## Industrial Policy in China: Quantification and Impact on Misallocation

## Source details

**Canonical URL:** [Industrial Policy in China: Quantification and Impact on Misallocation](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025155-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025155-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025155-source-pdf.pdf.json)

---

### Introduction — purpose, scope, and approach
- Purpose:
  - Industrial policies (IP) defined as policies aimed at changing the sectoral structure of the economy.
  - Two main aims:
    - Quantify the size of the main IP instruments in China.
    - Estimate the impact of IP on domestic factor misallocation and aggregate productivity using a structural model.
- Data and methods overview:
  - Financial statements of listed firms from WIND covering 2010-23.
  - Land registry covering the universe of land sales (about 1.6 million transactions).
  - Structural estimation uses IP counts from Juhasz et al. (2025), Hsieh and Klenow (2009) framework, and data from Orbis and KLEMS.
- Key scope limitations:
  - Analysis abstracts from potential benefits of IP (e.g., correcting market failures, knowledge spillovers).
  - Analysis abstracts from international spillovers of China’s IP.

### Quantification of Industrial Policy — data, measurement, and aggregate size
- Instruments quantified and measurement approaches:
  - Cash subsidies: directly reported by firms in WIND; subsidy rate = aggregate cash subsidies as a share of value added by sector.
  - Tax benefits: sector-level gap = 25 percent (corporate income tax top statutory rate) minus effective corporate income tax rate (corporate income tax payments over total earnings before taxes from WIND).
  - Subsidized credit: difference in effective nominal interest rates (r = interest payments / interest-bearing liabilities (previous year)) for manufacturing vs other sectors, estimated via firm-level pooled OLS (2010-23) with controls.
  - Subsidized land: manufacturing unit price p_i,t compared to benchmark p̂_i,t = average price of nearby non-manufacturing transactions within a 1 kilometer radius in the same year (real RMB per square meter, deflated with GDP deflator).
- Main 2023 aggregate quantification results (Figure 4; 2023 values):
  - Total IP support equivalent to up to 4.4 percent of GDP as of 2023.
  - Breakdown in 2023:
    - Cash subsidies: 2.0 percent of GDP
    - Tax benefits: 1.5 percent of GDP
    - Land subsidies: 0.5 percent of GDP
    - Subsidized credit: 0.4 percent of GDP
- Temporal and distributional patterns:
  - Total size of IP broadly stable over time.
  - Tax subsidies increased post-pandemic; other instruments slightly diminished.
  - SOEs tend to benefit from lower interest rates and higher cash subsidy rates after controlling for sector.
  - Tax benefits are higher for private firms (POEs).
  - Private firms dominate sectors favored by IP, indicating IP extends beyond SOE support.
- Detailed measurement findings (selected):
  - Aggregate cash subsidy rate declined from 2.4 percent in 2013 to 2.0 percent in 2023.
  - Aggregate tax benefit rates increased from 4.4 percent in 2013 to 6.3 percent in 2023.
  - Manufacturing firms benefit from effective interest rates about 0.4 percentage points below others (manufacturing coefficients: -0.43***, -0.4***, -0.42***, -0.32*** across specifications).
  - Central SOE coefficient on effective rates: -0.55*** (central SOEs tend to benefit from lower rates).
  - Median manufacturing land price roughly stable and slightly above 200 RMB per square meter; median nearby non-manufacturing price consistently above 600 RMB per square meter — implied manufacturing price discount of at least 2/3.
  - Aggregate land sales in manufacturing as share of GDP = 0.3 percent in 2023.
- Caveats on aggregation and extrapolation:
  - Extrapolation from listed to non-listed firms relies on strong assumptions.
  - Calculations do not account for general equilibrium effects (e.g., impact on interest rates or tax revenues).
  - Other IP instruments not covered could imply underestimation.
  - Example structural-model comparison: structural model suggests cash and credit subsidies amounted to 1.5 percent of GDP in 2018, compared with 2.2 percent of GDP estimated here using listed-firm data — extrapolation here may be an upper bound.
  - Illustrative extreme assumption: if non-listed firms received no support, total IP would amount to about 1 percent of GDP.

### Impact on misallocation and productivity — model, empirical tests, and results
- Theoretical framework:
  - Firms produce with Cobb-Douglas: y_i = a_i k_i^{α_s} l_i^{1−α_s}.
  - Monopolistic competition with CES demand; TFPR defined as TFPR_i = p_i y_i / (k_i^{α_s} l_i^{1−α_s}).
  - In equilibrium TFPR_i ∝ 1/(1+ω_i); TFPR differences across firms reveal distortions and factor misallocation.
  - Aggregate TFP approximation (log):
    - log TFP ≅ (1/(σ−1)) log(∑ a_i^{σ−1}) − (σ/2) var(log TFPR_i) (exact if a and TFPR jointly log-normal).
- Empirical hypotheses (four):
  - H1: Higher subsidies → lower sector-level TFPR.
  - H2: Trade/regulatory entry barriers → higher sector-level TFPR.
  - H3: More IP measures → higher TFPR dispersion within sector.
  - H4: IP does not affect firm-level TFP.
- Data and identification:
  - IP counts from GTA (Juhasz et al. (2025)), reclassified into subsidized credit, other subsidies, tax relief, trade barriers, regulatory barriers.
  - Sector-level regressions (NACE 4-digit) use IP intensity defined as IP_{2009−18,s}^x = log(1 + ∑_{t=2009}^{2018} NIP_{t,s}^x).
  - Firm-level TFPR estimated from value added, fixed assets, labor, and sectoral capital shares α_{s,t} from KLEMS.
  - Firm-level TFP from Diez et al. (2021) on Orbis using Ackerberg et al. (2015).
- Key empirical findings — between-sector effects:
  - Subsidies (subsidized credit and other subsidies) tend to lower sector TFPR.
  - Trade and regulatory barriers tend to increase sector TFPR.
  - Subsidized credit coefficient larger in absolute terms and more statistically significant than other subsidies; other subsidies are twice as frequent so total impact larger.
  - Tax effects ambiguous and statistically insignificant.
  - Robustness: similar results using average outcome 2015-18 or median TFPR by sector.
- Key empirical findings — within-sector effects:
  - Sectors with some IP (2009-18) have interquartile range of log(TFPR) shifted right.
  - Sectors with some IP have 14 percent higher within-sector TFPR dispersion (statistically significant).
  - Similar results using IP intensity or standard deviation of TFPR.
- Contribution to misallocation and cross-country comparison:
  - IP explains 24 percent of overall between-sector TFPR dispersion.
  - IP explains 4 percent of total within-sector misallocation.
  - Overall misallocation between firms is three times as large as between sectors.
  - G7 benchmark (France, Germany, Italy, Japan, United Kingdom): relationship between IP and misallocation not statistically significant for 2009-18; G7 countries have less overall misallocation than China.
- Firm-level TFP:
  - No statistically significant relationship between sector-level IP intensity and average firm-level TFP under any specification.
  - Null result holds for sector-level average (or median) change in within-firm log(TFP) between 2008 and 2018.
- Aggregate TFP and GDP implications:
  - Using σ = 4 in equation (1):
    - IP-induced misallocation estimated to lower aggregate TFP in China by 1.2 percent.
      - Of the 1.2 percent, 1.0 percentage points from between-sector misallocation and 0.2 from within-sector misallocation.
    - Assuming capital adjusts endogenously, the TFP loss could imply a GDP level loss of up to 2 percent.
  - Note: results may be a lower bound due to potential attenuation bias from imperfect IP measurement.

### Annex II — implicit fiscal cost from structural estimation and cross-checks
- Method:
  - Structural model yields analytic relation between TFPR distortions and subsidies (ω): log TFPR ∝ −ω.
  - Combining this relation with regression estimates of IP subsidies produces an implicit fiscal cost calculation that includes cash and credit subsidies and excludes tax benefits (tax coefficient close to zero and opposite sign).
- Quantitative cross-checks:
  - Estimated fiscal cost of subsidies using structural-estimation cross-check = 1.5 percent of value added in 2018.
  - Comparison: Section 2 estimate for listed firms = 2.2 percent of GDP in 2018.
    - Interpretation: listed firms may capture more subsidies; extrapolation from listed firms may overstate total support.
    - Caveat: Annex II estimate may be subject to attenuation bias due to IP counts being an imperfect measure; result described as tentative.
- Related empirical evidence on industrial champions:
  - Leaders (largest revenue share in each sector) typically have higher TFP than the average firm.
  - Leaders typically have lower TFPR than the average firm, implying they produce at inefficiently high levels.
  - TFPR gap particularly pronounced for SOE leaders — consistent with SOE advantages and implicit guarantees.
- Aggregate recap:
  - Equivalent fiscal cost of IP in China quantified at about 4 percent of GDP per year (instrument-by-instrument using listed firms and land registry).
  - Misallocation from IP estimated to reduce aggregate productivity by about 1.2 percent and GDP by up to 2 percent.

### Policy-relevant implications and recommendations
- Increase transparency around IP, especially local government IP measures, to improve policymaking and address trading-partner concerns.
- Scale down IP to reduce factor misallocation and fiscal costs; pursue IP cautiously and only to tackle well-defined market failures.
- When IP is implemented, prefer budgetary tools because they tend to be more transparent and less distortionary than indirect measures such as credit allocations or regulation.
- Research and data priorities:
  - Further work needed to quantify other IP instruments (e.g., Government Guided Funds, Public-Private Partnerships).
  - Better data reporting would facilitate more accurate characterization of Chinese IP and its economic effects.
  - Recognize strong assumptions used to overcome data gaps (e.g., extrapolating to non-listed firms, reliance on IP counts for structural estimation).

*IMF Working Paper — Industrial Policy in China: Quantification and Impact on Misallocation (Working Paper No. WP/2025/155).*

### 1. Introduction ........................................................................................................

### 1. Introduction

### Purpose and scope
- Industrial policies (IP) are defined as policies aimed at changing the sectoral structure of the economy.
- The paper has two main aims:
  - Quantify the size of the main IP instruments in China.
  - Estimate the impact of IP on domestic factor misallocation and aggregate productivity using a structural model.

### Quantification of IP instruments: data and methods
- Data sources and period:
  - Financial statements of listed firms from WIND covering 2010-23.
  - Land registry covering the universe of land sales (about 1.6 million transactions).
- Four IP instruments quantified and measurement approaches:
  - Cash subsidies: directly reported by firms.
  - Tax benefits: measured as the gap between the statutory tax rate and effective corporate income tax rates computed from financial statements at the sector level.
  - Subsidized credit: identified as the difference in effective interest rates for firms in the manufacturing sector versus other sectors, after controlling for other financial variables at the firm level and firm type.
  - Subsidized land: estimated by comparing unit prices of land sold by the government to manufacturing firms with prices for nearby firms (within a one-kilometer radius) in other sectors and the same year.

### Main quantification results
- Equivalent fiscal cost of IP in China as of 2023:
  - Total IP: 4.4 percent of GDP.
  - Cash subsidies: 2.0 percent of GDP.
  - Tax benefits: 1.5 percent of GDP.
  - Land subsidies: 0.5 percent of GDP.
  - Subsidized credit: 0.4 percent of GDP.
- Temporal evolution:
  - Total size of IP broadly stable over time.
  - Tax subsidies have grown in importance in the aftermath of the pandemic.
  - Use of other instruments has slightly diminished.
- Distributional observations:
  - State-owned enterprises (SOEs) tend to benefit from lower interest rates and higher cash subsidy rates (after controlling for sector).
  - Tax benefits are higher for private firms.
  - Private firms are dominant in the sectors favored by IP, indicating IP extends beyond SOE support.

### Impact on misallocation and productivity: approach and findings
- Structural estimation framework:
  - Uses IP counts from Juhasz et al. (2025).
  - Measures misallocation and aggregate productivity using the Hsieh and Klenow (2009) model and data from Orbis and KLEMS.
  - Estimates cumulative effect of IP policies implemented between 2009 and 2018 on sector-level misallocation outcomes in 2018.
  - Approximates impact of IP on aggregate total factor productivity (TFP) through changes in allocative efficiency; captures general equilibrium effects and covers a wider sample of firms and IP tools (including trade and regulatory barriers).
- Key empirical findings:
  - Different IP instruments affect allocation in opposite directions:
    - Subsidies are associated with excess production relative to a no-distortions benchmark.
    - Trade and regulatory barriers limit production, possibly by increasing market power of incumbents.
  - Factor misallocation from IP is estimated to reduce domestic aggregate TFP by about 1.2 percent relative to a no IP baseline.
  - This channel could reduce the level of GDP by up to 2 percent.
  - Industrial champions (market-leading firms) derive their position from both higher productivity and policies encouraging their production relative to the average firm in the sector.
  - No evidence that sector-level IP intensity significantly affects average firm-level TFP in the same sector; potential benefits (e.g., knowledge spillovers) may be offset by negative effects such as decreased competition or within-firm factor misallocation.
- Scope limitations:
  - The analysis abstracts from potential benefits of IP (e.g., correcting market failures, knowledge spillovers) that could lead to positive TFP or welfare effects.
  - The analysis abstracts from international spillovers of China’s IP.

### Relation to existing literature
- Builds on literature focused on SOE advantages and individual policy instruments or sectors.
- Contributes by:
  - Defining IP based on sector-level support consistent with international literature.
  - Providing a comprehensive estimate of the size of IP across various instruments for the whole economy.
  - Estimating impacts on both between-sector and within-sector misallocation, finding between-sector misallocation impacts substantially larger than within-sector impacts.
- Complements recent work measuring IP policy counts in China.

### Organization of the paper
- Section 2: describes the quantification of IP instruments.
- Section 3: presents the structural estimation of the impact of industrial policy on misallocation and productivity.
- Section 4: concludes with policy takeaways.

*IMF Working Papers — Industrial Policy in China: Quantification and Impact on Misallocation (Introduction).*

### 2. Quantification of Industrial Policy

### 2. Quantification of Industrial Policy

### Data and scope
- Instruments quantified: cash subsidies, tax benefits, subsidized credit, and subsidized land.  
- Data sources and coverage:
  - Financial statements of listed firms from WIND, covering more than 5,000 listed firms over the 2010-23 period (cash subsidies, tax benefits, subsidized credit).
  - Land registry data from the Ministry of Natural Resources (universe of government land sales, about 1.6 million transactions) over 2010-23 (land subsidies).  
- Annex I contains more detailed data descriptions and summary statistics for key variables.

### Cash subsidies — measurement and findings
- Measurement: reported as a separate income-statement item for listed firms in WIND; subsidy rate defined as aggregate cash subsidies as a share of value added by sector.
- Sector patterns:
  - Most subsidized sectors include semiconductors, high-tech manufacturing, and automobiles.
  - Least subsidized sectors include consumer goods, services, real estate, and energy.
  - Sectoral ranking of subsidy rates is stable over time.
- Aggregate evolution:
  - Aggregate cash subsidy rate declined from 2.4 percent in 2013 to 2.0 percent in 2023.
- Firm-type differences:
  - Privately-owned enterprises (POEs) have higher subsidy rates for most of the period, driven by a larger POE share in favored sectors.
  - After controlling for sector, SOEs have an average subsidy rate 0.7 percentage points higher than POEs (statistically significant).

### Tax benefits — measurement and findings
- Measurement: sector-level tax benefits = 25 percent (corporate income tax top statutory rate) minus effective corporate income tax rate (corporate income tax payments over total earnings before taxes from WIND).
- Scope: captures asset-specific investment incentives and tax discounts where incidence varies across sectors; excludes other taxes with sectoral rates (e.g., value added taxes, social security contributions).
- Sector correlation:
  - Ranking of sectors by tax benefits is highly correlated with ranking by cash subsidies (6 out of 7 sectors at top and bottom thirds overlap).
- Aggregate evolution:
  - Aggregate tax benefit rates increased from 4.4 percent in 2013 to 6.3 percent in 2023.
- Data validation notes:
  - Estimates are close to zero or negative for some sectors (reassurance against systematic overstatement).
  - Effective tax rates would be lower if deferred tax expenses were excluded, implying a higher estimated tax benefit.
- Firm-type differences:
  - POEs receive larger tax benefits as a percent of profits than SOEs; sectoral distribution explains about half of the POE advantage.
  - Possible explanations: stronger tax compliance by SOEs and lower need for tax incentives to align SOE production with government goals.

### Subsidized credit — measurement and findings
- Measurement: credit subsidies proxied by differences in effective nominal interest rates across firms/sectors not explained by controls. Effective nominal interest rate r = interest payments / interest-bearing liabilities (previous year) from WIND.
- Regression specification (firm-level pooled OLS, 2010-23) includes:
  - Sector categorical variables, firm-type dummies (central SOEs, local SOEs, foreign firms, POEs excluded), controls z (leverage, log assets, short-term debt share, intangible asset share), year fixed effects; financial controls lagged.
- Key regression findings (Table 1 summary):
  - Most-favored sectors (top by cash subsidies and tax benefits) coefficient: -0.23** (relative to other sectors).
  - Least-favored sectors coefficient: -0.17 (not statistically significant).
  - Manufacturing (relative to other sectors) coefficients across specifications: -0.43***, -0.4***, -0.42***, -0.32*** (benchmark: manufacturing firms benefit from effective rates 0.4 percentage points below others).
  - Central SOE coefficient: -0.55*** (or similar across specs) — central SOEs tend to benefit from lower rates.
  - Local SOE coefficient: -0.19** (or similar).
  - Foreign Firm coefficient: -0.39**.
  - Debt/Assets (lag) coefficient: -0.01** (higher leverage associated with slightly lower rates).
  - Log Assets (lag) coefficient: -0.13*** (larger firms get lower rates).
  - Short-Term Debt Share (lag) coefficient: 1.02*** (higher short-term debt share associated with higher effective rates).
  - Intangible Asset Share (lag) coefficients: not statistically significant in main specs (e.g., 0.6, 0.37, 0.41, 0.4, 0.25 across columns).
  - Return on Assets (lag) entered in some specs with coefficient -1.83***.
  - Industry Price-to-Earnings (lag) coefficient: 0.00 (not significant).
  - R2 values around 0.05–0.06.
  - Number of observations: 32,159 (except 31,811 in one specification).
- Robustness and interpretation:
  - Manufacturing discount robust to excluding firm-type control, adding profitability controls, using current denominator for interest-bearing liabilities, and adding an industry-risk proxy.
  - Discount becomes smaller and statistically insignificant when financial controls are removed.
  - No statistically significant difference found for zero-interest net accounts payable by manufacturing firms.
- Interpretation: SOEs, particularly central SOEs, likely benefit from lower rates due to lower perceived risk and implicit government guarantees; bank behavior (SOE banks assisting higher-risk firms) may explain some results.

### Subsidized land — measurement and findings
- Measurement approach:
  - Compare unit price p_i,t of land sold by government to manufacturing firms with benchmark price p̂_i,t = average price of nearby non-manufacturing transactions within a 1 kilometer radius in the same year (prices in real RMB per square meter, deflated with GDP deflator).
  - Dataset: land registry of government land sales, 1.6 million transactions between 2010-23; exclude land for public uses, top/bottom percentile unit prices, and transactions with fewer than two nearby non-manufacturing transactions (N < 2).
  - After filters, ~130,000 manufacturing observations remain (about one third of manufacturing transactions) for which equivalent prices computed.
- Findings from Figure 3 and analysis:
  - Median manufacturing land price roughly stable and slightly above 200 RMB per square meter.
  - Median unit price for nearby non-manufacturing sectors consistently above 600 RMB per square meter.
  - Implied manufacturing price discount of at least 2/3 relative to nearby non-manufacturing land.
  - Manufacturing discount increased during the property market boom that ended in 2021 and eased with the subsequent correction — suggesting government shielding of manufacturing firms from generalized land value rises.
- Controls for land development:
  - Regression controls: manufacturing dummy, log area, year fixed effects, city × land-grade categorical interactions (land grade = government rating of land development).
  - Inclusion of land grade accounts for less than 10 percent of the discount: manufacturing gap coefficient falls from 616 to 567 RMB per square meter.
  - Conclusion: differences in land development do not explain the bulk of the manufacturing discount.
- Firm-type differences:
  - Merge with Orbis (SOE defined as public authorities as global ultimate owner with ≥ 50 percent ownership) shows no statistically significant average unit price differences between SOEs and POEs; year-to-year ranking fluctuates — no evidence of systematic SOE favoritism in industrial land prices.

### Total size of Industrial Policy (IP)
- Aggregation and extrapolation methods:
  - Cash subsidies: aggregate subsidies for listed firms as share of their combined value added, assume same subsidy rate for non-listed firms.
  - Subsidized credit: subsidy rate by year from regression interacting manufacturing dummy with year fixed effects (Table 1); multiply coefficients by aggregate interest-bearing liabilities of manufacturing sector divided by total value added in WIND; applied to all manufacturing firms including non-listed.
  - Tax benefits: WIND tax benefit rate extrapolated to non-listed firms, correcting for disproportionate share of corporate income taxes contributed by listed firms relative to their value-added share.
  - Land subsidies: assume shadow market price for manufacturing properties equals price of nearby non-manufacturing properties p̂_i,t; aggregate land subsidy rate = median discount ((p̂_i,t − p_i,t) / p_i,t) times aggregate land sales in manufacturing as share of GDP (equal to 0.3 percent in 2023).
- Aggregate results (Figure 4; 2023 values):
  - Total IP support equivalent to up to 4.4 percent of GDP as of 2023.
  - Breakdown in 2023:
    - Cash subsidies: 2.0 percent of GDP
    - Tax benefits: 1.5 percent of GDP
    - Land subsidies: 0.5 percent of GDP
    - Subsidized credit: 0.4 percent of GDP
  - Time-series note: total IP broadly stable over time; composition shifted with tax benefits rising after the pandemic and other instruments slightly diminishing.
- Comparative context:
  - State aid provided by EU countries (cash subsidies, tax benefits, credit subsidies) ≈ 1.5 percent of GDP in 2022.
  - Criscuolo et al. (2023) find a similar average across 9 OECD economies.
  - Conclusion: IP appears to be used more intensively in China.

### Caveats and limitations
- Extrapolation to non-listed firms is large and relies on strong assumptions due to lack of non-listed firm data.
- Calculations do not account for general equilibrium effects (e.g., higher cash subsidies affecting interest rates or tax revenues), which could imply potential overstatement of total support.
- Other IP instruments not covered (e.g., sector-specific taxes beyond corporate income tax, subsidized equity via Government Guided Funds) could imply the estimates are an underestimate.
- Annex II structural model (Section 3) provides alternative estimates including non-listed firms; example: structural model suggests cash and credit subsidies amounted to 1.5 percent of GDP in 2018, compared to 2.2 percent of GDP estimated here using listed-firm data — indicating extrapolation here might be an upper bound.  
- Illustrative extreme assumption: if non-listed firms received no support, total IP would amount to about 1 percent of GDP.

*IMF Working Paper — Chapter 2: Quantification of Industrial Policy*

### 3. Impact on Misallocation and Productivity

### 3. Impact on Misallocation and Productivity

### Model and theoretical framework
- Firms indexed by i produce with Cobb-Douglas production function: 푦ᵢ = 푎ᵢ 푘ᵢ^{α_s} 푙ᵢ^{1−α_s}, where 푎 is firm-specific TFP, 푘 capital, 푙 labor, and α_s the sector-specific capital share.
- Monopolistic competition with CES demand (elasticity σ) implies price pᵢ ∝ yᵢ^{−σ}.
- Profit maximization with firm-specific production subsidy/distortion ωᵢ: 휋ᵢ = (1+ωᵢ) pᵢ yᵢ − w lᵢ − r kᵢ.
- Firm total factor productivity in revenues (TFPR) defined as TFPRᵢ = pᵢ yᵢ / (kᵢ^{α_s} lᵢ^{1−α_s}).
- In equilibrium TFPRᵢ ∝ 1/(1+ωᵢ): TFPR differences across firms reveal distortions and factor misallocation; TFPR does not depend on physical productivity a.
- Aggregate TFP approximation (log scale):
  - log TFP ≅ (1/(σ−1)) log(∑ aᵢ^{σ−1}) − (σ/2) var(log TFPRᵢ). (Approximation exact if a and TFPR jointly log-normal.)

### Empirical approach and hypotheses
- Four empirical hypotheses:
  - Hypothesis 1: Sectors with higher subsidies tend to feature lower sector-level TFPR.
  - Hypothesis 2: Sectors subject to entry barriers from trade or regulatory policies tend to feature higher sector-level TFPR.
  - Hypothesis 3: Sectors with more IP measures tend to have higher TFPR dispersion within the sector.
  - Hypothesis 4: IP does not affect firm-level TFP.
- Sector-level regression specification (NACE 4-digit, controls and lag):
  - y_{2018,s} = β x * IP_{2009−18,s}^x + δ * y_{2008,s} + γ_{s̅} + ε_s.
- IP intensity defined as IP_{2009−18,s}^x = log(1 + ∑_{t=2009}^{2018} NIP_{t,s}^x), where NIP is number of IP measures of type x covering sector s in year t.
- Dependent variables:
  - For Hypotheses 1 and 2: sector-level average log(TFPR_{i,t}).
  - For Hypothesis 3: interquartile range of log(TFPR_{i,t}) within sector.
  - For Hypothesis 4: sector-level average log(TFP_{i,t}).
- Time range 2009-18 driven by IP counts data availability and Orbis sample limitations.
- Caution: IP intensity likely endogenous; results are associations not causal estimates.

### Data and measurement
- IP counts from GTA database (Juhasz et al. (2025) text-based analysis). GTA measures nationwide policies only and counts measures irrespective of size.
- IP measures reclassified into five categories: subsidized credit, other subsidies, tax relief, trade barriers, regulatory barriers.
- Firm-level TFPR estimation:
  - TFPR̂ᵢ = value addedᵢ / (fixed assetsᵢ^{α_{s,t}} (laborᵢ)^{1−α_{s,t}}).
  - Value added = gross output − intermediate inputs.
  - Labor input proxied by firm’s cost of goods sold × sector-level share of labor costs in cost of goods sold (from KLEMS) when cost of employees missing.
  - Capital share α_{s,t} estimated from average capital to value added across bottom-quarter countries by cross-sectoral TFPR dispersion in KLEMS (countries listed: Sweden, Denmark, Czechia, Japan, and Italy).
- Firm-level TFP from Diez et al. (2021) using Ackerberg et al. (2015) on Orbis.

### Results — between-sector misallocation
- Regression with five IP types as separate regressors shows:
  - Subsidies (subsidized credit and other subsidies) tend to lower sector TFPR.
  - Trade and regulatory barriers tend to increase sector TFPR.
  - Coefficient for subsidized credit larger in absolute terms and more statistically significant than other subsidies; other subsidies are twice as frequent, so total impact larger.
  - Tax effects ambiguous and statistically insignificant.
- Robustness: similar results if dependent variable is average outcome over 2015-18 or median log(TFPR) by sector.

### Results — within-sector misallocation
- Sectors with at least some IP between 2009 and 2018 have the distribution of the interquartile range of log(TFPR) shifted right relative to sectors with no IP measures.
- Formal test (sector dummy for any IP): sectors with some IP have 14 percent higher within-sector TFPR dispersion; difference statistically significant.
- Similar results using IP intensity instead of a dummy or using standard deviation of TFPR instead of interquartile range.
- Comparison: Chen et al. (2022) estimates a 6 percent increase in within-sector variance of log(TFPR) for industries targeted by IP in the 10th 5-Year Plan (2001-05).

### Contribution to overall misallocation and cross-country comparison
- Contribution of IP to between-sector TFPR dispersion:
  - IP explains 24 percent of overall between-sector TFPR dispersion (figure constructed from regression residual comparisons as described).
- Contribution of IP to within-sector misallocation:
  - IP explains 4 percent of total within-sector misallocation.
- Overall misallocation between firms is three times as large as between sectors.
- International benchmark (G7 countries with large Orbis samples: France, Germany, Italy, Japan, United Kingdom):
  - Relationship between IP and misallocation (Hypotheses 1-3) not statistically significant for G7 countries over 2009-18.
  - G7 countries have less overall misallocation than China; part of the gap may be explained by IP-induced misallocation in China, especially for between-sector misallocation.

### Firm-level TFP
- No statistically significant relationship between IP and average firm-level TFP of a sector under any specification.
- Null result holds for sector-level average (or median) change in within-firm log(TFP) between 2008 and 2018.
- Implication: calculation of aggregate TFP impact focuses on the misallocation channel, abstracting from firm-level TFP effects.

### Aggregate TFP impact and GDP implication
- Calculation uses regression results and equation (1) with σ = 4 (as in Hsieh and Klenow (2009)).
- IP-induced misallocation estimated to lower aggregate TFP in China by 1.2 percent.
  - Of the 1.2 percent, 1.0 percentage points come from between-sector misallocation and 0.2 from within-sector misallocation.
- Assuming capital adjusts endogenously, the TFP loss could imply a GDP level loss of up to 2 percent.
- Note on interpretation: results may be a lower bound due to potential mismeasurement of IP intensity causing attenuation bias.

*Source: IMF Working Papers — Industrial Policy in China: Quantification and Impact on Misallocation (Section 3).*

### Annex II uses the model to infer the implicit fiscal cost of IP subsidies (including cash and credit), providing a

### Annex II. Implicit Fiscal Cost from Structural Estimation

### Model and Method
- The structural model in Section 3 yields an analytical relation between TFPR distortions and subsidies (휔):
  - log푇퐹푃푅∝≈−휔 .
- Combining this relation with the estimated impact of IP subsidies on sector-level log(TFPR) from regression specification (2) produces the implicit fiscal cost calculation:
  - ∑−(훽
푐푎푠ℎ
̂
∗퐼푃
2009−18,푠
푐푎푠ℎ
+훽
푐푟푒푑푖푡
̂
∗퐼푃
2009−18,푠
푐푟푒푑푖푡
)∗푣푎푙푢ｅ 푎𝑑𝑑𝑒𝑑
2018,푠
푠
- The calculation includes both cash and credit subsidies and explicitly excludes tax benefits because their estimated coefficient is close to zero and has the opposite sign to the coefficients for subsidies (see Figure 6).

### Key Quantitative Results from Annex II
- The estimated fiscal cost of subsidies using the structural-estimation cross-check is 1.5 percent of value added in 2018.
- This 1.5 percent estimate is compared with the estimate for listed firms in Section 2 of 2.2 percent in 2018.
  - Interpretation offered: listed firms may capture more subsidies (e.g., larger administrative capacity or closer political connections), so extrapolating listed-firm subsidy rates to non-listed firms in Section 3 may overstate IP size.
  - Caveat: the Annex II estimate may be subject to attenuation bias because it relies on an imperfect measure of IP (narrative IP counts); hence the result is described as tentative.

### Related Empirical Evidence and Context (from main text)
- Industrial champions (sector leaders, defined as the firm with the largest revenue share in each sector):
  - In most sectors, leaders have higher TFP than the average firm (Figure 9.1).
  - Leaders typically have lower TFPR than the average firm, implying they are producing at inefficiently high levels (Figure 9.2).
  - The TFPR gap is particularly pronounced for SOE leaders, consistent with Jurzyk and Ruane (2021).
  - Conclusion: market leaders often owe their position to both higher productivity and policies encouraging production (which may include subsidies and, for SOEs, implicit credit guarantees), rather than to barriers to competition.
- Aggregate conclusions reported in the paper:
  - Equivalent fiscal cost of IP in China is quantified at about 4 percent of GDP per year (instrument-by-instrument estimation using listed firms and the land registry).
  - Misallocation from IP is estimated to reduce aggregate productivity by about 1.2 percent, and GDP by up to 2 percent.

### Policy-relevant Implications (as stated in the paper)
- Increase transparency around IP, especially on local government IP measures, to improve policymaking and address trading-partner concerns.
- Scale down IP to reduce factor misallocation and fiscal costs; pursue IP cautiously and only to tackle well-defined market failures.
- When IP is implemented, prefer budgetary tools because they tend to be more transparent and less distortionary than indirect measures such as credit allocations or regulation.
- Data limitations and methodological caveats highlighted:
  - Further research is needed to quantify other IP instruments (e.g., Government Guided Funds, Public-Private Partnerships).
  - Strong assumptions were made to overcome data gaps (e.g., extrapolating results to non-listed firms, relying on IP counts data for structural estimation).
  - Better data reporting would facilitate more accurate characterization of Chinese IP and its economic effects.

*Source: IMF Working Paper — Industrial Policy in China: Quantification and Impact on Misallocation (Working Paper No. WP/2025/155), Annex II.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025155-source-pdf.pdf_