## imf-research-perspectives-2025

## Source details

**Canonical URL:** [imf-research-perspectives-2025](https://www.imf.org/-/media/files/research/research-perspective/2025/imf-research-perspectives-2025.pdf)

## Other formats

- [Markdown version](/-/media/files/research/research-perspective/2025/imf-research-perspectives-2025.pdf.md)
- [Structured JSON version](/-/media/files/research/research-perspective/2025/imf-research-perspectives-2025.pdf.json)

---

### Framework and measurement of occupational barriers
- Builds on a choice-theoretic model in the spirit of Hsieh and others (2019); individuals choose occupations and education to maximize lifetime utility, with group-specific and idiosyncratic talents and preferences.
- Defines a composite measure of female occupational barriers as a selection-adjusted gender wage gap: the geometric mean ratio of women’s to men’s earnings, corrected for relative participation odds, aggregated across occupations by earnings share.
- The measure captures frictions beyond conventional wage gaps (discriminatory practices, social norms, costs to human capital accumulation) and allows consistent cross-country comparison under the assumption that innate comparative advantages of women relative to men are the same everywhere.
- Empirical basis: harmonized income microdata from the Luxembourg Income Study covering income surveys from 52 economies; figures and counterfactuals reported for a 43-economy subsample using 2015 microdata.

### Cross-country patterns and correlation with gender indices
- Stark disparities between high-income economies and EMDEs: in Egypt and India, data-implied barriers facing women are almost three times as high as in the US in 2015—levels comparable to the US in the 1960s.
- Some former centrally planned EMDEs exhibit among the lowest estimated barriers, reflecting historically mandated workforce participation regardless of gender.
- Composite barrier measure correlations:
  - Regressing the occupational barrier on outcome- and institution-based indices yields R-squared values ranging from 0.30 to 0.47.
  - Indices referenced include the Gender Development Index (UNDP 2015), Global Gender Gap Index (WEF 2015), Gender Inequality Index (UNDP 2015), and Women, Business and the Law Index (World Bank 2016).

### Simulated gains from lowering barriers to US 2015 levels
- Counterfactual: lowering female occupational barriers to 2015 US levels.
- Almost all economies in the sample experience income gains; gains are closely correlated with the size of barriers relative to the US.
- Representative outcomes:
  - Egypt: real per capita GDP increases by 26 percent.
  - India: real per capita GDP increases by 24 percent.
  - Some advanced economies (Czech Republic, Germany, Iceland, Lithuania) could expect income increases of about 5 percent.
- Decomposition of gains:
  - Two channels: participation (more women move from home sector to market work) and productivity (improved talent allocation across occupations).
  - For the median sample economy, about two-thirds of income gains are derived from improved productivity, and about one-third from increased participation.
- Distributional note:
  - For many EMDEs, income gains are double-digit; gains for AEs are relatively modest but potentially meaningful amid demographic headwinds (declining fertility and aging populations).

### Role of sectoral composition and robustness to physical-strength occupations
- Composite barriers aggregate across occupations using earnings weights; high composite barriers can reflect high within-occupation barriers and/or an economic structure skewed toward female-unfriendly occupations.
- Shift-share decomposition: large EMDEs with the highest barriers tend to have both higher within-occupation barriers and unfavorable sectoral composition versus the US.
- Robustness check for “brawn-intensive” occupations:
  - Alternative assumption treats skilled agriculture, crafts, and machine operation as potentially reflecting male comparative advantage rather than barriers.
  - Under this assumption, variation in barriers narrows somewhat but relative ranking of economies remains virtually unchanged.
  - Eliminating all barriers under this alternative assumption still yields sizable per capita GDP gains (~9.3 percent).

### Welfare effects
- Steady-state per capita welfare effects are broadly in line with GDP effects but tend to be slightly smaller.
- Reason: welfare accounts for the trade-off between higher market consumption and lower consumption of home services as more women shift into market work; welfare quantifies the net social gain after this trade-off.
- Figures present log change in steady-state real GDP per capita and log change in steady-state welfare from moving to US barriers, by gender (men, women, overall).

### Policy implications and conclusions (gender and occupational barriers)
- Reducing female occupational barriers can be a powerful source of growth and international income convergence, primarily via improved talent allocation rather than just higher participation.
- For economies facing demographic headwinds, lowering barriers provides a lever to boost medium-term growth and sustain labor force participation.
- Context-specific policy responses could include:
  - Education reforms,
  - Legal reforms,
  - Public support for families and childcare.
- For EMDEs in particular, addressing occupational segregation is both an equity objective and an opportunity to tap a significant engine of growth that can reinforce development trajectories.

### The Productivity Paradox — key empirical findings (Lehr)
- US R&D investment maintained a steady 3 percent of GDP over recent decades.
- TFP growth fell from 0.5 percent annually during 1975–95 to 0.3 percent during 2005–18.
- Distribution of firm-level R&D returns (log):
  - P25 = 1.12
  - P50 = 1.7
  - P75 = 2.27
  - SD = 0.93
- Measured macroeconomic impacts of R&D misallocation:
  - Poor allocation reduced US productivity growth by an estimated 18 percent during 1975–2014.
  - Comparing 2000–14 to 1975–90, worsening R&D allocation accounts for an 11 percent reduction in growth.
- Methodology highlights:
  - R&D return measured as value created per R&D dollar using estimated valuations of a firm’s new patents.
  - Adjustments account for divergences between private and public impacts of innovation using patent quality measures, including patent citations.

### Monopsony and inventor labor markets — empirical magnitudes (Lehr 2025b)
- On average, R&D workers receive 70 percent of the value they bring to the firm, compared with 100 percent under competitive conditions.
- Size-dependent monopsony:
  - R&D workers at firms with above-median R&D employment receive 57 percent of the value they contribute.
  - R&D workers at firms with below-median R&D employment receive 100 percent (compensated at competitive levels).
- Aggregate impacts from a quantitative model:
  - Labor market power in innovation reduces US economic growth by 0.20 percentage point annually (a 13 percent reduction).
  - This reduction translates to 11 percent lower welfare.
  - Monopsony reduces aggregate inventor employment by about 2 percent.
  - Misallocation accounts for 90 percent of the total growth impact; even without employment loss, allocation distortions would cut growth by 0.18 percentage point.

### Policy recommendations on innovation allocation
- Allocation-conscious R&D policy:
  - Target firms with particularly high expected returns on R&D rather than uniform subsidies; use past returns as an input due to their strong predictive power.
- Strengthening inventor mobility:
  - Enforce bans on noncompete agreements, prevent wage-fixing conspiracies, and ensure portability of benefits to enhance inventors' outside options.
- Antitrust enforcement in innovation markets:
  - Incorporate impacts on inventor labor markets into merger review, not just product market competition.
- Note: Some restriction on worker mobility can increase firms’ incentives to invest in worker human capital (Acemoglu and Pischke 1999).

### South Korea big push — empirical findings (Shim and Choi)
- Policy context:
  - Between 1973 and 1979, the Korean government temporarily supported firms in heavy manufacturing to adopt modern technologies.
- Aggregate adoption and industrialization effects:
  - Heavy manufacturing share of GDP more than doubled from 6 percent to 13 percent during the program window.
  - Number of new technology contracts in heavy manufacturing sectors quadrupled during the program window.
  - The share of heavy manufacturing in GDP, employment, and exports continued to rise after the policy ended.
- Firm-level causal effects (winners vs losers design):
  - Firms that successfully adopted modern technologies experienced:
    - After seven years, sales increased by 60%.
    - Total factor productivity (TFP) were almost double those of similar firms whose contracts fell through.
- Local spillovers and complementarities:
  - A 1 percentage point increase in the share of local adopters raised nonadopting firms’ sales by 2.7 percent and their revenue TFP by 1.6 percent.
  - A 1 percentage point increase in the share of local adopters raised a firm’s adoption probability by 0.85 percentage point (about 14 percent of the average annual adoption probability in the final year of the program).
  - Complementarities were stronger where market size was larger.
- Modeling approach:
  - A dynamic model in which spillovers occur with a one-period lag: a larger local base of adopters in period t–1 increases productivity and lowers the cost of adoption in period t.

### Heavy-manufacturing model setup and counterfactual calibration
- Adoption costs are fixed; adoption modeled as a one‑off investment with returns larger when spillovers and firm scale are larger.
- Analytical result: if spillovers and private returns lie in a middle range, the model admits multiple steady states: a low adoption “preindustrialized” state and a high adoption “industrialized” state.
- Counterfactual simulations calibrated to Korea:
  - Without the 1970s big push subsidies, South Korea would have converged to a less industrialized steady state.
  - Heavy-manufacturing share of GDP would have been 27 percent lower in the absence of the policy (compared with the baseline with the policy).
  - Export intensity would have been 39 percent lower in the absence of the policy (compared with the baseline with the policy).
  - The big push policy raised aggregate welfare by 14.6 percent.
  - Welfare gains were uneven across regions, with welfare changes ranging from -1.2 percent to 83.8 percent.
- Scenario analysis on market access and complementary policies:
  - Three scenarios where scale was temporarily reduced: flat foreign demand, higher import tariffs, no major highway improvement.
  - Each scenario weakens the impact of the subsidies; holding down foreign demand is especially powerful.
  - Combined scenario (flat foreign demand, higher import tariffs, and no highway improvement): the big push does not occur even with the same subsidies.

### Online platforms and firm-to-firm learning (China)
- Data and linkage:
  - Over 20 million online posts on a Chinese social media platform for trade professionals were analyzed and linked to official customs data.
- Export gains from platform adoption (estimates over time):
  - First year after access: exports increase by 3 percent on average.
  - Second year after access: exports increase by 8 percent on average.
  - Third year after access: exports increase by 11 percent on average.
- Network effects and external economies of scale:
  - A 10 percent increase in the local pool of information is linked to an extra 0.3 percent in export growth for a new member.
  - Export effect grows over time, implying ongoing learning and better use of the network.
- Information flows and geography:
  - A 10 percent increase in physical distance between two cities reduces the volume of online messages between firms in those locations by about 0.7 percent.
  - Upstream suppliers share about 0.6 percent more knowledge with a buyer when that buyer's purchases from the supplier increase by 10 percent.
  - Empirical scatterplots: association between knowledge share and downstreamness slope = 0.058 (0.021), R-squared = 0.018; near-zero association with upstreamness slope = -0.005 (0.018), R-squared = 0.000.
- Digital divide findings:
  - Cities with higher GDP per capita and better internet infrastructure have higher rates of platform adoption.
  - Less developed regions adopt the platform less, potentially worsening regional inequality.
- Model of endogenous learning and sharing:
  - A general equilibrium trade model with multiple regions and industries, supply-chain linkages, heterogeneous firms choosing to learn and to share information.
  - Firms can pay a "learning fee" to access a shared pool of knowledge; the export boost from joining the platform determined the "value of an idea" in calibration.

### Policy implications for digital platforms and trade
- Governments should complement trade‑cost reductions with support for knowledge networks that empower firms:
  - Invest in platforms that make peer-to-peer knowledge sharing easier, more trusted, and more discoverable.
  - Design platforms that reward knowledge sharing (verified answers, case studies, tutorials, recognition mechanisms).
  - Address geographic gaps via strategic partnerships, "twinning" programs, and hybrid online-offline forums to bridge the digital divide.
- Targeted interventions can leverage firms' natural incentives to share knowledge—especially along supply chains—rather than uniform subsidies for all firms.

### Conclusion: Establish Information Networks and conference takeaways
- Social media platforms are becoming essential digital economic infrastructure for businesses engaged in international trade; by allowing companies to learn directly from peers, these networks reduce information barriers and raise exports.
- Evidence on the digital adoption gap:
  - Platform firm share in total exporters (%) 2017 vs. Log GDP per capita 2017: Slope = 4.851 (0.677); R-squared = 0.151.
  - Platform firm share in total exporters (%) 2017 vs. Internet broadband connections/population (%) 2017: Slope = 0.132 (0.037); R-squared = 0.049.
- Key lessons:
  - Spread of business knowledge is driven by economic incentives within supply chains.
  - Geography and local business relationships continue to matter.
  - Unequal access to digital tools can widen regional divides.
- Conference highlights (IMF Research Department events):
  - The IMF Research Department held the 26th Jacques Polak Annual Research Conference, November 6–7, 2025, on “The Evolving Landscape of Global Trade and Financial Integration.”
  - The joint IMF-FCDO Conference marked 13 years of collaboration on the Macroeconomic Research in Low-Income Countries (MRLIC) program; discussions emphasized geoeconomic fragmentation, demographic transitions, rapid technological change, and extreme weather events.

*Source: Li, Zhang, and Zymek, forthcoming; Luxembourg Income Study database; IMF Research Perspectives.*

### 2010. These gains came not only from higher labor

### imf-research-perspectives-2025 - 2010. These gains came not only from higher labor

### Framework and measurement of occupational barriers
- Builds on a choice-theoretic model in the spirit of Hsieh and others (2019); individuals choose occupations and education to maximize lifetime utility, with group-specific and idiosyncratic talents and preferences.
- Defines a composite measure of female occupational barriers as a selection-adjusted gender wage gap: the geometric mean ratio of women’s to men’s earnings, corrected for relative participation odds, aggregated across occupations by earnings share.
- The measure captures frictions beyond conventional wage gaps (discriminatory practices, social norms, costs to human capital accumulation) and allows consistent cross-country comparison under the assumption that innate comparative advantages of women relative to men are the same everywhere.
- Empirical basis: harmonized income microdata from the Luxembourg Income Study covering income surveys from 52 economies; figures and counterfactuals reported for a 43-economy subsample using 2015 microdata.

### Cross-country patterns and correlation with gender indices
- Results show stark disparities between high-income economies and EMDEs; in Egypt and India, data-implied barriers facing women are almost three times as high as in the US in 2015—levels comparable to the US in the 1960s.
- Some former centrally planned EMDEs exhibit among the lowest estimated barriers, reflecting historically mandated workforce participation regardless of gender.
- The composite barrier measure is strongly correlated with established gender inequality indices:
  - Regressing the occupational barrier on outcome- and institution-based indices yields R-squared values ranging from 0.30 to 0.47.
  - Indices referenced include the Gender Development Index (UNDP 2015), Global Gender Gap Index (WEF 2015), Gender Inequality Index (UNDP 2015), and Women, Business and the Law Index (World Bank 2016).

### Simulated gains from lowering barriers to US 2015 levels
- Counterfactual: lowering female occupational barriers to 2015 US levels.
- Almost all economies in the sample experience income gains; gains are closely correlated with the size of barriers relative to the US.
- Representative outcomes:
  - Egypt: real per capita GDP increases by 26 percent.
  - India: real per capita GDP increases by 24 percent.
  - Some advanced economies (Czech Republic, Germany, Iceland, Lithuania) could expect income increases of about 5 percent.
- Decomposition of gains:
  - Two channels: participation (more women move from home sector to market work) and productivity (improved talent allocation across occupations).
  - For the median sample economy, about two-thirds of income gains are derived from improved productivity, and about one-third from increased participation.
- Distributional note: for many EMDEs, income gains are double-digit; gains for AEs are relatively modest but potentially meaningful amid demographic headwinds (declining fertility and aging populations).

### Role of sectoral composition and robustness to physical-strength occupations
- Composite barriers aggregate across occupations using earnings weights; high composite barriers can reflect:
  - High within-occupation barriers, and/or
  - Economic structure skewed toward female-unfriendly occupations.
- Shift-share decomposition: large EMDEs with the highest barriers tend to have both higher within-occupation barriers and unfavorable sectoral composition versus the US.
- Robustness check for “brawn-intensive” occupations:
  - Alternative assumption treats skilled agriculture, crafts, and machine operation as potentially reflecting male comparative advantage rather than barriers.
  - Under this assumption, variation in barriers narrows somewhat but relative ranking of economies remains virtually unchanged.
  - Eliminating all barriers under this alternative assumption still yields sizable per capita GDP gains (~9.3 percent).

### Welfare effects
- Steady-state per capita welfare effects are broadly in line with GDP effects but tend to be slightly smaller.
- Reason: welfare accounts for the trade-off between higher market consumption and lower consumption of home services as more women shift into market work; welfare quantifies the net social gain after this trade-off.
- Figures present log change in steady-state real GDP per capita and log change in steady-state welfare from moving to US barriers, by gender (men, women, overall).

### Policy implications and conclusions
- Reducing female occupational barriers can be a powerful source of growth and international income convergence, primarily via improved talent allocation rather than just higher participation.
- For economies facing demographic headwinds, lowering barriers provides a lever to boost medium-term growth and sustain labor force participation.
- Policy responses should be context-specific and could include:
  - Education reforms,
  - Legal reforms,
  - Public support for families and childcare.
- For EMDEs in particular, addressing occupational segregation is both an equity objective and an opportunity to tap a significant engine of growth that can reinforce development trajectories.

*Source: Li, Zhang, and Zymek, forthcoming; Luxembourg Income Study database; IMF Research Perspectives.*

### REFERENCES

### REFERENCES

### Cited works
- Hsieh, C. T., E. Hurst, C. I. Jones, and P. Klenow. 2019. “The Allocation of Talent and US Economic Growth.” Econometrica 87 (5): 1439-74.
- Li, N., L. Zhang, and R. Zymek. Forthcoming, “The Allocation of Talent and Income Differences across Countries,” Working paper.
- United Nations Development Programme (UNDP). 2015. Human Development Report 2015: Work for Human Development. Technical Report, New York, NY.
- World Bank. 2016. Women, Business and the Law 2016: Getting to Equal. Technical Report, Washington, DC.
- World Economic Forum (WEF). 2015. The Global Gender Gap Report 2015. Technical Insight Report, Geneva.
- Acemoglu, Daron, and Jorn-Steffen Pischke. 1999. “The Structure of Wages and Investment in General Training.” Journal of Political Economy, 107 (3): 539–72.
- Aghion, Philippe, Antonin Bergeaud, Timo Boppart, Peter J. Klenow, and Huiyu Li. 2025b. “Good Rents versus Bad Rents: R&D Misallocation and Growth.” NBER Working Paper 34190.
- de Ridder, Maarten. 2024. “Market Power and Innovation in the Intangible Economy.” American Economic Review 1 (114): 199–251.
- Kogan, Leonid, Dimitris Papanikolaou, Amit Seru, and Noah Stoffman. 2017. “Technological Innovation, Resource Allocation, and Growth.” Quarterly Journal of Economics 132 (2): 665–712.
- Lehr, Nils H. 2025a. “Did R&D Misallocation Contribute to Slower Growth?” IMF Working Paper 25/183, International Monetary Fund, Washington, DC.
- Lehr, Nils H. 2025b. “Does Monopsony Matter for Innovation?” Working Paper.

### The Productivity Paradox — key empirical findings (Lehr)
- US R&D investment maintained a steady 3 percent of GDP over recent decades.
- TFP growth fell from 0.5 percent annually during 1975–95 to 0.3 percent during 2005–18.
- Distribution of firm-level R&D returns (log):
  - P25 = 1.12
  - P50 = 1.7
  - P75 = 2.27
  - SD = 0.93
- Measured macroeconomic impacts of R&D misallocation:
  - Poor allocation reduced US productivity growth by an estimated 18 percent during 1975–2014.
  - Comparing 2000–14 to 1975–90, worsening R&D allocation accounts for an 11 percent reduction in growth.
- Methodology highlights:
  - R&D return measured as value created per R&D dollar using estimated valuations of a firm’s new patents.
  - Adjustments account for divergences between private and public impacts of innovation using patent quality measures, including patent citations.

### Monopsony and inventor labor markets — empirical magnitudes (Lehr 2025b)
- On average, R&D workers receive 70 percent of the value they bring to the firm, compared with 100 percent under competitive conditions.
- Size-dependent monopsony:
  - R&D workers at firms with above-median R&D employment receive 57 percent of the value they contribute.
  - R&D workers at firms with below-median R&D employment receive 100 percent (compensated at competitive levels).
- Aggregate impacts from a quantitative model:
  - Labor market power in innovation reduces US economic growth by 0.20 percentage point annually (a 13 percent reduction).
  - This reduction translates to 11 percent lower welfare.
  - Monopsony reduces aggregate inventor employment by about 2 percent.
  - Misallocation accounts for 90 percent of the total growth impact; even without employment loss, allocation distortions would cut growth by 0.18 percentage point.

### Policy recommendations on innovation allocation
- Allocation-conscious R&D policy:
  - Target firms with particularly high expected returns on R&D rather than uniform subsidies; use past returns as an input due to their strong predictive power.
- Strengthening inventor mobility:
  - Enforce bans on noncompete agreements, prevent wage-fixing conspiracies, and ensure portability of benefits to enhance inventors' outside options.
- Antitrust enforcement in innovation markets:
  - Incorporate impacts on inventor labor markets into merger review, not just product market competition.
- Note: Some restriction on worker mobility can increase firms’ incentives to invest in worker human capital (Acemoglu and Pischke 1999).

### South Korea big push — empirical findings (Shim and Choi)
- Policy context:
  - Between 1973 and 1979, the Korean government temporarily supported firms in heavy manufacturing to adopt modern technologies.
- Aggregate adoption and industrialization effects:
  - Heavy manufacturing share of GDP more than doubled from 6 percent to 13 percent during the program window.
  - Number of new technology contracts in heavy manufacturing sectors quadrupled during the program window.
  - The share of heavy manufacturing in GDP, employment, and exports continued to rise after the policy ended.
- Firm-level causal effects (winners vs losers design):
  - Firms that successfully adopted modern technologies experienced:
    - After seven years, sales increased by 60%.
    - Total factor productivity (TFP) were almost double those of similar firms whose contracts fell through.
- Local spillovers and complementarities:
  - A 1 percentage point increase in the share of local adopters raised nonadopting firms’ sales by 2.7 percent and their revenue TFP by 1.6 percent.
  - A 1 percentage point increase in the share of local adopters raised a firm’s adoption probability by 0.85 percentage point (about 14 percent of the average annual adoption probability in the final year of the program).
  - Complementarities were stronger where market size was larger.
- Modeling approach:
  - A dynamic model in which spillovers occur with a one-period lag: a larger local base of adopters in period t–1 increases productivity and lowers the cost of adoption in period t.

*IMF RESEARCH perspectives, International Monetary Fund.*

### 1. Heavy-Manufacturing

### 1. Heavy-Manufacturing

### Model setup and dynamic complementarities
- Adoption costs are fixed: the decision to adopt is modeled as a one‑off investment whose returns are larger the stronger the spillovers and the larger the scale of a firm.
- Model ingredients generate dynamic complementarities: more adoption today makes adoption tomorrow more attractive.
- Analytical result: if spillovers and private returns lie in a middle range—neither too weak nor too strong—the model admits multiple steady states: a low adoption “preindustrialized” state and a high adoption “industrialized” state.
- Policy implication from the model: a temporary technology adoption subsidy can have a permanent effect by pushing the economy from the low to the high steady state when multiple steady states exist.

### Empirical calibration and counterfactual simulations (South Korea’s 1970s big push)
- Calibration matches causal estimates of:
  - (1) direct gains and local spill-overs from technology adoption (pinning down private returns and externalities),
  - (2) cross-regional patterns linking adoption to market access.
- Counterfactual question: What would have been Korea’s growth trajectory without the 1970s technology adoption subsidies?
- Counterfactual result (quantitative simulations):
  - Without the 1970s big push subsidies, South Korea would have converged to a less industrialized steady state.
  - Heavy-manufacturing share of GDP would have been 27 percent lower in the absence of the policy (compared with the baseline with the policy).
  - Export intensity would have been 39 percent lower in the absence of the policy (compared with the baseline with the policy).
  - The big push policy raised aggregate welfare by 14.6 percent.
  - Welfare gains were uneven across regions, with welfare changes ranging from -1.2 percent to 83.8 percent.
- Mechanisms behind regional disparities:
  - Internal trade linkages transmit lower prices, benefiting consumers and input-using firms elsewhere.
  - Stronger domestic competition compresses profits in some regions.

### Role of market access and complementary policies: scenario analysis
- Scale of a firm depends on access to international markets and input costs; complementarity strength rises with firm scale.
- Three “What if?” scenarios examined where scale was temporarily reduced during the policy years:
  - Foreign demand held flat rather than growing
  - Higher import tariffs (raising the cost of imported intermediates)
  - No major highway improvement (higher internal trade costs)
- Each scenario weakens the impact of the subsidies; holding down foreign demand is especially powerful.
- Combined scenario (flat foreign demand, higher import tariffs, and no highway improvement): the big push does not occur even with the same subsidies.
- Takeaway: Market access and complementary policies magnify (or mute) the power of temporary adoption subsidies.

### Empirical patterns on adopters (winners versus losers)
- Estimated effects of adopting new technology (winners versus losers design) are presented for:
  - Log sales
  - Revenue TFP
- Specifications include match-year and match-firm fixed effects; shaded areas represent 95 percent confidence intervals based on standard errors clustered at the match and firm levels.

### Conclusion on industrialization via a big push
- Empirical and quantitative evidence from South Korea’s 1970s experience suggests:
  - A big push policy can be successful by addressing coordination failures and facilitating diffusion of advanced technologies.
  - Complementary policies ensuring sufficient market access are important for sustainable industrial growth.

---

### Online platforms and firm-to-firm learning: evidence from a Chinese trade social media platform
- Data and linkage:
  - Over 20 million online posts on a Chinese social media platform for trade professionals were analyzed.
  - Digital conversations were linked to official customs data to measure platform effects on exports.
- Identification strategy:
  - Firms that join the platform were matched to nearly identical firms that had not joined yet (controls matched on industry, size, export history, and other characteristics).
  - Multiple pre-treatment checks confirm similar trajectories before platform access.
- Export gains from platform adoption (estimates over time):
  - First year after access: exports increase by 3 percent on average.
  - Second year after access: exports increase by 8 percent on average.
  - Third year after access: exports increase by 11 percent on average.
- Network effects and external economies of scale:
  - A 10 percent increase in the local pool of information is linked to an extra 0.3 percent in export growth for a new member.
  - The export effect grows over time, suggesting ongoing learning and better use of the network.
- How information flows:
  - Geography still matters: a 10 percent increase in physical distance between two cities reduces the volume of online messages between firms in those locations by about 0.7 percent.
  - Information flows are strategic and shaped by supply chains:
    - Upstream suppliers are more likely to share information with downstream customers.
    - When a buyer's purchases from a supplier increase by 10 percent, the supplier shares about 0.6 percent more knowledge with that buyer.
  - Empirical scatterplots show positive association between knowledge share and downstreamness (slope = 0.058 (0.021), R-squared = 0.018) and near-zero association with upstreamness (slope = -0.005 (0.018), R-squared = 0.000).
- Digital divide findings:
  - Cities with higher GDP per capita and better internet infrastructure have higher rates of platform adoption.
  - Less developed regions adopt the platform less, potentially worsening regional inequality.
- Model of endogenous learning and sharing:
  - A general equilibrium trade model with multiple regions and industries, supply-chain linkages, heterogeneous firms choosing to learn and to share information.
  - Firms can pay a "learning fee" to access a shared pool of knowledge; the export boost from joining the platform determined the "value of an idea" in calibration.
  - The model incorporates strategic sharing behavior that maximizes local network profits and is calibrated to match empirical patterns.

### Policy implications for digital platforms and trade
- Governments should complement trade‑cost reductions with support for knowledge networks that empower firms:
  - Invest in platforms that make peer-to-peer knowledge sharing easier, more trusted, and more discoverable.
  - Design platforms that reward knowledge sharing (verified answers, case studies, tutorials, recognition mechanisms).
  - Address geographic gaps via strategic partnerships, "twinning" programs, and hybrid online-offline forums to bridge the digital divide.
- Targeted interventions can leverage firms' natural incentives to share knowledge—especially along supply chains—rather than uniform subsidies for all firms.

*International Monetary Fund — IMF RESEARCH perspectives (Heavy-Manufacturing).*

### Conclusion: Establish Information Networks

### Conclusion: Establish Information Networks

### Social media and digital economic infrastructure
- Social media platforms are becoming an essential digital economic infrastructure for businesses engaged in international trade.
- By allowing companies to learn directly from their peers, these networks reduce information barriers, leading to real, sustained increases in exports.

### Evidence on the digital adoption gap
- Figure statistics:
  - Platform firm share in total exporters (%) 2017 vs. Log GDP per capita 2017: Slope = 4.851 (0.677); R-squared = 0.151.
  - Platform firm share in total exporters (%) 2017 vs. Internet broadband connections/population (%) 2017: Slope = 0.132 (0.037); R-squared = 0.049.
- Note: The scatterplots display platform firm share in total exporters against the city-level logarithm GDP per capita and internet access.

### Key lessons from the research
- The spread of business knowledge is not random; it is driven by clear economic incentives within supply chains.
- Geography and local business relationships continue to matter even in a hyperconnected world.
- Unequal access to digital tools can widen the gap between prosperous and lagging regions.

### Policy implications and recommendations
- Trade policy should complement tariff reductions with investments in digital infrastructure that enable firms to build knowledge networks.
- Supporting access to platforms or relevant knowledge bases—especially for businesses in underserved areas—could be a high-impact, low-cost way to foster more inclusive economic growth.
- In the 21st century, a smart trade policy must also be a smart information policy.

### Conference and event takeaways
- The IMF Research Department held the 26th Jacques Polak Annual Research Conference, November 6–7, 2025, on “The Evolving Landscape of Global Trade and Financial Integration.”
  - The policy panel “A World in Transition: Are We Ready to Adapt?” featured Hélène Rey, Kristin Forbes, Philip Lane, Sethaput Suthiwartnarueput, and Pierre-Olivier Gourinchas.
  - The Mundell-Fleming Lecture was delivered by Ricardo Reis, who presented a framework to think through financial repression and provided estimates of the revenues these measures can generate; while repression revenues are likely to average zero over the long run, Reis showed that they can be sizable over a decade or two.
  - Contact: Jiaqian Chen — JChen@IMF.org
- The joint IMF-FCDO Conference marked 13 years of collaboration on the Macroeconomic Research in Low-Income Countries (MRLIC) program.
  - Discussions focused on geoeconomic fragmentation, demographic transitions, rapid technological change, and extreme weather events as they relate to low-income countries (LICs).
  - The conference highlighted the need for partnerships to support macroeconomic research on LICs and outlined areas for future research.
  - Opening remarks were given by IMF Managing Director Kristalina Georgieva and FCDO chief economist Dennis Novy; closing highlights were provided by IMF Deputy Managing Director Nigel Clarke and IMF UK Executive Director Veda Poon.
  - Contact: Petia Topalova — ptopalova@IMF.org

---


_Source: https://www.imf.org/-/media/files/research/research-perspective/2025/imf-research-perspectives-2025.pdf_