## Government Debt and Growth: The Role of R&D ## Source details **Canonical URL:** [Government Debt and Growth: The Role of R&D](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025036-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025036-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025036-print-pdf.pdf.json) --- ### 1. Introduction — framing, mechanisms, and scope - Framing: - Government debt ratios have been increasing in the advanced economies (AEs) since the 2000s and reached uncharted levels in the wake of the Covid-19 pandemic. - The paper connects rising government debt and a downward trend in economic growth in AEs since the early 2000s, largely driven by the slowdown in productivity growth. - Central hypothesis: high levels of government debt hinder R&D activities in AEs, and through this R&D channel contribute to lower long-term economic growth. - Identification strategy and scope: - Industry-level panels for 2-digit manufacturing industries (ISIC Rev. 3, 15-36) from 36 AEs over 1980-2022 (UNIDO). - Exploits within-country variation in industries’ reliance on R&D activities, following a Rajan and Zingales (1998)-style approach. - Baseline industry R&D intensity (RDI) benchmarked using R&D expenditures of large, listed firms from the US. - Fixed effects: country-industry, country-year, industry-year to mitigate omitted-variable bias and reverse causality. - Proposed mechanisms: - Uncertainty channel: high government debt raises policy uncertainty; R&D is sensitive because it is long-horizon, lumpy, risky, and involves non-redeployable assets. - Confidence and demand channel: fiscal choices tied to high debt reduce consumer/investor confidence and demand for high-tech products. - Financial/crowding-out channel: high debt may crowd out private finance, disproportionately affecting R&D-intensive industries that rely more on external finance. ### 2. Data — composition, key measures, and sample - Industry-level data: - Source: UNIDO; coverage: 2-digit manufacturing industries (ISIC Rev. 3, 15-36). - Variables: industry value added (USD), industry output (USD), number of employees, labor productivity (value added / number of employees). - Growth rates = log difference * 100, winsorized at the 2.5th and 97.5th percentiles. - Industry value added share in total manufacturing (percent) used as control. - Industry RDI and alternative proxies: - Baseline RDI = share of R&D expenditures in overall capital expenditures; firm-level data from Compustat averaged over 1980–1999; industry proxy = median across firms in each 2-digit industry. - Key RDI summary statistics: Mean RDI = 0.97; Standard deviation of RDI = 1.34. - 75th percentile industry: machinery (ISIC 29); 25th percentile industry: paper products (ISIC 21). - Alternative RDI proxies: R&D/sales (1990-2000), OECD-STAN R&D/production (1995-2005), Choi et al. (2018b) RDI (1970-2000); correlations across proxies are high. - Industry-level measures (selected entries, exact values preserved): - Chemicals and chemical products, 24, RDI = 5.456, Physical capital intensity = 1.768, Asset tangibility = 0.206, Dependence on external finance = 3.868, Liquidity needs = 0.141 - Medical, precision, optical instruments, 33, RDI = 3.168, Physical capital intensity = 0.549, Asset tangibility = 0.155, Dependence on external finance = 1.241, Liquidity needs = 0.235 - Office, accounting, computing machinery, 30, RDI = 2.845, Physical capital intensity = 0.752, Asset tangibility = 0.133, Dependence on external finance = 1.056, Liquidity needs = 0.190 - Wearing apparel, fur, 18, RDI = 0.000, Physical capital intensity = 0.440, Asset tangibility = 0.116, Dependence on external finance = -0.298, Liquidity needs = 0.213 - Coke, refined petroleum, nuclear fuel, 23, RDI = 0.062, Physical capital intensity = 2.496, Asset tangibility = 0.617, Dependence on external finance = -0.236, Liquidity needs = 0.076 - Country-level data and controls: - Government debt: IMF’s Global Debt Database; defined as total stock of debt liabilities issued by the central or general government (percent of GDP). - Controls: real GDP per capita (log), trade (percent of GDP), real GDP (log), inflation (CPI), population aging (old-age dependency ratio), financial development index (ranges between 0 and 1), Chinn-Ito index (ranges between 0 and 1), Polity index (ranges between -10 and 10). - Patenting trends: patent applications filed through PCT or national patent offices (WIPO). - Sample: - Main analysis sample: 36 AEs over 1980-2022 (list provided in source). - Robustness: drop the US; tests with 63 EMs. - Data imputation: Chinn-Ito 2022 filled with 2021; polity 2019-2022 filled with 2018. ### 3. Methodology — empirical specifications and identification - Main empirical specification (Equation (1)): - Δlog(VA_{c,j,t}) = α VA share_{c,j,t−1} + β RDI_{j} × Government debt_{c,t−1} + θ_{c,t} + θ_{c,j} + θ_{j,t} + ε_{c,j,t} - Dependent variable: growth rate of industry value added (percent). - β measures how the impact of government debt on growth depends on industry RDI; negative β implies larger adverse impact on R&D-intensive industries. - Identification assumptions: - Persistent cross-industry differences in reliance on R&D carry over across AEs; US-based RDI is a valid proxy for industries in other AEs. - Industry-level growth is unlikely to materially alter national government debt, mitigating reverse causality. - Fixed effects and controls: - Country-year θ_{c,t}, country-industry θ_{c,j}, industry-year θ_{j,t}. - Lagged VA share_{c,j,t−1} controls for initial industry size. - Standard errors clustered at the country-industry level; robustness checks with alternative clustering schemes reported. - Extensions and tests: - Local projections (Jorda 2005) to estimate cumulative effects up to 10 years. - Equation (2): include I_{j} × Government debt to test alternative industry channels (physical capital intensity, asset tangibility, dependence on external finance, liquidity needs). - Equation (3): include RDI_{j} × X_{c,t−1} to control for differential country-level confounders (GDP, GDP per capita, trade, inflation, polity, financial development, financial openness, aging). - Equation (4): triple interactions RDI × Government debt × Z_{c,t−1} to test whether financial development/openness mitigates the effect. ### 4. Main empirical findings — coefficients, magnitudes, persistence, and heterogeneity - Core coefficient (AEs sample): - RDI_{j} × Government debt_{c,t−1} = -0.014*** (0.005) (Table 2, AEs). - Quantified magnitudes and examples: - Machinery industry (75th percentile of R&D intensity) growth rate becomes 0.5 percentage points lower compared to paper products (25th percentile) when located in a country with high government debt (75th percentile) rather than low debt (25th percentile). - Additional forgone growth in the top 5 industries (ISIC 30-33, 24) due to higher debt, compared to the bottom 5 industries (ISIC 15, 18, 20, 23, 27), is 2 percentage points per annum. - Average industry growth rate in the sample is 2.9 percent per annum. - Statistical highlights and robustness (exact reported values preserved): - Table 2 (AEs): RDI_{j} × Government debt_{c,t−1} = -0.014*** (0.005); VA share_{c,j,t−1} = -1.138*** (0.100); Observations = 23,539; R-squared = 0.470. - Table 3: RDI_{j} × Government debt_{c,t−1} ranges from -0.014*** (0.005) to -0.021** (0.009) when controlling for other industry characteristics; Observations = 23,539; R-squared between 0.470 and 0.471. - Table 4: RDI_{j} × Government debt_{c,t−1} ranges from -0.012** (0.005) to -0.016*** (0.005) when including country-level controls; VA share_{c,j,t−1} approximately -1.17 to -1.22 (all ***); Observations vary (e.g., 22,877; 21,255; 22,240); R-squared up to 0.499. - Table 5 (robustness panels): alternative RDI measures and outcome variables produce RDI×debt coefficients including -0.019*** (0.007), -0.287*** (0.095), -0.009** (0.004), -0.030** (0.015). - Table 5 Panel B quartile test (nonlinearity): RDI_{j} × 3rd quartile = -0.786** (0.394); RDI_{j} × 4th quartile = -1.396*** (0.533). - Table 6 (excluding country-year fixed effects): RDI_{j} × Government debt_{c,t−1} = -0.010** (0.005); Government debt_{c,t−1} = -0.002 (0.005) in column (1) and 0.006 (0.006) in column (2) (insignificant); Observations = 23,539; R-squared = 0.304. - Persistence and cumulative losses (local projections, 10-year horizon): - Accumulated loss in industry value added over 10 years is 3.7 percentage points larger in machinery compared to paper products due to high levels of debt (25th vs 75th percentiles). - Accumulated disproportionate effect comparing top and bottom 5 RDI industries over 10 years = 16.4 percentage points. - Benchmark: average decadal change in industry value added in the sample is 30 percent. - Heterogeneity: role of finance and openness: - Triple-interaction results (Table 7): - RDI_{j} × Government debt_{c,t−1} = -0.037*** (0.013) with financial development tests; = -0.033*** (0.008) with financial openness tests. - Triple interaction (RDI × Government debt × financial development) = 0.036** (0.017). - Triple interaction (RDI × Government debt × Chinn-Ito openness) = 0.024*** (0.008). - RDI_{j} × Z_{c,t−1} = -0.415 (1.465) and -0.746 (0.738) (insignificant). - VA share_{c,j,t−1} ≈ -1.175*** (0.107) and -1.222*** (0.119). - Observations = 22,877 and 22,240; R-squared = 0.473 and 0.475. - Quantitative threshold: impact of debt on R&D-intensive industries becomes statistically insignificant only for financial development index values above 0.83 (upper bound 1); about 87 percent of country-year observations fall below this threshold. For financial openness, the effect remains statistically significant even at the index upper bound (1). - Advanced vs Emerging Markets: - The RDI × debt interaction is negligible in EMs: coefficient -0.000 (0.005) (EMs column), implying the R&D channel of government debt is pronounced in AEs but not detected in EMs. - Robustness summary: - Results robust to controlling for multiple country-level macroeconomic and structural factors. - Findings are not driven by outliers or a few richer/poorer countries. - Dropping the US does not alter results. - Weighted regressions show results are not driven by industry size. - Results hold across subperiods: pre-Global Financial Crisis, post-Global Financial Crisis, and pre-Covid-19. - Interpretation limits: - Interaction coefficients quantify the differential impact on high-RDI versus low-RDI industries, not the overall average effect across all industries. - When country-year fixed effects are dropped and lagged government debt is included, government debt itself is statistically insignificant on average, while the negative interaction with RDI remains. ### 5. Conclusion — implications, trade-offs, and future research directions - Central conclusions: - Government debt leads to a decline in growth, particularly in industries that rely heavily on R&D activities, with effects persistent for upwards of 10 years. - The documented RDI channel helps explain part of the observed link between high government debt and low growth in AEs. - More developed or open financial systems tend to mitigate the negative impact of debt on innovative activities. - Policy implications and trade-offs: - Fiscal stabilization policies can stabilize economies short-term but can generate high debt levels that likely pose a drag on long-term growth by adversely impacting innovative activities. - Policies favoring financial deepening and openness can help alleviate adverse effects of debt on future growth. - Heterogeneity and further research: - The RDI channel is pronounced in AEs but not detected in EMs; this suggests different mechanisms may operate in EMs. - Future research should: (i) understand which channels are important in EMs for the debt–growth link; (ii) provide a theoretical explanation for the AE/EM distinction documented in this study. *Source: IMF Working Paper — “Government Debt and Growth: The Role of R&D” (content unit: wpiea2025036-print-pdf)* ### 1. Introduction ........................................................................................................ ### Government Debt and Growth: The Role of R&D ### Document structure and sections - 1. Introduction ..................................................................................................................................................... 3 - 2. Data .................................................................................................................................................................. 9 - 3. Metholodogy .................................................................................................................................................. 13 - 4. Stylized Facts ................................................................................................................................................ 16 - 5. Results ........................................................................................................................................................... 17 - 6. Conclusion ..................................................................................................................................................... 25 - References ......................................................................................................................................................... 27 - Appendix ............................................................................................................................................................ 33 ### Publication series and imprint - IMF WORKING PAPERS - INTERNATIONAL MONETARY FUND 3 *Source: wpiea2025036-print-pdf - 1. Introduction ........................................................................................................ (Canonical URL: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025036-print-pdf.pdf)* ### 1. Introduction ### 1. Introduction ### Framing and research question - Government debt ratios have been increasing in the advanced economies (AEs) since the 2000s and reached uncharted levels in the wake of the Covid-19 pandemic. - The paper connects two observed phenomena: rising government debt and a downward trend in economic growth in AEs since the early 2000s, largely driven by the slowdown in productivity growth. - Central hypothesis: high levels of government debt hinder R&D activities in AEs, and through this R&D channel contribute to lower long-term economic growth. ### Data, identification strategy, and scope - Data: industry-level panels for 2-digit manufacturing industries (ISIC Rev. 3, 15-36) from 36 AEs over the period 1980-2022 (UNIDO). - Identification: exploits within-country variation in industries’ reliance on R&D activities, following an approach similar to Rajan and Zingales (1998). - Baseline industry R&D intensity is calculated using R&D expenditures of large, listed firms from a benchmark country (the US). - Country-industry fixed effects, country-year fixed effects, and industry-year fixed effects are used to mitigate omitted-variable bias and reverse causality. - Rationale: industry-level growth is less likely to drive national government debt ratios, helping to address reverse causality concerns inherent in macro studies. ### Proposed mechanism(s) - Uncertainty channel: - High government debt is associated with higher uncertainty about economic policies and political environment. - R&D investment is particularly sensitive to uncertainty because it: (i) yields outcomes with longer delays and higher risk; (ii) requires lumpy, large-scale initial investments; (iii) involves highly specialized, irreversible, and low-redeployability assets. - Under uncertainty, firms defer or reallocate investment away from R&D toward more redeployable, reversible projects. - Confidence and demand channel: - Fiscal choices tied to high debt can depress consumer and investor confidence, reducing demand for high-tech products and increasing investor risk aversion toward long-term R&D projects. - Financial/crowding-out channel (examined separately): - High government debt can crowd out private finance, potentially reducing investment generally and especially for R&D-intensive industries that rely more on external finance and have low collateralizability. ### Main empirical findings - Higher government debt leads to a decline in growth, particularly in industries that heavily rely on R&D activities. - Persistence: the differential negative effect of government debt on R&D-intensive industries is long-lasting—persistent for upwards of 10 years. - Nonlinearity: the negative effect on high-tech industries’ growth is more pronounced at higher debt levels. - Cross-region specificity: the R&D channel of government debt is pronounced in AEs but not in emerging markets (EMs). - Magnitude examples (differential effects): - Growth rate of machinery industry (a high-tech industry, at the 75th percentile of R&D intensity) becomes 0.5 percentage points lower compared to paper products (a low-tech industry, at the 25th percentile of R&D intensity), when located in a country with high government debt (at the 75th percentile of the sample) rather than in a country with lower debt (at the 25th percentile of the sample). - The additional forgone growth in the top 5 industries with the highest R&D intensity due to higher debt, compared to the decline in growth in the bottom 5 industries, is 2 percentage points per annum. - Context: the average industry growth rate in the sample is 2.9 percent per annum, on average. - Financial system mitigation: - The negative impact of government debt on growth in R&D-intensive industries is mitigated in countries with more developed or open financial systems. ### Robustness and checks - Results are robust to controlling for several other country-level macroeconomic and structural factors. - Findings are not driven by outliers or a few richer/poorer countries. - Dropping the US from the sample does not alter results (addresses potential endogeneity from using US-based R&D intensity benchmarks). - Weighted regressions show results are not driven by industry size. - Results hold across subperiods: pre-Global Financial Crisis, post-Global Financial Crisis, and pre-Covid-19. ### Contribution relative to existing literature - Addresses causality concerns in the government debt–growth nexus by using granular industry-level data and a Rajan and Zingales (1998)-style identification leveraging cross-industry R&D intensity within countries. - Identifies a specific mechanism—government debt’s disproportionate adverse impact on R&D and innovation—that helps explain observed weak economic performance following episodes of high debt. - Complements related work (e.g., Croce et al. 2019 for US firm-level evidence; Fan et al. 2022 for local government debt and firm innovation in China) by providing cross-country, industry-level evidence in AEs. *IMF Working Paper — “Government Debt and Growth: The Role of R&D”, Section 1. Introduction* ### 2. Data ### 2. Data ### 2.1. Industry-level Data - Data source: UNIDO database (compiled mainly based on industrial surveys). - Coverage: 2-digit manufacturing industries (ISIC Rev. 3, 15-36). - Variables used: - Industry value added (USD). - Industry output (USD). - Number of employees. - Industry labor productivity = value added / number of employees. - Growth rates of industry value added, output and labor productivity = log difference * 100, winsorized at the 2.5th and 97.5th percentile levels. - Industry value added share in total manufacturing (control variable) = industry value added / total manufacturing value added (in percent). - Summary statistics of the variables used in the analysis are presented in the Appendix (Table A1). ### 2.2. Industry R&D Intensity and Other Characteristics - Objective: Explore the disproportionate effect of government debt on growth in industries that heavily engage in R&D. - Identification strategy: Rajan and Zingales (1998)-style cross-industry variation approach using an exogenous industry-level R&D intensity proxy benchmarked from US publicly listed firms. - Rationale for US-based proxy: - US financial markets are highly developed, alleviating financing-side distortions that could generate underinvestment in R&D. - Favorable legal framework and enforcement institutions (intellectual property protection, lower corruption) reduce appropriability problems. - Large technology consumption market and advanced technology frontier. - Large, listed US firms likely have easier access to capital markets, stronger intellectual property protection, and better capacity to exploit market opportunities. - Baseline RDI measure: - RDI = share of R&D expenditures in overall capital expenditures. - Data source: Compustat database for publicly listed US companies. - Firm-level ratio averaged over 1980–1999; industry proxy = median across firms in each 2-digit industry. - Key summary statistics for RDI: - Mean RDI = 0.97. - Standard deviation of RDI = 1.34. - 75th percentile industry: machinery (ISIC 29). - 25th percentile industry: paper products (ISIC 21). - Robustness: Alternative industry characteristics accounted for to test whether RDI proxies for other features: - Physical capital intensity = total real capital stock / value added (NBER-CES Manufacturing Industry Database). - Asset tangibility = net property, plant and equipment / total book-value of assets. - Dependence on external finance = share of capital expenditures not financed with cash flow from operations. - Liquidity needs = total inventories / annual sales. - These measures are computed at the firm level (averaged over 1980-1999) and then aggregated to industry medians (adopted from Igan et al. (2022); asset tangibility per Braun and Larrain (2005); dependence per Rajan and Zingales (1998); liquidity needs per Raddatz (2006); physical capital intensity per Nunn (2007) and Ciccone and Papaioannou (2009)). - Alternative RDI proxies used in robustness: - Ratio of R&D expenditures to sales for the median US firm in each 2-digit industry over 1990-2000 (from Compustat; adopted from Brown et al. (2017)). - OECD-STAN based ratio of R&D expenditures to production in each industry over 1995-2005 (adopted from Seitz and Watzinger (2017)). - RDI measure from Choi et al. (2018b), calculated over 1970-2000. - Correlations between these different proxies are high (Table A2). ### 2.3. Industry-level measures (Table 1 key values) - Notes: Measures adopted from Igan et al. (2022). R&D intensity = R&D expenditure / capital expenditure. Physical capital intensity = total capital stock / value added. Asset tangibility = net property, plant and equipment / total book-value of assets. Dependence on external finance = share of capital expenditures not financed with cash flow from operations. Liquidity needs = total inventories / annual sales. Industry-level proxies based on US data. - Industry-level measures by industry (Industry, ISIC, R&D intensity (RDI), Physical capital intensity, Asset tangibility, Dependence on external finance, Liquidity needs): - Food and beverages, 15, 0.180, 1.466, 0.363, -0.309, 0.102 - Tobacco products, 16, 0.402, 1.047, 0.219, -2.897, 0.252 - Textiles, 17, 0.349, 1.287, 0.329, -0.157, 0.178 - Wearing apparel, fur, 18, 0.000, 0.440, 0.116, -0.298, 0.213 - Leather, leather products and footwear, 19, 0.554, 0.630, 0.128, -0.735, 0.219 - Wood products (excl. furniture), 20, 0.164, 1.126, 0.293, -0.250, 0.114 - Paper and paper products, 21, 0.266, 1.674, 0.510, -0.385, 0.116 - Printing and publishing, 22, 0.602, 0.911, 0.267, -0.752, 0.069 - Coke, refined petroleum, nuclear fuel, 23, 0.062, 2.496, 0.617, -0.236, 0.076 - Chemicals and chemical products, 24, 5.456, 1.768, 0.206, 3.868, 0.141 - Rubber and plastics products, 25, 0.368, 1.253, 0.363, -0.273, 0.126 - Non-metallic mineral products, 26, 0.337, 1.600, 0.421, -0.350, 0.145 - Basic metals, 27, 0.168, 2.272, 0.397, -0.210, 0.168 - Fabricated metal products, 28, 0.351, 1.137, 0.278, -0.566, 0.178 - Machinery and equipment n.e.c., 29, 0.926, 0.995, 0.216, -0.431, 0.203 - Office, accounting, computing machinery, 30, 2.845, 0.752, 0.133, 1.056, 0.190 - Electrical machinery and apparatus, 31, 1.245, 0.938, 0.245, -0.025, 0.194 - Radio, TV, communication equipment, 32, 2.353, 1.108, 0.179, 0.317, 0.194 - Medical, precision, optical instruments, 33, 3.168, 0.549, 0.155, 1.241, 0.235 - Motor vehicles, trailers, semi-trailers, 34, 0.441, 1.365, 0.270, -0.140, 0.142 - Other transport equipment, 35, 0.614, 0.914, 0.250, -0.248, 0.213 - Furniture; manufacturing n.e.c., 36, 0.511, 0.764, 0.204, -0.337, 0.176 - Illustrative ordering and highlights from Table 1: - Highest RDI: Chemicals and chemical products (ISIC 24) with RDI = 5.456. - Next highest RDI: Medical, precision, optical instruments (ISIC 33) with RDI = 3.168; Office, accounting, computing machinery (ISIC 30) with RDI = 2.845; Radio, TV, communication equipment (ISIC 32) with RDI = 2.353; Electrical machinery and apparatus (ISIC 31) with RDI = 1.245. - Lowest RDI: Wearing apparel, fur (ISIC 18) with RDI = 0.000; Coke, refined petroleum, nuclear fuel (ISIC 23) with RDI = 0.062; Food and beverages (ISIC 15) with RDI = 0.180; Basic metals (ISIC 27) with RDI = 0.168; Wood products (ISIC 20) with RDI = 0.164. ### 2.3. Country-level Data - Government debt: - Data source: IMF’s Global Debt Database. - Definition: total stock of debt liabilities issued by the central or general government (as percent of GDP). - Macroeconomic, financial, institutional and structural controls used in robustness: - Real GDP per capita (constant US dollars, used in logarithm) — proxy for economic development (source: World Bank WDI). - Trade (as percent of GDP) — proxy for trade openness (source: WDI). - Real GDP (constant USD, used in logarithm) — proxy for economy size (source: WDI). - Inflation rate (CPI) (source: WDI). - Population aging proxy = ratio of older dependents (older than 65) to working-age population (ages 15-64) (source: WDI). - Financial development index (IMF): composite index combining depth, access and efficiency (ranges between 0 and 1; higher = greater financial development; index constructed via principal component analysis; see Svirydzenka 2016). - Chinn-Ito index (Chinn and Ito 2006): de jure financial openness measure, ranges between 0 and 1 (higher = more open). - Polity index (Polity V dataset, Center for Systemic Peace): ranges between -10 and 10 (higher = greater institutional quality). - Patenting trends (stylized facts): number of patent applications filed through PCT or national patent offices (World Intellectual Property Organization) from WDI. ### 2.4. Sample - Main analysis sample: AEs with available data over 1980-2022. - Main sample consists of 36 AEs: Australia, Austria, Belgium, Canada, Hong Kong SAR, Taiwan Province of China, Croatia, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, Korea, Latvia, Lithuania, Luxembourg, Malta, New Zealand, Norway, Portugal, Singapore, Slovakia, Slovenia, Spain, Sweden, Switzerland, UK, USA. - Robustness checks: - Drop the US from the sample to address potential endogeneity concerns (since RDI is calculated using US data). - Test results in a sample consisting of EMs (63 countries). - Data imputation note: - For Chinn-Ito index, 2022 values filled with 2021 values. - For polity index, period 2019-2022 filled with 2018 values. - These datasets stop at 2021 and 2018, respectively; associated results are not affected by this step. *Source: IMF Working Paper — chapter "2. Data" (wpiea2025036-print-pdf).* ### 3. Methodology ### 3. Methodology ### Objective and hypothesis - Objective: Investigate the differential effect of government debt levels on growth based on the variation in industries’ RDI. - Hypothesis: If high government debt levels hinder innovative activities, the growth impact of debt should be more pronounced in industries which heavily rely on R&D activities. ### Main empirical specification - Primary regression (Equation (1)): - Δlog(푉퐴_{c,j,t}) = 훼 푉퐴 푠ℎ푎푟푒_{c,j,t−1} + 훽 푅퐷퐼_{j} × Government debt_{c,t−1} + 휃_{c,t} + 휃_{c,j} + 휃_{j,t} + 휖_{c,j,t} - Definitions: - c, j, t: countries, 2-digit manufacturing industries, and years, respectively. - Dependent variable Δlog(푉퐴_{c,j,t}): growth rate of industry value added at year t (in percent). - 푅퐷퐼_{j}: proxy for industry’s R&D intensity (allows comparison across 2-digit manufacturing industries). - Government debt_{c,t−1}: central government debt as share of GDP at year t-1. - 푉퐴 푠ℎ푎푟푒_{c,j,t−1}: lagged value of the value added share of each industry in total manufacturing value added in its country (control). - Interpretation: - Coefficient 훽 on the interaction 푅퐷퐼_{j} × Government debt_{c,t−1} gauges how the impact of government debt on growth depends on industries’ RDI. - A negative 훽 implies government debt reduces growth more in R&D-intensive industries. ### Identification assumptions - Two implicit assumptions (analogous to Rajan and Zingales (1998)): - Some industries inherently and persistently rely more on R&D activities than others, driven by production methods and processes. - These differences carry over AEs, so an industry’s RDI as adopted from the US serves as a good proxy for its RDI in other AEs. ### Fixed effects and controls - Fixed effects included: - Country-year 휃_{c,t}: absorbs common time-variant economic, political, institutional developments, and country-specific shocks (including demand and supply). - Country-industry 휃_{c,j}: accounts for underlying persistent differences in industry growth within countries. - Industry-year 휃_{j,t}: absorbs industry-specific global developments, shocks, trends, or growth opportunities at an annual frequency. - Role of lagged size control: - Lagged value added share 푉퐴 푠ℎ푎푟푒_{c,j,t−1} controls for initial relative size of industries (larger industries may have less room to grow). - Standard errors: - Standard errors are clustered at the country-industry level. - Robustness: results stay virtually the same if standard errors are (i) robust to heteroskedasticity (not clustered), (ii) three-way clustered (country-industry, country-year, industry-year), (iii) two-way clustered with any combinations, or (iv) one-way clustered at the country-year or industry-year levels. ### Use of industry-level data and reverse causality - Advantage: industry-level identification mitigates reverse causality concerns because growth dynamics of a R&D-intensive (2-digit) manufacturing industry are unlikely to materially alter a country’s government debt pattern. ### Long-run dynamics: local projections - Method: - Use local projections (Jorda 2005) to test persistence of differential effect. - Left-hand side: cumulative change of value added (in percent) from year t-1 to t, t-1 to t+1, ..., t-1 to t+10 in separate regressions. - Report coefficient estimate of the interaction term from those regressions. - Horizon: cumulative changes over a 10-year period. ### Tests for alternative industry channels (Equation (2)) - Extended specification: - Δlog(푉퐴_{c,j,t}) = 훼 푉퐴 푠ℎ푎푟푒_{c,j,t−1} + 훽 푅퐷퐼_{j} × Government debt_{c,t−1} + 훾 I_{j} × Government debt_{c,t−1} + 휃_{c,t} + 휃_{c,j} + 휃_{j,t} + 휖_{c,j,t} - I_{j}: industry characteristics — physical capital intensity, asset tangibility, dependence on external finance, liquidity needs. - Purpose: - 훾 captures alternative channels through which government debt can affect industry growth. - If these alternative channels do not undermine the R&D channel, 훽 should be similar to the baseline estimate. ### Tests for country-level confounders (Equation (3)) - Extended specification: - Δlog(푉퐴_{c,j,t}) = 훼 푉퐴 푠ℎ푎푟푒_{c,j,t−1} + 훽 푅퐷퐼_{j} × Government debt_{c,t−1} + 훾 푅퐷퐼_{j} × X_{c,t−1} + 휃_{c,t} + 휃_{c,j} + 휃_{j,t} + 휖_{c,j,t} - X_{c,t−1}: various country-level variables (GDP, GDP per capita, trade, inflation, proxies for institutional quality, financial development, financial openness and population aging). - Purpose: - 훾 captures differential impacts of country-level variables on R&D-intensive industries. - If government debt is not merely a proxy for these factors, 훽 should remain similar to the baseline. ### Role of finance: triple interaction (Equation (4)) - Extended specification: - Δlog(푉퐴_{c,j,t}) = 훼 푉퐴 푠ℎ푎푟푒_{c,j,t−1} + 훽 푅퐷퐼_{j} × Government debt_{c,t−1} + 훾 푅퐷퐼_{j} × Z_{c,t−1} + 휆 푅퐷퐼_{j} × Government debt_{c,t−1} × Z_{c,t−1} + 휃_{c,t} + 휃_{c,j} + 휃_{j,t} + 휖_{c,j,t} - Z_{c,t−1}: proxies for financial development or openness. - Interpretation: - If 훽 is negative, a positive 휆 implies financial development or openness mitigates the negative impact of government debt on R&D-intensive industries. *Source: IMF Working Paper — 3. Methodology* ### 26.4 percent of GDP, while this is 74.7 percent at the 75 ### wpiea2025036-print-pdf - 26.4 percent of GDP, while this is 74.7 percent at the 75th percentile of the sample. ### Main empirical findings - The estimated coefficient of the interaction term between industry R&D intensity (RDI) and government debt is -0.014 (standard error (0.005), *** p<0.01) in the advanced-economies (AEs) sample, implying a disproportionate negative effect of higher government debt on growth in R&D-intensive industries (Table 2, AEs column). - Quantified examples: - The growth rate of machinery industry (high-tech) becomes 0.5 percentage points lower compared to paper products (low-tech) when located in a country with high government debt (at the 75th percentile) rather than low debt (at the 25th percentile).12 - The average growth loss in the top 5 industries with the largest RDI (ISIC 30-33, 24) due to higher debt, compared to the bottom 5 industries with the lowest RDI (ISIC 15, 18, 20, 23, 27), is about 2 percentage points on average (calculated using the difference in the average values of the RDI measure across those industries at the top and the bottom of the distribution). - Context: average industry growth in the sample is about 2.9 percent per annum (Table A.1 in the Appendix). - The negative and statistically significant coefficient on industry’s initial value added share (e.g., -1.138*** (0.100) in AEs) suggests smaller industries tend to exhibit higher growth, indicating convergence across industries and countries over time. - The RDI×debt interaction is negligible in EMs (coefficient -0.000 (0.005) in EMs column), implying the R&D channel of government debt is not pronounced in EMs; thus the analysis focuses on AEs for most tests. ### Evidence from core tables (selected statistics preserved exactly) - Table 2 (AEs): - RDIj × Government debtc,t−1 = -0.014*** (0.005) - VA sharec,j,t−1 = -1.138*** (0.100) - Observations = 23,539 - R-squared = 0.470 - Table 3 (robustness to other industry characteristics): - RDIj × Government debtc,t−1 ranges: -0.014*** (0.005) to -0.021** (0.009) across columns testing physical capital intensity, asset tangibility, dependence on external finance, liquidity needs. - VA sharec,j,t−1 = -1.138*** (0.100) across those specifications. - Observations = 23,539; R-squared between 0.470 and 0.471. - Table 4 (country-level controls): - RDIj × Government debtc,t−1 ranges from -0.012** (0.005) to -0.016*** (0.005) across columns including GDP per capita, GDP, polity, financial development, financial openness, trade, inflation, aging. - VA sharec,j,t−1 approximately -1.17 to -1.22 (all ***). - Observations vary by column (e.g., 22,877; 21,255; 22,240); R-squared up to 0.499 in some specifications. - Table 5 (additional tests, Panel A–C highlights): - Panel A: alternative RDI measures and outcome variables produce RDI×debt coefficients including -0.019*** (0.007), -0.287*** (0.095), -0.009** (0.004), -0.030** (0.015). - Panel B: central government debt winsorized yields -0.019*** (0.006); general government debt yields -0.015*** (0.005); weighted regressions and country exclusions show similar negative interaction terms. - Panel B quartile test (nonlinearity): RDIj × 3rd quartile = -0.786** (0.394); RDIj × 4th quartile = -1.396*** (0.533). - Panel C (time-span tests): RDI×debt over 3-year windows and subperiods includes estimates such as -0.042*** (0.013) and -0.050** (0.020); VA share coefficient ranges from -1.198*** (0.106) to -3.209*** (0.292) depending on horizon and overlap. - Table 6 (excluding country-year fixed effects): - RDIj × Government debtc,t−1 = -0.010** (0.005) - Government debtc,t−1 = -0.002 (0.005) in column (1) and 0.006 (0.006) in column (2) (statistically insignificant). - VA sharec,j,t−1 = -0.989*** (0.095) and -0.998*** (0.095). - Observations = 23,539; R-squared = 0.304. - Table 7 (role of finance): - RDIj × Government debtc,t−1 = -0.037*** (0.013) with financial development triple interaction tests; -0.033*** (0.008) with financial openness triple interaction tests. - Triple interaction (RDI × Government debt × Zc,t−1) = 0.036** (0.017) for financial development proxy; = 0.024*** (0.008) for Chinn-Ito openness proxy. - RDIj × Zc,t−1 = -0.415 (1.465) and -0.746 (0.738) (statistically insignificant). - VA sharec,j,t−1 ≈ -1.175*** (0.107) and -1.222*** (0.119). - Observations = 22,877 and 22,240; R-squared = 0.473 and 0.475. ### Robustness and alternative explanations tested - Industry-level alternatives considered and ruled out as sole drivers: physical capital intensity, asset tangibility, dependence on external finance, liquidity needs (Table 3). - Country-level alternatives considered and ruled out as sole drivers: real GDP per capita, real GDP, polity index (institutional quality), financial development index, Chinn-Ito capital account openness, trade as percent of GDP, inflation rate, old-age dependency ratio (Table 4). - Robustness checks (Table 5): - Winsorization of RDI at 5th and 95th percentiles; alternative RDI measures from Brown et al. (2017), Seitz and Watzinger (2017), Choi et al. (2018b). - Alternative dependent variables: output growth and labor productivity growth. - Winsorization/non-winsorization of value added growth. - Winsorization of central government debt at 2.5th and 97.5th percentiles; use of general government debt. - Drop US; weighted regressions by log(real GDP per capita) and its inverse; weights by industry value added share and 100 minus it. - Quartile analysis shows nonlinearity: disproportionate negative impact becomes particularly pronounced at higher debt quartiles (3rd and 4th quartiles). - Time-span analyses: overlapping 3-year windows; average debt over last 3 years; pre- and post-GFC periods (1980-2007; 2011-2022); 1980-2019 excluding Covid-19; 1990-2022. ### Heterogeneity: role of financial development and openness - Main interaction (RDI × debt) remains negative and statistically significant when including financial development and openness interactions, indicating financial factors do not eliminate the R&D channel of government debt. - Triple interactions (RDI × debt × financial development) and (RDI × debt × financial openness) are positive and statistically significant (0.036** (0.017) and 0.024*** (0.008), respectively), implying greater financial development or openness reduces the sensitivity of R&D-intensive industries to government debt. - Quantitative threshold: combining interaction terms implies the impact of debt on R&D-intensive industries becomes statistically insignificant only for values of the index on financial development above 0.83 (upper bound 1); about 87 percent of country-year observations have an index below this threshold. For financial openness, the effect of government debt stays statistically significant (and negative) even at the upper bound of the index (1). ### Durability and cumulative effects - Local projections over a 10-year horizon indicate the impact is long-lasting (Figure 2 description). - Accumulated estimates: - Accumulated loss in industry value added over 10 years is 3.7 percentage points larger in machinery compared to paper products due to high levels of debt (comparing the 25th and 75th percentiles). - Accumulated disproportionate effect comparing top and bottom 5 RDI industries over 10 years = 16.4 percentage points. - Benchmark: average decadal change in industry value added in the sample is 30 percent. - Interpretation: results are consistent with R&D projects being cancelled or shelved for prolonged periods following high government debt, with no evidence for recovery in the growth loss in high-tech industries over the 10-year horizon. ### Interpretation and limits of inference from presented results - The estimated interaction coefficients quantify the differential impact of higher debt levels on growth in industries with relatively higher RDI, not the overall effect on all industries. - When country-year fixed effects are dropped and lagged government debt is included, government debt itself is statistically insignificant on average, while the negative interaction with RDI remains, suggesting the overall direction of the impact of government debt on growth is likely negative and mainly works through the R&D channel. - The RDI channel is pronounced in AEs but not detected in EMs; a possible explanation is EMs being subject to a wider set of shocks (e.g., violent conflicts) that have larger impacts on R&D activities than government debt. _Italic: Source — IMF Working Paper: Government Debt and Growth: The Role of R&D (content unit: wpiea2025036-print-pdf)_ ### 6. Conclusion ### 6. Conclusion ### Central question and context - Whether government debt poses a drag on future economic growth is a central question with important policy implications. - The increasing trend in debt ratios across AEs since the early 2000s has been coupled with a slowdown in growth, which was largely driven by the slowdown in productivity growth. - The environment characterized by high government debt ratios and low growth has become particularly relevant for AEs recently, due to the soaring of already-high debt levels in the wake of the Covid-19 pandemic. - It is important to understand the channels shaping the interplay between rising debt levels and slowing growth. ### Main findings and mechanism - The paper builds causality in the government debt–growth nexus by showing evidence on a particular mechanism: government debt adversely impacts innovative activities. - Based on granular data from manufacturing industries in AEs over the period 1980-2022: - Government debt leads to a decline in growth, particularly in industries which rely heavily on R&D activities. - The differential effect of debt on R&D-intensive industries is persistent for upwards of 10 years. - The pattern is not driven solely by specific events during the sample period, such as the Global Financial Crisis or the Covid-19 pandemic. - More developed or open financial systems tend to mitigate the negative impact of debt on innovative activities. ### Policy implications and trade-offs - The findings offer a specific explanation for the observed link between high government debt and low growth in AEs, given the role of technological progress in productivity growth and long-term economic performance. - Policy trade-off highlighted: - Fiscal stabilization policies in response to economic shocks can help stabilize economies in the short-term. - However, such policies can lead to high levels of government debt that likely pose a drag on long-term growth by adversely impacting innovative activities. - There is suggestive evidence that policies favoring financial deepening and openness can help alleviate the adverse effects of debt on future growth. ### Heterogeneity and directions for future research - The documented pattern does not hold in EMs, unlike in AEs: - This does not imply government debt does not affect growth in EMs. - It suggests that the R&D channel through which debt hinders growth in AEs does not seem to be at play in other economies. - Future research priorities: - Understand which channels are important in EMs for the link between government debt and growth. - Provide a theoretical explanation for the distinction between AEs and other economies found by this study. *IMF Working Paper: Government Debt and Growth: The Role of R&D — 6. Conclusion.* --- _Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025036-print-pdf.pdf_