## _wp13113

## Source details

**Canonical URL:** [_wp13113](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13113.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13113.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13113.pdf.json)

---

### I. Introduction and Research Questions
- Focus: the role of net foreign asset (NFA) / net foreign liability (NFL) positions and their composition in triggering major external crises.
- Three core questions:
  - Can a proximate threshold be identified beyond which further build-up of net external liabilities sharply raises external-crisis risk?
  - Does the composition of external assets and liabilities (debt, portfolio equity, foreign direct investment, reserve vs non-reserve assets) affect crisis risk?
  - How well does an econometric early-warning/probit model that includes these variables and a few controls perform in- and out-of-sample, including over the 2008-2011 crises?

### II. Data, Crisis Definition, and Sample
- Sample and exclusions:
  - Initial sample: 70 countries (of which 41 are emerging markets) spanning 1970-2011.
  - Ireland dropped due to distortions from a sizable mutual fund industry affecting debt/equity breakdowns.
  - Iceland dropped after 2000 because NFL jumped from around 110 percent in 2007 to close to 700 percent of GDP at end-2008.
- Baseline crisis definition and event construction:
  - Includes defaults and rescheduling events (as per Bein and Calomiris (2001) and Standard & Poors, compiled in Borensztein and Panizza, 2008, and updated by the authors).
  - Includes resort to large multilateral/IMF support: IMF loans at least twice as large as the country's quota, measured using all net disbursements from program inception to end.
  - Treats events as discrete watershed-like occurrences; excludes observations that are ramifications of the initial major crisis outbreak up to the year preceding market re-entry.
- Additional sample summary (alternative tabulation):
  - Baseline sample: close to 2000 observations and 61 crisis events -> unconditional probability of crisis of 3 percent.
  - Event-analysis window: 11-year window centered on crisis year (t=0), spanning 5 years prior and after.
  - Estimation: pooled probit/logit with country fixed effects and time effects; explanatory variables lagged one year.
  - Receiver operating characteristic (ROC) curve and area under ROC (AUROC) used for model selection and comparison.

### III. Key Findings on Thresholds, Composition, and Predictors
- Thresholds and dynamics:
  - Evidence that crisis risk increases sharply as net foreign liabilities (NFL) exceed 50 percent of GDP.
  - Evidence that crisis risk increases sharply whenever the NFL/GDP ratio rises some 20 percentage points above the country-specific historical mean.
  - Crisis risk associated when NFA in the 50 to 60 percent of GDP range; recent crises closer to 60 percent.
  - Conditional estimates: external crises associated with NFA/GDP ratios between 15 and 20 percent below mean in run-up to crisis; effects statistically significant at 5 percent.
- Composition of liabilities:
  - Crisis risk rises as the composition of NFL is tilted toward debt liabilities.
  - Effects of portfolio equity liabilities are more mixed and generally weaker.
  - Higher FDI liabilities tend, if anything, to reduce crisis risk.
- Other predictors and assets:
  - Current account deficits have a higher predictive power than any other individual regressor in most specifications.
  - Predictive power is marginally higher for unconditional levels of the current account relative to deviations from a model-based current account “norm” (using standard specifications).
  - Higher foreign exchange reserves reduce crisis risk by more than other asset holdings.
  - REER gap: pre-crisis appreciation followed by depreciation (nearly 20 percent from peak to trough) associated with crises.
  - Global financial conditions: VIX and corporate spread tighten markedly around post-2007 crises and appear as significant common triggering factors.
  - Fiscal balance (cyclically adjusted) and output gap deteriorate pre-crisis.

### IV. Model Selection and Predictive Performance
- AUROC progression for nested models (exact reported steps):
  - 1) NFA only: 0.73
  - 2) Net Debt, Net Portfolio, FDI: 0.75
  - 3) Adding Reserves: 0.76
  - 4) Adding Per capita Income viz US: 0.82
  - 5) Adding Current Account/GDP: 0.86
  - 6) Adding REER gap: 0.88
  - 7) Adding VIX: 0.89
  - 8) Adding Fiscal Balance Gap: 0.90
- Final favored specification AUROC = 0.90 (baseline external crisis definition, 1970-2011).
- In-sample predictive classification (baseline probit):
  - At 20% cut-off:
    - Correctly predicts 33 out of 61 crises.
    - Correctly classifies 99% of non-crisis observations.
    - Overall correctly classified = 97%.
  - At 10.5% cut-off (max signal-to-noise ratio):
    - Correctly predicts 42 out of 61 crises.
    - False alarms = 5% of tranquil observations.
    - Overall correctly classified = 94%.

### V. Estimated Coefficients and Statistical Significance (highlights)
- Net Foreign Assets/GDP:
  - Univariate coefficient: -0.990***.
- Net external debt assets /GDP:
  - Examples: -1.750*** (debt-only column); -1.223** (column (7) with full controls).
- Net Foreign Direct Investment/GDP:
  - Positive and significant in richer specifications (example: 1.193*** in column (7)).
- FX reserves/GDP:
  - Strongly negative and significant in richer specifications (example: -3.984*** in column (7)).
- CA balance/GDP (2-year MA):
  - Strongly negative and highly significant (examples: -8.492*** in earlier columns; -10.40*** when included).
- REER gap and VIX: positive and significant predictors.
- Fiscal Gap: negative coefficient (larger fiscal deficit -> higher crisis risk) and significant in richer specs (example: -5.069** in column (7)).
- Note on inference: Robust standard errors clustered at country level; significance denoted as *** p<0.01, ** p<0.05, * p<0.1.

### VI. Elasticities and Economic Magnitudes (favored specification examples)
- Net external debt assets/GDP:
  - SD = 0.20; dP/dx at mean = -0.008; year prior to crisis = -0.28; when P>0.1 = -0.32.
  - Interpretation: one standard deviation increase in net external debt to GDP (a 20 percentage point rise) raises probability of external crises by over 6 percent in the specified computation context (as described in the source).
- FX reserves/GDP:
  - SD = 0.07; dP/dx at mean = -0.027; year prior to crisis = -0.91; when P>0.1 = -1.05.
  - Example interpretation: an increase in central bank reserves of 7 percent of GDP reduces crisis risk by some 7 percent (holding net debt unchanged).
- CA balance/GDP (2-year MA):
  - SD = 0.04; dP/dx at mean = -0.072; year prior to crisis = -2.37; when P>0.1 = -2.75.

### VII. Model-Based Thresholds (tipping-point estimates)
- Univariate NFA/GDP threshold maximizing signal-to-noise ratio: -49%.
- Baseline multivariate tipping points (Table 4):
  - Net Foreign Assets/GDP: -53%
  - Net Debt Assets/GDP: -35%
  - Net Portfolio Assets/GDP: 0%
  - Net FDI Assets/GDP: -18%
  - Net Reserves Assets/GDP: 4.6%
- Summary interpretation:
  - Net foreign liabilities in excess of 50 percent of GDP in absolute terms and higher than 20 percent of the country-specific historical mean are associated with steeper crisis risk.
  - Tipping point typically associated with net external debt liabilities above 35 percent of GDP.

### VIII. Robustness and Sensitivity
- Robust across:
  - Alternative crisis definitions (AUROC examples: 0.90, 0.90, 0.89, 0.85, 0.91 across definitions/samples).
  - Breakdowns into gross debt assets and liabilities: net debt remains significant; coefficients on gross debt assets and liabilities are virtually same magnitude but opposite sign.
  - Addition of many other controls (public debt/GDP, credit growth, US corporate spread, institutional quality, capital controls, trade openness): core results for net debt, FX reserves, CA balance, REER gap, VIX, and fiscal gap remain sizeable and significant in most specifications.
- Sensitivities:
  - FDI coefficient sensitive to outlier countries with very large net FDI liabilities (financial centers); adding an FDI outlier dummy reduces significance.
  - Current account gap (deviation from fundamentals) is collinear with unconditional current account; unconditional 2-year MA CA performs better as predictor.

### IX. Out-of-Sample (1970-2006 estimated, predicting 2007-2011) Performance
- Using model estimated over 1970-2006 and a 20% cut-off:
  - Correctly predicts crises for Greece and Portugal; flags Spain as high-risk since 2008.
  - Correctly predicts Dominican Republic, Jamaica, Latvia, Romania (at 18% threshold), and Serbia.
  - Misses crises in Ecuador, Hungary, and Ukraine; Pakistan predicted at 7% (above unconditional 3% but below 20% threshold).
  - Overall: parsimonious model does a good job predicting the bulk of 2008-2011 crises using pre-crisis fundamentals.

### X. Policy-Relevant Conclusions and Recommendations
- Net external debt liabilities are the most important component of NFA for external crisis risk: highly statistically significant and stable across specifications.
- Quantitative thresholds for concern:
  - Net foreign liabilities > 50 percent of GDP (absolute) and > 20 percent above country-specific historical mean -> associated with steeper crisis risk.
  - Net external debt liabilities around or above 35 percent of GDP mark a typical tipping point.
- Current account deficits (speed of accumulation of liabilities) strongly increase crisis risk; unconditional measures (2-year MA) perform well as predictors.
- Reserve accumulation has a precautionary role: substantial reserve increases (e.g., 7 percent of GDP) materially reduce crisis probability.
- Net FDI liabilities, controlling for current account and other factors, do not increase crisis risk and can be associated with lower risk (consistent with "good cholesterol" view of FDI).
- Practical implication: parsimonious monitoring frameworks using NFA decomposition, CA balance, reserves, REER gap, fiscal gap, and global financial volatility (VIX) can achieve high predictive accuracy (AUROC ~0.90) and reasonable in- and out-of-sample performance.

### XI. Appendices (selected)
- Appendix 1: Sample listings for Overall Sample and Crisis Sample (baseline and broader). Selected country-year examples reproduced verbatim in the source (e.g., Argentina 1995, Greece 2010, Portugal 2011, Romania 2009, Turkey 2008, Ukraine 2008, United Kingdom 2009, etc.).
- Appendix 2: Estimates of Current Account Gaps — Methodology:
  - Based on macroeconomic balance approach and inter-temporal Saving-Investment model; residuals from baseline panel specification used as country current account gap.
- Appendix 2: Panel Estimates — Table A2 (selected reported coefficients and fit statistics):
  - Lagged NFA/Y (Column (1)): 0.0452*** (0.00346)
  - Relative PPP GDPpc (Column (1)): 0.0467*** (0.00563)
  - Oil Balance Dummy (Column (1)): 0.287*** (0.0471)
  - Old Age Dependency Ratio (Column (1)): -0.143*** (0.0246)
  - Population Growth (Column (1)): -0.402** (0.174)
  - Polity Index (Column (1)): -0.000747*** (0.000189)
  - Trend Growth (Column (1)): -0.262*** (0.0574)
  - General Gov. Balance (cyc.adj) (Column (1)): 0.375*** (0.0580)
  - Quinn Index of Capital Controls (Column (1)): 0.0226*** (0.00469)
  - Constant (Column (1)): 0.0103*** (0.00227)
  - Observations (Column (1)): 2,300; R-squared (Column (1)): 0.319
  - Additional reported columns and variables included in the source; residuals from specification (1) are used as the measure of the country’s current account gap in the main text.

*Source: IMF working paper (_wp13113) — extracted content provided.*

### References .............................................................................................................

### References (excerpt)

### I. Introduction and Research Questions
- Focus: the role of net foreign asset (NFA) / net foreign liability (NFL) positions and their composition in triggering major external crises.
- Three core questions:
  - Can a proximate threshold be identified beyond which further build-up of net external liabilities sharply raises external-crisis risk?
  - Does the composition of external assets and liabilities (debt, portfolio equity, foreign direct investment, reserve vs non-reserve assets) affect crisis risk?
  - How well does an econometric early-warning/probit model that includes these variables and a few controls perform in- and out-of-sample, including over the 2008-2011 crises?

### II. Data and Crisis Definition
- Sample:
  - Initial sample: 70 countries (of which 41 are emerging markets) spanning 1970-2011.
  - Ireland dropped due to distortions from a sizable mutual fund industry affecting debt/equity breakdowns.
  - Iceland dropped after 2000 because NFL jumped from around 110 percent in 2007 to close to 700 percent of GDP at end-2008.
- Baseline external-crisis definition:
  - Includes defaults and rescheduling events (as per Bein and Calomiris (2001) and Standard & Poors, compiled in Borensztein and Panizza, 2008, and updated by the authors).
  - Includes resort to large multilateral/IMF support, defined as IMF loans that are at least twice as large as the respective country's quota in the IMF, when all net disbursements are computed from program's inception to end.
  - Treats events as discrete watershed-like occurrences; excludes observations that are ramifications of the initial major crisis outbreak up to the year preceding market re-entry.

### III. Key Findings (enumerated)
- Thresholds and dynamics:
  - Evidence that crisis risk increases sharply as net foreign liabilities (NFL) exceed 50 percent of GDP.
  - Evidence that crisis risk increases sharply whenever the NFL/GDP ratio rises some 20 percentage points above the country-specific historical mean.
- Composition of liabilities:
  - Crisis risk rises as the composition of NFL is tilted toward debt liabilities.
  - Effects of portfolio equity liabilities are more mixed and generally weaker.
  - Higher FDI liabilities tend, if anything, to reduce crisis risk.
- Other predictors and assets:
  - Current account deficits have a higher predictive power than any other individual regressor in most specifications.
  - Predictive power is marginally higher for unconditional levels of the current account relative to deviations from a model-based current account “norm” (using standard specifications).
  - Higher foreign exchange reserves reduce crisis risk by more than other asset holdings.
- Model performance:
  - A multivariate but reasonably parsimonious probit model including the above controls has substantial predictive power, in and out of sample—particularly regarding the 2008-2011 crises.
  - Many other variables featured in previous crisis literature do not add significant explanatory or predictive power in this framework.

### IV. Relation to Literature and Conceptual Implications
- Builds on and extends early-warning system (EWS) and external-sustainability literatures:
  - Novelty: use of disaggregated net foreign assets and composition controls, a longer sample (1970-2011), focus on major external crises “stricto sensu”, and ROC-based model selection with extensive out-of-sample evaluation.
  - Consistent with sovereign-debt models that emphasize the ratio of external debt liabilities to GDP as a key gauge of default risk.
  - Compares with Frankel and Saravelos (2012): both find external debt significant; this analysis finds current account and combined models increase predictive power, with current account the most powerful single predictor.

### V. Policy-Relevant Interpretations
- Tipping-point implications:
  - National fiscal and macro-prudential policies that affect current accounts and external-liability accumulation should consider the identified thresholds (NFL > 50 percent of GDP and increases of 20 percentage points above country mean) as markers of sharply rising external-crisis risk.
- Liability composition:
  - Policies and risk assessments should differentiate debt liabilities from FDI and portfolio equity given their distinct associations with crisis risk.
- Reserves and precautionary buffers:
  - Maintaining higher foreign exchange reserves is associated with lower crisis risk and supports the rationale for reserves as a crisis-prevention device.

*Source: IMF working paper (References and selected sections provided).*

### 1992. In that case, we do not treat credit events associated with partial repayments and partial

### _wp13113 - 1992. In that case, we do not treat credit events associated with partial repayments and partial

### Crisis definition, sample, and methodology
- Baseline sample: close to 2000 observations and 61 crisis events -> unconditional probability of crisis of 3 percent.
- Crisis episodes exclude interim years between initial default and market re-entry; market re-entry defined as either the year after S&P classifies the default to have ended or—when crisis categorization involves large IMF lending—when country liabilities vis-à-vis the IMF are brought down to below 200 percent of quota or, if remaining above 200 percent, decline by two consecutive years.
- Event-analysis window: 11-year window centered on crisis year (t=0), spanning 5 years prior and after.
- Estimation: pooled probit/logit with country fixed effects and time effects; explanatory variables lagged one year. Receiver operating characteristic (ROC) curve used for model selection; area under ROC (AUROC) used to compare models.

### Crisis dynamics (unconditional and conditional)
- Net Foreign Liabilities (NFL / NFA):
  - Crisis risk associated when NFA in the 50 to 60 percent of GDP range; recent crises closer to 60 percent.
  - Conditional (fixed and time effects): external crises associated with NFA/GDP ratios between 15 and 20 percent below mean in run-up to crisis; effects statistically significant at 5 percent.
- Net debt accumulation drives the NFA effect: reduction in net debt assets between 15 to 20 percent of GDP on average precedes crises.
- FX reserves and current account:
  - Crises tend to occur in countries with reserves lower than the mean by about 2 percent of GDP.
  - Crises associated with current account deficits around 3 percent of GDP larger than country-specific/global mean and deteriorating; typical starting current account deficits of around 4 percent of GDP pre-crisis.
- Other predictors with consistent dynamics:
  - Real effective exchange rate (REER) gap: pre-crisis appreciation followed by depreciation (nearly 20 percent from peak to trough).
  - Fiscal balance (cyclically adjusted) and output gap: deterioration pre-crisis, similar between pre- and post-2007 events.
  - Global financial conditions: VIX and corporate spread tighten markedly around post-2007 crises; VIX and corporate spread appear as significant common triggering factors.

### Model selection and predictive performance
- Univariate probit with lagged NFA/GDP only: AUROC = 0.73.
- Disaggregating NFA into components and adding covariates improves AUROC to 0.90 for a parsimonious multivariate model. Table of AUROC progression:
  - 1) NFA only: 0.73
  - 2) Net Debt, Net Portfolio, FDI: 0.75
  - 3) Adding Reserves: 0.76
  - 4) Adding Per capita Income viz US: 0.82
  - 5) Adding Current Account/GDP: 0.86
  - 6) Adding REER gap: 0.88
  - 7) Adding VIX: 0.89
  - 8) Adding Fiscal Balance Gap: 0.90
- Final favored specification AUROC = 0.90 (baseline external crisis definition, 1970-2011).
- In-sample predictive classification (baseline probit):
  - At 20% cut-off:
    - Correctly predicts 33 out of 61 crises.
    - Correctly classifies 99% of non-crisis observations.
    - Overall correctly classified = 97%.
  - At 10.5% cut-off (max signal-to-noise ratio):
    - Correctly predicts 42 out of 61 crises.
    - False alarms = 5% of tranquil observations.
    - Overall correctly classified = 94%.

### Key estimated coefficients and statistical significance (baseline probit results highlights)
- Net Foreign Assets/GDP: coefficient -0.990*** (in univariate; Table 2).
- Net external debt assets /GDP: coefficients across specifications (examples):
  - -1.750*** (column with debt only)
  - -1.223** (column (7) with full controls)
- Net Foreign Direct Investment/GDP: positive and significant in richer specifications (e.g., 1.193*** in column (7)).
- FX reserves/GDP: strongly negative and significant in richer specifications (e.g., -3.984*** in column (7)).
- CA balance/GDP (2-year MA): strongly negative and highly significant (e.g., -8.492*** in earlier columns; -10.40*** when included).
- REER gap and VIX: positive and significant predictors of crisis risk.
- Fiscal Gap: negative coefficient (i.e., larger fiscal deficit -> higher crisis risk) and significant in richer specs (e.g., -5.069** in column (7)).

Note: Robust standard errors clustered at country level; significance denoted as *** p<0.01, ** p<0.05, * p<0.1.

### Elasticities and economic magnitudes (favored specification)
- Table 3 elasticity examples (dP/dx; SD; elasticity at mean; elasticity year prior to crisis; elasticity when P>0.1):
  - Net external debt assets/GDP: SD = 0.20; dP/dx at mean = -0.008; year prior to crisis = -0.28; when P>0.1 = -0.32.
    - Interpretation given: one standard deviation increase in net external debt to GDP (a 20 percentage point rise) raises probability of external crises by over 6 percent in specified computation context.
  - FX reserves/GDP: SD = 0.07; dP/dx at mean = -0.027; year prior to crisis = -0.91; when P>0.1 = -1.05.
    - Example interpretation: an increase in central bank reserves of 7 percent of GDP (sample SD) reduces crisis risk by some 7 percent (holding net debt unchanged).
  - CA balance/GDP (2-year MA): SD = 0.04; dP/dx at mean = -0.072; year prior to crisis = -2.37; when P>0.1 = -2.75.

### Model-based threshold (tipping point) estimates
- Univariate NFA/GDP threshold maximizing signal-to-noise ratio: -49%.
- Baseline multivariate tipping points (Table 4):
  - Net Foreign Assets/GDP: -53%
  - Net Debt Assets/GDP: -35%
  - Net Portfolio Assets/GDP: 0%
  - Net FDI Assets/GDP: -18%
  - Net Reserves Assets/GDP: 4.6%
- Summary threshold interpretation in conclusions:
  - Net foreign liabilities in excess of 50 percent of GDP in absolute terms and higher than 20 percent of the country-specific historical mean are associated with steeper crisis risk.
  - Tipping point typically associated with net external debt liabilities above 35 percent of GDP.

### Robustness and sensitivity checks
- Results robust across:
  - Alternative crisis definitions (broader and narrower): AUROC remains high (examples in Table 7: 0.90, 0.90, 0.89, 0.85, 0.91 across definitions/samples).
  - Breakdowns into gross debt assets and liabilities: net debt remains the significant indicator; coefficients on gross debt assets and liabilities are virtually same magnitude but opposite sign.
  - Addition of many other controls (public debt/GDP, credit growth, US corporate spread, institutional quality, capital controls, trade openness): core results for net debt, FX reserves, CA balance, REER gap, VIX, and fiscal gap remain sizeable and significant in most specifications.
- Some sensitivities:
  - FDI coefficient sensitive to outlier countries with very large net FDI liabilities (financial centers); adding an FDI outlier dummy reduces significance.
  - Current account gap (deviation from fundamentals) is collinear with unconditional current account; unconditional 2-year MA CA performs better in these regressions.

### Out-of-sample (1970-2006 estimated, predicting 2007-2011) performance
- Using model estimated over 1970-2006 and a 20% cut-off:
  - Correctly predicts crises for Greece and Portugal; flags Spain as high-risk since 2008.
  - Correctly predicts Dominican Republic, Jamaica, Latvia, Romania (at 18% threshold), and Serbia.
  - Misses crises in Ecuador, Hungary, and Ukraine; Pakistan predicted at 7% (above unconditional 3% but below 20% threshold).
  - Overall assessment: parsimonious model does a good job predicting the bulk of 2008-2011 crises using pre-crisis fundamentals.

### Key policy-relevant conclusions
- Net external debt liabilities are the most important component of NFA for external crisis risk; they are highly statistically significant and stable across specifications.
- Quantitative thresholds for concern:
  - Net foreign liabilities > 50 percent of GDP (absolute) and > 20 percent above country-specific historical mean -> associated with steeper crisis risk.
  - Net external debt liabilities around or above 35 percent of GDP mark a typical tipping point.
- Current account deficits (speed of accumulation of liabilities) strongly increase crisis risk; unconditional measures (2-year MA) perform well as predictors.
- Reserve accumulation has a precautionary role: substantial reserve increases (e.g., 7 percent of GDP) materially reduce crisis probability.
- Net FDI liabilities, controlling for current account and other factors, do not increase crisis risk and can be associated with lower risk (consistent with "good cholesterol" view of FDI).
- Parsimonious monitoring frameworks using NFA decomposition, CA balance, reserves, REER gap, fiscal gap, and global financial volatility (VIX) can achieve high predictive accuracy (AUROC ~0.90) and reasonable in- and out-of-sample performance.

*Source: _wp13113 - extracted content (pages provided).*

### REFERENCES

### _wp13113 - REFERENCES

### Appendix 1: Sample
- Provides the country-year listing for:
  - Overall Sample
  - Crisis Sample (baseline definition)
  - Crisis Sample (broader crisis definition)
- Selected entries (verbatim examples from the table):
  - Australia 1930, Argentina 1982, Argentina 1975
  - Argentina 1995, Argentina 2001, Belize 2006
  - Brazil 1983, Brazil 1999, Brazil 2001
  - Chile 1972, Chile 1983, Costa Rica 1981
  - Dominican Republic 1982, Dominican Republic 2003, Dominican Republic 2009
  - Ecuador 1983, Ecuador 1999, Ecuador 2008
  - Greece 2010, Egypt 1984, Finland 1993
  - Iceland 1975, Iceland 1983, India 1984
  - Korea 1975, Korea 1980, Korea 1997
  - Mexico 1982, Mexico 1995, Panama 1983
  - Peru 1978, Peru 1982, Philippines 1976
  - Poland 1981, Portugal 1977, Portugal 2011
  - Romania 1999, Romania 2009, Serbia 2009
  - South Africa 1985, Thailand 1981, Thailand 1985
  - Turkey 1976, Turkey 1994, Turkey 2000, Turkey 2008
  - Ukraine 1998, Ukraine 2008, United Kingdom 2009
  - Uruguay 1972, Uruguay 1983, Uruguay 2002
  - Venezuela 1983, Venezuela 1996, Venezuela 2002

### Appendix 2: Estimates of Current Account Gaps — Methodology
- Framework:
  - Based on the macroeconomic balance approach and the inter-temporal Saving-Investment model (Obstfeld and Rogoff, 1996), with empirical implementation following Chinn & Prasad (2003) and others.
  - Model components combined into four equations:
    - I-S behavioral relation (equation (1)).
    - BOP constraint (equation (2)).
    - Solvency constraint (equation (3)).
    - Multilateral constraint (equation (4)), where i ω is country i's share of world GDP and globally 1
      , i
      N
      i
      j
      j
      ω
      =
      =
      ∀
      ∑
- Policy rule linking country real interest rate r to output gap and world interest rate:
  - Floating/IT regime representation (equation (5a)).
  - Peg representation (equation (5.b)).
- Reduced-form expression for current account ratio to GDP (equation (6)), where regressors include:
  - s_x: consumption/saving shifters (income per capita, demographics, expected income, social insurance, budget balance).
  - I_x: investment shifters (income per capita, TFP/trend growth, governance).
  - CF_x: capital account shifters (global risk aversion, capital controls).
  - RS_x: reserve accumulation shifters (precautionary and policy factors including capital controls).
- Multilateral constraint implies variables measured relative to current GDP–weighted world averages; the world interest rate term drops out; in long-run equilibrium output gap term drops out.
- Empirical implementation:
  - Baseline estimate of equation (6) uses standard proxies and annual panel data for 1970-2011.
  - Residuals of specification (1) used as the measure of the country’s current account gap in the main text.

### Appendix 2: Panel Estimates — Table A2 (selected results)
- Table title: Panel Estimates of Current Account Norms
- Estimation reported across specifications (columns (1) to (6)). Selected coefficients (with robust standard errors in parentheses) — all numbers reproduced exactly as in the source:
  - Lagged NFA/Y:
    - Column (1): 0.0452*** (0.00346)
    - Column (2): 0.0455*** (0.00351)
    - Column (3): 0.0442*** (0.00398)
    - Column (4): 0.0465*** (0.00353)
    - Column (5): 0.0452*** (0.00346)
    - Column (6): 0.0473*** (0.00370)
  - Relative PPP GDPpc:
    - Column (1): 0.0467*** (0.00563)
    - Column (2): 0.0480*** (0.00592)
    - Column (3): 0.0459*** (0.00584)
    - Column (4): 0.0445*** (0.00582)
    - Column (5): 0.0465*** (0.00566)
    - Column (6): 0.0185*** (0.00592)
  - Oil Balance Dummy:
    - Column (1): 0.287*** (0.0471)
    - Column (2): 0.285*** (0.0471)
    - Column (3): 0.291*** (0.0475)
    - Column (4): 0.277*** (0.0466)
    - Column (5): 0.287*** (0.0472)
    - Column (6): 0.356*** (0.0700)
  - Old Age Dependency Ratio:
    - Column (1): -0.143*** (0.0246)
    - Column (2): -0.145*** (0.0250)
    - Column (3): -0.140*** (0.0250)
    - Column (4): -0.140*** (0.0257)
    - Column (5): -0.142*** (0.0246)
    - Column (6): -0.0802*** (0.0245)
  - Population Growth:
    - Column (1): -0.402** (0.174)
    - Column (2): -0.413** (0.177)
    - Column (3): -0.407** (0.175)
    - Column (4): -0.423** (0.184)
    - Column (5): -0.400** (0.174)
    - Column (6): -0.460** (0.191)
  - Polity Index:
    - Column (1): -0.000747*** (0.000189)
    - Column (2): -0.000769*** (0.000194)
    - Column (3): -0.000726*** (0.000186)
    - Column (4): -0.000781*** (0.000206)
    - Column (5): -0.000746*** (0.000190)
    - Column (6): -0.000214 (0.000192)
  - Trend Growth:
    - Column (1): -0.262*** (0.0574)
    - Column (2): -0.264*** (0.0575)
    - Column (3): -0.260*** (0.0588)
    - Column (4): -0.265*** (0.0555)
    - Column (5): -0.262*** (0.0574)
    - Column (6): -0.291*** (0.0675)
  - General Gov. Balance (cyc.adj):
    - Column (1): 0.375*** (0.0580)
    - Column (2): 0.375*** (0.0583)
    - Column (3): 0.369*** (0.0556)
    - Column (4): 0.456*** (0.0578)
    - Column (5): 0.377*** (0.0585)
    - Column (6): 0.452*** (0.0616)
  - Quinn Index of Capital Controls:
    - Column (1): 0.0226*** (0.00469)
    - Column (2): 0.0228*** (0.00473)
    - Column (3): 0.0232*** (0.00462)
    - Column (4): 0.0224*** (0.00499)
    - Column (5): 0.0227*** (0.00480)
    - Column (6): 0.0160*** (0.00575)
  - Additional variables reported (selected):
    - Aging Speed: -0.0131 (0.0203) — included in one specification
    - Financial Center Dummy: 0.00395 (0.00654)
    - Trade Openness (5-year MA): -0.00222 (0.00598)
    - Reserve Currency Dummy: 0.000841 (0.00300)
    - Social Protection Index: -0.0132 (0.00938)
  - Constant:
    - Column (1): 0.0103*** (0.00227)
    - Column (2): 0.0106*** (0.00231)
    - Column (3): 0.00943*** (0.00270)
    - Column (4): 0.0104*** (0.00313)
    - Column (5): 0.0102*** (0.00239)
    - Column (6): 0.00498** (0.00232)
- Sample and fit statistics (exact values):
  - Observations:
    - Column (1): 2,300
    - Column (2): 2,300
    - Column (3): 2,300
    - Column (4): 2,300
    - Column (5): 1,342
    - Column (6): 1,891
  - R-squared:
    - Column (1): 0.319
    - Column (2): 0.319
    - Column (3): 0.319
    - Column (4): 0.341
    - Column (5): 0.319
    - Column (6): 0.344
- Notes:
  - Robust SEs in parentheses.
  - Significance markers: *** p<0.01, ** p<0.05, * p<0.1.
  - The first column is used as the baseline specification; the residuals from specification (1) are used as the measure of the country’s current account gap in the main text.

*Document: _wp13113 - REFERENCES*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13113.pdf_