## wp18269 — Section 7 concludes (with linked Sections 5.1 and 6.2 findings)

## Source details

**Canonical URL:** [wp18269 — Section 7 concludes (with linked Sections 5.1 and 6.2 findings)](https://www.imf.org/-/media/files/publications/wp/2018/wp18269.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18269.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18269.pdf.json)

---

### Conceptual framework
- Structural mapping:
  - Domestic knowledge stock proxied by cumulative discounted flow R_ct of domestic R&D up to time t.
  - Fraction φ_cl of knowledge generated in foreign country l available to domestic economy c; 1−φ_cl interpreted as barriers to knowledge flow.
  - Total foreign-generated knowledge available: R_Fct = ∑_l φ_cl R_lt.
- Usable knowledge and production functions:
  - K_ct = (R_ct)^β (R_Fct)^χ
  - P_ct = X_Pct (K_ct)^θ
  - A_ct = X_Act (K_ct)^ψ, with X_Act = A_ct−1^ρ and ρ∈[0,1] a possible persistence parameter.
- Common linearized empirical relation:
  - log Y_ct = X_Yct + γ_Y log R_ct + μ_Y log R_Fct, where Y_ct ∈ {P_ct, A_ct}; μ_Y measures importance of usable foreign knowledge.

### Sectoral extension and empirical specification
- Sectoral foreign knowledge stock: R_Fcit = ∑_{l∈G5} φ_cli R_lit.
- Estimated specification:
  - log Y_cit = X_Yct + γ_Y log R_clit + μ_Y log R_Fcit, with country-time fixed effects X_Yct.
- Focus:
  - Source countries restricted to G5; recipient index c spans non-G5 countries (advanced and emerging).
  - Emphasis on within-sector knowledge flows (source sector = recipient sector).

### Data and measurement
- Patents and citations: PATSTAT; use international patent families (require application in at least two distinct patent offices); family attributed to country of first inventor; citations counted only if within four years of publication; self-citations excluded.
- Sector-level R&D: OECD ANBERD; R&D stocks built with perpetual inventory method in constant PPP USD.
- Productivity regressions data: 2017 EU KLEMS (industry value added, employment, capital stock), sample shrinks to mainly advanced economies.
- Patent family attribution reduces cross-country comparability issues from differing patenting cultures.

### Determinants of knowledge flows (gravity model) and stylized evolution
- Gravity specification for φ_cli:
  - φ_cli = exp[a + f_ci + ̃f_li + b_1 (diff.country_cl) + b_2 (diff.border_cl) + b_3 (diff.lang_cl) + b_4 (dist.int_cl) + b_5 (tech.controls_cli) + ε_cli]; within-country-sector baseline normalized to 1.
- Key regressors: diff.country, diff.border, diff.lang, dist.int, tech.spec, tech.dev (log-difference in R&D or value added per worker).
- Estimation: PPML; cited countries restricted to G5; citing countries include 23 advanced and 9 emerging economies (sectoral R&D availability limits sample).
- Main regularities:
  - Coefficients on diff.country, diff.lang, and tech.spec negative and statistically significant across subperiods (barriers reduce citation intensity).
  - tech.dev.RnD becomes statistically indistinguishable from zero in the last period (weakening role of physical and technical distance over time).
- Predicted within-country-sector citation intensity normalized to 100 percent; crossing a national border reduces diffusion by roughly 1/2.

### Evolution of predicted knowledge flow intensity from G5
- For emerging economies, average intensity ̄φ_gG5 rose from about 10 percent in 1995-1999 to about 16 percent in 2010-2014; 95-percent confidence bands indicate this increase is statistically significant (1995-1999 band [8.4,12.0]; 2010-2014 band [11.5,22.4]).
- For non-G5 advanced countries, intensity ̄φ_AEG5 hovered around 20 percent with no statistically significant change.
- Early subperiods: ̄φ_EMG5 statistically much smaller than ̄φ_AEG5; by 2010-2014 the two estimates are statistically indistinguishable.
- Interpretation: evidence of knowledge globalization over the last two decades; international barriers have weakened and emerging markets’ integration with G5 knowledge has deepened.

### Robustness and extensions on knowledge flows
- Excluding China: results largely unaffected.
- Allowing cited countries beyond G5: diffusion from non-G5 is weaker than from G5.
- Cross-sectoral diffusion: weaker than within-sector diffusion.
- Alternative tech.dev proxy using value-added per employee increases EM sample but yields less precise estimates; point estimates for ̄φ_AEG5 and ̄φ_EMG5 change little.
- Comparison: paper’s 20 percent estimate for ̄φ_AEG5 roughly comparable to Peri (2005)’s 20-25 percent (samples/definitions differ).

### Impact on innovation (Section 5.1)
- Empirical setup:
  - Dependent variable: log count of international patent families P_cit.
  - Estimation: OLS with country-year fixed effects.
  - Sample: eleven manufacturing sectors in 27 countries over 1995-2014; sub-samples: 15 AEs and 12 EMs.
- Main findings:
  - Foreign knowledge flows significantly stimulate domestic innovation (μ in equation (4) positive).
  - Estimated elasticities:
    - Elasticity of innovation to foreign R&D: 0.35.
    - Elasticity of innovation to domestic R&D: about 0.45 (column 1).
  - Peri (2005) reference: elasticity to foreign R&D 0.4-0.47; elasticity to domestic R&D 0.74-0.81.
  - Split-sample: foreign R&D coefficient similar between AEs and EMs; in EMs foreign R&D has roughly the same impact on domestic patenting as domestic R&D.
- Time evolution:
  - Coefficient on foreign R&D allowed to vary by five-year periods shows a steady and statistically significant increase over time; increase larger for emerging economies.
  - Result robust to balanced sample and allowing all coefficients to vary by sub-period.
- Non-stationarity and alternative estimator:
  - Panel unit root tests find little evidence of unit roots; panel cointegration tests generally point to cointegration between patent flows and R&D series.
  - DOLS (Kao and Chiang (2001)) with two lags and one lead yields similar estimates; OLS retained for remainder given inconclusive unit root tests and similar DOLS results.
- Robustness checks:
  - Alternative patent measures (families with at least one application at top 3 patent offices): results similar.
  - Alternative weighting using time-varying bilateral trade links: results robust.
  - Alternative fixed effects (sector-year): coefficients on foreign and domestic R&D become significantly larger.
  - Expanded EM sample using aggregate R&D × sector intensity interaction (U.S. R&D per employee average 1995-2014): correlation with sector-level R&D stock about 0.49; weighted foreign R&D coefficient remains significant and time-evolution holds.

### Impact on productivity (Section 6.2 and linked results)
- Productivity specification:
  - Augmented production function with domestic and foreign R&D capital; domestic and foreign R&D lagged by one year.
  - Sample: 11 manufacturing industries in 9 (mostly advanced) countries over 1995-2014 (reduced sample due to capital stock data).
- Main productivity results (Table 4):
  - Foreign knowledge boosts domestic productivity modestly compared with innovation results.
  - Estimated TFP responses:
    - A one percent increase in the weighted foreign knowledge stock associated with about 0.05 percent increase in TFP of the receiving country-sector.
    - A one percent increase in the domestic R&D stock associated with about 0.06 percent increase in TFP.
  - Allowing foreign R&D coefficient to vary by five-year periods shows diffusion to TFP has strengthened over the past two decades; increases statistically significant.
- Regression evidence on productivity effects (competition interaction):
  - OLS shows no statistically significant effect of trade competition with China on industry productivity (column 1).
  - Instrumented estimation of Chinese import penetration (column 2) suggests increased trade competition from China helps boost production efficiency and diffusion to TFP.
  - Instrumentation critical due to reverse causality risk (lower productivity sectors more likely to have higher Chinese import penetration).
  - Replacing trade competition with global concentration (column 3):
    - Main effect of sectoral concentration on productivity not statistically significant.
    - Higher concentration associated with reduced diffusion of foreign knowledge to productivity (interaction negative and significant).
  - Quadratic term inclusion produces statistically non-significant coefficient for the quadratic term in productivity regressions.

### Interaction with international competition (innovation and diffusion)
- Competition measures:
  - Import penetration from China: goods imports from China as share of receiving country-sector gross output (available 1998-2014).
  - Global market concentration: global market share of four largest firms at two-digit ISIC level (Diez et al. (2018) Orbis-based, available 2000-15).
- Findings on China trade competition and innovation:
  - Stronger trade competition with China boosts domestic innovation and technology diffusion in both AEs and EMs: positive and statistically significant main and interaction effects (Table 5 columns 1–2).
  - Results hold when expanding EM sample using aggregate-R&D × sector intensity interaction (column 3).
  - Instrumental variables approach (import penetration instrumented with U.S. import penetration for non-G5 advanced economies following Autor et al. (2014)):
    - Coefficients on trade with China and its interaction with weighted foreign R&D remain strongly statistically significant with larger magnitude (column 4).
    - First-stage statistics:
      - Kleibergen-Paap rk LM statistic (under-identification test): 22.4, p-value 0.
      - Kleibergen-Paap rk Wald F statistic (weak identification test): 23.702.
  - Non-linearity: quadratic specifications show an “inverted-U” between competition and innovation (Figure 3); majority of observations lie in the upward-sloping part (stronger competition initially increases innovation, relation becoming negative at higher competitive pressure).
- Findings on global concentration and innovation:
  - Higher global concentration has a negative and statistically significant effect on innovation and on diffusion from G5 leaders for both non-G5 advanced economies and emerging markets (Table 6 columns 1–3).
  - Quadratic terms show some evidence of inverted-U; non-linearity statistically significant but quantitatively small (Figure 4).
  - Most observations lie on the downward-sloping part (less concentration associated with higher productivity).
  - China excluded from EM sample when using global concentration due to its outsized role.

### Robustness summary (innovation and productivity)
- Results robust to:
  - Excluding China.
  - Allowing cited countries beyond G5 (diffusion weaker from non-G5).
  - Cross-sectoral diffusion allowed (weaker than within-sector).
  - Alternative tech.dev proxies (value-added per employee) — larger EM sample but less precision.
  - Alternative patent measures and weighting schemes.
  - Alternative fixed effects (sector-year) — larger coefficients.
  - DOLS vs OLS — similar estimates for foreign R&D effects on patents.

### Key quantitative takeaways and broader conclusions (Section 7)
- Over the entire sample, only 15 percent of the G5s domestic knowledge diffuses internationally.
- Intensity of knowledge diffusion heterogeneous:
  - Advanced economies: diffusion roughly stable at around 20 percent.
  - Emerging economies: diffusion increased from 10 percent in 1995-1999 to 16 percent in 2010-2014.
- Effectiveness of foreign knowledge once diffused:
  - Foreign knowledge is about 80 percent as effective as domestically generated knowledge in raising domestic innovation.
- Sub-sample analysis:
  - Statistically significant increase in estimated coefficients over time, change especially large for emerging economies.
- Role of international competition:
  - Heightened international competitive forces had a positive impact on domestic innovation and on its sensitivity to foreign-generated knowledge.
- Magnitudes to preserve:
  - Innovation elasticities: foreign R&D 0.35; domestic R&D about 0.45.
  - Productivity responses: weighted foreign knowledge +0.05 percent TFP per 1 percent; domestic R&D +0.06 percent TFP per 1 percent.
  - Predicted flow intensities: emerging economies ~10 percent → ~16 percent (1995-99 → 2010-14); non-G5 advanced ~20 percent stable.
  - International diffusion overall: 15 percent of G5 domestic knowledge diffuses internationally.
  - Foreign knowledge effectiveness relative to domestic: about 80 percent.

*Source: wp18269 — Section 7 concludes; linked content from Sections 5.1 and 6.2.*

### Section 7 concludes.

### Section 7 concludes.

### Conceptual framework
- The paper links knowledge stocks originating domestically and abroad to domestic innovation outcomes (patenting P_ct or productivity A_ct) using a simple structural model that nests common empirical specifications.
- Knowledge stocks:
  - Domestic knowledge stock proxied by cumulative discounted flow R_ct of domestic R&D up to time t.
  - Fraction φ_cl of knowledge generated in foreign country l available to domestic economy c; 1−φ_cl interpreted as barriers to knowledge flow.
  - Total foreign-generated knowledge available: R_Fct = ∑_l φ_cl R_lt.
- Usable knowledge and innovation production functions:
  - K_ct = (R_ct)^β (R_Fct)^χ
  - P_ct = X_Pct (K_ct)^θ
  - A_ct = X_Act (K_ct)^ψ
  - Aggregate factors X_Pct and X_Act capture factors like propensity to patent and persistence in productivity (e.g., X_Act = A_ct−1^ρ with ρ∈[0,1]).
- Linearized empirical relation used in much of the literature:
  - log Y_ct = X_Yct + γ_Y log R_ct + μ_Y log R_Fct, where Y_ct ∈ {P_ct, A_ct}.
  - μ_Y measures importance of usable foreign knowledge for domestic innovation.

### Sectoral extension and empirical specification
- Augments framework with sectoral dimension: allows study of how R&D stocks R_Fcit in foreign sector i influence domestic sector outcome Y_cit.
- Sectoral foreign knowledge stock: R_Fcit = ∑_{l∈G5} φ_cli R_lit.
- Empirical specification estimated:
  - log Y_cit = X_Yct + γ_Y log R_clit + μ_Y log R_Fcit
  - Country-time fixed effects X_Yct used to control time-varying aggregate unobservables.
- Source countries restricted to G5 to reduce endogeneity; recipient index c spans a wide set of non-G5 countries (advanced and emerging).
- Focus is on within-sector knowledge flows (source sector = recipient sector), which are empirically largest.

### Effect of international competition
- Extended empirical model for non-G5 countries includes competition proxy Z_cit:
  - log Y_cit = X_Yct + θ Z_cit + ̃γ_Y log R_cit + ̃μ_Y log R_Fcit + δ Z_cit R_Fcit
- Interpretation:
  - θ captures direct impact of international competition on innovation (if Z_cit exogenous).
  - Total impact of foreign knowledge on innovation = ̃μ_Y + δ Z_cit; δ measures how competition modifies knowledge diffusion.

### Data
- Main data sources:
  - Patents and cross-patent citations from PATSTAT.
  - Sector-level business R&D spending in constant PPP USD from OECD ANBERD; R&D stocks built with perpetual inventory method.
  - Industry value added, employment, capital stock from the 2017 EU KLEMS database for productivity regressions (sample shrinks to mainly advanced economies).
- Patent measures:
  - Use international patent families (patent family requires one application in at least two distinct patent offices) as quality-adjusted patent count.
  - Patent family attributed to country of first inventor, earliest publication year, main industrial sector.
  - Citation counts: attributed to application year; citations counted only if within four years of publication; self-citations excluded.
- Motivation: patent families reduce cross-country comparability issues arising from differing patenting cultures and incentives.

### Determinants of knowledge flows (gravity model)
- Gravity specification for predicted citation-based weights φ_cli (citing country-sector c,i; cited country-sector l,i):
  - φ_cli = exp[a + f_ci + ̃f_li + b_1 (diff.country_cl) + b_2 (diff.border_cl) + b_3 (diff.lang_cl) + b_4 (dist.int_cl) + b_5 (tech.controls_cli) + ε_cli]
  - Includes country-sector fixed effects f_ci and ̃f_li; regressors defined so same country-sector yields zeros (predicted within-country flow excluding fixed effects = 1).
- Regressors and interpretations:
  - diff.country, diff.border, diff.lang, dist.int (geographical distance), tech.spec (compositional technological specialization difference), tech.dev (difference in technological development measured as log-difference in R&D or value added per worker).
- Estimation approach:
  - Pseudo-Poisson-Maximum Likelihood (PPML) estimator used to handle heteroskedasticity, zeros, many dummies.
  - Cited countries restricted to G5; citing countries sample includes 23 advanced and 9 emerging economies (sample limited by sectoral R&D availability).
- Main empirical regularities:
  - Coefficients on diff.country, diff.lang, and tech.spec are negative and statistically significant across subperiods: these barriers reduce citation intensity.
  - Contiguity (diff.next) and distance (dist.int) coefficients vary across subperiods, partly due to collinearity.
  - Difference in technological development (tech.dev.RnD) becomes statistically indistinguishable from zero in the last period—consistent with weakening role of physical and technical distance over time.
- Predicted citation frequencies ˆφ_clit are constructed from estimated coefficients; they measure relative share of knowledge that diffuses from cited to citing country-sector.

### Key empirical findings on intensity and evolution of knowledge flows
- Within-country-sector predicted citation intensity (baseline) normalized to 100 percent; crossing a national border reduces diffusion by roughly 1/2.
- Average predicted knowledge flow intensity from G5 to recipient groups over subperiods (computed as linear combinations of gravity coefficients and group averages):
  - For emerging economies, average intensity ̄φ_gG5 rose from about 10 percent in 1995-1999 to about 16 percent in 2010-2014. 95-percent confidence bands indicate this increase is statistically significant (1995-1999 band [8.4,12.0]; 2010-2014 band [11.5,22.4]).
  - For non-G5 advanced countries, intensity ̄φ_AEG5 hovered around 20 percent with no statistically significant change over the sample.
  - In early subperiods ̄φ_EMG5 was statistically much smaller than ̄φ_AEG5; by 2010-2014 the two estimates are statistically indistinguishable.
- Interpretation: evidence of knowledge globalization over the last two decades; international barriers to knowledge flows have weakened, deepening emerging markets’ integration with G5 knowledge.

### Robustness and extensions
- Robustness checks:
  - Excluding China: results largely unaffected.
  - Allowing cited countries beyond G5: diffusion from non-G5 is weaker than from G5.
  - Allowing cross-sectoral diffusion: diffusion is weaker across sectors than within sectors.
  - Alternative proxy for tech.dev using value-added per employee (instead of R&D) increases sample of emerging economies but yields less precise estimates; point estimates for ̄φ_AEG5 and ̄φ_EMG5 change little.
- Comparison note: the paper’s 20 percent estimate for ̄φ_AEG5 is roughly comparable to Peri (2005)’s 20-25 percent, but samples and definitions differ (Peri uses regions within advanced countries and an earlier period without industry dimension).

### Impact on innovation and productivity (second-step regressions)
- Second-step approach: construct foreign knowledge flows for each country-sector by weighting G5 R&D stocks with ˆφ_clit from the gravity model; then estimate sectoral regressions of type log Y_cit = X_Yct + γ_Y log R_clit + μ_Y log R_Fcit with Y = sectoral patents P_cit (primary focus) and, for restricted sample, sectoral productivity A_cit.
- Rationale:
  - Foreign technology can affect domestic patented innovations and productivity more directly via imported equipment, licensing, or non-patented channels.
  - Country-time fixed effects absorb time-varying aggregate factors (including institutional propensity to patent), improving identification of sectoral foreign knowledge effects.
- Data coverage constraints:
  - Productivity regressions using value added, labor, capital rely on 2017 EU KLEMS and thus focus mainly on advanced economies.

*Source: wp18269 - Section 7 concludes.*

### 5.1    Impact on innovation

### 5.1    Impact on innovation

### Empirical setup and sample
- Dependent variable: (log) count of international patent families in a given country-sector (Y_cit = P_cit).
- Estimation method: OLS with country-year fixed effects.
- Sample: eleven manufacturing sectors in 27 countries over 1995-2014.
- Sub-samples: 15 AEs and 12 EMs recipients.

### Main findings on innovation
- Foreign knowledge flows significantly stimulate domestic innovation (coefficient μ in equation (4)).
- Estimated elasticities:
  - Elasticity of innovation to foreign R&D: 0.35.
  - Elasticity of innovation to domestic R&D: about 0.45 (column 1).
- Comparison to Peri (2005):
  - Peri (2005) estimates elasticity to foreign R&D of 0.4-0.47, and elasticity to domestic R&D of 0.74-0.81.
- Split-sample results:
  - The foreign R&D coefficient is similar between AEs and EMs.
  - For emerging economies, foreign R&D has roughly the same impact on domestic patenting as domestic R&D (i.e., foreign R&D’s value relative to domestic R&D is much higher for EMs).
- Time evolution:
  - Allowing the coefficient on foreign R&D to vary by five-year periods (with 1995-99 excluded) shows a steady and statistically significant increase in the coefficient over time.
  - The rise over time holds for both advanced and emerging economies but is larger for emerging economies.
  - Result robust to a roughly balanced sample and to allowing all coefficients to vary over each sub-period.

### Non-stationarity and alternative estimator
- Panel unit root tests find little evidence of unit roots; panel cointegration tests generally point to cointegration between patent flows and domestic and foreign R&D series.
- Dynamic OLS (Kao and Chiang (2001)) reported in column 6:
  - Procedure adds lags and leads of changes in regressors; requires strongly balanced sample (smaller sample of country-sectors).
  - Chosen number of lags: two; number of leads: one.
  - Estimates similar to baseline; coefficient on the weighted foreign R&D stock remains sizable and statistically significant.
- Given inconclusive unit root tests and similar DOLS results, OLS retained for remainder of analysis.

### Robustness and sensitivity checks
- Alternative patent measures:
  - Results similar when using patent families with at least one application at one of the top 3 patent offices (United States Patent and Trademark Office, European Patent Office, and Japanese Patent Office).
- Alternative weighting schemes:
  - Baseline uses predicted share of knowledge flow φ̂ based on cross-patent citations.
  - Robust to using time-varying bilateral trade links between country-sectors, where trade weights = imports from origin country-sector as a share of gross output of importing country-sector.
- Alternative fixed effects:
  - Baseline uses country-year fixed effects.
  - Results robust to using sector-year fixed effects; coefficients on both foreign and domestic R&D become significantly larger under sector-year fixed effects.
- Expanded sample for emerging economies:
  - Replace domestic sector-level R&D with interaction of domestic aggregate R&D stock and sector’s “representative” R&D intensity (based on U.S. R&D spending per employee, average for 1995-2014).
  - Correlation between sector-level R&D stock and this interacted variable: about 0.49.
  - Allows inclusion of significantly larger number of EMs; coefficient on weighted foreign R&D stock remains statistically significant and time-evolution results hold.

### Impact on productivity (summary linking to innovation)
- Estimation: augmented production function with domestic and foreign R&D capital entering production; dependent variable sectoral output with capital and labor inputs as regressors. Domestic and foreign R&D variables lagged by one year.
- Sample: 11 manufacturing industries in 9 (mostly advanced) countries over 1995-2014 (sample reduced due to limited sector-level capital stock data).
- Main productivity results (Table 4):
  - Foreign knowledge boosts domestic productivity modestly compared with innovation results.
  - Estimated TFP responses:
    - A one percent increase in the weighted foreign knowledge stock is associated with about 0.05 percent increase in the TFP of the receiving country-sector.
    - A one percent increase in the domestic R&D stock is associated with about 0.06 percent increase in TFP.
  - Allowing the foreign R&D coefficient to vary over five-year periods shows diffusion to TFP has strengthened over the past two decades; increases are statistically significant.

### Interaction with international competition (overview from adjacent sections)
- Two measures constructed:
  - Import penetration from China: goods imports from China as a share of receiving country-sector gross output (available 1998-2014).
  - Global market concentration: global market share of the four largest firms (based on Diez et al. (2018) Orbis-based calculation) aggregated to two-digit ISIC level (available 2000-15).
- Evidence:
  - Trade competition with China has risen markedly across industries over past two decades.
  - Global market concentration in most industries appears to have declined by this measure.

### Findings on China trade competition and innovation (summary)
- Stronger trade competition with China is found to boost domestic innovation and technology diffusion in both AEs and EMs:
  - Positive and statistically significant main and interaction effects (columns 1 and 2 of Table 5).
  - Results broadly similar across AEs and EMs and hold when expanding EM sample using aggregate-R&D × sector intensity interaction (column 3).
- Instrumental variables approach:
  - Import penetration from China instrumented with U.S. import penetration for non-G5 advanced economies (following Autor et al. (2014)).
  - When instrumenting, coefficients on trade with China and its interaction with weighted foreign R&D remain strongly statistically significant but with larger magnitude (column 4), possibly indicating downward OLS bias.
  - First-stage statistics supporting instrument relevance:
    - Kleibergen-Paap rk LM statistic (under-identification test): 22.4, p-value 0.
    - Kleibergen-Paap rk Wald F statistic (weak identification test): 23.702.
- Non-linearity:
  - Running regressions with quadratic terms yields an “inverted-U” relationship between competition and innovation (Figure 3): stronger competition initially associated with more innovation, relation becoming negative at higher competitive pressure.
  - Vast majority of observations lie in upward-sloping part of inverted U (where more competition leads to more innovation).

### Findings on global concentration and innovation
- Using global concentration (top-four global market share) instead of China shock:
  - Higher global concentration has a negative and statistically significant effect on innovation and knowledge diffusion from G5 leaders for both non-G5 advanced economies and emerging markets (Table 6, columns 1–3).
  - Adding quadratic terms reveals some evidence of an inverted-U shape; non-linearity statistically significant but quantitatively small (Figure 4).
  - Most observations lie on the downward-sloping part of the inverted U (less concentration associated with higher productivity).
- China excluded from EM sample when using global concentration measure due to China’s outsized role in global concentration trends.

*Source: wp18269 - 5.1    Impact on innovation.*

### 6.2    Impact on productivity

### 6.2    Impact on productivity

### Regression evidence on productivity effects
- OLS estimation (column 1) reveals no statistically significant effect of trade competition with China on industry productivity.
- Instrumented estimation of Chinese import penetration (column 2) suggests that increased trade competition from China helps boost production efficiency and diffusion to TFP.
- Instrumentation is critical in this specification because of the clear risk of reversed causality: lower productivity sectors are more likely to have higher Chinese import penetration, leading to a downward OLS bias in the estimated coefficients.
- Replacing trade competition with the global concentration measure (column 3) yields:
  - The main effect of sectoral concentration on productivity is not statistically significant.
  - Higher concentration is associated with reduced diffusion of foreign knowledge to productivity (the interaction effect is negative and significant).
- Including a quadratic term in the regressions produces a statistically non-significant coefficient for the quadratic term.

### Key empirical takeaways (from section 6.2 and linked conclusions)
- Instrumental-variable estimates indicate trade competition from China can raise production efficiency and increase the diffusion of foreign knowledge into TFP, while OLS estimates may mask this effect due to reverse causality.
- Measures of global concentration do not show a direct effect on productivity, but higher concentration weakens the diffusion of foreign knowledge into productivity.

### Broader conclusions from the paper (Section 7)
- Over the entire sample, only 15 percent of the G5s domestic knowledge diffuses internationally.
- The intensity of knowledge diffusion displays geographical and temporal heterogeneity:
  - Advanced economies: diffusion has remained roughly stable at around 20 percent.
  - Emerging economies: diffusion progressively increased from 10 percent in the period 1995-1999 to 16 percent in the period 2010-2014.
- Foreign knowledge, once it diffuses to the domestic economy, is about 80 percent as effective as domestically generated knowledge in raising domestic innovation.
- Sub-sample analysis shows a statistically significant increase in the estimated coefficients, with the change being especially large for emerging economies.
- Heightened international competitive forces had a positive impact on domestic innovation and on its sensitivity to foreign-generated knowledge.

*Source: wp18269 - 6.2    Impact on productivity*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18269.pdf_