## wpiea2022192-print-pdf

## Source details

**Canonical URL:** [wpiea2022192-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022192-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022192-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022192-print-pdf.pdf.json)

---

### Overview
- Innovation is key to economic growth; R&D subsidies are a standard government tool to encourage innovation.
- This paper studies the effect of R&D subsidies on innovation inputs (R&D spending) and outputs (patents) using German firm-level data covering the 2000s.
- Heterogeneity analyzed by ownership (foreign MNC subsidiaries vs. domestic MNCs vs. domestic firms), firm size, and industry.
- Identification: treatment model with propensity score matching and regressions on a matched sample in repeated cross-sections.

### Data, variables, and empirical methodology
- Data sources and scope:
  - Mannheim Innovation Panel (MIP) survey waves: 2000, 2002, 2003, 2004, and 2006.
  - Patent applications from EPO between 1997 and 2011 (pre-sample patent stock and up to five years after survey year for outcomes).
  - Sample exclusions: firms with less than 5 employees or with no product/process innovation.
  - Final full sample (after dropping outliers): 5,717 observations; R&D regressions matched sample: 5,623 observations.
  - Patent regressions sample without outliers: 5,353 observations; matched sample: 5,263 observations.
- Key dependent and independent variables:
  - Dependent: R&D expenditures (reported) and number of firm EPO patent applications in the subsequent five years.
  - Central independent: R&D subsidy recipient (binary); firm ownership (foreign MNC, domestic MNC, domestic firm); firm size (small: 50 or less employees; medium: 51-500; large: more than 500); six industry dummies (low-tech manufacturing; medium-tech manufacturing; high-tech manufacturing; distributive services; knowledge-intensive services; technological services).
  - Survey indicates subsidy receipt but does not provide subsidy amount.
- Controls: number of employees, firm age (years since founded in Germany), patent applications in year of observation and in 1997–1999, share of employees with college education, continuous R&D dummy, exports share of sales, process innovation dummy, and time dummies.
- Estimation approach:
  - OLS regressions on unmatched full sample and matched sample.
  - Propensity score matching: probit to estimate propensity; nearest neighbor matching; common support by dropping lowest 5 percent of treated observations.
  - Matched-sample weighted regressions using odds ratio based on propensity score.
  - Dependent variables transformed as ln(1 + variable). Robust standard errors clustered on industry.

### Descriptive statistics and context
- German context:
  - R&D intensity of the German economy is about 3 percent of GDP.
  - FDI outflow positions about 44 percent of GDP in 2017 and FDI inflow positions about 26 percent of GDP in 2017.
  - No R&D tax credits; government R&D subsidies are application-based grants (federal and state levels).
- Sample-level descriptive statistics:
  - In the full sample, firms spend on average €1.6 million on R&D and file about 2.3 patents in a 5-year period.
  - About one-third of firms received an R&D subsidy from state or federal governments.
  - Firm size distribution: about half of firms are small (less than 50 employees); about 13 percent are large (>500 employees).
  - Ownership distribution: foreign MNC subsidiaries comprise about 10 percent of firms; domestic MNCs about 10 percent.
  - Industry distribution: about one-third of firms in low-tech manufacturing; about 18 percent each in medium-tech manufacturing and technological services; 10 percent each in high-tech manufacturing and knowledge-intensive services; 14 percent in distributive services.
- Descriptive patterns:
  - MNCs (domestic and foreign) do more R&D and file more patents on average than domestic firms.
  - Domestic MNCs tend to outperform foreign MNC subsidiaries on average, except in some industries (e.g., high-tech manufacturing where foreign MNCs slightly lead).
  - Firms receiving a subsidy tend to have larger R&D spending and more patenting across ownership, size, and industry; effect more pronounced in large firms, domestic MNCs, foreign MNC subsidiaries, manufacturing, and technological services.

### Key empirical findings
- Aggregate/subsample treatment effects:
  - Matching estimator: average treatment effect (ATE) ≈ 0.1; average treatment effect on the treated (ATT) ≈ 0.2.
  - Interpretation: R&D spending increases by about €0.2 million conditional on the receipt of the subsidy (ATT) and by €0.1 million for a representative firm (ATE).
  - Weighted regression on matched sample shows similar ATE.
  - Regression without firm controls: coefficient on subsidy dummy ≈ 0.1 and statistically significant; with firm controls, ATE falls to about 0.04 (Table 3, columns 1–4).
- Ownership heterogeneity (R&D spending):
  - Interactions of subsidy with foreign and domestic MNC dummies produce insignificant coefficients in pooled R&D regressions (Table 3, column 5) — no overall difference between domestic firms and MNCs in extra R&D investment from a subsidy receipt.
  - Industry-interacted regressions: MNCs show larger responses in medium- and high-tech manufacturing:
    - Medium-tech manufacturing: MNC response 0.10***.
    - High-tech manufacturing: domestic MNC response about 0.47*** and foreign MNC response 0.41*** (Table 3 industry*MNC*subsidy interactions).
- Ownership heterogeneity (patents):
  - Subsidy impact on patents larger for foreign MNC subsidiaries than domestic MNCs or domestic firms, even after accounting for R&D spending.
  - Selected patent interaction coefficients (Table 4 / Appendix Table 10):
    - Foreign MNC subsidiary * R&D subsidy: 0.27** and 0.31** in different specifications.
    - Low-tech manuf. * R&D subsidy * Foreign MNC: 0.43***.
    - High-tech manuf. * R&D subsidy * Foreign MNC: 0.59***.
    - Technological services * R&D subsidy * Foreign MNC: 0.39***.
- Industry heterogeneity:
  - R&D spending impacts:
    - Domestic firms respond more (0.05) in low-tech manufacturing, knowledge-intensive services, and technological services than MNCs (Table 3, column 6).
    - MNCs respond more in medium-tech manufacturing (0.10) and especially high-tech manufacturing (~0.4) than domestic firms.
    - Knowledge-intensive services: domestic MNCs undertake on average €0.2 million less R&D than foreign MNCs and domestic firms (Table 9).
  - Patents:
    - Government R&D subsidy alone does not have a statistically significant effect on 5-year ahead patents in pooled specifications (Table 4, column 1; Appendix Table 8 shows subsidy coefficients range from -0.01 to -0.02).
    - R&D spending (log) is positively associated with 5-year ahead patents: coefficients reported include 0.15***, 0.13*, 0.17**, 0.13, 0.13* across specifications.
    - Interaction patterns show larger patent effects for foreign MNC subsidiaries in several industries (see interaction coefficients above).
- Size heterogeneity:
  - Firm size interactions generally not significant for R&D increases (Table 3, column 7).
  - Exception: firms with up to 50 employees * R&D subsidy coefficient is -0.09*** (Table 3) / -0.08** (Appendix Table 7), indicating small firms receiving a subsidy invest about €0.1 million less in R&D than comparable non-recipients.
  - For patents, notable interaction: firm up to 50 empl. * R&D subsidy * Domestic MNC = 0.35*** (Appendix Table 10).

### Mechanisms and interpretation
- Potential mechanisms:
  - Alleviation of financial constraints increases R&D and subsequently patenting.
  - MNCs’ larger R&D/patenting responses in medium- and high-tech manufacturing suggest stronger innovation requirements to compete internationally.
  - Higher patenting by foreign MNCs may reflect more efficient allocation toward patentable technologies or signaling to headquarters for expanded innovation mandates.
  - Knowledge spillovers and technology transfers across MNC global networks may enable subsidiaries to pursue frontier innovation and achieve better patenting outcomes.
- Theoretical rationale for R&D support:
  - Market failures and positive externalities (knowledge spillovers, imperfect information about future returns, market frictions, resource constraints) lead to underinvestment in R&D relative to socially optimal levels.
  - Governments can address systemic problems in the innovation ecosystem (technological transitions, lock-in problems, innovation network deficiencies).
  - System of innovation approach: interactions among firms, universities, and public research organizations matter for policy design.

### Robustness, estimation notes, and Appendix highlights
- Matching estimator and weighted matched regressions produce broadly similar results.
- Transformation ln(1 + variable) used to handle zeros; zero-inflated negative binomial estimations attempted but did not converge.
- Common support enforced by dropping 5 percent of treated observations with lowest control density.
- Appendix descriptive statistics:
  - Appendix Table 4 and Appendix Table 5 provide detailed means and standard deviations by industry and size (examples: Low-tech manufacturing R&D expenditures mean 0.69; Medium-tech manufacturing R&D expenditures mean 4.40; Small firms R&D expenditures mean 0.07; Large firms R&D expenditures mean 11.31).
  - Appendix Table 6 shows balance statistics after nearest neighbor matching (propensity score mean, treated 0.42; control 0.41).
- Appendix regression summaries (selected):
  - Appendix Table 7 (Unweighted R&D regressions, matched): National government R&D subsidy (d) coefficients range from 0.11*** to 0.03** across columns; Firm with up to 50 empl. * R&D subsidy: -0.08**.
  - Appendix Table 8 (Unweighted patents, matched): R&D spending (log) coefficients: 0.09**, 0.11**, 0.09** in reported columns; Foreign MNC subsidiary * R&D subsidy interaction positive in some specifications.
  - Appendix Table 9 (Weighted R&D regressions): National government R&D subsidy (d) coefficients 0.08***, 0.09***, 0.10***, 0.04*** across columns; High-tech manuf. * R&D subsidy * Domestic MNC: 0.47***; High-tech manuf. * R&D subsidy * Foreign MNC: 0.41***.
  - Appendix Table 10 (Weighted patents): R&D spending (log) coefficients include 0.15***, 0.13*, 0.17**; Foreign MNC subsidiary * R&D subsidy: 0.27** and 0.31** in different specifications; High-tech manuf. * R&D subsidy * Foreign MNC: 0.59***.

### Policy-relevant implications and recommendations
- Overall: R&D subsidies raise R&D spending and increase future patenting on average, but impacts vary by ownership and industry.
- Targeting recommendations:
  - Industry heterogeneity: larger returns in medium- and high-tech manufacturing for MNCs suggest targeted support may amplify innovation where international competition and frontier technologies matter.
  - Ownership effects: foreign MNC subsidiaries show relatively larger patenting responses—policies that attract and integrate foreign R&D activity may yield spillover benefits.
  - Size considerations: small firms may reallocate subsidy-induced resources toward commercialization or other activities rather than immediate R&D expansion; complementary measures may be needed to translate subsidies into R&D for small firms.
- Inclusion of foreign affiliates: subsidy providers should not necessarily exclude foreign affiliates because increasing R&D of foreign MNC subsidiaries enlarges the knowledge pool within a host country and potentially spills over to domestic firms; to mitigate potential negative effects, facilitators could encourage collaboration between MNC subsidiaries and domestic firms.
- Note: subsidy amounts are not observed in the data; evidence indicates a positive multiplier on R&D spending but not necessarily greater than one.

*Source: wpiea2022192-print-pdf — https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022192-print-pdf.pdf*

### Introduction ........................................................................................................ Er

### wpiea2022192-print-pdf - Introduction ........................................................................................................ Er

### Main Sections
- Introduction
- Data, Descriptive Statistics, and Empirical Methodology — 6
- Results — 17
- Conclusion — 21
- References — 22
- Appendixes — 27

### Tables (enumerated)
- Table 1. R&D Spending (in million euros): Descriptive Statistics by Ownership, Industry, and Size — 10
- Table 2. Patent Applications in the Next 5 Years: Descriptive Statistics by Ownership, Industry, and Size — 11
- Table 3. R&D Spending: Weighted Regressions, Matched Sample — 18
- Table 4. Patents: Weighted Regressions, Matched Sample — 19

### Figures (enumerated)
- Figure 1. R&D Spending (log[1+R&D]) by Ownership — 12
- Figure 2. R&D Spending (log[1+R&D]) by Ownership and Subsidy (=1) — 12
- Figure 3. R&D Spending (log[1+R&D]) by Size and Subsidy (=1) — 13
- Figure 4. R&D Spending (log[1+R&D] and 5-Year Ahead Patents(log[1+Patents]) by Ownership and Subsidy (Red) — 13
- Figure 5. R&D Spending (log[1+R&D]) and 5-Year Ahead Patents (log[1+Patents]) by Industry and Subsidy (Red) — 14
- Figure 6. R&D Spending (log[1+R&D]) and 5-Year Ahead Patents (log[1+Patents]) by Size and Subsidy (Red) — 14

*Source: wpiea2022192-print-pdf - Introduction ........................................................................................................ Er*

### Introduction

### Introduction

### Overview
- Innovation is key to economic growth; R&D subsidies are a standard government tool to encourage innovation.
- This paper studies the effect of R&D subsidies on innovation inputs (R&D spending) and outputs (patents) using German firm-level data covering the 2000s.
- The analysis examines heterogeneity by ownership (foreign MNC subsidiaries vs. domestic MNCs vs. domestic firms), firm size, and industry.
- Identification uses a treatment model with propensity score matching and regressions on a matched sample in repeated cross-sections.

### Data and empirical methodology
- Data sources and scope:
  - Mannheim Innovation Panel (MIP) survey waves used: 2000, 2002, 2003, 2004, and 2006.
  - Patent applications from EPO between 1997 and 2011 to construct pre-sample patent stock and up to five years after survey year for outcomes.
  - Firms with less than 5 employees or with no product/process innovation excluded.
  - Final full sample (after dropping outliers): 5,717 observations; 5,623 can be matched with control firms for R&D regressions.
  - Patent regressions sample without outliers: 5,353 observations; matched sample: 5,263 observations.
- Key variable definitions:
  - Dependent variables: R&D expenditures (reported) and number of firm EPO patent applications in the subsequent five years.
  - Central independent variables: R&D subsidy recipient (binary), firm ownership (foreign MNC, domestic MNC, domestic firm), firm size (small: 50 or less employees; medium: 51-500; large: more than 500), and six industry dummies (low-tech manufacturing; medium-tech manufacturing; high-tech manufacturing; distributive services; knowledge-intensive services; technological services).
  - Note: Survey indicates subsidy receipt but does not provide subsidy amount.
- Controls included: number of employees, firm age (years since founded in Germany), patent applications in year of observation and in 1997–1999, share of employees with college education, continuous R&D dummy, exports share of sales, process innovation dummy, and time dummies.
- Estimation approach:
  - OLS regressions on unmatched full sample and matched sample.
  - Propensity score matching (probit to estimate propensity; nearest neighbor matching; common support by dropping lowest 5 percent of treated observations).
  - Matched-sample weighted regressions using odds ratio based on propensity score.
  - Dependent variables transformed as ln(1 + variable) to handle zeros and skewness.
  - Robust standard errors clustered on industry.

### Descriptive statistics and context
- German context:
  - R&D intensity of the German economy is about 3 percent of GDP.
  - FDI outflow positions of about 44 percent of GDP in 2017 and FDI inflow positions of about 26 percent of GDP in 2017.
  - No R&D tax credits; government R&D subsidies are application-based grants (federal and state levels).
- Sample-level descriptive statistics:
  - In the full sample, firms spend on average €1.6 million on R&D and file about 2.3 patents in a 5-year period.
  - About one-third of firms received an R&D subsidy from state or federal governments.
  - Firm size distribution: about half of firms are small (less than 50 employees); about 13 percent are large (>500 employees).
  - Ownership distribution: foreign MNC subsidiaries comprise about 10 percent of firms; domestic MNCs about 10 percent.
  - Industry distribution: about one-third of firms in low-tech manufacturing; about 18 percent each in medium-tech manufacturing and technological services; 10 percent each in high-tech manufacturing and knowledge-intensive services; 14 percent in distributive services.
- Descriptive patterns:
  - MNCs (domestic and foreign) do more R&D and file more patents on average than domestic firms.
  - Domestic MNCs tend to outperform foreign MNC subsidiaries on average, except in some industries (e.g., high-tech manufacturing where foreign MNCs slightly lead).
  - Firms receiving a subsidy tend to have larger R&D spending and more patenting across ownership, size, and industry. The effect is more pronounced in large firms, domestic MNCs, foreign MNC subsidiaries, manufacturing, and technological services.

### Key empirical findings
- Aggregate/subsample treatment effects:
  - Matching estimator: average treatment effect (ATE) ≈ 0.1; average treatment effect on the treated (ATT) ≈ 0.2.
  - Interpretation: R&D spending increases by about €0.2 million conditional on the receipt of the subsidy (ATT) and by €0.1 million for a representative firm (ATE).
  - Weighted regression on matched sample shows similar ATE.
  - Regression without firm controls: coefficient on subsidy dummy ≈ 0.1 and statistically significant; with firm controls, ATE falls to about 0.04 (Table 3, columns 1–4).
- Heterogeneity by ownership:
  - Interactions of subsidy with foreign and domestic MNC dummies produce insignificant coefficients in pooled R&D regressions (Table 3, column 5), suggesting no overall difference between domestic firms and MNCs in extra R&D investment from a subsidy receipt.
  - In industry-interacted regressions, MNCs show larger responses in medium- and high-tech manufacturing:
    - Medium-tech manufacturing: MNC response 0.10*** (interaction Low/Medium/High-tech manufacturing and MNC noted in Table 3).
    - High-tech manufacturing: MNC response about 0.47*** for domestic MNC and 0.41*** for foreign MNC in Table 3 (industry*MNC*subsidy interactions).
  - For patents, subsidy impact is larger for foreign MNC subsidiaries than domestic MNCs or domestic firms (Table 4 interactions), even after accounting for R&D spending.
  - Specific patent interactions (Table 4):
    - Foreign MNC subsidiary * R&D subsidy: 0.27** (column 2) and 0.31** (column 5) in various specifications.
    - Low-tech manuf. * R&D subsidy * Foreign MNC: 0.43***.
    - High-tech manuf. * R&D subsidy * Foreign MNC: 0.59***.
    - Technological services * R&D subsidy * Foreign MNC: 0.39***.
- Heterogeneity by industry:
  - R&D spending impacts:
    - Domestic firms respond more (0.05) in low-tech manufacturing, knowledge-intensive services, and technological services than MNCs (Table 3, column 6).
    - MNCs respond more in medium-tech manufacturing (0.10) and especially high-tech manufacturing (~0.4) than domestic firms.
    - The only differential between domestic and foreign MNCs is in knowledge-intensive services, where domestic MNCs undertake on average €0.2 million less R&D than foreign MNCs and domestic firms.
  - Patents:
    - Government R&D subsidy alone does not have a statistically significant effect on 5-year ahead patents (Table 4, column 1).
    - R&D spending (log) is positively associated with 5-year ahead patents: coefficient 0.15*** (Table 4, column 1), 0.13* (column 2), 0.17** (column 3), 0.13 (column 4), 0.13* (column 5).
    - Interaction patterns show larger patent effects for foreign MNC subsidiaries in several industries (see interactions listed above).
- Heterogeneity by firm size:
  - Firm size interactions generally not significant for R&D increases (Table 3, column 7).
  - Exception: firms with up to 50 employees * R&D subsidy coefficient is -0.09***, indicating small firms receiving a subsidy invest about €0.1 million less in R&D than comparable non-recipients (Table 3).
  - For patents, some significant size*ownership*subsidy interactions (e.g., firm up to 50 empl. * R&D subsidy * Domestic MNC = 0.35*** in Table 4).

### Mechanisms and interpretation
- Possible mechanisms for subsidy effects:
  - Alleviation of financial constraints increases R&D and subsequently patenting.
  - MNCs’ larger R&D/patenting responses in medium- and high-tech manufacturing suggest stronger innovation requirements to compete in international markets.
  - Higher patenting by foreign MNCs compared to domestic MNCs could reflect more efficient allocation toward patentable technologies or signaling to headquarters to pursue innovation in foreign locations.
  - Knowledge spillovers and technology transfers across MNC global networks may enable subsidiaries to pursue frontier innovation and achieve better patenting outcomes.
- Theoretical justification for R&D support:
  - Market failures and positive externalities (knowledge spillovers, imperfect information about future returns, market frictions, resource constraints) lead firms to underinvest in R&D relative to socially optimal levels (Arrow 1962; Klette et al., 2000).
  - Governments may also address systemic problems in the innovation ecosystem: technological transitions, lock-in problems, and innovation network deficiencies.
  - The system of innovation approach emphasizes interactions among firms, universities, and public research organizations.

### Literature context and robustness
- Consistent with a large literature finding positive effects of R&D programs on innovation (R&D spending and patents) with some exceptions and nuanced effects (examples cited in text).
- Prior literature typically finds no or only partial crowding out of R&D investments; elasticity of R&D to tax credits often equal to or greater than one in some studies.
- Robustness and estimation notes:
  - Matching estimator and weighted matched regressions produce broadly similar results.
  - Transformation ln(1 + variable) used to handle zeros; zero-inflated negative binomial estimations attempted but did not converge.
  - Common support enforced by dropping 5 percent of treated observations with lowest control density.

### Policy-relevant implications (from the analysis)
- R&D subsidies raise R&D spending and increase future patenting on average, but impacts vary by ownership and industry.
- Subsidy design and targeting could consider:
  - Industry heterogeneity: larger returns in medium- and high-tech manufacturing for MNCs suggest targeted support may amplify innovation where international competition and frontier technologies matter.
  - Ownership effects: foreign MNC subsidiaries show relatively larger patenting responses—policies that attract and integrate foreign R&D activity may have spillover benefits.
  - Size considerations: small firms may reallocate subsidy-induced resources toward commercialization or other activities rather than immediate R&D expansion; complementary measures may be needed to translate subsidies into R&D for small firms.
- Because the subsidy amount is not observed, evidence indicates a positive multiplier on R&D spending but not necessarily greater than one.

*Source: wpiea2022192-print-pdf - Introduction — https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022192-print-pdf.pdf*

### 0.3 extra patents for subsidy recipients (Table 4, column 2). However, the impact of the subsidy does not

### wpiea2022192-print-pdf - 0.3 extra patents for subsidy recipients (Table 4, column 2). However, the impact of the subsidy does not

### Main empirical findings on innovation outcomes
- R&D subsidy recipients have on average "0.3 extra patents" (Table 4, column 2).
- The impact of the subsidy on patenting "does not depend on the level of R&D spending" (Table 4, column 3).
- Accounting for industry interactions, the subsidy shows:
  - "larger impact for foreign MNCs than other firms in low-tech manufacturing and especially in high-tech manufacturing and technological services" (Table 4, column 4).
  - Only in medium-tech manufacturing is "the impact of the subsidy for domestic MNCs ... larger than for foreign MNCs, about 0.2 extra patents."
- Ignoring industry interactions, "only domestic MNCs with less than 50 employees ... tend to increase their patent applications upon receiving a subsidy."
- Overall: "foreign MNCs tend to benefit more from R&D subsidies than domestic MNCs and domestic firms."

### Heterogeneity by industry, ownership, and size
- Industry and ownership matter more than firm size for subsidy impacts:
  - Impact on R&D spending is "largely similar between foreign and domestic MNCs" and is larger than for domestic firms in "medium-tech and high-tech manufacturing."
  - "Domestic firms tend to benefit more in low-tech manufacturing and services than MNCs."
  - Impact on future patents is "mostly larger for foreign MNCs than domestic MNCs and domestic firms in low-tech and high-tech manufacturing and technological services."
  - Exception: "Only in medium-tech manufacturing, domestic MNCs tend to benefit more than other firms."
- An unexpected result: "the impact of the subsidy on future patent applications is slightly negative for domestic firm subsidy recipients than domestic firm non-recipients."

### Interpretation and suggested mechanisms
- MNCs (both domestic and foreign) are "more responsive to an R&D subsidy than domestic firms"—potentially because they "compete internationally, and innovation is key to staying competitive."
- Possible explanations for stronger patent responses by foreign MNCs:
  - Foreign subsidiaries may "use a host country subsidy more effectively or apply it toward more patentable technologies or products."
  - Subsidies may be used to "signal their headquarters for a new or extended innovation mandate," enabling more vigorous research and patenting.
  - Consistent with international technology sourcing: conducting R&D in technological frontier countries allows firms to "source technology, benefit from knowledge spillovers, and improve innovation outcomes."

### Policy implications and recommendations
- Targeting matters: "the effect of subsidies depends on the industry, and our results suggest that a targeted subsidy could be more efficient."
  - In "medium- and high-tech industries, there should be a greater focus on both foreign and domestic MNCs."
  - "Domestic firms should be more favored in low-tech manufacturing, knowledge-intensive services, and technological services."
- Inclusion of foreign affiliates: "subsidy providers should not necessarily exclude foreign affiliates" because increasing R&D of foreign MNC subsidiaries "enlarges the knowledge pool within a host country, potentially spilling over to domestic firms."
- To mitigate potential negative effects of subsidizing foreign MNC subsidiaries, the government could "facilitate collaboration between MNC subsidiaries and domestic firms to enable knowledge and technology transfer."

### Additional notes
- Detailed estimation results referenced are in "Appendix Tables 9-10."
- The study emphasizes accounting for differential impacts by industry and ownership when evaluating R&D subsidy programs.

*Source: wpiea2022192-print-pdf*

### Appendix Table 4. Descriptive Statistics by Industry

### Appendix Table 4. Descriptive Statistics by Industry

### Descriptive statistics — Low-tech manufacturing, Medium-tech manufacturing, High-tech manufacturing (selected variables)
- No. of patent applications in the next 5 years:
  - Low-tech manufacturing: Sample mean 0.75; Std. Dev. 4.20; Min 0; Max 68
  - Medium-tech manufacturing: Sample mean 7.65; Std. Dev. 53.39; Min 0; Max 1069
  - High-tech manufacturing: Sample mean 5.34; Std. Dev. 38.62; Min 0; Max 574
- R&D expenditures (€ mil):
  - Low-tech manufacturing: Mean 0.69; Std. Dev. 12.47; Min 0; Max 469.88
  - Medium-tech manufacturing: Mean 4.40; Std. Dev. 43.30; Min 0; Max 714.79
  - High-tech manufacturing: Mean 4.19; Std. Dev. 30.16; Min 0; Max 426.75
- Patent stock:
  - Low-tech manufacturing: Mean 0.80; Std. Dev. 4.06; Min 0; Max 61.62
  - Medium-tech manufacturing: Mean 5.39; Std. Dev. 30.28; Min 0; Max 492.49
  - High-tech manufacturing: Mean 4.05; Std. Dev. 19.18; Min 0; Max 206.06
- National government R&D subsidy (d):
  - Low-tech manufacturing: Mean 0.25; Std. Dev. 0.44; Min 0; Max 1
  - Medium-tech manufacturing: Mean 0.42; Std. Dev. 0.49; Min 0; Max 1
  - High-tech manufacturing: Mean 0.52; Std. Dev. 0.50; Min 0; Max 1
- Domestic MNC (d):
  - Low-tech manufacturing: Mean 0.11; Std. Dev. 0.31; Min 0; Max 1
  - Medium-tech manufacturing: Mean 0.18; Std. Dev. 0.38; Min 0; Max 1
  - High-tech manufacturing: Mean 0.13; Std. Dev. 0.33; Min 0; Max 1
- Foreign MNC subsidiary (d):
  - Low-tech manufacturing: Mean 0.09; Std. Dev. 0.28; Min 0; Max 1
  - Medium-tech manufacturing: Mean 0.15; Std. Dev. 0.35; Min 0; Max 1
  - High-tech manufacturing: Mean 0.15; Std. Dev. 0.36; Min 0; Max 1
- Firms with up to 50 empl. (d):
  - Low-tech manufacturing: Mean 0.37; Std. Dev. 0.48; Min 0; Max 1
  - Medium-tech manufacturing: Mean 0.34; Std. Dev. 0.47; Min 0; Max 1
  - High-tech manufacturing: Mean 0.50; Std. Dev. 0.50; Min 0; Max 1
- Firms with 500+ empl. (d):
  - Low-tech manufacturing: Mean 0.13; Std. Dev. 0.34; Min 0; Max 1
  - Medium-tech manufacturing: Mean 0.18; Std. Dev. 0.38; Min 0; Max 1
  - High-tech manufacturing: Mean 0.11; Std. Dev. 0.32; Min 0; Max 1
- Sales (log):
  - Low-tech manufacturing: Mean 2.44; Std. Dev. 1.82; Min -4.20; Max 10.52
  - Medium-tech manufacturing: Mean 2.76; Std. Dev. 2.00; Min -2.96; Max 10.11
  - High-tech manufacturing: Mean 1.98; Std. Dev. 2.02; Min -2.30; Max 10.31
- No of employees (levels):
  - Low-tech manufacturing: Mean 419.37; Std. Dev. 4664.01; Min 5; Max 193000
  - Medium-tech manufacturing: Mean 703.38; Std. Dev. 3926.42; Min 5; Max 53144
  - High-tech manufacturing: Mean 575.18; Std. Dev. 4716.15; Min 5; Max 104000
- Company age (years):
  - Low-tech manufacturing: Mean 38.78; Std. Dev. 36.21; Min 0; Max 150
  - Medium-tech manufacturing: Mean 34.99; Std. Dev. 35.06; Min 0; Max 149
  - High-tech manufacturing: Mean 25.09; Std. Dev. 28.13; Min 0; Max 148
- College educ. empl. (share):
  - Low-tech manufacturing: Mean 10.97; Std. Dev. 12.24; Min 0; Max 100
  - Medium-tech manufacturing: Mean 21.86; Std. Dev. 20.28; Min 0; Max 100
  - High-tech manufacturing: Mean 33.53; Std. Dev. 23.79; Min 0; Max 100
- Contin. R&D activities (d):
  - Low-tech manufacturing: Mean 0.37; Std. Dev. 0.48; Min 0; Max 1
  - Medium-tech manufacturing: Mean 0.66; Std. Dev. 0.47; Min 0; Max 1
  - High-tech manufacturing: Mean 0.74; Std. Dev. 0.44; Min 0; Max 1
- Share of exports in sales:
  - Low-tech manufacturing: Mean 0.36; Std. Dev. 0.45; Min 0; Max 1.96
  - Medium-tech manufacturing: Mean 0.70; Std. Dev. 0.55; Min 0; Max 2.06
  - High-tech manufacturing: Mean 0.57; Std. Dev. 0.54; Min 0; Max 2.13
- Process innovator (d):
  - Low-tech manufacturing: Mean 0.71; Std. Dev. 0.46; Min 0; Max 1
  - Medium-tech manufacturing: Mean 0.60; Std. Dev. 0.49; Min 0; Max 1
  - High-tech manufacturing: Mean 0.54; Std. Dev. 0.50; Min 0; Max 1
- Industry dummies (indicators shown as 1/0 where applicable):
  - Low-tech manuf. (d): Low-tech group shows 1.00; Medium/high-tech groups show 0.00 as appropriate.
- Year dummies (selected):
  - Year 2000 (d): Low-tech 0.22; Medium-tech 0.23; High-tech 0.17
  - Year 2004 (d): Low-tech 0.33; Medium-tech 0.29; High-tech 0.30
  - Year 2006 (d): Low-tech 0.18; Medium-tech 0.14; High-tech 0.16
- No of observations by industry section: 1,938 (Low-tech), 1,087 (Medium-tech), 594 (High-tech)

### Descriptive statistics — Distributive services, Knowledge-intensive services, Technological services (selected variables)
- No. of patent applications in the next 5 years:
  - Distributive services: Mean 0.31; Std. Dev. 5.81; Min 0; Max 164
  - Knowledge-intensive services: Mean 0.03; Std. Dev. 0.43; Min 0; Max 9
  - Technological services: Mean 0.49; Std. Dev. 2.93; Min 0; Max 68
- R&D expenditures (€ mil):
  - Distributive services: Mean 0.53; Std. Dev. 7.43; Min 0; Max 196.06
  - Knowledge-intensive services: Mean 0.38; Std. Dev. 3.00; Min 0; Max 60.84
  - Technological services: Mean 0.43; Std. Dev. 1.89; Min 0; Max 29.17
- Patent stock:
  - Distributive services: Mean 0.17; Std. Dev. 2.22; Min 0; Max 58.53
  - Knowledge-intensive services: Mean 0.10; Std. Dev. 1.17; Min 0; Max 23.36
  - Technological services: Mean 0.77; Std. Dev. 5.47; Min 0; Max 112.38
- National government R&D subsidy (d):
  - Distributive services: Mean 0.15; Std. Dev. 0.36; Min 0; Max 1
  - Knowledge-intensive services: Mean 0.09; Std. Dev. 0.29; Min 0; Max 1
  - Technological services: Mean 0.51; Std. Dev. 0.50; Min 0; Max 1
- Firms with up to 50 empl. (d):
  - Distributive services: Mean 0.53; Std. Dev. 0.50; Min 0; Max 1
  - Knowledge-intensive services: Mean 0.43; Std. Dev. 0.50; Min 0; Max 1
  - Technological services: Mean 0.74; Std. Dev. 0.44; Min 0; Max 1
- Share of exports in sales:
  - Distributive services: Mean 0.13; Std. Dev. 0.34; Min 0; Max 2.00
  - Knowledge-intensive services: Mean 0.04; Std. Dev. 0.16; Min 0; Max 1.54
  - Technological services: Mean 0.19; Std. Dev. 0.37; Min 0; Max 1.96
- Process innovator (d):
  - Distributive services: Mean 0.69; Std. Dev. 0.46
  - Knowledge-intensive services: Mean 0.76; Std. Dev. 0.43
  - Technological services: Mean 0.60; Std. Dev. 0.49
- No of observations by services: Distributive services 833; Knowledge-intensive services 580; Technological services 1,020

### Appendix Table 5. Descriptive Statistics by Size (selected variables)

- Size categories: Small (up to 50 empl.), Medium (51-500 empl.), Large (500+ empl.)
- No. of patent applications in the next 5 years:
  - Small: Mean 0.15; Std. Dev. 1.56; Min 0; Max 68
  - Medium: Mean 1.27; Std. Dev. 22.08; Min 0; Max 1069
  - Large: Mean 13.18; Std. Dev. 60.16; Min 0; Max 711
- R&D expenditures (€ mil):
  - Small: Mean 0.07; Std. Dev. 0.19; Min 0; Max 3.43
  - Medium: Mean 0.32; Std. Dev. 0.85; Min 0; Max 16.87
  - Large: Mean 11.31; Std. Dev. 60.91; Min 0; Max 714.79
- Patent stock:
  - Small: Mean 0.16; Std. Dev. 1.00; Min 0; Max 20.38
  - Medium: Mean 1.06; Std. Dev. 8.72; Min 0; Max 399.99
  - Large: Mean 10.00; Std. Dev. 36.96; Min 0; Max 492.49
- National government R&D subsidy (d):
  - Small: Mean 0.35; Std. Dev. 0.48
  - Medium: Mean 0.30; Std. Dev. 0.46
  - Large: Mean 0.29; Std. Dev. 0.45
- Domestic MNC (d):
  - Small: Mean 0.02; Std. Dev. 0.16
  - Medium: Mean 0.13; Std. Dev. 0.33
  - Large: Mean 0.37; Std. Dev. 0.48
- Foreign MNC subsidiary (d):
  - Small: Mean 0.03; Std. Dev. 0.17
  - Medium: Mean 0.14; Std. Dev. 0.34
  - Large: Mean 0.20; Std. Dev. 0.40
- No of employees (levels):
  - Small: Mean 21.20; Std. Dev. 12.91; Min 5; Max 50
  - Medium: Mean 172.54; Std. Dev. 113.56; Min 51; Max 500
  - Large: Mean 3209.88; Std. Dev. 13143.56; Min 501; Max 240000
- Company age (years):
  - Small: Mean 20.35; Std. Dev. 22.94
  - Medium: Mean 37.29; Std. Dev. 35.75
  - Large: Mean 52.34; Std. Dev. 41.56
- Contin. R&D activities (d):
  - Small: Mean 0.40; Std. Dev. 0.49
  - Medium: Mean 0.46; Std. Dev. 0.50
  - Large: Mean 0.63; Std. Dev. 0.48
- No of observations by size: Small 2,828; Medium 2,451; Large 773

### Appendix Table 6. Nearest Neighbor Matching: Mean Comparison (selected balance statistics)
- Propensity score:
  - Mean, treated 0.42; Mean, control 0.41; t-test 1.24; P < t 0.21
- Foreign MNC subsidiary (d): Mean, treated 0.10; Mean, control 0.10; t-test 1.00
- Domestic MNC (d): Mean, treated 0.09; Mean, control 0.09; t-test 1.00
- No of employees (log): Mean, treated 3.99; Mean, control 3.99; t-test -0.03; P < t 0.98
- Company age (years): Mean, treated 18.93; Mean, control 19.26; t-test -0.53; P < t 0.60
- Patent stock (ln): Mean, treated -3.17; Mean, control -3.24; t-test 0.32
- Share of exports to sales (ratio): Mean, treated 0.43; Mean, control 0.40; t-test 1.45; P < t 0.15
- Industry and year dummies shown with balance statistics (examples):
  - Medium high-tech manuf. (d): Mean, treated 0.24; Mean, control 0.24; t-test 1.00
  - High-tech manuf. (d): Mean, treated 0.16; Mean, control 0.16; t-test 1.00
  - Year 2004 (d): Mean, treated 0.22; Mean, control 0.22; t-test -0.04; P < t 0.97
  - Year 2006 (d): Mean, treated 0.23; Mean, control 0.22; t-test 0.04; P < t 0.97

### Appendix Table 7. R&D Spending: Unweighted Regressions, Matched Sample (selected coefficient estimates)
- Dependent variable: Log (1 + R&D spending). OLS; Observations 5,623.
- National government R&D subsidy (d):
  - Column (1): 0.11*** (0.01)
  - Column (2): 0.11*** (0.01)
  - Column (3): 0.09*** (0.01)
  - Column (4): 0.03** (0.01)
  - Column (5): -0.00 (0.01)
  - Column (6): -0.01 (0.01)
  - Column (7): 0.03 (0.02)
- Interaction and heterogeneity coefficients (selected):
  - Domestic MNC * R&D subsidy: 0.17** (0.05) in one specification
  - Foreign MNC subsidiary * R&D subsidy: 0.15* (0.06) in one specification
  - Low-tech manuf. * R&D subsidy: 0.02* (0.01)
  - High-tech manuf. * R&D subsidy: -0.06*** (0.01)
  - Knowledge-intens. services * R&D subsidy: 0.04* (0.02)
  - Technological services * R&D subsidy: 0.03** (0.01)
  - Low-tech manuf. * R&D subsidy * Domestic MNC: 0.06** (0.02)
  - Medium high-tech manuf. * R&D subsidy * Domestic MNC: 0.20*** (0.03)
  - High-tech manuf. * R&D subsidy * Domestic MNC: 0.41*** (0.03)
  - Firm with up to 50 empl. * R&D subsidy: -0.08** (0.02)
  - Firm with 500+ empl. * R&D subsidy: 0.22* (0.10)
- Adjusted R-squared across columns: 0.02, 0.13, 0.31, 0.39, 0.39, 0.40, 0.44
- Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

### Appendix Table 8. Patents: Unweighted Regressions, Matched Sample (selected coefficient estimates)
- Dependent variable: Log (1 + 5-year ahead patents). OLS; Observations 5,263.
- National government R&D subsidy (d): coefficients range from -0.01 to -0.02 across specifications; standard errors (0.01)–(0.02).
- R&D spending (log):
  - Column (2): 0.09** (0.02)
  - Column (3): 0.11** (0.03)
  - Column (7): 0.09** (0.02)
- Selected interaction effects:
  - Foreign MNC subsidiary * R&D subsidy: 0.12* (0.05) in one specification; 0.15** (0.06) in another
  - Low-tech manuf. * R&D subsidy: -0.02*** (0.00)
  - High-tech manuf. * R&D subsidy: -0.04*** to -0.05*** (0.01)
  - Low-tech manuf. * R&D subsidy * Domestic MNC: 0.22*** (0.04)
  - Low-tech manuf. * R&D subsidy * Foreign MNC: 0.27*** (0.01)
  - Medium high-tech manuf. * R&D subsidy * Foreign MNC: -0.04* to -0.09** (0.02–0.03)
  - High-tech manuf. * R&D subsidy * Foreign MNC: 0.36*** (0.02) and 0.30*** (0.03)
  - Technological services * R&D subsidy * Foreign MNC: 0.28*** (0.02) and 0.23*** (0.02)
- Adjusted R-squared across columns: 0.56 in all reported specifications.
- Note: Time, ownership, and industry dummies and other controls included. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

### Appendix Table 9. R&D Spending: Weighted Regressions, Matched Sample (selected coefficient estimates)
- Dependent variable: Log (1 + R&D spending). OLS; Observations 5,623.
- National government R&D subsidy (d):
  - Column (1): 0.08*** (0.02)
  - Column (2): 0.09*** (0.01)
  - Column (3): 0.10*** (0.01)
  - Column (4): 0.04*** (0.01)
  - Column (5): 0.02 (0.02)
  - Column (6): -0.02* (0.01)
  - Column (7): 0.05 (0.03)
- Industry-specific R&D subsidy interactions (selected):
  - Low-tech manuf. * R&D subsidy: 0.05*** (0.01)
  - Medium high-tech manuf. * R&D subsidy: 0.04** (0.01)
  - Knowledge-intens. services * R&D subsidy: 0.05*** (0.01)
  - Technological services * R&D subsidy: 0.05*** (0.01)
  - Medium high-tech manuf. * R&D subsidy * Domestic MNC: 0.10*** (0.02)
  - High-tech manuf. * R&D subsidy * Domestic MNC: 0.47*** (0.02)
  - Knowledge-intens. services * R&D subsidy * Domestic MNC: -0.20*** (0.02)
  - Medium high-tech manuf. * R&D subsidy * Foreign MNC: 0.09** (0.03)
  - High-tech manuf. * R&D subsidy * Foreign MNC: 0.41*** (0.04)
- Ownership and firm-level controls (selected coefficients):
  - Domestic MNC (d): ranges from 0.41*** to 0.08 across specifications (example Column (1) 0.41*** (0.08))
  - Foreign MNC subsidiary (d): ranges from 0.35*** to 0.03 across specifications (example Column (1) 0.35*** (0.08))
  - Patent stock (log): 0.01–0.01* (0.01)
  - Sales (log): 0.05** (0.02) in multiple columns
  - No of employees (log): 0.07*** (0.02) in multiple columns
  - College educ. empl. (share): 0.00** to 0.00* (0.00)
  - No of patent applications (log): 0.17** (0.05) in some columns; 0.12** (0.05) in others
  - Contin. R&D activities (d): 0.13*** (0.02) across columns
- Adjusted R-squared across columns: 0.01, 0.14, 0.37, 0.44, 0.45, 0.46, 0.50
- Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

### Appendix Table 10. Patents: Weighted Regressions, Matched Sample (selected coefficient estimates)
- Dependent variable: Log (1 + 5-year ahead patents). OLS; Observations 5,263.
- National government R&D subsidy (d): coefficients across columns: 0.01, -0.02, 0.00, 0.03, -0.01 (standard errors 0.02–0.03).
- R&D spending (log):
  - Column (1): 0.15*** (0.03)
  - Column (2): 0.13* (0.06)
  - Column (3): 0.17** (0.05)
  - Column (5): 0.13* (0.06)
- Interaction effects (selected):
  - Foreign MNC subsidiary * R&D subsidy: 0.27** (0.09) in one specification; 0.31** (0.08) in another
  - Low-tech manuf. * R&D subsidy: -0.05*** (0.00)
  - Medium high-tech manuf. * R&D subsidy: -0.07*** (0.01)
  - High-tech manuf. * R&D subsidy: -0.09*** (0.02)
  - Low-tech manuf. * R&D subsidy * Domestic MNC: 0.18* (0.08)
  - Low-tech manuf. * R&D subsidy * Foreign MNC: 0.43*** (0.03)
  - High-tech manuf. * R&D subsidy * Foreign MNC: 0.59*** (0.07)
  - Technological services * R&D subsidy * Foreign MNC: 0.39*** (0.05)
  - Firm with up to 50 empl. * R&D subsidy * Domestic MNC: 0.35*** (0.08)
- Covariates (selected):
  - Patent stock (log): 0.08*** (0.00) across columns
  - No of patent applications (log): 0.89*** (0.06) to 0.88*** (0.07)
  - Exports to sales (share): 0.06** (0.02) across columns
  - Knowledge-intens. services (d): -0.01* to -0.02** (0.00–0.01)
- Adjusted R-squared across reported columns: 0.55 in each column shown
- Note: Time, ownership, and industry dummies and other controls included. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

*Promoting Innovation: The Differential Impact of R&D Subsidies — Working Paper No. WP/2022/XX*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022192-print-pdf.pdf_