## Chasing the Dream: Industry-Level Productivity Developments in Europe — IMF Working Paper WP/24/258 (December 2024)

## Source details

**Canonical URL:** [Chasing the Dream: Industry-Level Productivity Developments in Europe — IMF Working Paper WP/24/258 (December 2024)](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024258-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024258-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024258-print-pdf.pdf.json)

---

### Abstract and purpose
- Study scope: industry-level TFP (total factor productivity) growth across 28 countries in Europe over the period 1995–2020.
- Primary dataset: EU-KLEMS.
- Main aim: explore patterns and sources of TFP growth at the industry level to inform policies that strengthen growth prospects.

### Major empirical findings (core points)
- TFP growth drivers:
  - "TFP growth is driven largely by the extent to which countries are involved in scientific and technological innovation as the leader country or benefiting from stronger knowledge spillovers."
  - "The technological gap is associated with TFP growth as countries move towards the technological frontier by adopting new innovations and technologies."
  - "Increased investment in information and communications technology (ICT) capital and research and development (R&D) contributes significantly to higher TFP growth."
  - "The impact of human capital tends to be stronger when a country is closer to the technological frontier."
- Additional empirical notes:
  - Interaction results: both ICT and non-ICT capital expenditures tend to moderate the negative effect of the technological gap on TFP growth.
  - Subsample finding: "the technological gap is an important driver of the TFP slowdown in post-GFC period."
  - Figure 2 sample size: 9,151 observations (binned scatter plot).

### Stylized facts and aggregate trends
- EU aggregate TFP growth:
  - Average of 0.7 percent between 1996 and 2007.
  - 0.1 percent over the period 2009–2019.
  - -2 percent in 2020 (COVID-19 pandemic).
- Cross-country variation in average TFP growth, 1996–2020:
  - Minimum: -2 percent in Greece.
  - Maximum: 2 percent in Slovakia.
- Aggregate sample summary:
  - Over 1996–2020, average TFP growth stood at 0.5 percent.
  - 11 out of 28 EU countries presented a negative TFP growth.
- Sectoral dispersion:
  - Accommodation and food services: TFP growth was 1.6 percentage points lower than the average TFP growth across all sectors during the sample period.
  - Agriculture: TFP growth was 1.4 percentage points higher than the average TFP growth across all sectors during the sample period.
- Technological gap (distance to frontier):
  - Average: -19.2 percent.
  - Range: from -67.7 percent to 0 percent.

### Data, variables, and measurement
- Sample and sources:
  - Unbalanced panel of annual observations on 26 industries in 28 EU countries during 1995–2020.
  - Data source: EU-KLEMS; additional country-level controls from WEO, WDI, and ICRG; authors' calculations.
- Capital categorization and statistics:
  - Seven types of capital aggregated using user cost of capital to produce capital service flows.
  - ICT assets include computers, software and telecommunication equipment.
  - Non-ICT assets proxied by transportation equipment.
  - ICT capital spending: average 12.6 percent of gross fixed investment (Min: 0 percent; Max: 44.6 percent).
  - Non-ICT capital spending: average 8.3 percent of gross fixed investment (Min: 0.3 percent; Max: 32.6 percent).
  - R&D expenditure as share of gross fixed investment: average 9.5 percent (Min: 0 percent; Max: 47.8 percent).
- Labor inputs:
  - Differentiated by educational attainment (primary, secondary, tertiary), age, and gender.
  - Measure of high-skilled labor: share of total working hours provided by workers with tertiary education.
- Dependent variable:
  - Industry-level TFP growth defined as a residual from a production-function decomposition:
    - ∆y_{ijt} = ∆V_{ijt} − ω^{K}_{ijt} ∆K_{ijt} − ω^{L}_{ijt} ∆L_{ijt}
    - where ∆y_{ijt} is TFP growth in country i and industry j at time t; V, K, and L denote value-added, capital, and labor; and the coefficients ω^{K}_{ijt} and ω^{L}_{ijt} are the average share of capital and labor inputs, respectively.
- Data processing:
  - Industry-level variables winsorized at the 5th and 95th percentiles.

### Industry- and country-level sample counts (reported)
- Industry-level variables:
  - TFP growth: 8,438 observations; sample mean shown in text as 0.5 percent.
  - Technological gap: 9,151 observations.
  - ICT capital spending: 11,879 observations.
  - Non-ICT capital spending: 14,612 observations.
  - R&D spending: 12,315 observations.
- Country-level variables:
  - Real GDP per capita: 18,021 observations.
  - Inflation: 17,969 observations.
  - Financial development: 14,369 observations.
  - Trade openness: 18,021 observations.
  - Bureaucratic quality: 17,633 observations.
  - Population: 18,021 observations.

### Econometric methodology (baseline specification)
- Baseline model:
  - Dependent variable: ∆y_ijt (TFP growth).
  - Key regressors:
    - ∆y^L_jt: TFP growth frontier in the EU (highest TFP growth in industry j at time t).
    - (y_ijt−1 − y^L_jt−1): technological gap (TFP difference to frontier).
    - X^k_ijt−1: industry- and country-level controls (industry-level: ICT capital spending, non-ICT capital spending, R&D spending, share of high-skilled labor; country-level: real GDP per capita, consumer price inflation, trade openness, domestic credit to the private sector, population, bureaucratic quality).
    - Interaction terms X^l_ijt−1 * (y_ijt−1 − y^L_jt−1) explored.
  - Fixed effects: industry-specific η_i, time effects μ_t, and country effects γ_j (country fixed effects included in some specifications; models also run replacing country fixed effects with country-level controls).
  - Robust standard errors clustered at the industry level.
- Time period and special samples:
  - 28 EU countries over 1995–2020.
  - High-skilled labor available from 2008 for a separate model covering 2008–2020.
  - Robustness subsamples: pre-GFC (1995–2007), post-GFC (2010–2020), excluding COVID-19 (1995–2019).

### Key baseline empirical results (standardized coefficients and interpretations)
- Sample sizes and fit (Table 2 examples):
  - Observations with country FE: 8,615 (All), 4,153 (Tradables), 3,644 (Non-tradables).
  - Observations with country-level controls: 7,505 (All), 3,584 (Tradables), 3,200 (Non-tradables).
  - R^2 reported ranges: 0.159, 0.174, 0.144, 0.162, 0.186, 0.138.
- TFP growth at frontier:
  - Column [1] (All, country FE): 0.271*** (t = 5.783).
  - Column [2] (Tradables): 0.277*** (t = 4.659).
  - Column [3] (Non-tradables): 0.207*** (t = 4.071).
  - Column [4] (All, with country controls): 0.247*** (t = 5.225).
  - Column [5] (Tradables, with country controls): 0.265*** (t = 4.581).
  - Column [6] (Non-tradables, with country controls): 0.144** (t = 2.395).
  - Interpretation: a one-standard deviation increase in TFP growth at frontier is associated with about a 0.25 standard deviation increase in TFP growth (column [4]); frontier effect larger for tradables than non-tradables.
- Technological gap:
  - Column [1]: -0.310*** (t = -9.342).
  - Column [2]: -0.340*** (t = -7.910).
  - Column [3]: -0.260*** (t = -5.556).
  - Column [4]: -0.357*** (t = -8.264).
  - Column [5]: -0.403*** (t = -6.974).
  - Column [6]: -0.271*** (t = -5.345).
  - Interpretation:
    - Technological gap has the largest economic magnitude among regressors: in column [4], a one-standard deviation widening in the technological gap is associated with an average decline of about 0.36 standard deviation in TFP growth.
    - Closing the technological gap would raise TFP growth, on average, by 2.3 percent across the whole sample of industries; and by 2.8 percent for tradables and 3.8 percent for non-tradables (figures as reported in text).
- Industry-level investment and intangible capital:
  - ICT capital spending:
    - Column [1]: 0.096*** (t = 2.837).
    - Column [2]: 0.053 (t = 1.140).
    - Column [3]: 0.150*** (t = 3.634).
    - Column [4]: 0.102** (t = 2.627).
    - Column [5]: 0.090 (t = 1.522).
    - Column [6]: 0.133** (t = 2.494).
    - Interpretation: ICT capital spending associated with higher TFP growth; in column [4], a 1 standard deviation increase in ICT investment increases TFP growth by about 0.1 standard deviation.
  - Non-ICT capital (transportation equipment proxy):
    - Coefficients generally small and statistically insignificant in baseline (examples: -0.012, -0.114, 0.033; -0.014, -0.134, 0.044).
    - Conclusion: non-ICT capital spending does not appear to matter for TFP growth in baseline estimations.
  - R&D spending:
    - Column [1]: 0.143** (t = 2.579).
    - Column [2]: 0.126* (t = 1.791).
    - Column [3]: 0.162 (t = 1.606).
    - Column [4]: 0.148** (t = 2.577).
    - Column [5]: 0.157* (t = 1.890).
    - Column [6]: 0.112 (t = 1.546).
    - Interpretation: R&D spending has a positive and statistically significant effect on TFP growth across all industries and tradables in specifications including country controls (coefficients about 0.15 in columns [4] and [5]).
- Country-level controls (selected, standardized, column [4]):
  - Real GDP per capita: 0.264* (t = 1.781).
  - Inflation: -0.684 (t = -1.407).
  - Financial development: -0.072 (t = -1.392).
  - Trade openness: 0.111 (t = 0.680).
  - Bureaucratic quality: 0.165** (t = 2.616).
  - Population: 0.635 (t = 0.533).

### High-skilled labor results (2008–2020, Table 3)
- TFP growth at frontier: 0.298*** to 0.292*** across columns.
- Technological gap: large negative coefficients (examples: -0.589***, -0.693***, -0.378***; -0.601***, -0.695***, -0.384***).
- ICT capital: mixed but some positive coefficients (examples: 0.120, 0.117, 0.185*).
- R&D spending: mixed and not consistently significant (examples: 0.106, 0.143, -0.001).
- Share of high-skilled labor: coefficients not statistically significant in baseline (examples: -0.055, -0.156, 0.031).
- Interpretation:
  - Industry-level intensity of high-skilled labor does not appear to have a statistically significant direct effect on TFP growth at conventional levels in the baseline.
  - Some evidence that the effect of high-skilled labor is stronger when a country is closer to the technological frontier and that human capital matters more in non-tradables than tradables after controlling for other factors.

### Interaction results (Table 4 summary)
- ICT capital spending × technological gap:
  - ICT capital spending moderates the negative effect of the technological gap on TFP growth for the full sample and especially for tradables (interaction terms statistically significant in text).
- Non-ICT capital × technological gap:
  - Non-ICT capital has a moderating effect on the technological gap statistically significant for the full sample, but not consistently for tradable/non-tradable breakdown.
- R&D × technological gap:
  - Interaction shows a weak significant and negative effect across specifications, except for all industries when country-level variables are included.
- High-skilled labor × technological gap:
  - Interaction significant for the tradable sector only.
- Interpretation:
  - ICT and non-ICT capital can mitigate the dampening effect of a larger technological gap on TFP growth; R&D interactions show weaker and sometimes negative moderating effects; high-skilled labor appears to help close the gap primarily in tradables.

### Robustness checks, visualization, and subsamples
- Robustness:
  - Results robust to replacing country fixed effects with country-level control variables.
  - Similar qualitative results obtained with inclusion of country-year and country-industry fixed effects.
  - Subsample periods estimated: pre-GFC (1995–2007), post-GFC (2010–2020), excluding COVID-19 (1995–2019).
  - Subsample evidence: technological gap is an important driver of the TFP slowdown in the post-GFC period.
- Visualization:
  - Figure 4: binned scatter plots of 9,151 observations show strong positive correlation between TFP growth and TFP growth at frontier and an inverse relationship with the technological gap.

### Dynamic modelling (System GMM) and results
- Purpose:
  - Address potential endogeneity and include lagged dependent variable.
- Implementation:
  - System GMM (Arellano and Bover, 1995; Blundell and Bond, 1998), one-step estimator; instrument strategy controls for instrument proliferation per Roodman (2009).
  - Instruments: lagged dependent variable, asset tangibility, innovative property, training.
- Specification tests (reported p-values):
  - AR(1): 0.000.
  - AR(2): 0.404 and 0.284 (no evidence of significant second-order autocorrelation).
  - Hansen J-test: 0.037 and 0.029 (validity of internal instruments).
- Main dynamic results (Table 6, standardized):
  - TFP growth at frontier: 0.341*** (column 1), 0.335*** (column 2).
  - Technological gap: -0.321*** (column 1), -0.371*** (column 2).
  - ICT capital: 0.006 (column 1), 0.050 (column 2) — positive but statistically insignificant.
  - Non-ICT capital: 0.357** (column 1), 0.284* (column 2) — positive and significant in dynamic specification.
  - R&D spending: 0.085 (column 1), 0.030 (column 2) — positive but statistically insignificant.
  - Number of observations: 8,215 (column 1) and 7,154 (column 2).

### Interpretation and narrative conclusions
- Robust findings:
  - TFP growth at the frontier exerts a positive and robust impact on industry-level TFP growth.
  - The technological gap is negatively associated with TFP growth, indicating convergence dynamics: pace of convergence in “follower” industries increases with distance to the frontier.
  - ICT and R&D generally contribute positively to TFP growth, though significance varies by period and specification; non-ICT capital shows mixed signs across periods but is positive and significant in the dynamic model.
  - Human capital does not show a consistent statistically significant direct effect, but interaction evidence suggests its importance when countries approach the technological frontier and in non-tradables.
  - Subsample analysis indicates the technological gap is an important driver of the TFP slowdown in the post-GFC period.
- Aggregate inference:
  - Increases in intangible capital (ICT and R&D) are associated with higher TFP growth and positive externalities across the economy by accelerating adoption of new technologies.

### Policy implications and recommendations
- Priority actions emphasized:
  - Narrow innovation and technology gaps vis-à-vis the frontier and expand the frontier.
  - Revamp tangible and intangible capital investment in new technologies (including ICT and R&D).
  - Strengthen human capital for rapid progress in science and technology (education and healthcare).
  - Create a conducive environment for higher business investment and better capital allocation by providing incentives for capital investment and R&D.
- Rationale:
  - Revamped capital investment and strengthened human capital can promote innovation and facilitate diffusion of technologies to countries below the frontier, generating positive spillovers across industries.
  - Human capital appears to matter more when countries are closer to the technological frontier and particularly in non-tradable sectors.

### Organization of the paper (section map)
- Section II: literature review.
- Section III: data description.
- Section IV: econometric strategy.
- Section V: empirical results and robustness checks.
- Section VI: summary, concluding remarks, and policy implications.

*Prepared by Serhan Cevik, Sadhna Naik, and Keyra Primus — IMF Working Paper WP/24/258 (December 2024).*

### Section 1

### Chasing the Dream: Industry-Level Productivity Developments in Europe — Section 1

### Abstract and purpose
- Study scope: industry-level TFP (total factor productivity) growth across 28 countries in Europe over the period 1995–2020.
- Primary dataset: EU-KLEMS.
- Main aim: explore patterns and sources of TFP growth at the industry level to inform policies that strengthen growth prospects.

### Major empirical findings (four core points)
- (i) "TFP growth is driven largely by the extent to which countries are involved in scientific and technological innovation as the leader country or benefiting from stronger knowledge spillovers."
- (ii) "The technological gap is associated with TFP growth as countries move towards the technological frontier by adopting new innovations and technologies."
- (iii) "Increased investment in information and communications technology (ICT) capital and research and development (R&D) contributes significantly to higher TFP growth."
- (iv) "The impact of human capital tends to be stronger when a country is closer to the technological frontier."

Additional empirical notes:
- Interaction results: both ICT and non-ICT capital expenditures tend to moderate the negative effect of the technological gap on TFP growth.
- Subsample finding: "the technological gap is an important driver of the TFP slowdown in post-GFC period."
- Figure 2 sample size: 9,151 observations (binned scatter plot).

### Stylized facts and aggregate trends
- EU aggregate TFP growth:
  - Average of 0.7 percent between 1996 and 2007.
  - 0.1 percent over the period 2009–2019.
  - -2 percent in 2020 (COVID-19 pandemic).
- Cross-country variation in average TFP growth, 1996–2020:
  - Minimum: -2 percent in Greece.
  - Maximum: 2 percent in Slovakia.
- Aggregate sample summary:
  - Over 1996–2020, average TFP growth stood at 0.5 percent.
  - 11 out of 28 EU countries presented a negative TFP growth.
- Sectoral dispersion:
  - Accommodation and food services: TFP growth was 1.6 percentage points lower than the average TFP growth across all sectors during the sample period.
  - Agriculture: TFP growth was 1.4 percentage points higher than the average TFP growth across all sectors during the sample period.
- Technological gap (distance to frontier):
  - Average: -19.2 percent.
  - Range: from -67.7 percent to 0 percent.

### Data, variables, and measurement
- Sample: unbalanced panel of annual observations on 26 industries in 28 EU countries during 1995–2020.
- Data source: EU-KLEMS (industry-level measures of growth, productivity, employment, capital formation, technological change).
- Capital categorization:
  - Seven types of capital aggregated using user cost of capital to produce capital service flows.
  - ICT assets include computers, software and telecommunication equipment.
  - Non-ICT assets proxied by transportation equipment.
  - ICT capital spending: average 12.6 percent of gross fixed investment (Min: 0 percent; Max: 44.6 percent).
  - Non-ICT capital spending: average 8.3 percent of gross fixed investment (Min: 0.3 percent; Max: 32.6 percent).
  - R&D expenditure as share of gross fixed investment: average 9.5 percent (Min: 0 percent; Max: 47.8 percent).
- Labor inputs:
  - Differentiated by educational attainment (primary, secondary, tertiary), age, and gender.
  - Measure of high-skilled labor: share of total working hours provided by workers with tertiary education.
- Dependent variable definition:
  - Industry-level TFP growth defined as a residual from a production-function decomposition:
    ∆푦_{푖푗푡} = ∆푉_{푖푗푡} − 휔^{퐾}_{푖푗푡} ∆퐾_{푖푗푡} − 휔^{퐿}_{푖푗푡} ∆퐿_{푖푗푡}
    where ∆푦_{푖푗푡} is TFP growth in country i and industry j at time t; V, K, and L denote value-added, capital, and labor; and the coefficients 휔^{퐾}_{푖푗푡} and 휔^{퐿}_{푖푗푡} are the average share of capital and labor inputs, respectively.
- Data processing:
  - Industry-level variables winsorized at the 5th and 95th percentiles.
- Additional country-level controls used: real GDP per capita, consumer price inflation, trade openness, domestic credit to the private sector, population, and bureaucratic quality (drawn from WEO, WDI, and ICRG).

### Policy implications and recommendations
- Priority actions emphasized by the analysis:
  - Narrow innovation and technology gaps vis-à-vis the frontier and expand the frontier.
  - Revamp tangible and intangible capital investment in new technologies (including ICT and R&D).
  - Strengthen human capital for rapid progress in science and technology (education and healthcare).
  - Create a conducive environment for higher business investment and better capital allocation by providing incentives for capital investment and R&D.
- Rationale:
  - Revamped capital investment and strengthened human capital can promote innovation and facilitate diffusion of technologies to countries below the frontier, generating positive spillovers across industries.
  - Human capital appears to matter more when countries are closer to the technological frontier and particularly in non-tradable sectors.

### Organization of the paper (section map)
- Section II: literature review.
- Section III: data description.
- Section IV: econometric strategy.
- Section V: empirical results and robustness checks.
- Section VI: summary, concluding remarks, and policy implications.

*Prepared by Serhan Cevik, Sadhna Naik, and Keyra Primus — IMF Working Paper WP/24/258 (December 2024).*

### Section 2

### wpiea2024258-print-pdf - Section 2

### Industry-level developments and summary statistics
- Industry-level variables (sample counts and summary values as reported):
  - TFP growth: 8,438 observations; sample mean shown in text as 0.5 percent.
  - Technological gap: 9,151 observations.
  - ICT capital spending: 11,879 observations.
  - Non-ICT capital spending: 14,612 observations.
  - R&D spending: 12,315 observations.
- Country-level variables (observation counts and example summary values reported):
  - Real GDP per capita: 18,021 observations.
  - Inflation: 17,969 observations.
  - Financial development: 14,369 observations.
  - Trade openness: 18,021 observations.
  - Bureaucratic quality: 17,633 observations.
  - Population: 18,021 observations.
- Sources: EU-KLEMS; IMF; World Bank; ICRG; and authors' calculations.

### Econometric methodology (baseline specification)
- Baseline model (industry i, industry j, time t):
  - Dependent variable: ∆y_ijt (TFP growth).
  - Key regressors:
    - ∆y^L_jt: TFP growth frontier in the EU (highest TFP growth in industry j at time t).
    - (y_ijt−1 − y^L_jt−1): technological gap (TFP difference to frontier).
    - X^k_ijt−1: vector of industry- and country-level control variables (industry-level: ICT capital spending, non-ICT capital spending, R&D spending, share of high-skilled labor; country-level: real GDP per capita, consumer price inflation, trade openness, domestic credit to the private sector, population, bureaucratic quality).
    - Interaction terms X^l_ijt−1 * (y_ijt−1 − y^L_jt−1) explored.
  - Fixed effects: industry-specific η_i, time effects μ_t, and country effects γ_j (country fixed effects included in some specifications; models also run replacing country fixed effects with country-level controls).
  - Robust standard errors clustered at the industry level.
- Time period and sample: 28 EU countries over 1995–2020 (with high-skilled labor available from 2008 for a separate model covering 2008–2020).
- Estimation presentation:
  - Table 2: baseline results (columns [1]–[3] include country fixed effects; columns [4]–[6] replace country fixed effects with country-level control variables).
  - Table 3: baseline with high-skilled workers (2008–2020).
  - Table 4: extended model including interaction terms (ICT, non-ICT, R&D, high-skilled labor with technological gap).
- Identification and interpretation notes:
  - Inclusion of fixed effects helps address endogeneity from omitted variables.
  - Caution on the coefficient of TFP growth at frontier when a country is the frontier (measurement error bias affects level interpretation, not cross-specification comparisons).
  - Robustness checks include subsamples: pre-GFC (1995–2007), post-GFC (2010–2020), and excluding COVID-19 (1995–2019).

### Empirical evidence — baseline findings (Table 2)
- Sample sizes and fit:
  - Number of observations: 8,615 (All), 4,153 (Tradables), 3,644 (Non-tradables) in country-fixed-effects specifications; 7,505, 3,584, 3,200 in specifications with country-level controls.
  - R^2 ranges reported: 0.159, 0.174, 0.144, 0.162, 0.186, 0.138 across columns.
- TFP growth at frontier (standardized coefficients):
  - Column [1] (All, country FE): 0.271*** (t = 5.783).
  - Column [2] (Tradables): 0.277*** (t = 4.659).
  - Column [3] (Non-tradables): 0.207*** (t = 4.071).
  - Column [4] (All, with country controls): 0.247*** (t = 5.225).
  - Column [5] (Tradables, with country controls): 0.265*** (t = 4.581).
  - Column [6] (Non-tradables, with country controls): 0.144** (t = 2.395).
- Interpretation:
  - A one-standard deviation increase in TFP growth at frontier is associated with, on average, about a 0.25 standard deviation increase in TFP growth (specification in column [4]).
  - The frontier effect is larger for tradable sectors than for non-tradables (e.g., 0.27 vs. 0.14 in the specification with country controls).
- Technological gap (standardized coefficients):
  - Column [1]: -0.310*** (t = -9.342).
  - Column [2]: -0.340*** (t = -7.910).
  - Column [3]: -0.260*** (t = -5.556).
  - Column [4]: -0.357*** (t = -8.264).
  - Column [5]: -0.403*** (t = -6.974).
  - Column [6]: -0.271*** (t = -5.345).
- Interpretation:
  - The technological gap has the largest economic magnitude among regressors: in column [4], a one-standard deviation widening in the technological gap is associated with an average decline of about 0.36 standard deviation in TFP growth.
  - The gap effect is larger in tradable sectors than in non-tradables (-0.40 vs. -0.27 in column [4] vs. [6]).
  - Closing the technological gap would raise TFP growth, on average, by 2.3 percent across the whole sample of industries; and by 2.8 percent for tradables and 3.8 percent for non-tradables (figures as reported in text).
- Industry-level investment and intangible capital:
  - ICT capital spending (standardized coefficients):
    - Column [1]: 0.096*** (t = 2.837).
    - Column [2]: 0.053 (t = 1.140).
    - Column [3]: 0.150*** (t = 3.634).
    - Column [4]: 0.102** (t = 2.627).
    - Column [5]: 0.090 (t = 1.522).
    - Column [6]: 0.133** (t = 2.494).
  - Interpretation:
    - ICT capital spending as a share of gross fixed investment is associated with higher TFP growth; the effect is larger and more statistically significant for non-tradables in some specifications.
    - In the specification with country controls (column [4]), a 1 standard deviation increase in ICT investment increases TFP growth by about 0.1 standard deviation.
  - Non-ICT capital (transportation equipment proxy):
    - Coefficients generally small and statistically insignificant in baseline (examples: -0.012, -0.114, 0.033; -0.014, -0.134, 0.044).
    - Conclusion: non-ICT capital spending does not appear to matter for TFP growth in baseline estimations.
  - R&D spending:
    - Column [1]: 0.143** (t = 2.579).
    - Column [2]: 0.126* (t = 1.791).
    - Column [3]: 0.162 (t = 1.606).
    - Column [4]: 0.148** (t = 2.577).
    - Column [5]: 0.157* (t = 1.890).
    - Column [6]: 0.112 (t = 1.546).
    - Interpretation: R&D spending as a share of gross fixed investment has a positive and statistically significant effect on TFP growth across all industries and tradables in the specification including country controls (columns [4] and [5]); estimated coefficient about 0.15 in those specifications.
  - Aggregate interpretation: increases in intangible capital (ICT and R&D) are associated with higher TFP growth and positive externalities across the economy by accelerating adoption of new technologies.
- Country-level controls (selected coefficients, standardized):
  - Real GDP per capita (with country controls, column [4]): 0.264* (t = 1.781) — positive but marginal significance.
  - Inflation (column [4]): -0.684 (t = -1.407) — mixed and statistically insignificant effects overall.
  - Financial development (column [4]): -0.072 (t = -1.392) — negative but insignificant.
  - Trade openness (column [4]): 0.111 (t = 0.680) — positive but insignificant.
  - Bureaucratic quality (column [4]): 0.165** (t = 2.616) — positive and statistically significant at the 5 percent level for full sample and for non-tradables.
  - Population (column [4]): 0.635 (t = 0.533) — positive coefficient but statistically insignificant.

### Findings with high-skilled labor (Table 3, 2008–2020)
- Baseline with share of high-skilled labor (standardized coefficients):
  - TFP growth at frontier: 0.298*** to 0.292*** across columns (t-statistics reported).
  - Technological gap: large negative coefficients (e.g., -0.589***, -0.693***, -0.378***; -0.601***, -0.695***, -0.384***).
  - ICT capital: examples include 0.120, 0.117, 0.185* and 0.113, 0.132, 0.124 (some significance for non-tradables).
  - R&D spending: coefficients small and mixed; 0.106, 0.143, -0.001 and 0.092, 0.134, -0.014 (not consistently significant).
  - Share of high-skilled labor: -0.055, -0.156, 0.031 and 0.003, -0.066, 0.045 (not statistically significant in baseline).
- Interpretation:
  - Human capital measured by industry-level intensity of high-skilled labor does not appear to have a statistically significant effect on TFP growth at conventional levels in the baseline (consistent with mixed prior findings).
  - Some evidence that the effect of high-skilled labor is stronger when a country is closer to the technological frontier and that human capital matters more in non-tradables than tradables after controlling for other factors.

### Interaction results (Table 4, summarized)
- ICT capital spending × technological gap:
  - ICT capital spending moderates the negative effect of the technological gap on TFP growth for the full sample and especially for tradables (interaction terms reported as statistically significant in text).
- Non-ICT capital × technological gap:
  - Non-ICT capital has a moderating effect on the technological gap that is statistically significant for the full sample, but not for the tradable/non-tradable breakdown.
- R&D × technological gap:
  - The interaction of R&D spending with the technological gap shows a weak significant and negative effect across specifications, except for all industries when country-level variables are included.
- High-skilled labor × technological gap:
  - The interaction of high-skilled labor with the technological gap is statistically significant for the tradable sector only.
- Interpretation:
  - ICT and non-ICT capital can mitigate the dampening effect of a larger technological gap on TFP growth; R&D interactions show weaker and sometimes negative moderating effects; high-skilled labor appears to help close the gap primarily in tradables.

### Robustness checks and visualization
- Robustness:
  - Models estimated for subsample periods: pre-GFC (1995–2007), post-GFC (2010–2020), and excluding COVID-19 (1995–2019); results used to check stability of baseline findings.
  - Results robust to replacing country fixed effects with country-level control variables.
  - Similar qualitative results obtained with the inclusion of country-year and country-industry fixed effects.
- Visualization:
  - Figure 4 presents binned scatter plots of 9,151 observations showing a strong positive correlation between TFP growth and TFP growth at frontier and an inverse relationship with the technological gap (higher frontier growth and narrower technological gap → higher TFP growth).

*Source: EU KLEMS; IMF; World Bank; ICRG; and authors' calculations.*

### Section 3

### wpiea2024258-print-pdf - Section 3

### Industry-level empirical findings (cross-sectional and interaction specifications)
- TFP growth at frontier is a positive and significant determinant of industry-level TFP growth across almost all specifications.
  - Table 4: TFP growth at frontier coefficients: 0.304***, 0.287***, 0.302***, 0.299***, 0.287***, 0.265*** (robust t-statistics in parentheses shown in table).
- Technological gap is negative and statistically significant across specifications.
  - Table 4: Technological gap coefficients: -0.790***, -1.069***, -0.410*, -0.797***, -1.104***, -0.372.
- ICT capital spending:
  - Table 4: ICT capital coefficients: 0.204**, 0.270**, 0.271**, 0.194**, 0.282**, 0.224** (positive; statistically significant in many specifications).
  - Table 5 (subsamples): ICT capital coefficients across Pre-GFC/Post-GFC/Excluding Covid columns include 0.051, 0.014, 0.147, 0.083, 0.139, 0.094, 0.129***, 0.119*, 0.155** (with significance varying by subsample).
- Non-ICT capital spending:
  - Table 4: mixed results — coefficients: -0.013, -0.210, 0.046, -0.055, -0.219, -0.010 (positive and significant in pre-GFC for all industries and non-tradables; turns negative in post-GFC and excluding pandemic for all industries and tradables).
  - Table 5 (subsamples): coefficients include 0.168**, 0.207, 0.181*, -0.018, -0.263, 0.038, -0.019, -0.126, 0.032 (variation by period and sector).
- R&D spending as a share of gross fixed investment:
  - Table 4: coefficients: 0.042, 0.025, -0.081, 0.032, 0.025, -0.111** (positive and significant in pre-GFC and in period excluding COVID-19 across all industries; not significant in post-GFC).
  - Table 5 (subsamples): coefficients include 0.207**, 0.146, 0.205, -0.009, 0.041, -0.048, 0.174**, 0.200*, 0.105.
- Human capital (share of high-skilled labor):
  - Table 4: coefficients: -0.047, 0.109, -0.063, -0.003, 0.251, -0.068 (not statistically significant in baseline; some evidence of stronger effect as countries move closer to frontier and in non-tradables).
- Interaction effects (technological gap interacted with industry factors):
  - Technological gap * ICT capital: 0.158**, 0.309**, 0.171**, 0.159**, 0.298**, 0.205**.
  - Technological gap * non_ICT capital: 0.078**, 0.032, -0.003, 0.059*, 0.027, -0.036.
  - Technological gap * R&D: -0.109*, -0.176*, -0.339*, -0.100, -0.165*, -0.414*.
  - Technological gap * High skilled labor: 0.129, 0.381**, -0.071, 0.131, 0.417**, -0.101.
- Country-level controls (selected coefficients from Table 4 and Table 5):
  - Real GDP per capita: Table 4 shows 0.575** (All), 0.860 (Tradables), 0.420 (Non-tradables). Table 5 subsamples include 1.160***, 1.674**, 0.513, 0.768***, 0.898, 0.684*, 0.254, 0.489, -0.017.
  - Trade openness: Table 4 shows 0.264, 0.762*, -0.132. Table 5 subsamples include 0.686**, 1.029**, 0.285, 0.437, 1.288***, -0.033, 0.086, 0.373, -0.158.
  - Bureaucratic quality: Table 4 shows 0.333, -0.072, 0.674**. Table 5 subsamples include 0.107*, 0.129, 0.106, 0.232, 0.024, 0.304, 0.183***, 0.146, 0.235**.
- Model fit and sample sizes:
  - Table 4: Number of observations: 4,275 (All), 2,196 (Tradables), 1,547 (Non-tradables); R^2: 0.225, 0.244, 0.219 (specifications with country fixed effects); other columns show similar R^2 values: 0.223, 0.246, 0.211.
  - Table 5: No of Observations: 2,673; 1,390; 1,078; 4,151; 1,858; 1,840; 7,155; 3,429; 3,043 across subsample columns; R^2 values include 0.129, 0.168, 0.151, 0.189, 0.217, 0.169, 0.158, 0.184, 0.125.

### System GMM (dynamic modelling and robustness)
- Purpose and implementation:
  - System GMM implemented (Arellano and Bover, 1995; Blundell and Bond, 1998) to address potential endogeneity from feedback between productivity and factor demand; allows inclusion of lagged dependent variable.
  - One-step version applied to avoid the downward bias in standard errors associated with the two-step variant in panels with small number of time-series observations.
  - Strategy to limit proliferation of instruments follows Roodman (2009) to mitigate weak/excessive instruments and guard against over-identification.
- Specification tests and diagnostics (reported p-values):
  - AR(1) p-values: 0.000 (first-difference equation shows high first-order autocorrelation as expected).
  - AR(2) p-values: 0.404 and 0.284 (no evidence for significant second-order autocorrelation).
  - Hansen J-test p-values: 0.037 and 0.029 (indicate validity of internal instruments used in the dynamic model).
- Instruments noted:
  - The lagged dependent variable, asset tangibility, innovative property, and training are specified as instruments.

### Dynamic model results (Table 6)
- Main coefficients (standardized):
  - TFP growth at frontier: 0.341*** (column 1), 0.335*** (column 2).
  - Technological gap: -0.321*** (column 1), -0.371*** (column 2).
  - ICT capital: 0.006 (column 1), 0.050 (column 2) — positive but statistically insignificant.
  - Non-ICT capital: 0.357** (column 1), 0.284* (column 2) — positive and significant in dynamic specification.
  - R&D spending: 0.085 (column 1), 0.030 (column 2) — positive but statistically insignificant.
- Number of observations for dynamic estimations:
  - 8,215 (column 1) and 7,154 (column 2).
- Specification tests (p-values) reported in Table 6:
  - AR(1): 0.000, 0.000.
  - AR(2): 0.404, 0.284.
  - Hansen J-test: 0.037, 0.029.

### Interpretation and narrative conclusions
- Summary of empirical interpretation:
  - TFP growth at the frontier has a robust positive impact on industry-level TFP growth.
  - The technological gap is negatively associated with TFP growth, indicating convergence dynamics: pace of convergence in “follower” industries increases with distance to the frontier.
  - ICT and R&D generally contribute positively to TFP growth, though significance varies by period and specification; non-ICT capital shows mixed signs across periods but is positive and significant in the dynamic model.
  - Industry-level tangible and intangible investment patterns suggest the impact of financial imbalances and resource misallocation prior to the GFC.
  - Human capital does not show a consistent statistically significant direct effect, but interaction evidence suggests its importance when countries approach the technological frontier and in non-tradables.
  - Subsample analysis indicates the technological gap is an important driver of the TFP slowdown in the post-GFC period.
- Policy implications emphasized in the section:
  - Revamp tangible and intangible capital investment in new technologies to generate higher productivity growth directly and indirectly by closing the technological gap.
  - Strengthen human capital accumulation through education and healthcare to support scientific and technological progress.
  - Create a conducive environment for higher business investment and better capital allocation by providing incentives for capital investment and R&D to promote innovation and diffusion of technologies to countries below the frontier, producing positive spillovers across industries.

*IMF Working Paper — Section 3 (as provided in the source content)*

### Section 4

### Section 4

### Core literature on growth theory and productivity
- Solow, R. (1956). “A Contribution to the Theory of Economic Growth,” Quarterly Journal of Economics, Vol. 70, pp. 65–94.  
- Swan, T. (1956). “Economic Growth and Capital Accumulation,” Economic Record, Vol. 32, pp. 334–361.  
- Romer, P. (1987). “Growth Based on Increasing Returns Due to Specialization,” American Economic Review, Vol. 77, pp. 56–62.  
- Romer, P. (1990). “Endogenous Technological Change,” American Economic Review, Vol. 98, pp. 71–102.  
- Lucas, R. (1988). “On the Mechanics of Economic Development,” Journal of Monetary Economics, Vol. 22, pp. 3–42.  
- Nelson, R., and E. Phelps (1966). "Investment in Humans, Technological Diffusion and Economic Growth," American Economic Review, Vol. 56, pp. 69–75.  
- Grossman, G., and E. Helpman (1991). “Trade, Knowledge Spillovers, and Growth,” European Economic Review, Vol. 35, pp. 517–526.  

### Total factor productivity, industry patterns, and misallocation
- Hulten, C. (1996). "Infrastructure Capital and Economic Growth: How Well You Use It May Be More Important Than How Much You Have," NBER Working Papers No. 5847 (Cambridge, MA: National Bureau of Economic Research).  
- Hulten, C. (2001). “Total Factor Productivity: A Short Biography,” in New Directions in Productivity Analysis, C. Hulten, E. Dean, and M. Harper, eds. (Chicago: University of Chicago Press).  
- Krüger, J. (2003). “The Global Trends of Total Factor Productivity: Evidence from the Nonparametric Malmquist Index Approach,” Oxford Economic Papers, Vol. 55, pp. 527–548.  
- Syverson, C. (2011). "What Determines Productivity?" Journal of Economic Literature, Vol. 49, pp. 326–365.  
- Hsieh, C.-T., and P. Klenow (2009). "Misallocation and Manufacturing TFP in China and India," Quarterly Journal of Economics, Vol. 124, pp. 1403–1448.  
- Restuccia, D., and R. Rogerson (2017). "The Causes and Costs of Misallocation," Journal of Economic Perspectives, Vol. 31, pp. 151–174.  
- Schiersch, A., H. Belitz, and M. Gornig (2015). “Why Is TFP Growth Sectorally Concentrated?” Applied Economics, Vol. 47, pp. 5933–5944.  

### Human capital, education, and productivity convergence
- Hanushek, E., and L. Woessmann (2015). The Knowledge Capital of Nations: Education and the Economics of Growth (Cambridge, MA: MIT Press).  
- Henderson, D., and R. Russell (2005). “Human Capital and Convergence: A Production-Frontier Approach,” International Economic Review, Vol. 46, pp. 1167–1205.  
- Kumar, S., and R. Russell (2002). “Technological Change, Technological Catch-Up, and Capital Deepening: Relative Contributions to Growth and Convergence,” American Economic Review, Vol. 92, pp. 527–548.  
- Wolff, E. (1991). “Capital Formation and Productivity Convergence Over the Long Run,” American Economic Review, Vol. 81, pp. 565–579.  
- Margaritis, D., R. Färe, and S. Grosskopf (2007). “Productivity, Convergence and Policy: A Study of OECD Countries and Industries,” Journal of Productivity Analysis, Vol. 28, pp. 87–105.  
- Maudos, J., J. Pastor, and L. Serrano (2000). “Convergence in OECD Countries: Technical Change, Efficiency and Productivity,” Applied Economics, Vol. 32, pp. 757–765.  
- Üngör, M. (2017). “Productivity Growth ad Labor Reallocation: Latin America versus East Asia," Review of Economic Dynamics, Vol. 24, pp. 25–42.  

### Institutions, regulation, trade, and finance
- North, D. (1990). Institutions, Institutional Change and Economic Performance (Cambridge: Cambridge University Press).  
- Knack, S., and P. Keefer (1995). "Institutions And Economic Performance: Cross‐Country Tests Using Alternative Institutional Measures," Economics and Politics, Vol. 7, pp. 207–227.  
- Rodrik, D., A. Subramanian, and F. Trebbi (2004). "Institutions Rule: The Primacy of Institutions Over Geography and Integration in Economic Development," Journal of Economic Growth, Vol. 9, pp. 131–165.  
- Nicoletti, G., and S. Scarpetti (2003). “Regulation, Productivity and Growth: OECD Evidence,” Economic Policy, Vol. 18, pp. 9–72.  
- Scarpetta, S. and T. Tressel (2002). “Productivity and Convergence in a Panel of OECD Industries: Do Regulations and Institutions Matter?” OECD Economics Working Paper No. 342 (Paris: Organization for Economic Co-operation and Development).  
- Miller, S., and M. Upadhyay (2000). "The Effects of Openness, Trade Orientation, and Human Capital on Total Factor Productivity," Journal of Development Economics, Vol. 63, pp. 399–423.  

### Infrastructure, capital, and measurement
- Munnell, A. (1992). "Infrastructure Investment and Economic Growth," Journal of Economic Perspectives, Vol. 6, pp. 189–198.  
- Straub, S. (2011). “Infrastructure and Development: A Critical Appraisal of the Macro-level Literature,” Journal of Development Studies, Vol. 47, pp. 683–708.  
- O’Mahony, M., and M. Timmer (2009). “Output, Input and Productivity Measures at the Industry level: The EU KLEMS Database,” Economic Journal, Vol. 119, pp. F374–F403.  
- Inklaar, R., and M. Timmer (2007). “Of Yeasts and Mushrooms: Patterns of Industry-Level Productivity Growth,” German Economic Review, Vol. 8, pp. 174–187.  
- McMorrow, K., R. Werner, and A. Turrini (2010). “Determinants of TFP Growth: A Close Look at Industries Driving the EU-US TFP Gap,” Structural Change and Economic Dynamics, Vol. 21, pp. 165–180.  

### Empirical methods, datasets, and practical guides
- Roodman, D., 2009, “How to Do xtabond2: An Introduction to Difference and System GMM in Stata,” Stata Journal, Vol. 9, pp. 86–136.  
- Guellec, D., and B. van Pottelsberghe de la Potterie (2004). “From R&D to Productivity Growth: Do the Institutional Settings and the Source of Funds of R&D Matter?” Oxford Bulletin of Economics and Statistics, Vol. 66, pp. 353–378.  

*Source: wpiea2024258-print-pdf - Section 4*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024258-print-pdf.pdf_