = (K_α_si s wL_1−α_s si )_γ_s X_1−γ_s si. α_s and γ_s are industry cost shares, and we set σ= 4.
- Empirical findings:
  - No evidence that TFPR varies over the life-cycle, nor that it changes between the 2000s and the 2010s.
  - TFPQ of young firms relative to older firms declined substantially in the 2010s.
  - In the 2000s TFPQ growth declined monotonically with age; in the 2010s TFPQ growth is U-shaped with 6-10 year old firms experiencing lower TFPQ growth than 16+ firms.
  - Data on intangible fixed assets for a subset of firms show young firms investing relatively less in process efficiency and quality improvements in the 2010s than in the previous decade.

### Six facts on China’s declining business dynamism
- Fact 1: The revenue share of young firms has declined (aggregate finding summarized in Conclusion).
- Fact 2: The life-cycle growth of young firms relative to older incumbents has weakened.
- Fact 3: The weakening life-cycle growth is related to lower physical productivity (TFPQ) and lower innovation (intangible investment) among young firms.
- Fact 4: Younger / smaller firms are more capital constrained than larger / older firms.
  - The average product of capital is measured as revenues / capital (revenues divided by total assets).
  - Smaller firms tend to have much higher capital productivity than larger firms in 2017-18.
- Fact 5: Capital reallocation to high marginal product of capital firms has declined.
  - Dispersion in total asset growth declined from 0.126 in 2003-2007 to 0.083 in 2013-2018.
  - Regression specification for capital growth responsiveness:
    - ∆k_ist = α + β k_ist + γ arpk_ist + FE_st + ν_ist
    - ∆k_ist is the growth rate of capital of firm i in sector s between year t and t+1. k_ist is log(capital) in year t and arpk_ist is log(ARPK) in year t. FE_st is a full set of sector-year fixed effects.
  - Responsiveness of capital growth to lagged ARPK declined from 0.096 in 2004-2007 to 0.059 in 2016-2018.
  - The decline is particularly marked among the youngest and oldest firms.
- Fact 6: SOEs have persistently lower revenue and capital productivity than POEs.
  - SOEs have consistently had lower revenue productivity than private firms, with an average revenue productivity gap of around 4-5 percent.
  - These gaps are almost exclusively explained by SOEs’ lower capital productivity.
  - Service-sector gaps (2013-2018) are somewhat smaller but still statistically and economically significant.

### Regional SOE intensity and its associations with dynamism
- Motivation:
  - Regional state presence may reduce firm creation and business dynamism; the SOE share of industrial assets fell from over 54 percent in 2003 to below 43 percent in 2008, then plateaued with SOEs still accounting for 39 percent of assets in 2019.
- Life-cycle growth and SOE intensity (province-level analysis):
  - Regression: lc_pt = α + β SOEI_pt + FE_p + FE_t + ν_pt where lc_pt is 3-year average growth of young firms relative to old firms and SOEI_pt is SOE intensity (asset share).
  - Being in a province with 10 percentage point higher SOE share of assets reduces 3-year revenue and input growth by 3.5 percent for young firms, relative to an average 3-year growth rate of 17.5 percent.
  - Province-level results: higher SOE intensity associated with weaker revenue growth, capital growth, and TFPQ growth for young firms. Industry-level regressions do not find an association.
  - Interpretation: negative spillovers of local SOE intensity on private firms may reflect local regulatory barriers or political economy mechanisms limiting young firm investment or expansion.
- Capital reallocation and SOE intensity:
  - Extended specification interacts capital responsiveness with a dummy for high province-level SOE intensity; regression includes province fixed effects.
  - No significant link in 2003-2007; in 2014-2018 provinces with high SOE intensity have significantly lower capital responsiveness.
  - Suggestive mechanism: local banking systems in high-SOE provinces may be less efficient at allocating capital to productive private firms because lending incentives favor SOEs, worsening allocation among private firms.

### Key numerical results and regression estimates (selected)
- Dispersion in total asset growth: 0.126 (2003-2007) → 0.083 (2013-2018).
- Responsiveness of capital growth to lagged ARPK: 0.096 (2004-2007) → 0.059 (2016-2018).
- Province-level life-cycle regressions (Table 1):
  - State presence coefficients on 3-year growth:
    - 3-year revenue growth: -0.00349** (standard error 0.00147)
    - 3-year asset growth: -0.00291** (standard error 0.00136)
    - 3-year TFPQ growth: -0.00120 (standard error 0.000867)
    - 3-year intangibles’ growth: -0.00503 (standard error 0.0110)
  - Sample observations and fit:
    - Observations: 362, 363, 362, 210 (columns 1–4)
    - R-squared: 0.268, 0.196, 0.254, 0.233
    - Age groups in these regressions: 1-5 (young) relative to 16+ (old)
- Province-level capital responsiveness (Table 2):
  - Years 2003-2007 (col 1): coefficient on log(Average Product of Capital) = 0.114*** (standard error 0.000923); interaction log(ARPK) x High State Presence = 0.00171 (standard error 0.00188)
  - Years 2014-2018 (col 2): coefficient on log(Average Product of Capital) = 0.0766*** (standard error 0.00161); interaction log(ARPK) x High State Presence = -0.0325*** (standard error 0.00395)
  - Observations: 601,764 (2003-2007), 186,697 (2014-2018)

### Conclusion and policy implications
- Summary of documented trends:
  - After strong productivity growth in the 2000s, China’s productivity growth stagnated in recent years.
  - Six empirical facts document a decline in business dynamism linked to weaker life-cycle growth of young firms, lower young-firm productivity and innovation, capital constraints for young/small firms, weaker capital reallocation to high-ARPK firms, and persistent SOE–private productivity gaps.
  - Higher regional SOE intensity is associated with weaker business dynamism and lower capital responsiveness in recent years.
- Policy recommendations emphasized in the text:
  - Undertake pro-market reforms to boost productivity growth.
  - SOE reform could boost productivity growth directly through better resource reallocation and indirectly by stimulating business dynamism.

*Source: wpiea2022032-print-pdf - 2012.  This can introduce a selection bias when we consider revenue growth for firms who cross the 20 (IMF working paper content provided).*

### References

### References (Content Unit: References)

### References list
- Bibliographic citations included in the source (authors and papers) are listed in the original References section (Akcigit and Ates; Alon et al.; Asker et al.; Bai, Lu and Tian; Berkowitz et al.; Biesebroeck et al.; Bils, Klenow and Ruane; Brandt and coauthors; Decker et al.; Dı́ez, Fan and Villegas-Sanchez; Eslava, Haltiwanger and Pinzón; Foster, Haltiwanger and Syverson; Garcia-Macia, Hsieh and Klenow; Gopinath et al.; Haltiwanger, Jarmin and Miranda; Hsieh and Song; Hsieh and Klenow; Jurzyk and Ruane; Pugsley, Sedlacek and Sterk; Pugsley and Şahin; Rosen et al.; Timmer et al.; Tombe and Zhu), as provided in the source.

### Data Appendix — dataset coverage and sample construction
- Main dataset: Bureau Van Dijk’s Orbis database; coverage of Chinese manufacturing sector from 2003 to 2018.
- Revenue inclusion thresholds by year in raw Orbis:
  - 2003 to 2010: firms with more than 5 million RMB in revenues.
  - 2011 and 2012: firms with revenues over a 20 million RMB threshold.
  - 2013 to 2018: no minimum revenue threshold.
- Sample restriction for comparability:
  - Authors restrict sample to firms with at least 5 million RMB in revenues in all years.
  - Robustness checks include setting threshold to 20 million RMB in all years.
- Capital variables:
  - Main capital variables: tangible fixed assets, fixed assets, and total assets.
  - Total assets are reported by all firms in all years; tangible fixed assets and fixed assets are missing in some years.
  - Imputation approach: missing tangible fixed assets and fixed assets are imputed using industry shares and the firm’s reported total assets.
  - Imputation specifics (footnote): “We impute tangible fixed assets from fixed assets in 2003, and fixed assets from tangible fixed assets in 2010. We impute both fixed assets and tangible fixed assets from total assets in 2014, 2017 and 2018.”
- Costs and revenues:
  - Costs constructed as cost of goods solds + other operating costs (includes labor costs and material inputs; materials and labor not separated due to reporting gaps).
  - Revenues measured as total operating revenues (includes sales and other revenues).
- Data cleaning rules:
  - Drop firms with missing or zero assets, revenues or costs.
  - Drop plants where revenues, assets or costs increase or decrease by a factor of 20 or more from year to year.
  - Trim the 0.5% tails of revenues/costs, revenues/assets and assets/costs.
- The final number of observations in each year and average revenues, costs and capital are reported in Table 3 (reproduced below).

### Table 3 — Observations and averages in cleaned Orbis manufacturing database (as reported)
- 2003: Observations 86,968; Revenues 91,636,403; Costs 85,481,774; Total Assets 94,817,986; Tangible Fixed Assets 36,034,866
- 2004: Observations 161,234; Revenues 73,647,919; Costs 69,086,726; Total Assets 69,417,834; Tangible Fixed Assets 26,456,472
- 2005: Observations 180,741; Revenues 92,110,839; Costs 86,920,457; Total Assets 79,579,964; Tangible Fixed Assets 30,092,239
- 2006: Observations 265,105; Revenues 101,337,358; Costs 94,970,617; Total Assets 82,154,171; Tangible Fixed Assets 29,952,293
- 2007: Observations 303,623; Revenues 116,731,065; Costs 107,981,800; Total Assets 92,800,197; Tangible Fixed Assets 32,772,664
- 2008: Observations 353,580; Revenues 111,435,921; Costs 104,431,020; Total Assets 87,078,285; Tangible Fixed Assets 30,141,470
- 2009: Observations 294,682; Revenues 132,614,342; Costs 123,542,275; Total Assets 108,783,969; Tangible Fixed Assets 36,732,682
- 2010: Observations 262,169; Revenues 153,797,953; Costs 142,712,045; Total Assets 125,190,272; Tangible Fixed Assets 40,524,759
- 2011: Observations 242,957; Revenues 276,088,841; Costs 256,496,768; Total Assets 213,095,978; Tangible Fixed Assets 71,524,902
- 2012: Observations 235,888; Revenues 292,644,457; Costs 273,609,596; Total Assets 230,397,440; Tangible Fixed Assets 73,899,523
- 2013: Observations 250,795; Revenues 29,5314,973; Costs 276,183,723; Total Assets 239,986,516; Tangible Fixed Assets 77,116,929
- 2014: Observations 241,776; Revenues 225,927,394; Costs 214,976,571; Total Assets 253,124,775; Tangible Fixed Assets 75,962,412
- 2015: Observations 98,698; Revenues 425,288,257; Costs 412,465,565; Total Assets 545,848,812; Tangible Fixed Assets 161,837,954
- 2016: Observations 125,465; Revenues 434,359,175; Costs 410,411,028; Total Assets 592,013,264; Tangible Fixed Assets 167,105,593
- 2017: Observations 155,087; Revenues 519,030,692; Costs 499,327,685; Total Assets 724,920,665; Tangible Fixed Assets 184,086,506
- 2018: Observations 280,115; Revenues 329,358,590; Costs 321,326,740; Total Assets 548,890,078; Tangible Fixed Assets 137,220,768
- Notes (from table): “This table reports the number of observations in the cleaned Orbis manufacturing database in each year, along with average revenues, costs, total assets and tangible fixed assets.”

### State Ownership Data — classification and comparisons
- Ownership classification sources:
  - Listed firms: WIND data classifying firms each year into private or state-owned.
  - Unlisted firms: Orbis historical ownership databases.
- State-owned firm definition (Orbis): Global Ultimate Owner controlling at least 50 percent of the company and entity type is “S” (i.e. the state).
- Use of last-observed POE/SOE classification for sector/province-level analysis; observed transitions:
  - 1.16 percent show a transition from POE to SOE.
  - 0.42 percent show a transition from SOE to POE.
  - Authors note transitions are overall relatively rare, supporting use of last-observed classification.
- Comparison with CEIC/NBS aggregates:
  - Figures compare Orbis-WIND bottom-up estimates of SOE assets / total assets and SOE revenues / total revenues with aggregates from NBS compiled by CEIC (restricted to NACE 2-digit sectors 5 to 36).
  - Bottom-up estimates broadly match official percentages in levels and show the decline of the early 2000s.
  - Since around the GFC, bottom-up estimates show a slight upward trend not present in official data.
  - Cross-sectional correlations: bottom-up state-presence estimates tend to be highly correlated with CEIC-derived measures; 2018 sector-level comparison largely consistent, with a few sector disagreements (machinery repair, water supply).

### Constructing Sectoral TFP Measures — production function, deflators, elasticities
- Sectoral production function (as posited in source; variables in logs):
  - Q_jt = A_jt (K_sjt^α_s L_jjt^(1−α_j))^γ_j X_jt^(1−γ_j)
  - Sectoral TFP residual formula (no capitalization indicates logs):
    - a_jt = q_jt − γ_j(α_j k_jt + (1−α_j) l_jt) − (1−γ_j) x_jt .(5)
- Aggregation to sector-level TFP:
  - a_st = sum_{j∈s} (P_jt Q_jt / P_st Q_st) a_sjt
- Data and mappings used:
  - Industry classification: Orbis uses NACE Rev. 2; WIOD uses ISIC Rev. 4. The mapping from WIOD sectors to NACE Rev. 2 is provided in Table 4 (detailed sector correspondences included in source).
  - WIOD’s global input-output data used extensively (Timmer, Dietzenbacher, Los, Stehrer and de Vries, 2015).
- Nominal variables and imputations:
  - Nominal gross output measured as Operating revenue; capital measured as Tangible fixed assets.
  - Ideal components (Cost of employees, material costs) often missing → use Cost of goods sold + Operating expenses to measure sum of nominal wage bill and nominal intermediate spending.
  - Impute components using WIOD’s Socio Economic Accounts data for compensation of employees (COMP) and intermediate-input spending (II), using the average for 2003-2014.
- Price indices and deflators:
  - Gross-output deflators: for most WIOD sectors, a corresponding PPI index from NBS (retrieved from CEIC) is available; for some services sectors use GDP deflator for closest-matching sector (retrieved from Haver). Mapping shown in Table 5 (sectoral gross output price indices).
  - Intermediate-input deflators: weighted average of gross-output deflators of all sectors, weights given by sector’s 2004-2014 purchases from each sector (WIOD purchase data).
  - Capital deflator: NBS Fixed Asset Investment price index (retrieved from CEIC).
  - Wage deflators: NBS sectoral wage deflators; mapping to WIOD sectors shown in Table 6.
- Production elasticities:
  - Rental rate of capital assumed: R = 0.2 (Bils et al., 2021).
  - Industry-specific production elasticities α_jt and γ_jt constructed as capital share of value added and capital-cum-labor share of gross output, over the 2004-2016 sample:
    - α_jt = α_j = R K_j / (R K_j + w L_j)
    - γ_jt = γ_j = (R K_j + w L_j) / (R K_j +  w L_j + P_x X_j)
  - Bars over variables denote average values (used in construction).

### Mappings and tables included in source
- Table 4: NACE Rev. 2 to WIOD correspondence (detailed per-sector mappings).
- Table 5: Sectoral gross output price indices (WIOD sector → Price index → Source: CEIC or Haver).
- Table 6: NBS wages to WIOD mapping (WIOD sector → NBS sectoral wage).

*China’s Declining Business Dynamism — Working Paper No. WP/22/32*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022032-print-pdf.pdf_