## _wp12165

## Source details

**Canonical URL:** [_wp12165](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12165.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12165.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12165.pdf.json)

---

### Data and measures of variables (Section 3)
- Data source and coverage
  - EU KLEMS database; industry output, employment, price, capital stock and investment data from 1970 to 2005 for 15 countries: Australia, Austria, Denmark, Finland, Germany, Italy, Japan, Korea, Netherland, UK, USA, Czech, Portugal, Slovenia, and Sweden (data for Czech, Portugal, Slovenia, Sweden available only from mid-1990s).
- Capital endowment and intensity measures
  - Overall capital endowment = log(total real fixed capital stock / total labor).
  - Three detailed capital categories: ICT, machinery, non-residential structure.
  - Capital-x endowment: K_ENDW_{x,j,t} = ln(K_{x,j,t} / L_{j,t}) × (K_{x,j,t} / K_{j,t}).
  - Main capital intensity measure: industry capital stock / real output (capital-output ratio); robustness alternative: capital’s income share in industry value-added.
  - Human capital intensity control: high-skill workers’ compensation as % of industry total compensation.
  - Capital intensities standardized (z-scores) within country and time for estimation.
- Aggregate labor income share trends (examples from Table 2)
  - AUS: 1975 = 0.727, 1995 = 0.629, 2005 = 0.596, % change = -18.019
  - AUT: 0.728, 0.666, 0.627, -13.874
  - USA: 0.619, 0.630, 0.603, -2.585
  - JPN: 0.589, 0.604, 0.535, -9.338
  - CZE: n.a., 0.567, 0.596, 5.115
- Summary statistics (Table 4A)
  - Overall Capital endowment: N=427, Mean=5.001, Std. Dev.=0.460, Min=3.426, Max=5.989
  - ICT capital endowment: N=427, Mean=-2.869, Std. Dev.=2.035, Min=-8.921, Max=1.165
  - Structure capital endowment: N=427, Mean=3.320, Std. Dev.=0.673, Min=1.131, Max=4.504
  - Machinery capital endowment: N=427, Mean=1.159, Std. Dev.=0.472, Min=-0.488, Max=2.441
  - Annual growth rate of GDP per worker: N=416, Mean=0.020, Std. Dev.=0.022, Min=-0.058, Max=0.103
  - Log GDP per worker: N=427, Mean=4.481, Std. Dev.=0.385, Min=3.353, Max=5.303
  - Industry variables (N=11033): Real output share Mean=0.033, Std. Dev.=0.023, Min=0.000, Max=0.234; Employment share Mean=0.033, Std. Dev.=0.028, Min=0.000, Max=0.183; Nominal output share Mean=0.033, Std. Dev.=0.022, Min=0.000, Max=0.137.
- Key correlations (Tables 4B & 4C)
  - Country-level:
    - Correlation(Log GDP per worker, Overall Capital endowment) = 0.83
    - Correlation(Log GDP per worker, ICT endowment) = 0.42
    - Correlation(Log GDP per worker, Structure endowment) = 0.66
    - Correlation(Log GDP per worker, Machinery endowment) = 0.68
  - Industry-level:
    - Overall capital intensity vs ICT intensity = 0.43
    - Overall capital intensity vs Structure intensity = 0.81
    - Overall capital intensity vs Machinery intensity = 0.19
    - Overall capital intensity vs Human capital intensity = 0.29

### Illustrative theoretical model (Section 6)
- Model setup and assumptions
  - Two-sector final good aggregator with elasticity of substitution ε ∈ [0, ∞).
  - Sectoral production: Cobb–Douglas Yi,t = A_t K_{i,t}^{a_i} L_{i,t}^{1-a_i}, shared A_t, Sector 1 more capital-intensive: a1 - a2 > 0.
  - Labor freely mobile: L1,t + L2,t = Lt.
  - Capital mobile but subject to adjustment cost G(·), quadratic specification G(·) = φ (K_{1,t}/K_{t} − s_t)^2 with φ ≥ 0.
  - Capital market clearing: K1,t + K2,t + G(·) = Kt.
- Key propositions and implications
  - If ε > 1, increasing aggregate capital stock K_t shifts capital toward the capital-intensive sector (Sector 1) and increases Sector 1’s labor share via ξ_t increasing in λ_t (equation (11) with Φ in (12)).
  - Structural adjustment cost φ > 0 attenuates reallocation: ∂(ln λ_t)/∂(ln K_t) decreases with φ (equation (13)).
  - Aggregate output effect: with φ = 0 allocation maximizes Y_t for given K_t; with φ > 0 the increase in Y_t from higher K_t is lower: ∂^2 Y_t/∂K_t^2 |_{φ>0} < ∂^2 Y_t/∂K_t^2 |_{φ=0}.
- Main theoretical conclusions (verbatim insights)
  - Increasing capital endowment is likely to be accompanied by structural change towards capital intensive sectors in terms of real output.
  - Change in employment and nominal output composition depends on ε; if ε > 1, nominal output and employment shares of capital-intensive industries also rise with capital endowment.
  - Structural adjustment costs that hinder cross-sector reallocation suppress sectoral structural change and can lead to suboptimal aggregate performance.

### Country-level analysis (Section IV)
- Measures of capital intensity of industrial structure
  - COR(K_x, industry size) = Spearman rank correlation between industry capital-x intensity score and industry size (real output share, employment share, nominal output share).
  - Weighted average measure: Σ_i (industry size_i × K_x score_i).
- Empirical patterns (Tables 5a and 5b)
  - Overall capital: capital intensity of industrial structure (real output, employment, nominal output) is positively correlated with capital endowment; all correlation coefficients significant at 1% level.
  - Magnitude ordering: highest correlation for real output structure, lowest for employment structure.
  - ICT and structure capital: positive correlations with capital endowment (mostly significant at 1%).
  - Machinery capital: correlations between machinery intensity and machinery endowment are negative across industry-size measures.
  - Interpretation consistent with ε > 1 for overall and ICT capital; machinery shows atypical pattern.
- Structural incoherence index (SI) and growth regressions
  - SI_{x,j,t} = | k_intensity_{x,j,t} − k_endw_{x,j,t} | (standardized z-scores); two versions:
    - Version 1: intensity via Spearman rank correlation.
    - Version 2: intensity via industry real-output-share-weighted average.
  - SI summary statistics (Table 6, selected)
    - SI (v1) Overall capital: Mean=0.877, Std. Dev.=0.541, Min=0.008, Max=3.004
    - SI (v2) Overall capital: Mean=0.839, Std. Dev.=0.730, Min=0.001, Max=3.261
    - SI (v1) Machinery: Mean=1.225, Std. Dev.=0.865, Min=0.015, Max=4.178
    - SI (v2) Machinery: Mean=1.186, Std. Dev.=0.765, Min=0.007, Max=3.869
- Country-level growth regression results (equation (15), Tables 7a and 7b)
  - Version 1 (SI v1) key coefficients (heteroskedasticity-robust SEs):
    - k=1 (annual): Overall capital SI coefficient = -0.010*** (Column 1).
    - k=4 (5-year): Overall capital SI = -0.042** (Column 3).
    - k=9 (10-year): Overall capital SI = -0.149** (Column 5).
    - Quantitative example: decreasing SI from 75th percentile (1.23) to 25th percentile (0.46) associated with 0.8 percentage point increase in annual GDP growth (≈24% of growth differential between 25th and 75th percentile country-years).
    - Machinery and Structure SI: negative and often significant depending on specification/time window; ICT SI not significant.
  - Version 2 (SI v2):
    - k=1: Overall capital SI = -0.011*** (Column 1).
    - k=4: Overall capital SI = -0.067** (Column 3).
    - k=9: Overall capital SI = -0.245*** (Column 5).
- Instrumental variables (Tables 8a and 8b)
  - 2SLS/IV using two-period lagged SI as instruments: results consistent or larger in magnitude for overall capital; machinery SI similar; structure SI becomes mostly insignificant in IV for detailed capitals.
  - Hansen J test p-values reported; IV specifications pass overidentification checks in many cases.
- Robustness: partial-regression plots (Figures 6a–6c) show results not driven by outliers.

### Industry-level analysis (Section V)
- Baseline specifications (equations (16) and (17))
  - Dependent vars: ln(real output share), ln(employment share), ln(nominal output share) at end of 5-year window.
  - Key regressors: lagged industry capital-x intensity score K_{x,ij,t−1} (standardized), country capital-x endowment K_ENDW_{x,j,t−1}, interaction K_{x,ij,t−1} × K_ENDW_{x,j,t−1}, endowment growth ΔK_ENDW_{x,j,t}, and interaction K × ΔK_ENDW.
  - Controls: lagged dependent variable ln Y_{ij,t−1}, country log per-worker output, 5-year industry TFP growth; country-industry fixed effect; estimation via Arellano–Bond difference GMM.
- Baseline results for overall capital (Table 8)
  - Interaction K × K_ENDW: positive and significant in most specifications:
    - Real output share: Fixed effect 0.135**, GMM 0.033*.
    - Employment share: Fixed effect 0.132**, GMM 0.058**.
    - Nominal output share: Fixed effect 0.127*, GMM 0.145***.
  - Interaction K × ΔK_ENDW: positive, GMM significant in employment and nominal output share regressions (employment GMM 0.071**; nominal GMM 0.531***).
  - TFP growth positive and significant in real output share regressions (TFP growth coefficient 0.007*** in fixed effect).
- Detailed capital types (Table 9)
  - ICT:
    - ICT × ICT_ENDW interaction positive and significant across real, employment, nominal share regressions (GMM and fixed effect).
    - ICT × ΔICT_ENDW positive and significant in GMM.
  - Structure:
    - STR × STR_ENDW mostly positive; significance varies.
    - STR × ΔSTR_ENDW mostly not significant.
  - Machinery:
    - MCH × MCH_ENDW coefficients mainly negative; negative and significant in employment and nominal output GMM regressions.
    - MCH × ΔMCH_ENDW negative and significant for employment/nominal in GMM.
- Interpretation
  - Overall capital and ICT: capital-intensive industries expand with higher capital endowment and with capital accumulation—consistent with ε > 1 predictions.
  - Machinery: behaves differently, with negative interaction coefficients in several specifications.

### Structural coherence and growth (industry-level three-way interaction)
- Identification
  - Country-time coherence captured by interaction coefficient a_{3,j,t} on K × K_ENDW in Equation (16); linked to aggregate growth via GROW_j = d_1 + d_2 a_{3,j}.
- Three-way interaction results (Table 10)
  - Overall capital:
    - K × K_ENDW × GROW: Fixed effect 0.231***, GMM 0.242*** — positive and significant.
    - K × K_ENDW two-way: Fixed effect 0.070***, GMM -0.035 (variation by estimator).
  - ICT:
    - ICT × ICT_ENDW × GROW positive and significant (Fixed effect 0.118**, GMM 0.023**).
  - Structure:
    - STR × STR_ENDW × GROW positive in GMM (0.060**).
  - Machinery:
    - MCH × MCH_ENDW × GROW positive and significant in GMM/fixed in some specifications despite two-way interactions being negative.
- Magnitude interpretation (overall capital example)
  - From Table 10 GMM: three-way coefficient ≈ 0.242 implies a 5-year aggregate output growth rate difference of 0.136 between highest structural coherence scenario and random-structure scenario (≈25% of gap between 5th and 95th percentile countries).

### Robustness (Section VI)
- Using capital income share as factor-intensity (Tables 11 and 12)
  - Overall capital: k × K_ENDW sometimes loses significance; k × ΔK_ENDW becomes significant (e.g., 0.391** in real output GMM).
  - ICT: ict × ICT_ENDW and ict × ΔICT_ENDW remain positive and significant.
  - Machinery: mch × MCH_ENDW negative and significant in some specifications (employment/nominal).
  - Three-way interactions generally remain positive and sometimes significant for overall, ICT, and machinery; structure less consistent.
- Additional controls and checks (Tables 13–15B)
  - Added controls: human capital intensity (HUM), HUM × K_ENDW, country GDP per worker (Y) and interactions, HighVA (industry value-added/gross output), TFP_GROW × GROW.
  - HUM × K_ENDW positive and significant in many specifications; HUM interactions alter some K_x interaction significances.
  - K_x × Y × GROW mostly insignificant except MCH × Y × GROW positive and highly significant (0.393*** in one specification).
  - HighVA × Y × GROW positive in some regressions (e.g., structure capital 0.166*).
  - Overall finding: main result—that higher structural coherence between industrial structure and capital endowment is associated with higher aggregate growth—remains robust to alternative factor-intensity measures, inclusion of human capital and income-level controls, and TFP-alignment considerations.
  - Note: Machinery capital results are more sensitive to specification and controls.

### Conclusion and policy implications (Section VII)
- Empirical summary
  - Overall capital: capital-intensive industries expand with higher initial capital endowment and with capital accumulation across real output, employment, nominal output—consistent with theory when ε > 1.
  - ICT capital: similar positive relationships for industry expansion with ICT endowment and accumulation.
  - Machinery and structure capital: mixed results; machinery often shows atypical (negative) interactions.
  - Structural incoherence (SI) associated with lower GDP growth: country-level example indicates reducing SI from 75th to 25th percentile associates with ≈0.8 percentage point higher annual growth (Version 1).
  - Industry-level three-way interactions support that greater coherence amplifies growth; quantitative magnitudes broadly consistent with country-level estimates (e.g., 5-year growth difference of 0.136).
- Policy recommendations (verbatim thematic recommendations)
  - Structural coherence matters for growth; policies should account for evolving endowment mixes when designing industrial policy.
  - Facilitate resource mobility across industries to enable structural evolution:
    - Invest in continuous education of the labor force.
    - Develop a flexible financial sector that provides easy access to funding to emerging industries and firms.
  - Industrial policy should recognize that an industry historically coherent with endowment fundamentals may lose coherence over time as endowments change; government support decisions should consider long-term endowment trajectories.
- Overall conclusion
  - Structural coherence between industry capital-intensity patterns and country capital endowment is an important determinant of economic growth; frictions or impediments to cross-sector reallocation can suppress structural adjustment and reduce aggregate performance.

*Source: _wp12165 - Section 3 discusses the data and defines measures of variables. Section 4 and 5 present the — IMF PDF*

### Section 3 discusses the data and defines measures of variables. Section 4 and 5 present the

### _wp12165 - Section 3 discusses the data and defines measures of variables. Section 4 and 5 present the

### Data and measures of variables (Section 3)
- Section 3 discusses the data and defines measures of variables.

### Empirical models and estimation results (Sections 4 and 5)
- Section 4 and 5 present the empirical models, at country and industry level respectively, and discuss the estimation results.

### Additional restrictions and robustness checks
- More restrictions to the industry-level estimation and robustness checks are added in

*Source: _wp12165 - Section 3 discusses the data and defines measures of variables. Section 4 and 5 present the — IMF PDF*

### Section 6. Section 7 concludes.

### _wp12165 - Section 6. Section 7 concludes.

### II. AN ILLUSTRATIVE MODEL
- Model setup
  - Two-sector final good produced by combining two sectoral goods with elasticity of substitution ε ∈ [0, ∞).
  - Final good aggregator: Yt = [(γ1 Y1,t^{(ε-1)/ε} + γ2 Y2,t^{(ε-1)/ε})^{ε/(ε-1)}], with γ1 + γ2 = 1.
  - Sectoral production: Cobb–Douglas Yi,t = A_t K_{i,t}^{a_i} L_{i,t}^{1-a_i}, with shared productivity A_t and Sector 1 more capital-intensive: a1 - a2 > 0.
- Prices and nominal vs real output ratios
  - If ε > 1, nominal output ratio moves in the same direction as real output ratio; if ε < 1, the opposite holds.
- Resource mobility and adjustment costs
  - Labor freely mobile: L1,t + L2,t = Lt (equation (2)).
  - Capital mobile but subject to adjustment cost G(K1,t/K2,t − s_t) with G(0)=0, G'(·)>0, G''(·)≥0; quadratic specification G(·) = φ (K_{1,t}/K_{t} − s_t)^2 with φ ≥ 0 (equation (4)).
  - Capital market clearing: K1,t + K2,t + G(·) = Kt (equation (3)).
- Planner’s intra-temporal allocation problem (equation (5)) and first-order conditions imply equalized marginal products across sectors (equations (6) and (7)).
- Key proposition (equation (11) with Φ in (12))
  - When ε > 1, increasing aggregate capital stock K_t shifts capital toward the capital-intensive sector (Sector 1) and increases Sector 1’s labor share via ξ_t increasing in λ_t.
  - Presence of structural adjustment cost φ > 0 attenuates this reallocation: ∂(ln λ_t)/∂(ln K_t) decreases with φ (equation (13)).
- Aggregate output effect
  - With φ = 0 allocation maximizes Y_t for given K_t; with φ > 0 the increase in Y_t from higher K_t is lower than in the frictionless case: ∂^2 Y_t/∂K_t^2 |_{φ>0} < ∂^2 Y_t/∂K_t^2 |_{φ=0}.
- Main theoretical conclusions (verbatim insights)
  - Increasing capital endowment is likely to be accompanied by structural change towards capital intensive sectors in terms of real output.
  - Change in employment and nominal output composition depends on ε; if ε > 1, nominal output and employment shares of capital-intensive industries also rise with capital endowment.
  - Structural adjustment costs that hinder cross-sector reallocation suppress sectoral structural change and can lead to suboptimal aggregate performance.

*Italic line: Source: _wp12165 - Section 6. Section 7 concludes.*

### III. DATA AND VARIABLES
- Data source and coverage
  - EU KLEMS database; industry output, employment, price, capital stock and investment data from 1970 to 2005 for 15 countries: Australia, Austria, Denmark, Finland, Germany, Italy, Japan, Korea, Netherland, UK, USA, Czech, Portugal, Slovenia, and Sweden (data for Czech, Portugal, Slovenia, Sweden available only from mid-1990s).
- Industry panel and summary patterns
  - Table 1: cross-country median growth rates of real output shares, employment shares, nominal output shares (1970–2005) and median overall capital intensity (capital stock/output) for 28 industries; industries sorted by median real output share growth.
  - Stylized pattern: shift toward services and some sophisticated manufacturing; employment composition shifts even more strongly to services; median nominal output share growth mostly aligned with employment shares except seven industries.
  - Positive correlation between industry median real output share growth and median overall capital intensity: correlation coefficient = 0.25 at 1% significance level.
- Capital types and endowment measures
  - Overall capital endowment = log(total real fixed capital stock / total labor).
  - Three detailed capital categories: ICT, machinery, non-residential structure.
  - Capital-x endowment defined as ln(K_x/L) × (K_x / K) (text formula: K_ENDW_{x,j,t} = ln(K_{x,j,t} / L_{j,t}) × (K_{x,j,t} / K_{j,t})).
  - Change in capital-x endowment decomposed into percentage change in K_x/L and difference between percentage changes of K_x/L and aggregate K/L (text decomposition).
- Capital intensity measures
  - Main measure: industry capital stock / real output (capital-output ratio).
  - Robustness alternative: capital’s income share in industry value-added.
  - Human capital intensity control: high-skill workers’ compensation as % of industry total compensation.
  - For estimation, capital intensities are standardized (z-scores) within country and time to capture within-country cross-industry variation.
- Aggregate labor income share trends (Figure 1, Table 2)
  - In 13 out of 15 countries labor share declined over sample period.
  - Table 2 shows country-level aggregate labor income share for 1975, 1995, 2005 and % change 1975–2005 (examples):
    - AUS: 1975 = 0.727, 1995 = 0.629, 2005 = 0.596, % change = -18.019
    - AUT: 0.728, 0.666, 0.627, -13.874
    - USA: 0.619, 0.630, 0.603, -2.585
    - JPN: 0.589, 0.604, 0.535, -9.338
    - CZE: n.a., 0.567, 0.596, 5.115
- Summary statistics (Table 4A)
  - # observations and means:
    - Overall Capital endowment: N=427, Mean=5.001, Std. Dev.=0.460, Min=3.426, Max=5.989
    - ICT capital endowment: N=427, Mean=-2.869, Std. Dev.=2.035, Min=-8.921, Max=1.165
    - Structure capital endowment: N=427, Mean=3.320, Std. Dev.=0.673, Min=1.131, Max=4.504
    - Machinery capital endowment: N=427, Mean=1.159, Std. Dev.=0.472, Min=-0.488, Max=2.441
    - Annual growth rate of GDP per worker: N=416, Mean=0.020, Std. Dev.=0.022, Min=-0.058, Max=0.103
    - Log GDP per worker: N=427, Mean=4.481, Std. Dev.=0.385, Min=3.353, Max=5.303
  - Industry variables (N=11033): Real output share Mean=0.033, Std. Dev.=0.023, Min=0.000, Max=0.234; Employment share Mean=0.033, Std. Dev.=0.028, Min=0.000, Max=0.183; Nominal output share Mean=0.033, Std. Dev.=0.022, Min=0.000, Max=0.137.
- Correlations (Table 4B & 4C)
  - Country-level correlations:
    - Correlation(Log GDP per worker, Overall Capital endowment) = 0.83
    - Correlation(Log GDP per worker, ICT endowment) = 0.42
    - Correlation(Log GDP per worker, Structure endowment) = 0.66
    - Correlation(Log GDP per worker, Machinery endowment) = 0.68
  - Industry-level correlations:
    - Overall capital intensity vs ICT intensity = 0.43
    - Overall capital intensity vs Structure intensity = 0.81
    - Overall capital intensity vs Machinery intensity = 0.19
    - Overall capital intensity vs Human capital intensity = 0.29

### IV. COUNTRY LEVEL ANALYSIS
#### A. Capital Endowment and Industrial Structure
- Measures of capital intensity of industrial structure
  - Correlation measure: COR(K_x, industry size) = Spearman rank correlation between industry capital-x intensity score and industry size (real output share, employment share, nominal output share).
  - Weighted average measure: Σ_i (industry size_i × K_x score_i) over n industries.
- Empirical patterns (Tables 5a and 5b)
  - For overall capital, capital intensity of industrial structure (real output, employment, nominal output) is positively correlated with capital endowment; all correlation coefficients significant at 1% level.
  - Magnitude ordering: highest correlation for real output structure, lowest for employment structure.
  - For ICT and structure capital, capital intensity of industrial structure positively correlated with capital endowment (mostly significant at 1%).
  - For machinery capital, correlations between capital intensity of industrial structure and machinery endowment are negative across industry-size measures.
- Interpretation
  - Overall pattern consistent with ε > 1 (elasticity of substitution above unity) so capital-intensive industries grow with higher capital endowment.
  - Cross-country and time heterogeneity exist; countries with similar capital abundance may have different industrial structures.

#### B. Structural Coherence and Growth
- Structural incoherence (SI) index (equation (14))
  - SI_{x,j,t} = | k_intensity_{x,j,t} − k_endw_{x,j,t} | where lower-case denotes standardized (z-score) values; two SI versions:
    - Version 1: capital intensity measured as Spearman rank correlation between industry real output share and industry capital intensity.
    - Version 2: capital intensity measured as industry real-output-share-weighted average of industry capital intensities.
  - Table 6 (summary statistics of SI):
    - SI (v1) Overall capital: Mean=0.877, Std. Dev.=0.541, Min=0.008, Max=3.004
    - SI (v1) ICT: Mean=0.748, Std. Dev.=0.613, Min=0.007, Max=2.907
    - SI (v1) Machinery: Mean=1.225, Std. Dev.=0.865, Min=0.015, Max=4.178
    - SI (v1) Structure: Mean=1.004, Std. Dev.=0.644, Min=0.002, Max=2.908
    - SI (v2) Overall capital: Mean=0.839, Std. Dev.=0.730, Min=0.001, Max=3.261
    - SI (v2) ICT: Mean=0.674, Std. Dev.=0.606, Min=0.014, Max=2.525
    - SI (v2) Machinery: Mean=1.186, Std. Dev.=0.765, Min=0.007, Max=3.869
    - SI (v2) Structure: Mean=1.013, Std. Dev.=0.599, Min=0.014, Max=3.425
- Time-series patterns (Figures 4 & 5)
  - Japan: SI increased from near zero in 1970s to above 3 in 2005.
  - US and Germany: consistently lower-than-average SI; US SI decreased notably in 1990s.
  - Rising SI in countries such as Italy and Denmark driven mainly by lack of upward movement in capital intensity of industrial structure despite capital accumulation.
- Country-level growth regressions (equation (15))
  - Dependent variable: real GDP growth over interval (GROW_{j,t−k,t}).
  - Key explanatory: average SI over same period; controls Z include initial GDP, average investment intensity, average human capital intensity (shares of high and medium skilled in total labor compensation); country fixed effects included.
- Main regression results (Tables 7a and 7b; standard errors heteroskedasticity-robust)
  - Version 1 (SI v1) (Table 7a)
    - k=1 (annual): Overall capital SI coefficient = -0.010*** (Column 1) — significant; interpretation: higher SI → lower annual GDP growth.
    - k=4 (5-year): Overall capital SI = -0.042** (Column 3).
    - k=9 (10-year): Overall capital SI = -0.149** (Column 5).
    - Quantitative example (Version 1): decreasing SI from 75th percentile (1.23) to 25th percentile (0.46) associated with 0.8 percentage point increase in annual GDP growth (≈24% of growth differential between 25th and 75th percentile country-years).
    - Machinery and Structure SI: negative and often significant depending on specification/time window; ICT SI not significant.
  - Version 2 (SI v2) (Table 7b)
    - k=1: Overall capital SI = -0.011*** (Column 1).
    - k=4: Overall capital SI = -0.067** (Column 3).
    - k=9: Overall capital SI = -0.245*** (Column 5).
    - Machinery SI significant and negative in several specifications; Structure SI mostly negative and sometimes significant; ICT SI not significant.
- Instrumental variables (Tables 8a and 8b)
  - 2SLS/IV using two-period lagged SI as instruments: results consistent or larger in magnitude for overall capital; machinery SI similar; structure SI becomes mostly insignificant in IV for detailed capitals.
  - Hansen J test p-values reported; IV specifications pass overidentification checks in many cases.
- Robustness plots (Figures 6a–6c)
  - Partial regression plots for SI show results not driven by outliers in annual, 5-year, and 10-year regressions.

### V. INDUSTRY LEVEL ANALYSIS
#### A. Capital Endowment and Industrial Structure
- Baseline industry-level specifications (equations (16) and (17))
  - Dependent variables: ln(real output share), ln(employment share), ln(nominal output share) of industry i in country j at end of 5-year window.
  - Key regressors: lagged industry capital-x intensity score (standardized) K_{x,ij,t−1}, country capital-x endowment K_ENDW_{x,j,t−1}, interaction K_{x,ij,t−1} × K_ENDW_{x,j,t−1}, endowment growth ΔK_ENDW_{x,j,t}, and interaction K × ΔK_ENDW.
  - Controls Z' include lagged dependent variable ln Y_{ij,t−1}, country log per-worker output, 5-year industry TFP growth; error has country-industry fixed effect; estimation via Arellano–Bond difference GMM.
- Baseline results for overall capital (Table 8)
  - Interaction K × K_ENDW: positive and significant in most specifications:
    - Real output share: Fixed effect 0.135**; GMM 0.033* (columns (1)-(2) under log Real output share).
    - Employment share: Fixed effect 0.132**; GMM 0.058**.
    - Nominal output share: Fixed effect 0.127*; GMM 0.145***.
  - Interaction K × ΔK_ENDW: positive, GMM significant in employment and nominal output share regressions (e.g., employment GMM 0.071**; nominal GMM 0.531***).
  - K_ENDW and ΔK_ENDW direct terms varied in sign and significance across specifications; TFP growth positive and significant in real output share regressions (TFP growth coefficient 0.007*** in fixed effect).
  - Diagnostics: Arellano–Bond second-order autocorrelation p-values and Hansen J test p-values reported; instrument choices limited/collapsed to maintain validity.
- Detailed capital types (Table 9)
  - ICT:
    - ICT × ICT_ENDW interaction positive and significant across real, employment, nominal share regressions (GMM and fixed effect).
    - ICT × ΔICT_ENDW also positive and significant in GMM.
  - Structure (non-residential structure capital):
    - STR × STR_ENDW mostly positive; significance varies (significant in some employment regressions).
    - STR × ΔSTR_ENDW mostly not significant.
  - Machinery:
    - MCH × MCH_ENDW coefficients mainly negative (often insignificant in real output but significantly negative in employment and nominal output GMM regressions).
    - MCH × ΔMCH_ENDW negative and significant for employment/nominal in GMM.
- Interpretation
  - For overall capital and ICT, capital-intensive industries expand with higher capital endowment and with capital accumulation—consistent with ε > 1.
  - Machinery capital behaves differently: higher machinery endowment is not associated with expansion of machinery-intensive industries in the same way; interaction coefficients are negative in several specifications.

#### B. Structural Coherence and Economic Growth (industry-level approach)
- Identification of coherence via coefficient a_{3,j,t} from Equation (16): interaction coefficient on K × K_ENDW at country-time level interpreted as country-time measure of coherence between industrial structure and capital endowment.
- Linking coherence to growth (equations (19) and (20))
  - Assume GROW_j = f(a_{3,j}) = d_1 + d_2 a_{3,j}; expect d_2 > 0 if higher coherence promotes growth.
  - Equation (20) embeds this into the industry-level model through three-way interaction terms K × K_ENDW × GROW and related terms.
- Baseline three-way interaction results (Table 10)
  - Overall capital:
    - K × K_ENDW × GROW: Fixed effect 0.231***, GMM 0.242*** — positive and significant.
    - K × K_ENDW main two-way interaction: Fixed effect 0.070***, GMM -0.035 (GMM two-way varies).
  - ICT:
    - ICT × ICT_ENDW × GROW positive and significant (Fixed effect 0.118**; GMM 0.023**).
    - ICT two-way ICT × ICT_ENDW positive and significant.
  - Structure:
    - STR × STR_ENDW × GROW positive in GMM (0.060**) though fixed effect not significant.
  - Machinery:
    - MCH × MCH_ENDW × GROW positive and significant in GMM (0.333*** fixed/GMM variation).
    - MCH two-way interaction negative or insignificant.
- Magnitude interpretation (overall capital example)
  - From Table 10 GMM: d_2 ≈ 0.242 (three-way interaction) implies f coefficient around 4.13 (text arithmetic: d_2 = 1/f? text gives d_2 = 0.242 implies f ≈ 4.13).
  - Combined estimates indicate a 5-year aggregate output growth rate difference of 0.136 between highest structural coherence scenario and random structural change (a_3 = 0), amounting to ≈25% of gap between 5th and 95th percentile countries in the data.
- Robustness: alternative capital intensity measure (capital income share)
  - Re-estimations using capital income share as factor-intensity (Tables 11 and 12) largely preserve signs and many significances:
    - Interaction K × ΔK_ENDW becomes more significant in some real output regressions; initial endowment interaction may lose significance in some specifications.
    - Three-way interactions in structural-coherence regressions remain positive and significant for most capital categories except structure in some specifications.

### VI. ROBUSTNESS
#### A. Using income share to measure factor intensity
- Alternative measure: capital-x intensity measured by capital income share in industry value-added (variables denoted k, ict, str, mch in lowercase).
- Table 11 (Arellano–Bond GMM for Equation (17) with income-share measure)
  - Overall capital: k × K_ENDW coefficients become insignificant; k × ΔK_ENDW becomes significant (0.391**) in real output share regression (GMM).
  - ICT: ict × ICT_ENDW positive and significant; ict × ΔICT_ENDW significant.
  - Structure: str × STR_ENDW and str × ΔSTR_ENDW show increased significance in some nominal output regressions.
  - Machinery: mch × MCH_ENDW negative and significant in some specifications (employment/nominal).
  - Diagnostics: A-B 2 test p-values and Hansen J test p-values mostly acceptable except A-B 2 test p-value = 0.002 in one nominal-output regression (possible serial correlation concern).
- Table 12 (Equation (20) with income-share measure)
  - Three-way interactions k × K_ENDW × GROW, ict × ICT_ENDW × GROW, mch × MCH_ENDW × GROW generally positive and sometimes significant; str × STR_ENDW × GROW less consistent.
  - Magnitude of implied f changes: smaller for overall and ICT, larger for machinery relative to baseline.

#### B. Further robustness checks and additional controls
- Controls added to Equation (17) and Equation (20):
  - Human capital intensity (HUM) and HUM × K_ENDW interactions.
  - Country GDP per worker (Y) and interactions K × Y and HighVA × Y (HighVA = industry value-added / gross output).
  - Industry TFP growth interactions and allocative-efficiency measures (TFP_GROW × GROW).
- Table 13 (overall capital with HUM and Y/HighVA controls)
  - HUM × K_ENDW positive and significant across size regressions; adding HUM increases significance of some K × K_ENDW and K × ΔK_ENDW terms.
  - K × Y interactions generally insignificant; HighVA × Y interactions positive and significant in some regressions.
- Table 14 (detailed capitals with HUM and Y/HighVA controls)
  - HUM × ICT_ENDW interaction significant in employment share regression.
  - ICT × ICT_ENDW and ICT × ΔICT_ENDW largely remain positive and significant.
  - Machinery interactions become sensitive to inclusion of Y and HUM: some signs flip or significance changes.
- Table 15A and 15B (structural coherence regressions with additional interactions)
  - Adding HUM × K_ENDW × GROW and HUM × K_x_ENDW × GROW:
    - HUM × ICT_ENDW × GROW and HUM × STR_ENDW × GROW negative (HUM × ICT_ENDW × GROW -0.286***), while HUM × MCH_ENDW × GROW and HUM × K_ENDW × GROW positive but not significant.
    - Main coherence three-way interactions K_x × K_x_ENDW × GROW remain positive and significant for overall, ICT, structure, machinery in many specifications.
  - Adding K_x × Y × GROW, HighVA × Y × GROW, TFP_GROW × GROW:
    - K_x × Y × GROW mostly negative or insignificant except MCH × Y × GROW positive and highly significant (0.393***).
    - HighVA × Y × GROW positive in some regressions (e.g., structure capital 0.166*).
    - TFP_GROW × GROW positive in overall capital regression (0.010*).
- Overall robustness conclusion
  - Main finding—that higher structural coherence between industrial structure and capital endowment is associated with higher aggregate growth—remains robust to alternative factor-intensity measures, inclusion of human capital and income-level controls, and consideration of allocative-efficiency (TFP alignment) effects.
  - Machinery capital results are more sensitive and show different patterns, possibly reflecting measurement or country-development correlations.

### VII. CONCLUSION
- Research scope and questions
  - Examines how factor endowment changes (overall capital and three capital types: ICT, machinery, structure) relate to structural change across industries and whether a country’s industrial structure being coherent with its capital endowment (structural coherence) relates to aggregate growth.
- Empirical findings summary
  - Overall capital: capital-intensive industries expand with higher initial capital endowment and with capital accumulation (real output, employment, nominal output), consistent with theoretical predictions when ε > 1.
  - ICT capital: similar positive relationships for industry expansion with ICT endowment and ICT accumulation.
  - Machinery and structure capital: partial and mixed confirmations; machinery shows atypical (often negative) interactions in several specifications.
  - Structural incoherence (SI) at country level: higher SI (misalignment between industrial structure capital intensity and capital endowment) is associated with lower GDP growth; country-level estimates suggest reducing SI from 75th to 25th percentile associates with ≈0.8 percentage point higher annual growth (Version 1 example).
  - Industry-level three-way interaction evidence: K × K_ENDW × GROW positive and significant (Table 10), consistent with hypothesis that greater coherence amplifies growth.
  - Quantitative magnitudes: industry-level and country-level approaches produce similar assessments of coherence impact; e.g., a 5-year aggregate output growth rate difference of 0.136 between highest coherence and random-structure scenarios (≈25% of gap between 5th and 95th percentile countries).
- Policy implications (verbatim thematic recommendations)
  - Structural coherence matters for growth; policies should account for evolving endowment mixes when designing industrial policy.
  - Facilitate resource mobility across industries to enable structural evolution:
    - Invest in continuous education of the labor force.
    - Develop a flexible financial sector that provides easy access to funding to emerging industries and firms.
  - Industrial policy should recognize that an industry historically coherent with endowment fundamentals may lose coherence over time as endowments change; government support decisions should consider long-term endowment trajectories.
- Overall conclusion
  - Structural coherence between industry capital-intensity patterns and country capital endowment is an important determinant of economic growth; frictions or impediments to cross-sector reallocation can suppress structural adjustment and reduce aggregate performance.

*Italic line: Source: _wp12165 - Section 6. Section 7 concludes.*

### References

### References

### Works cited
- 1. Acemoglu, D. and Guerrieri, V.: Capital deepening and non-balanced economic growth. Journal of Political Economy, vol. 116, no.3, 467-498 (2008)
- 2. Arellano, M. and Bond, S.R.: Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations.  Review of Economic Studies 58, 277-297 (1991)
- 3. Arnold, J., Nicoletti,G. and Scarpetta, S.: Regulation, allocative efficiency and productivity in OECD countries: industry and firm-level evidence.  OECD Working Paper (2008)
- 4. Arpaia, A., Pérez, E. and Pichelmann, K.: Understanding labor income share dynamics in Europe.  European Commission Working Paper (2009)
- 5. Baily, M.: Macroeconomic implications of the new economy.  In: Economic Policy for the Information Economy.  Federal Reserve Bank of Kansas City (2001)
- 6. Bartelsman, E., Haltiwanger, J. and Scarpetta, S.: Cross country differences in productivity: the role of allocative efficiency.  Working paper (2008)
- 7. Bentolila, S. and Saint-Paul, G.: Explaining movements in the labor share. The B.E. Journal of Macroeconomics, Vol. 3 (1) (2003)
- 8. Blanchard, O.: The medium run. Brookings Papers on Economic Activity, 2, 89-158 (1997)
- 9. Blum, B.: Endowments, Output, and the Bias of Directed Innovation.  Review of Economic Studies (forthcoming)
- 10. Buera, F. J. and Kaboski, J. P.: The rise of the service economy. NBER Working Paper (2009)
- 11. Chenery, H. B.: Structural change and Development Policy.  Oxford University Press (1979)
- 12. Ciccone, A. and Papaioannou, E.: Human capital, the structure of production, and growth.  Review of Economics and Statistics, 91(1), 66-82 (2009)
- 13. de Serres, A., Scarpetta, S. and de la Maisonneuve, C.: Sectoral shifts in Europe and the united states: how they affect aggregate labour shares and the properties of wage equations.  OECD Working Paper (2002)
- 14. Duarte, M. and Restuccia, D.: The role of the structural transformation in aggregate productivity.  Quarterly Journal of Economics, v 125(1), 129–173 (2010)
- 15. Echevarria, C.: Changes in sectoral composition associated with economic growth.  International Economics Review 38: 431-452
- 16. Fitzgerald, D. and Hallak, J.C.: Specialization, factor accumulation and development.  Journal of International Economics, 64, 277-302 (2004)
- 17. Harrigan, J.: Technology, factor supplies and international specialization: testing the neoclassical model.  American Economic Review, 87(4), 475-494 (1997)
- 18. Jones, C.I.: Growth, Capital shares, and a new perspective on production functions, Working Paper (2003)
- 19. Ju, J., Lin, J.Y. and Wang, Y.: Endowment structures, industrial dynamics, and economic growth.  World Bank Working Paper (2009)
- 20. Kongsamut, P., Rebelo, S. and Xie, D.: Beyond balanced growth. Review of Economic Studies 68, 869-882
- 21. Laitner, J.: Structural change and economic growth. Review of Economic Studies 67, 545-561 (2000)
- 22. Lewis, E.: Immigration, skill mix, and the choice of technique.  Federal Reserve Bank of Philadelphia Working Paper (2006)
- 23. Ngai, L.R. and Pissarides, C. A.: Structural change in a multi-sector model of growth. The American Economic Review 97, 429-443 (2007)
- 24. Oulton, N.: Must the growth rate decline? Baumol's unbalanced growth revisited. Oxford Economic Papers 53, 605-627 (2001)
- 25. Rogerson, R.: Structural transformation and the deterioration of European labor market outcomes.  NBER Working Paper No.12889 (2007)
- 26. Romalis, J.: Factor proportions and the structure of commodity trade.  American Economic Review, 94(1), 67-97 (2004)
- 27. Roodman, D.: A Note on the Theme of Too Many Instruments.  CGD Working Paper 125 (2007)
- 28. Schott, P.: One size fits all?  Theory, evidence and implications of cones of diversification.  American Economic Review 93(3), 686-708 (2003)
- 29. van Ark, B., O’Mahony, M. and Timmer, M.P.: The productivity gap between europe and the united states: trends and causes.  Journal of Economic Perspectives, Volume 22(1), 25-44 (2008)

*Source: _wp12165 - References*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12165.pdf_