## 1. Number of Countries in Database by Income Group by Year

## Source details

**Canonical URL:** [1. Number of Countries in Database by Income Group by Year](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021134-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021134-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021134-print-pdf.pdf.json)

---

### Major findings on spillover coverage (2010–2019)
- Overall discussion of spillovers in IMF Article IV staff reports declined over the period 2010–2019.
- Spillover discussion spiked in specific years that overlapped with major monetary policy moves by the Federal Reserve (FED) and European Central Bank (ECB) and other non-monetary policy events (examples cited include FED monetary easing from 2009; ECB unconventional monetary policy deployment in 2010; FED “Operation Twist” in 2011 and 2012; FED announcement of future tapering in 2013 and “taper tantrum”; FED ending of Quantitative Easing (QE) in 2014 and onset of normalization in interest rates in 2015; ECB expanded asset purchase program (APP) in 2015 and 2016; Brexit-related risks and global trade tensions in 2018-19).
- Spillover coverage remains prominent in staff reports for systemically important economies: the US, euro area, and China.
- Spillovers are predominantly covered in staff reports of advanced economies (AEs) and more frequently in the context of the real, financial, and external sectors.
- Additional econometric analysis shows that staff reports covering countries with high trade openness and lower inflation are more likely to cover spillovers.

### Institutional and policy context
- Coverage of spillovers in Article IV staff reports was mainstreamed by the 2012 Integrated Surveillance Decision (ISD).
- The 2014 Triennial Surveillance Review (TSR) recommends more systematic analysis of outward spillovers and spillbacks in systemic countries, greater quantification of impacts based on global risk scenarios, and discussion of policy implications.
- The 2015 Guidance Note for Surveillance under Article IV Consultations emphasizes examining the most significant actual and potential outward spillovers irrespective of transmission channels.
- The 2018 Interim Surveillance Review (ISR) highlights increased depth of global spillover coverage in flagship reports, while noting scope for more prominent coverage of outward spillovers in Article IV consultations.
- The 2020 Comprehensive Surveillance Review (CSR) proposes new modalities and procedures for conducting bilateral and multilateral surveillance to capture new economic realities, including changing sources of spillovers during COVID-19.

### Data, scope, and preprocessing
- Database coverage: more than 80,000 paragraphs, across 1280 IMF staff reports, for 193 countries/territories.
- Coverage variation drivers: non-published (confidential) staff reports; 12-month vs 24-month Article IV consultation cycles (countries with IMF-supported programs may be on 24-month cycles).
- Computational environment: training and prediction performed on Amazon Web Services (AWS).
- BERT pre-training corpora: Wikipedia (2500 million words) and Book Corpus (800 million words).

### BERT model training, tuning, and evaluation
- Training dataset: 1,356 paragraphs from Article IVs, with 379 paragraphs (28 percent) manually labeled as containing a discussion of spillovers (label = 1) and the remainder labeled 0.
- Test dataset: 717 paragraphs from staff reports, of which roughly 27 percent mention spillovers.
- Fine-tuning hyperparameters:
  - Number of epochs: 3
  - Batch size: 16
  - Optimizer: Adam (adaptive learning rate)
- Performance on test set:
  - Accuracy: 0.9
  - F1-score: 0.78
  - Evaluation loss: 0.29
- Comparative model performance (bag-of-words features; 70/30 train-test split; cross-validation):
  - LR: 0.88
  - SVM: 0.84
  - RF: 0.79
  - All comparative models perform worse than the BERT model.
- Full-sample prediction result: of the roughly 80,000 paragraphs in the dataset, the BERT model determines that 3,495 (4.4 percent) are on the subject of spillovers.

### Topic assignment and sector coverage methodology
- Sector taxonomy: external, financial, fiscal, monetary, and real (based on IMF Enterprise Business Vocabulary (EBV) terms).
- Topic assignment pipeline:
  - Extract sector-associated terms from EBV.
  - Train Word2Vec on IMF Staff Reports to expand sector term lists (terms most similar to sector vectors).
  - Match expanded term lists to paragraph contents to assign sector(s) to paragraphs (double assignments allowed).
- Accuracy reference for dictionary+Word2Vec approach: Fayad et al. find this method achieves 88 percent accuracy.

### Key empirical patterns and descriptive findings
- Time trends:
  - Overall discussion of spillovers in Article IV staff reports shows a declining trend over 2010–2019, with notable spikes during major global spillover events.
  - Solid line measure: ratio of staff reports discussing spillovers at least once over total staff reports.
  - Dashed line measure: average (within groups) of ratio of number of paragraphs discussing spillovers over total number of paragraphs in a staff report (considered more indicative of coverage).
- Income-group patterns:
  - Spillovers are discussed more in staff reports covering advanced economies (AEs) and least in those covering low-income countries (LICs); emerging markets (EMs) typically fall between AEs and LICs.
  - Over time, the ratio of staff reports discussing spillovers at least once has increased somewhat for LICs (though paragraph-level coverage in LICs remains low).
- Regional patterns:
  - Spillovers are discussed most in staff reports covering Asian-Pacific (APD) and European (EUR) regions.
  - Slight increases over the last two years in the sample for Middle-East and Central Asia (MCD), Western Hemisphere (WHD), and APD.
- Sectoral coverage (all paragraphs):
  - 37 percent discuss the real sector
  - 28 percent discuss fiscal
  - 22 percent discuss financial
  - 16 percent discuss external
  - 6 percent discuss monetary
  - Note: shares sum to more than 100 percent because paragraphs can be assigned multiple topics.
- Sectoral composition of paragraphs that discuss spillovers:
  - 45 percent occur in the context of the external sector
  - 34 percent in the real sector
  - 25 percent in the financial sector
  - 6 percent in the monetary sector
  - Additional note: only 2 percent of the full sample considers monetary issues.

### Regression evidence: probability of discussing spillovers (main results)
- Econometric approach:
  - Probit model: dependent variable Y_{i,t} = 1 if spillovers are discussed in country i’s Article IV staff report in year t; 0 otherwise.
  - Controls: AE and EM dummies (LICs as baseline), ln(GDP), GDP growth rate, trade openness, capital account openness (Chinn-Ito index), public debt (gross), inflation (CPI change). Controls are lagged by one year. Year fixed effects included. Robust standard errors reported.
  - OLS variants and Seemingly Unrelated Regressions (SUR) used as robustness checks.
- Probit / OLS headline coefficients (Table 2, selected):
  - AE: Probit Model 1 coefficient = 0.685*** (standard error 0.139); OLS Model 3 coefficient = 0.192*** (standard error 0.039)
  - EM: Probit Model 1 coefficient = 0.444*** (standard error 0.096); OLS Model 3 coefficient = 0.153*** (standard error 0.033)
  - Trade Openness: Probit Model 2 coefficient = 0.004*** (standard error 0.001); OLS Model 4 coefficient = 0.001*** (standard error 0.000)
  - Inflation: Probit Model 2 coefficient = -0.023** (standard error 0.009); OLS Model 4 coefficient = -0.006* (standard error 0.003)
  - Ln GDP: Probit Model 1 coefficient = 0.120*** (standard error 0.018); Probit Model 2 coefficient = 0.214*** (standard error 0.026)
  - GDP Growth: Probit Model 1 coefficient = -0.015** (standard error 0.008)
  - Observations: 1,272 (Probit Model 1), 1,006 (Probit Model 2)
  - Pseudo R2: 0.117 (Model 1), 0.170 (Model 2)
- Interpretation of primary regression findings:
  - Staff reports covering AEs and EMs are more likely to discuss spillovers relative to LICs.
  - Larger economies (higher ln GDP) are more likely to discuss spillovers.
  - Greater trade openness is associated with a higher probability of discussing spillovers.
  - Lower inflation is associated with a higher probability of discussing spillovers.
  - IMF-supported program status (MONA dummy) is not statistically significant in the main specifications (un-tabulated).
  - Year dummies and event dummies confirm that 2010–2016 were years with heavier spillover discussion.

### Sector-specific regression findings (selected)
- Probit results (Table 3, selected coefficients and significance):
  - Real (Model 1): AE = 0.303* (0.180); EM = 0.277** (0.113); Trade Openness = 0.002** (0.001); CapAcct Openness = 0.279*** (0.100); Ln GDP = 0.094*** (0.023); Inflation = -0.018** (0.009)
  - Fiscal (Model 2): AE = -0.391 (0.251); EM = -0.712*** (0.194); CapAcct Openness = 0.685*** (0.172)
  - Financial (Model 3): AE = 0.340* (0.196); EM = 0.268** (0.130); Trade Openness = 0.002** (0.001); Ln GDP = 0.161*** (0.026)
  - Monetary (Model 4): Trade Openness = 0.003*** (0.001); Ln GDP = 0.180*** (0.042)
  - External (Model 5): AE = -0.133 (0.186); EM = 0.201* (0.115); Trade Openness = 0.003*** (0.001); Ln GDP = 0.222*** (0.024)
- SUR results (Table 4) confirm probit findings with similar signs and significance:
  - Real (SUR Model 1): AE = 0.114* (0.067); EM = 0.100** (0.042); Trade Openness = 0.001** (0.000); CapAcct Openness = 0.103*** (0.037); Ln GDP = 0.035*** (0.008); Inflation = -0.006** (0.003)
  - Financial (SUR Model 3): AE = 0.133** (0.063); EM = 0.071* (0.039); Trade Openness = 0.001** (0.000); Ln GDP = 0.053*** (0.008)

### Percentage-of-paragraphs regressions (Table 5)
- Dependent variable: ratio of number of paragraphs discussing spillovers over total paragraphs in a country’s staff report.
- Selected OLS coefficients:
  - AE: Model 1 coefficient = 0.038*** (0.005)
  - EM: Model 1 coefficient = 0.013*** (0.002)
  - IMF-supported Program: Model 2 coefficient = -0.012*** (0.003)
  - Trade Openness: Model 2 coefficient = 0.000*** (standard error 0.000)
  - Ln GDP: Model 1 coefficient = 0.004*** (0.001); Model 2 coefficient = 0.007*** (0.001)
  - Inflation: Model 2 coefficient = -0.001*** (0.000)
  - Observations: 1,272 (Model 1 and Model 2); 1,006 (Model 3)
  - R2: 0.238 (Model 1); 0.192 (Model 2); 0.321 (Model 3)
- Interpretation:
  - Staff reports for AEs and EMs have a significantly higher proportion of paragraphs devoted to spillovers.
  - Staff reports covering countries with an IMF-supported program contain fewer paragraphs that discuss spillovers (significant at the 1 percent level).

### Conclusions and policy-relevant takeaways
- Overall result: Discussion of spillovers in IMF Article IV staff reports declined over 2010–2019 but spikes during major global spillover events and remains elevated for systemically important economies.
- Spillover discussion is concentrated in staff reports covering AEs and EMs and is most common in the external and real sector contexts.
- Econometric confirmation: larger country size, greater trade and capital account openness, lower inflation, and lower growth are associated with higher likelihood or greater intensity of spillover discussion.
- Practical recommendation emphasized in the text:
  - Strengthening coordination between IMF country teams to ensure spillover discussions remain prominent not only in source (systemic) countries but also in recipient countries.

*Source: IMF staff calculations.*

### 1. Number of Countries in Database by Income Group by Year ______________________ 6

### 1. Number of Countries in Database by Income Group by Year

### Major findings on spillover coverage (2010–2019)
- Overall discussion of spillovers in IMF Article IV staff reports declined over the period 2010–2019.
- Spillover discussion spiked in specific years that overlapped with major monetary policy moves by the Federal Reserve (FED) and European Central Bank (ECB) and other non-monetary policy events (examples cited include FED monetary easing from 2009; ECB unconventional monetary policy deployment in 2010; FED “Operation Twist” in 2011 and 2012; FED announcement of future tapering in 2013 and “taper tantrum”; FED ending of Quantitative Easing (QE) in 2014 and onset of normalization in interest rates in 2015; ECB expanded asset purchase program (APP) in 2015 and 2016; Brexit-related risks and global trade tensions in 2018-19).
- Spillover coverage remains prominent in staff reports for systemically important economies: the US, euro area, and China.
- Spillovers are predominantly covered in staff reports of advanced economies (AEs) and more frequently in the context of the real, financial, and external sectors.
- Additional econometric analysis shows that staff reports covering countries with high trade openness and lower inflation are more likely to cover spillovers.

### Institutional and policy context
- Coverage of spillovers in Article IV staff reports was mainstreamed by the 2012 Integrated Surveillance Decision (ISD), which requires discussion of outward spillovers induced by policies employed by systemic economies.
- The 2014 Triennial Surveillance Review (TSR) recommends more systematic analysis of outward spillovers and spillbacks in systemic countries, greater quantification of impacts based on global risk scenarios, and discussion of policy implications.
- The 2015 Guidance Note for Surveillance under Article IV Consultations emphasizes examining the most significant actual and potential outward spillovers irrespective of transmission channels.
- The 2018 Interim Surveillance Review (ISR) highlights increased depth of global spillover coverage in flagship reports, while noting scope for more prominent coverage of outward spillovers in Article IV consultations.
- The 2020 Comprehensive Surveillance Review (CSR) proposes new modalities and procedures for conducting bilateral and multilateral surveillance to capture new economic realities, including changing sources of spillovers during COVID-19.

### Methodology and data
- A state-of-the-art deep learning model (BERT) was trained to recognize when a paragraph discusses spillovers, going beyond keyword search to capture complex phrasings (example: “The ECB’s tightening of monetary policy would have widespread effects across Europe”).
- The model was deployed on a database containing IMF Article IV staff reports for the period 2010 - 2019.
- The topic model developed in Fayad et al (2020) was used to assign a sector to each paragraph, classifying paragraphs into external, fiscal, financial, monetary, or real sectors.
- The resulting dataset contains paragraphs with metadata indicating whether the paragraph discusses cross-border spillovers, its sector, the covered country, level of development, geographic region, and publication year, enabling summary statistics, trend visualization, and econometric analysis.

### Implications for surveillance and policy
- Given the likelihood of a sustained period of macro-financial spillovers in the post-pandemic landscape and the multi-speed nature of recovery, discussing scope and effects of spillovers should be a key component of Fund surveillance.
- Strengthening quantification of outward spillovers, systematic analysis for systemic economies, and discussion of policy implications remain priorities per TSR, Guidance Note, ISR, and CSR recommendations.
- Monitoring spillover coverage across sectors and country characteristics (trade openness, inflation) can inform the targeting and depth of Article IV discussions.

*Content unit: wpiea2021134-print-pdf - 1. Number of Countries in Database by Income Group by Year ______________________ 6*

### 2019. The database contains more than 80,000 paragraphs, across 1280 IMF staff reports, for

### wpiea2021134-print-pdf - 2019. The database contains more than 80,000 paragraphs, across 1280 IMF staff reports, for

### Data, scope, and preprocessing
- Database coverage: more than 80,000 paragraphs, across 1280 IMF staff reports, for 193 countries/territories.
- Coverage variation drivers: non-published (confidential) staff reports; 12-month vs 24-month Article IV consultation cycles (countries with IMF-supported programs may be on 24-month cycles).
- Computational environment: training and prediction performed on Amazon Web Services (AWS).
- BERT pre-training corpora: Wikipedia (2500 million words) and Book Corpus (800 million words).

### BERT model training, tuning, and evaluation
- Training dataset: 1,356 paragraphs from Article IVs, with 379 paragraphs (28 percent) manually labeled as containing a discussion of spillovers (label = 1) and the remainder labeled 0.
- Test dataset: 717 paragraphs from staff reports, of which roughly 27 percent mention spillovers.
- Fine-tuning hyperparameters:
  - Number of epochs: 3
  - Batch size: 16
  - Optimizer: Adam (adaptive learning rate)
- Performance on test set:
  - Accuracy: 0.9 (i.e., correctly labels 90 percent of paragraphs in the test dataset)
  - F1-score: 0.78
  - Evaluation loss: 0.29
- Comparative model performance (bag-of-words features; 70/30 train-test split; cross-validation):
  - LR: 0.88
  - SVM: 0.84
  - RF: 0.79
  - All comparative models perform worse than the BERT model.
- Full-sample prediction result: of the roughly 80,000 paragraphs in the dataset, the BERT model determines that 3,495 (4.4 percent) are on the subject of spillovers.

### Topic assignment and sector coverage methodology
- Sector taxonomy: external, financial, fiscal, monetary, and real (based on IMF Enterprise Business Vocabulary (EBV) terms).
- Topic assignment pipeline:
  - Extract sector-associated terms from EBV.
  - Train Word2Vec on IMF Staff Reports to expand sector term lists (terms most similar to sector vectors).
  - Match expanded term lists to paragraph contents to assign sector(s) to paragraphs (double assignments allowed).
- Accuracy reference for dictionary+Word2Vec approach: Fayad et al. find this method achieves 88 percent accuracy (noting potential errors and results interpreted with caution).

### Key empirical patterns and descriptive findings
- Time trends:
  - Overall discussion of spillovers in Article IV staff reports shows a declining trend over 2010–2019, with notable spikes during major global spillover events (e.g., FED and ECB unconventional monetary policy periods; global trade tensions; Brexit).
  - Solid line measure: ratio of staff reports discussing spillovers at least once over total staff reports.
  - Dashed line measure: average (within groups) of ratio of number of paragraphs discussing spillovers over total number of paragraphs in a staff report (considered more indicative of coverage).
- Income-group patterns:
  - Spillovers are discussed more in staff reports covering advanced economies (AEs) and least in those covering low-income countries (LICs); emerging markets (EMs) typically fall between AEs and LICs.
  - Over time, the ratio of staff reports discussing spillovers at least once has increased somewhat for LICs (though paragraph-level coverage in LICs remains low).
- Regional patterns:
  - Spillovers are discussed most in staff reports covering Asian-Pacific (APD) and European (EUR) regions.
  - Slight increases over the last two years in the sample for Middle-East and Central Asia (MCD), Western Hemisphere (WHD), and APD.
- Sectoral coverage (all paragraphs):
  - 37 percent discuss the real sector
  - 28 percent discuss fiscal
  - 22 percent discuss financial
  - 16 percent discuss external
  - 6 percent discuss monetary
  - Note: shares sum to more than 100 percent because paragraphs can be assigned multiple topics.
- Sectoral composition of paragraphs that discuss spillovers:
  - 45 percent occur in the context of the external sector
  - 34 percent in the real sector
  - 25 percent in the financial sector
  - 6 percent in the monetary sector
  - Additional note: only 2 percent of the full sample considers monetary issues.

### Regression evidence: probability of discussing spillovers (main results)
- Econometric approach:
  - Probit model: dependent variable Y_{i,t} = 1 if spillovers are discussed in country i’s Article IV staff report in year t; 0 otherwise.
  - Controls: AE and EM dummies (LICs as baseline), ln(GDP), GDP growth rate, trade openness, capital account openness (Chinn-Ito index), public debt (gross), inflation (CPI change). Controls are lagged by one year. Year fixed effects included. Robust standard errors reported.
  - OLS variants and Seemingly Unrelated Regressions (SUR) used as robustness checks.
- Probit / OLS headline coefficients (Table 2, selected):
  - AE: Probit Model 1 coefficient = 0.685*** (standard error 0.139); OLS Model 3 coefficient = 0.192*** (standard error 0.039)
  - EM: Probit Model 1 coefficient = 0.444*** (standard error 0.096); OLS Model 3 coefficient = 0.153*** (standard error 0.033)
  - Trade Openness: Probit Model 2 coefficient = 0.004*** (standard error 0.001); OLS Model 4 coefficient = 0.001*** (standard error 0.000)
  - Inflation: Probit Model 2 coefficient = -0.023** (standard error 0.009); OLS Model 4 coefficient = -0.006* (standard error 0.003)
  - Ln GDP: Probit Model 1 coefficient = 0.120*** (standard error 0.018); Probit Model 2 coefficient = 0.214*** (standard error 0.026)
  - GDP Growth: Probit Model 1 coefficient = -0.015** (standard error 0.008)
  - Observations: 1,272 (Probit Model 1), 1,006 (Probit Model 2)
  - Pseudo R2: 0.117 (Model 1), 0.170 (Model 2)
- Interpretation of primary regression findings:
  - Staff reports covering AEs and EMs are more likely to discuss spillovers relative to LICs.
  - Larger economies (higher ln GDP) are more likely to discuss spillovers.
  - Greater trade openness is associated with a higher probability of discussing spillovers.
  - Lower inflation is associated with a higher probability of discussing spillovers.
  - IMF-supported program status (MONA dummy) is not statistically significant in the main specifications (un-tabulated), suggesting program status may not predict spillover discussion frequency.
  - Year dummies and event dummies confirm that 2010–2016 were years with heavier spillover discussion; time fixed effects used in main analysis.

### Sector-specific regression findings (Tables 3 and 4; selected)
- Probit results (Table 3, selected coefficients and significance):
  - Real (Model 1): AE = 0.303* (0.180); EM = 0.277** (0.113); Trade Openness = 0.002** (0.001); CapAcct Openness = 0.279*** (0.100); Ln GDP = 0.094*** (0.023); Inflation = -0.018** (0.009)
  - Fiscal (Model 2): AE = -0.391 (0.251); EM = -0.712*** (0.194); CapAcct Openness = 0.685*** (0.172)
  - Financial (Model 3): AE = 0.340* (0.196); EM = 0.268** (0.130); Trade Openness = 0.002** (0.001); Ln GDP = 0.161*** (0.026)
  - Monetary (Model 4): Trade Openness = 0.003*** (0.001); Ln GDP = 0.180*** (0.042)
  - External (Model 5): AE = -0.133 (0.186); EM = 0.201* (0.115); Trade Openness = 0.003*** (0.001); Ln GDP = 0.222*** (0.024)
- SUR results (Table 4) confirm probit findings with similar signs and significance:
  - Real (SUR Model 1): AE = 0.114* (0.067); EM = 0.100** (0.042); Trade Openness = 0.001** (0.000); CapAcct Openness = 0.103*** (0.037); Ln GDP = 0.035*** (0.008); Inflation = -0.006** (0.003)
  - Financial (SUR Model 3): AE = 0.133** (0.063); EM = 0.071* (0.039); Trade Openness = 0.001** (0.000); Ln GDP = 0.053*** (0.008)

### Percentage-of-paragraphs regressions (Table 5)
- Dependent variable: ratio of number of paragraphs discussing spillovers over total paragraphs in a country’s staff report.
- Selected OLS coefficients:
  - AE: Model 1 coefficient = 0.038*** (0.005)
  - EM: Model 1 coefficient = 0.013*** (0.002)
  - IMF-supported Program: Model 2 coefficient = -0.012*** (0.003)
  - Trade Openness: Model 2 coefficient = 0.000*** (standard error 0.000)
  - Ln GDP: Model 1 coefficient = 0.004*** (0.001); Model 2 coefficient = 0.007*** (0.001)
  - Inflation: Model 2 coefficient = -0.001*** (0.000)
  - Observations: 1,272 (Model 1 and Model 2); 1,006 (Model 3)
  - R2: 0.238 (Model 1); 0.192 (Model 2); 0.321 (Model 3)
- Interpretation:
  - Staff reports for AEs and EMs have a significantly higher proportion of paragraphs devoted to spillovers.
  - Staff reports covering countries with an IMF-supported program contain fewer paragraphs that discuss spillovers (significant at the 1 percent level).

### Conclusions and policy-relevant takeaways
- Overall result: Discussion of spillovers in IMF Article IV staff reports declined over 2010–2019 but spikes during major global spillover events and remains elevated for systemically important economies.
- Spillover discussion is concentrated in staff reports covering AEs and EMs and is most common in the external and real sector contexts.
- Econometric confirmation: larger country size, greater trade and capital account openness, lower inflation, and lower growth are associated with higher likelihood or greater intensity of spillover discussion.
- Practical recommendation emphasized in the text:
  - Strengthening coordination between IMF country teams to ensure spillover discussions remain prominent not only in source (systemic) countries but also in recipient countries.

*Source: IMF staff calculations.*

### References

### wpiea2021134-print-pdf - References

### References

- Arrow, K.J., 1962. Economic Welfare and the Allocation of Resources to Invention, The Rate and Direction of Inventive Activity: Economic and Social Factors, edited by the Universities-National Bureau Committee for Economic Research and the Committee on Economic Growth of the Social Science Research Councils, 609– 626.  
- Arrow, K.J., 1973. Social Responsibility and Economic Efficiency. Public Policy, 21, 303-317. 
- Blagrave, P., Ho, G., Koloskova, K. and Vesperoni, E., 2017. Fiscal Spillovers : The Importance of Macroeconomic and Policy Conditions in Transmission. Spillover Notes, 11, IMF. 
- Coase, R.H., 1960. The Problem of Social Cost. Journal of Law and Economics, 3(1), 1-44. 
- Devlin, J., Chang, M., Lee, K., and Toutanova, K., 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In North American Association for Computational Linguistics (NAACL). 
- Diamond, D.W., Hu, Y. and Rajan, R.G., 2020. The Spillovers from Easy Liquidity and the Implications for Multilateralism. IMF Economic Review, 68, 4–34. 
- Fayad, G., Huang, C., Shibuya, Y. and Zhao, P., 2019, “How Do Member Countries Receive IMF Policy Advice: Results form a State-of-the-art Sentiment Index”, IMF Working Paper 20/7. 
- Krugman, P., 1991a. Geography and Trade. Cambridge: MIT Press. 
- Krugman, P., 1991b. Increasing Returns and Economic Geography. Journal of Political Economy, 99(3), 483-499. 
- IMF, 2012. Modernizing the Legal Framework for Surveillance - An Integrated Surveillance Decision. 
- IMF, 2014. Triennial Surveillance Review. 
- IMF, 2015. Guidance Note for Surveillance under Article IV Consultation. 
- IMF, 2016. Spillovers from Migration and China's Transition. World Economic Outlook, October.  
- IMF, 2018. Interim Surveillance Review. 
- Marshall, A., 1890. Principles of Economics. MacMillan & Co., London and New York.  
-  
- 24
- Mikolov, T., Chen, K., Corrado, G., Dean, J., 2013. Efficient Estimation of Word Representations in Vector Space. arXiv: 1301:3781. 
- Pigou, A.C., 1920. The Economic of Welfare. Macmillan & Co., London.  
- Porter, M.E., 1990. The Competitive Advantage of Nations. Harvard Business Review, 68(2), 73–93. 
- Romer, P.M., 1986. Increasing Returns and Long-Run Growth. Journal of Political Economy, 94(5), 1002-1037.

*Source: wpiea2021134-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021134-print-pdf.pdf_