## How Far Has Globalization Gone? A Tale of Two Regions

## Source details

**Canonical URL:** [How Far Has Globalization Gone? A Tale of Two Regions](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023255-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023255-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023255-print-pdf.pdf.json)

---

### Key findings on the evolution of trade globalization
- Trade globalization increased between 1995 and 2018 across the world, with growth concentrated before the global financial crisis (GFC) and largely stalling thereafter; Asia and Latin America did not lag behind.
- The aggregate picture masks substantial heterogeneity across countries and sectors:
  - Asia: growing trade globalization concentrated in China, Vietnam, Cambodia (mostly manufacturing and agriculture), and in India (services).
  - Latin America: gains concentrated in Mexico (agriculture and manufacturing), Chile and Peru (mining), with Brazil showing some signs of increasing trade globalization in agriculture.
- The paper estimates an indicator called border thickness that captures the cost of trading internationally relative to trading domestically; lower border thickness indicates increased globalization.

### Correlates of border thickness and trade policy (Table 1 and Table 2 results)
- Aggregate correlations (Table 1; each column reports a regression of border thickness on a single trade policy variable, with time and country fixed effects; MATR refers to aggregate trade restrictions excluding tariffs):
  - MFN tariffs (standardized): -0.269*** (0.0478); -0.211*** (0.0436)
  - MATR (standardized): -0.0330 (0.0216); -0.0141 (0.0207)
  - WTO dummy: 0.397*** (0.0475); 0.302*** (0.0527)
  - PTA count (standardized): 0.0519*** (0.0180); 0.0477** (0.0186)
  - Constants and model fit:
    - Constant (various specifications): 0.0798; 0.202**; 0.394***; -0.272***; 0.0968
    - Observations: 447 (in each column)
    - Adjusted R-squared: 0.769; 0.767; 0.824; 0.811; 0.847
  - Robust standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

- Sectoral correlations (Table 2; regressions of sectoral border thickness on policy variables; sectors: Agriculture (1), Mining (2), Manufacturing (3), Services (4)):
  - MFN tariffs (standardized):
    - Agriculture (1): -0.223*** (0.0558)
    - Mining (2): -0.190*** (0.0542)
    - Manufacturing (3): -0.187*** (0.0408)
    - Services (4): -0.145*** (0.0472)
  - MATR (standardized):
    - Agriculture (1): 0.0109 (0.0475)
    - Mining (2): 0.118** (0.0589)
    - Manufacturing (3): -0.00287 (0.0313)
    - Services (4): -0.0640** (0.0266)
  - WTO dummy:
    - Agriculture (1): 0.467*** (0.102)
    - Mining (2): -0.0786 (0.0997)
    - Manufacturing (3): 0.422*** (0.0649)
    - Services (4): 0.203*** (0.0508)
  - PTA count (standardized):
    - Agriculture (1): 0.0397 (0.0358)
    - Mining (2): 0.230*** (0.0524)
    - Manufacturing (3): 0.0503** (0.0229)
    - Services (4): 0.0108 (0.0222)
  - Constants and model fit (sectoral):
    - Constant (1–4): -0.256; 0.0187; -0.0513; 0.265* 
    - Observations: 447; 419; 447; 447
    - Adjusted R-squared: 0.813; 0.620; 0.811; 0.732
  - Robust standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

- Interpretation of correlations:
  - Border thickness is lower (greater globalization) in countries pursuing policies that reduce trade costs or foster international integration.
  - Lower MFN tariffs and WTO membership are associated with higher globalization across sectors.
  - Non-tariff restrictions (MATR) do not show a statistically significant correlation with relative border thickness at the aggregate level, though sectoral effects vary.
  - The impact of trade agreements (PTA count) is sector-dependent.

### Quantified GDP effects of globalization (general equilibrium simulations)
- Method:
  - Multi-sector multi-country quantitative trade model (similar to Caliendo and Parro (2015)); 66 countries and four aggregate sectors; counterfactual sets trade costs to their 1995 level to compute how much higher GDP was in 2018 due to globalization.
- Decomposition of GDP effects (Figure 6):
  - The overall effect of globalization is decomposed into:
    - Common globalization trend (coefficient γ): same change in trade costs for all countries; impact on GDP varies because of sector sizes, within-country sector linkages, trade openness, and country size.
    - Idiosyncratic deviation (coefficient δ): country-specific changes from the common trend; tends to dominate quantitatively.
  - Regional patterns:
    - Latin America: impact of the common trend is more homogeneous (but lower on average); idiosyncratic gains notable for Mexico and Peru.
    - Asia: impact of the common trend is more heterogeneous; countries benefiting disproportionately from the common trend include Brunei, Cambodia, Singapore, and Vietnam; China and India benefit less from the common trend. Idiosyncratic globalization led to large GDP gains for China, Cambodia, Singapore, and Vietnam; Hong Kong SAR, Malaysia, and Taiwan province of China show idiosyncratic effects in the opposite direction.

- Sectoral drivers of idiosyncratic GDP effects (Figure 7 and related discussion):
  - Latin America:
    - Mexico and Costa Rica: positive impact almost entirely explained by globalization in manufacturing.
    - Peru: most benefits explained by globalization in mining.
    - Argentina: decline in globalization explained by less globalization in agriculture (consistent with policy changes that raised barriers to agricultural exports).
    - Services: relatively minor role in idiosyncratic globalization in Latin America.
  - Asia:
    - China and Vietnam: gains almost entirely explained by a more globalized manufacturing sector.
    - Cambodia: most gains explained by manufacturing.
    - Malaysia and Taiwan province of China: decline in overall globalization explained by less globalized manufacturing.
    - Mining: contribution is negative for most Asian countries, in contrast to generally positive effects in Latin America.
    - Services: main driver of GDP increase in Singapore and of GDP decrease in Hong Kong SAR (consistent with anecdotal evidence of service-sector firm relocation).
    - Agriculture: generally minor role in Asia, with occasional positive effects where detected.

### Conclusions and implications
- The structural gravity estimates indicate a strong but heterogeneous increase in trade globalization across Asia and Latin America from 1995 to 2018.
- Trade policy matters:
  - Lower MFN tariffs and WTO membership are associated with increased globalization.
  - Effects of non-tariff measures and trade agreements vary by sector.
- The globalization process has had a significant impact on long-run GDP for many countries, with no single sector explaining cross-country variation:
  - Manufacturing globalization is a key driver of positive GDP differentials in several countries (Mexico, Costa Rica, Cambodia, China, Vietnam).
  - Agriculture played a smaller overall role but was crucial in explaining Argentina's relative decline.
  - Mining globalization tends to be more favorable in Latin America than in Asia.
  - Services globalization explains contrasting outcomes for Singapore (positive) and Hong Kong SAR (negative).
- The analysis does not address firm-level effects or the political economy dynamics of globalization; these areas warrant further research within the Asian and Latin American context.

### Annex 1. Variance Decomposition Exercise — role of trade policies
- Objective:
  - Investigate the role of trade policy variables in explaining differences in relative border thickness in LATAM and Asia.
  - Group trade policies into: bilateral policies (trade agreements), trade-related multilateral policies (WTO membership and MFN tariffs), and non-tariff restrictions.
- Methodology:
  - First-stage regressions of border thickness estimates on different combinations of trade policy variables and country and year fixed effects to construct predicted values.
  - Second stage regresses the predicted border thickness on the original border thickness estimate from Equation (1); the second-stage coefficient gauges the fraction of variability captured.
  - Benchmark: predicted values using first-stage controls of only country and time fixed effects (no trade policy variables).
- Table A1 — Second-stage coefficients (coefficient on original border thickness estimate) and standard errors:
  - Aggregate trade:
    - (1) 0.790*** (0.0193)
    - (2) 0.861*** (0.0164)
    - (3) 0.792*** (0.0192)
    - (4) 0.790*** (0.0193)
    - (5) 0.861*** (0.0164)
    - (6) 0.863*** (0.0163)
  - Agriculture:
    - (1) 0.807*** (0.0187)
    - (2) 0.832*** (0.0177)
    - (3) 0.807*** (0.0187)
    - (4) 0.807*** (0.0187)
    - (5) 0.832*** (0.0177)
    - (6) 0.833*** (0.0177)
  - Mining:
    - (1) 0.641*** (0.0235)
    - (2) 0.650*** (0.0234)
    - (3) 0.653*** (0.0233)
    - (4) 0.641*** (0.0235)
    - (5) 0.650*** (0.0234)
    - (6) 0.660*** (0.0232)
  - Manufacturing:
    - (1) 0.775*** (0.0198)
    - (2) 0.829*** (0.0178)
    - (3) 0.776*** (0.0197)
    - (4) 0.776*** (0.0198)
    - (5) 0.829*** (0.0178)
    - (6) 0.831*** (0.0178)
  - Services:
    - (1) 0.720*** (0.0213)
    - (2) 0.757*** (0.0203)
    - (3) 0.721*** (0.0213)
    - (4) 0.726*** (0.0211)
    - (5) 0.762*** (0.0202)
    - (6) 0.762*** (0.0202)
- Key findings from variance decomposition:
  - Multilateral policy variables play an important role in explaining differences in border thickness estimates in LATAM and Asian countries, especially for aggregate trade and manufacturing, and to a lesser extent agriculture and services.
  - The coefficient from the model that includes only trade agreements is roughly the same as that of the baseline model for all flows except mining.
  - Controlling for multilateral variables increases the coefficient (and R-squared) by 2 to 7 percentage points, depending on the flow.

*International Monetary Fund — IMF Working Papers: How Far Has Globalization Gone? A Tale of Two Regions*

### 1995.  Changes  in  non-tariff  restrictions  to  trade  have  been  more  modest.  However,  Figure  5  also

### How Far Has Globalization Gone? A Tale of Two Regions

### Key findings on the evolution of trade globalization
- Trade globalization increased between 1995 and 2018 across the world, with growth concentrated before the global financial crisis (GFC) and largely stalling thereafter; Asia and Latin America did not lag behind.
- The aggregate picture masks substantial heterogeneity across countries and sectors:
  - Asia: growing trade globalization concentrated in China, Vietnam, Cambodia (mostly manufacturing and agriculture), and in India (services).
  - Latin America: gains concentrated in Mexico (agriculture and manufacturing), Chile and Peru (mining), with Brazil showing some signs of increasing trade globalization in agriculture.
- The paper estimates an indicator called border thickness that captures the cost of trading internationally relative to trading domestically; lower border thickness indicates increased globalization.

### Correlates of border thickness and trade policy (Table 1 and Table 2 results)
- Aggregate correlations (Table 1; each column reports a regression of border thickness on a single trade policy variable, with time and country fixed effects; MATR refers to aggregate trade restrictions excluding tariffs):
  - MFN tariffs (standardized): -0.269*** (0.0478); -0.211*** (0.0436)
  - MATR (standardized): -0.0330 (0.0216); -0.0141 (0.0207)
  - WTO dummy: 0.397*** (0.0475); 0.302*** (0.0527)
  - PTA count (standardized): 0.0519*** (0.0180); 0.0477** (0.0186)
  - Constants and model fit:
    - Constant (various specifications): 0.0798; 0.202**; 0.394***; -0.272***; 0.0968 (reported alongside standard errors where shown)
    - Observations: 447 (in each column)
    - Adjusted R-squared: 0.769; 0.767; 0.824; 0.811; 0.847
  - Robust standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

- Sectoral correlations (Table 2; regressions of sectoral border thickness on policy variables; sectors: Agriculture (1), Mining (2), Manufacturing (3), Services (4)):
  - MFN tariffs (standardized):
    - Agriculture (1): -0.223*** (0.0558)
    - Mining (2): -0.190*** (0.0542)
    - Manufacturing (3): -0.187*** (0.0408)
    - Services (4): -0.145*** (0.0472)
  - MATR (standardized):
    - Agriculture (1): 0.0109 (0.0475)
    - Mining (2): 0.118** (0.0589)
    - Manufacturing (3): -0.00287 (0.0313)
    - Services (4): -0.0640** (0.0266)
  - WTO dummy:
    - Agriculture (1): 0.467*** (0.102)
    - Mining (2): -0.0786 (0.0997)
    - Manufacturing (3): 0.422*** (0.0649)
    - Services (4): 0.203*** (0.0508)
  - PTA count (standardized):
    - Agriculture (1): 0.0397 (0.0358)
    - Mining (2): 0.230*** (0.0524)
    - Manufacturing (3): 0.0503** (0.0229)
    - Services (4): 0.0108 (0.0222)
  - Constants and model fit (sectoral):
    - Constant (1–4): -0.256; 0.0187; -0.0513; 0.265* (standard errors reported in table)
    - Observations: 447; 419; 447; 447
    - Adjusted R-squared: 0.813; 0.620; 0.811; 0.732
  - Robust standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

- Interpretation of correlations:
  - Border thickness is lower (greater globalization) in countries pursuing policies that reduce trade costs or foster international integration.
  - Lower MFN tariffs and WTO membership are associated with higher globalization across sectors.
  - Non-tariff restrictions (MATR) do not show a statistically significant correlation with relative border thickness at the aggregate level, though sectoral effects vary.
  - The impact of trade agreements (PTA count) is sector-dependent.

### Quantified GDP effects of globalization (general equilibrium simulations)
- Method: Multi-sector multi-country quantitative trade model (similar to Caliendo and Parro (2015)); 66 countries and four aggregate sectors; counterfactual sets trade costs to their 1995 level to compute how much higher GDP was in 2018 due to globalization.
- Decomposition of GDP effects (Figure 6):
  - The overall effect of globalization is decomposed into:
    - Common globalization trend (coefficient γ): same change in trade costs for all countries; impact on GDP varies because of sector sizes, within-country sector linkages, trade openness, and country size.
    - Idiosyncratic deviation (coefficient δ): country-specific changes from the common trend; tends to dominate quantitatively.
  - Regional patterns:
    - Latin America: impact of the common trend is more homogeneous (but lower on average); idiosyncratic gains notable for Mexico and Peru.
    - Asia: impact of the common trend is more heterogeneous; countries benefiting disproportionately from the common trend include Brunei, Cambodia, Singapore, and Vietnam; China and India benefit less from the common trend. Idiosyncratic globalization led to large GDP gains for China, Cambodia, Singapore, and Vietnam; Hong Kong SAR, Malaysia, and Taiwan province of China show idiosyncratic effects in the opposite direction.

- Sectoral drivers of idiosyncratic GDP effects (Figure 7 and related discussion):
  - Latin America:
    - Mexico and Costa Rica: positive impact almost entirely explained by globalization in manufacturing.
    - Peru: most benefits explained by globalization in mining.
    - Argentina: decline in globalization explained by less globalization in agriculture (consistent with policy changes that raised barriers to agricultural exports).
    - Services: relatively minor role in idiosyncratic globalization in Latin America.
  - Asia:
    - China and Vietnam: gains almost entirely explained by a more globalized manufacturing sector.
    - Cambodia: most gains explained by manufacturing.
    - Malaysia and Taiwan province of China: decline in overall globalization explained by less globalized manufacturing.
    - Mining: contribution is negative for most Asian countries, in contrast to generally positive effects in Latin America.
    - Services: main driver of GDP increase in Singapore and of GDP decrease in Hong Kong SAR (consistent with anecdotal evidence of service-sector firm relocation).
    - Agriculture: generally minor role in Asia, with occasional positive effects where detected.

### Conclusions and implications
- The structural gravity estimates indicate a strong but heterogeneous increase in trade globalization across Asia and Latin America from 1995 to 2018.
- Trade policy matters: lower MFN tariffs and WTO membership are associated with increased globalization; effects of non-tariff measures and trade agreements vary by sector.
- The globalization process has had a significant impact on long-run GDP for many countries, with no single sector explaining cross-country variation:
  - Manufacturing globalization is a key driver of positive GDP differentials in several countries (Mexico, Costa Rica, Cambodia, China, Vietnam).
  - Agriculture played a smaller overall role but was crucial in explaining Argentina's relative decline.
  - Mining globalization tends to be more favorable in Latin America than in Asia.
  - Services globalization explains contrasting outcomes for Singapore (positive) and Hong Kong SAR (negative).
- The analysis does not address firm-level effects or the political economy dynamics of globalization; these areas warrant further research within the Asian and Latin American context.

*International Monetary Fund — IMF Working Papers: How Far Has Globalization Gone? A Tale of Two Regions*

### References

### wpiea2023255-print-pdf - References

### References
- Bibliographic list of works cited covering: trade theory and gravity models; trade policy and trade agreements; trade costs, globalization, and regional integration; sectoral studies (tourism, garments, electronics, agriculture); measurement of trade restrictions and non-tariff barriers; and methodological contributions to gravity estimation and variance decomposition.  
- Notable cited authors and works (as presented): ADB (2017); Alfaro-Ureña A., I. Manelici, and J.P. Vazquez (2022); Anderson J.E., and E. van Wincoop (2003); Anderson J.E., I. Borchert, A. Mattoo, and Y.V. Yotov (2018); Ariu A. (2022); Artuc, E., G. Porto, and B. Rijkers (2019); Baier S.L., J.H. Bergstrand and E. Vidal (2007); Baier S.L., Y.V. Yotov and T. Zylkin (2019); Baldwin R. (2022); Baldwin R. and D. Taglioni (2006); Battacharya, R. and S. Pienknagura (forthcoming); Bergstrand J.H., M. Larch and Y.V. Yotov (2015); Borchert I. and Y.V. Yotov (2017); Borchert I., M. Larch, S. Shikher, and Y.V. Yotov (2021); Breinlich H., D. Novy, and J.M.C. Santos Silva (2022); Cabrillac B. et al. (2016); Caliendo L. and F. Parro (2015, 2022); Camarero M. et al. (2016); Campos R. and J. Timini (2022); Campos R., J. Timini, and E. Vidal (2021); Costinot A. and A. Rodriguez-Clare (2014, 2018); Duran, Jose, N. Mulder, and O. Onodera (2008); Egger P., M. Larch, and K. Staub (2012); Egger P. H., and F. Tarlea (2015); Egger P.H., M. Larch and Y.V. Yotov (2022); El-Dahrawy Sánchez-Albornoz A. and J. Timini (2021); Estefania-Flores J. et al. (2022); Felbermayr G. et al. (2022); Felbermayr G. and Y. Yotov (2021); Fernandez-Stark K., P. Bamber and G. Gereffi (2016); Gereffi G., S. Frederick, and P. Bamber (2019); Hannan S.A. (2017); Hanson G.H. (2020); Head K. and T. Mayer (2014); Heid B., M. Larch, and Y.V. Yotov (2021); Hill H. (2000); Irwin D.A. (2020, 2022); Jacks D. S., C. M. Meissner, and D. Novy (2008, 2010, 2011); Kataryniuk I., J. Perez, and F. Viani (2021); Kee H.L., A. Nicita and M. Olarreaga (2009); Kohl T. (2014); Lee J. and I. Park (2005); Loayza N. and J. Rigolini (2016); Manchin and Pelkmans-Balaoing (2007); McCallum J. (1995); Merchán F. and M. Mesquita Moreira (2019); Mesquita Moreira M. (ed.) (2018); Monfort B. (2008); Pellegrina H.S. (2022); Pierola M.D., A.M. Fernandes, and T. Farole (2018); Rasiah R. (2009); Rasiah R., V. Gopal, and P. Sanjivee (2013); Salinas G. (2021); Santos Silva J.M.C. and S. Tenreyro (2006); Sedik T.S. (2018); Steinberg J.B. (2020); Thuong N.T.T. (2018); Yotov Y.V. (2012, 2022); Yotov Y.V., R. Piermartini, J. Monteiro and M. Larch (2016).

### Annex 1. Variance Decomposition Exercise
- Objective:
  - Further investigate the role of trade policy variables in explaining differences in relative border thickness in LATAM and Asia.
  - Group trade policies into: bilateral policies (trade agreements), trade-related multilateral policies (WTO membership and MFN tariffs), and non-tariff restrictions.
- Methodology:
  - Following Felbermeyer and Yotov (2021), run first-stage regressions of border thickness estimates on different combinations of trade policy variables and country and year fixed effects to construct predicted values.
  - In a second stage, regress the predicted border thickness on the original border thickness estimate from Equation (1).
  - The second-stage coefficient on the border thickness estimate gauges the fraction of variability captured by the chosen combination of trade policies.
  - Benchmark: coefficients from predicted values using first-stage controls of only country and time fixed effects (no trade policy variables).
- Table A1 — Decomposing border thickness estimates—the role of trade policies:
  - Columns correspond to first-stage models:
    - (1) baseline (country and time FEs)
    - (2) multilateral (WTO dum + MFN)
    - (3) bilateral (RTAs)
    - (4) non-tariff barriers
    - (5) multilateral + MATR
    - (6) All
  - Second-stage coefficients (coefficient on original border thickness estimate) and standard errors (in parentheses):
    - Aggregate trade:
      - (1) 0.790*** (0.0193)
      - (2) 0.861*** (0.0164)
      - (3) 0.792*** (0.0192)
      - (4) 0.790*** (0.0193)
      - (5) 0.861*** (0.0164)
      - (6) 0.863*** (0.0163)
    - Agriculture:
      - (1) 0.807*** (0.0187)
      - (2) 0.832*** (0.0177)
      - (3) 0.807*** (0.0187)
      - (4) 0.807*** (0.0187)
      - (5) 0.832*** (0.0177)
      - (6) 0.833*** (0.0177)
    - Mining:
      - (1) 0.641*** (0.0235)
      - (2) 0.650*** (0.0234)
      - (3) 0.653*** (0.0233)
      - (4) 0.641*** (0.0235)
      - (5) 0.650*** (0.0234)
      - (6) 0.660*** (0.0232)
    - Manufacturing:
      - (1) 0.775*** (0.0198)
      - (2) 0.829*** (0.0178)
      - (3) 0.776*** (0.0197)
      - (4) 0.776*** (0.0198)
      - (5) 0.829*** (0.0178)
      - (6) 0.831*** (0.0178)
    - Services:
      - (1) 0.720*** (0.0213)
      - (2) 0.757*** (0.0203)
      - (3) 0.721*** (0.0213)
      - (4) 0.726*** (0.0211)
      - (5) 0.762*** (0.0202)
      - (6) 0.762*** (0.0202)
  - Notes in table:
    - Standard errors in parentheses
    - *** p<0.01, ** p<0.05, * p<0.1
    - Predicted border thickness is constructed from first-stage regressions of border thickness estimates obtained from Equation (1) on different combinations of trade policy variables and country and year fixed effects. Country coverage varies by year depending on availability of trade policy variables.
- Key findings:
  - Results confirm the important role of multilateral policy variables in explaining differences in border thickness estimates in LATAM and Asian countries, especially for aggregate trade and manufacturing, and to a lesser extent agriculture and services.
  - The coefficient of the second-stage regression for the model that includes only trade agreements is roughly the same as that of the baseline model for all flows except mining.
  - The predicted values stemming from the model that controls for multilateral variables increases the coefficient (and R-squared of the regression) by 2 to 7 percentage points, depending on the flow.

*Italic: Source — wpiea2023255-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023255-print-pdf.pdf_