## ch2

## Source details

**Canonical URL:** [ch2](https://www.imf.org/-/media/files/publications/esr/2020/english/ch2.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/esr/2020/english/ch2.pdf.md)
- [Structured JSON version](/-/media/files/publications/esr/2020/english/ch2.pdf.json)

---

### Background and motivation
- External assets and liabilities more than tripled as a share of GDP from the early 1990s to the years preceding the Great Lockdown.
- The sharp increase in gross and net external positions—referred to as the rise of “stock imbalances”—raises sustainability questions for debtor economies and macrofinancial vulnerabilities to domestic and global shocks.
- The initial tightening in global financial conditions and large terms-of-trade fluctuations caused by the coronavirus (COVID-19) outbreak and the Great Lockdown led to sharp currency and current account movements; in most cases the exchange rate acted as a shock absorber, while a few countries resorted to foreign exchange intervention and capital flow management measures.

### Research questions
- How do the size and composition of the various types of external assets and liabilities relate to the risk of external stress episodes? Is the relationship for emerging market and developing economies different from that for other (advanced) economies?
- What is the role of other well-studied variables, such as the level of global financial risk aversion and external current account balances, in explaining the likelihood of external stress episodes? How do these factors combine with the structure of the IIP in amplifying or mitigating risks?
- When an external stress event occurs, how does the size and composition of the IIP relate to the impact on output, the current account, and the exchange rate? How do external stress events impact creditor economies?

### Data, sample, and definitions
- Sample: 73 advanced and emerging market and developing economies during 1991–2018.
- External stress episode definition: an event that involves either a sovereign external debt default, debt restructuring, or recourse to an IMF arrangement. Sources include updated Das, Papaioannou, and Trebesch (2011); Asonuma and Trebesch (2016); and Paris Club reports.
- Currency-denomination data for external debt assets and liabilities available starting in 1991.
- Large external crises subset: cases of IMF financial assistance exceeding 200 percent of quota.

### Main findings (high-level)
- Not all IIP components relate equally to the likelihood of external stress episodes.
- The net IIP declines in the run-up to an external stress episode; the more negative it becomes, the greater the likelihood of external stress materializing.
- Gross external debt liabilities are stronger predictors of external stress than equity liabilities or private external debt assets.
- Larger stock of foreign official reserves mitigates the likelihood of an external stress episode, although with diminishing effects.
- For the full sample, external debt liabilities are strong predictors of stress irrespective of currency denomination; foreign-currency-denominated debt liabilities are particularly relevant predictors for emerging market and developing economies.
- Traditional predictors remain important: large current account deficits and higher levels of global risk aversion increase external financing risks.
- The nature of external vulnerabilities has rotated over time:
  - before the Asian financial crisis low reserves were central;
  - ahead of the global financial crisis large current account deficits were central;
  - in the years preceding the Great Lockdown elevated gross external debt liabilities and their foreign-currency-denominated component were central vulnerabilities for emerging market and developing economies.
- Preexisting external vulnerabilities amplify macroeconomic costs of external stress episodes.

### Key quantitative impacts of external stress episodes
- Identified external stress cases: 128.
- Real GDP:
  - For countries with large current account deficits, elevated foreign-currency-denominated debt, and low levels of reserves, real GDP falls by about 4.1 percent within two years of an external stress episode.
  - For countries with more limited external vulnerabilities, the decline in real GDP levels is typically about 1 percent.
- Real effective exchange rate: depreciates by about 10 percent within the first year of an external stress episode in countries with high preexisting vulnerabilities.
- Current account balance: rises by more than 2 percent of GDP within the first year of an external stress episode in countries with high preexisting vulnerabilities.
- Creditor economies: following large global crises, on average in the decade after the global financial crisis, a 1 percent of GDP rise in the current account surplus has been associated with a 0.5 percent of GDP valuation loss in creditor economies.

### IIP dynamics around stress episodes
- Event-study controls: country and time fixed effects to capture differences in country average IIP levels and common shocks.
- Run-up to stress episode:
  - Net IIP declines, driven predominantly by a sharp rise in foreign-currency-denominated external debt liabilities as a share of GDP (partly reflecting currency depreciation dynamics).
  - Domestic-currency-denominated debt liabilities also increase ahead of the stress episode, but by a smaller magnitude.
  - Equity assets and liabilities decline gradually.
  - Foreign-currency-denominated external debt assets increase; private foreign-currency-denominated external debt assets increase ahead of the stress episode (likely reflecting private capital flight and valuation effects).
  - Official foreign exchange reserves decline sharply just ahead of the stress episode.
- After onset of stress:
  - Net IIP typically rises, driven primarily by a significant drop in foreign-currency-denominated external debt liabilities (deleveraging and restructuring).
  - Official foreign exchange reserves typically decline in the aftermath and then bounce back.
- For large external crises, dynamics are starker: larger drops in net IIP ahead of crisis, larger rises in foreign-currency-denominated debt liabilities, and sharper declines in gross equity and official reserve assets that rebound to levels well below precrisis peaks.

### Estimating external stress probabilities: methodology
- Approach: pooled probit model where the dependent variable equals 1 for a stress episode in a given country-year and 0 otherwise.
- Explanatory variables include:
  - IIP components (gross and net), including currency denomination of assets and liabilities.
  - Current account balance.
  - Global risk aversion.
  - Real effective exchange rate gap (deviations from average of previous five years).
  - Income per capita relative to the United States.
  - Credit gap (constructed analogously to the real exchange rate gap).
  - Degree of financial development (index including market depth, access, and efficiency).
- Sample and specification broadly similar to Catão and Milesi-Ferretti (2014) with extensions for currency denomination.

### Estimation results (summary)
- Lower net IIP (larger net debtor position) is associated with higher probability of external stress.
- Disaggregated IIP findings:
  - Both higher foreign-currency and domestic-currency external debt liabilities increase the probability of external stress events.
  - Gross positions (assets versus liabilities) provide useful information beyond net positions; coefficients for symmetric external debt categories differ for assets and liabilities.
  - Higher levels of foreign exchange reserves reduce the occurrence of stress episodes.
  - Private external debt assets do not appear to play a mitigating role, possibly reflecting capital flight dynamics that often accompany deteriorating conditions.

### Box 2.1 — Predicting external crises: overview and methodology
- Objective: Predict external crises using alternative definitions, including sudden stop episodes with high growth impact and exchange rate market pressure episodes.
- Techniques: Signal extraction and machine-learning techniques; probit estimation for external stress events; VXO (precursor of the VIX) used to maximize data coverage as a proxy for global risk aversion.
- Estimation period: 1991–2018. Analysis covers full sample (advanced + EMDEs) and an EMDE subsample.
- Country-specific variables are lagged by one year. Current account/GDP included as a two-year moving average.
- About 80 predictive indicators considered, spanning policy regimes; imbalances and mismatches; asset price booms and busts; global liquidity and contagion; and political shocks.

### Box 2.1 — Key predictors and statistical findings
- Central finding: External debt is a strong predictor of external stress episodes; stock vulnerabilities such as external debt measures are reliable predictors across crisis types.
- Other significant predictors:
  - Larger current account deficits are associated with higher external stress.
  - Global risk aversion increases the likelihood of external stress.
  - Higher levels of foreign exchange reserves play a mitigating role.
- Differences between samples:
  - Foreign-currency-denominated debt liabilities have a statistically significant relationship with external stress risk for EMDEs, whereas domestic-currency-denominated debt liabilities do not (in EMDEs).
  - Private external debt assets denominated in foreign currency reduce the probability of a stress episode in EMDEs.
  - Equity assets and liabilities and external debt assets denominated in domestic currency do not play a statistically significant role in EMDEs.

### Box 2.1 — Probit estimates (selected coefficients; Estimation period: 1991–2018)
- Number of Observations:
  - Full Sample columns: 1,838 and 1,828.
  - EMDE Sample columns: 1,014 and 1,004.
- Probit coefficients (selected; significance markers preserved as in source):
  - NIIP/GDP: Full Sample: –0.27* ; EMDE Sample: –0.58**
  - Debt Assets: Foreign Currency/GDP: Full Sample: 0.40 ; EMDE Sample: –0.13
  - Debt Liabilities: Foreign Currency/GDP: Full Sample: 0.44*** ; EMDE Sample: 1.78***
  - Debt Liabilities: Domestic Currency/GDP: Full Sample: 0.75** ; EMDE Sample: 1.32
  - Equity Assets/GDP: Full Sample: 0.34 ; EMDE Sample: –0.52
  - Equity Liabilities/GDP: Full Sample: –0.66*** ; EMDE Sample: –0.56
  - FX Reserves/GDP: Full Sample: –5.22*** ; EMDE Sample: –5.47***
  - Current Account/GDP: Full Sample: –5.45***–4.61*** ; EMDE Sample: –6.89***–5.10***
  - Global Risk Aversion (VXO): Full Sample: 0.02**0.02** ; EMDE Sample: 0.02***0.02***
  - Constant: Full Sample: –0.11–0.61** ; EMDE Sample: –0.67**–1.24***
- Note: Country-specific variables are lagged by one year. EMDE = emerging market and developing economies; FX = foreign exchange; NIIP = net international investment position; VXO = Chicago Board Options Exchange Volatility Index.

### Box 2.1 — Predicted probabilities and economic significance (selected)
- Method: hold variables at sample means and change variables of interest to reveal nonlinear effects.
- EMDE sample marginal effects:
  - Foreign-currency-denominated debt liabilities:
    - Increase from 40 percent of GDP to 60 percent of GDP → predicted probability of external stress increases by 5 percentage points (EMDE sample).
    - Same rise in full sample → predicted probability increases by 0.2 percentage points.
  - Current account balance:
    - Decline from a surplus of 5 percent of GDP to a deficit of 5 percent of GDP → predicted probability of external stress increases by 5.3 percentage points (EMDE sample).
    - Same change in full sample → probability rises by 1.1 percentage points.
  - Official foreign exchange reserves (nonlinear effects):
    - Predicted external stress probability near zero when reserves are above 40 percent of GDP.
    - Decline in reserves from 20 percent to 10 percent of GDP → predicted external stress probability increases by 6.5 percentage points (EMDE sample).
    - Further decline from 10 percent to 0 percent of GDP → predicted external stress probability increases by an additional 12.6 percentage points (EMDE sample).
    - Corresponding values for entire sample: 0.7 percent and 2.1 percent, respectively.
- Interaction effects and heterogeneity:
  - A combination of two or more external vulnerabilities greatly increases the probability of external stress for EMDEs.
  - Example: With foreign-currency-denominated debt at 40 percent of GDP, predicted probability ranges from 2–12 percent depending on whether reserves and current account are at the 75th percentile or 25th percentile of the sample.
- Under heightened global risk aversion:
  - When global risk aversion reaches peak values seen during the global financial crisis or the Great Lockdown, the predicted external stress episode probability for an EMDE with an average level of preexisting vulnerabilities rises to about 40 percent—more than double the estimated probability for less vulnerable EMDEs.

### Box 2.1 — External stress drivers over time (EMDE configuration)
- 1996: external risks mostly associated with low foreign exchange reserves and, to a lesser extent, large current account deficits.
  - Not in vulnerable zone: 32 (percent of sample)
- 2007: external risks mainly reflected current account deficits and, to a lesser extent, foreign-currency-denominated debt liabilities; low reserves were less of a vulnerability.
  - Not in vulnerable zone: 52
- 2018: elevated foreign-currency-denominated debt liabilities became a central vulnerability for EMDEs but were often mitigated by relatively small current account deficits and relatively high foreign exchange reserves.
  - Not in vulnerable zone: 50

### Consequences for debtor economies (local projections)
- Method: Local projections following Jordà (2005); countries classified by preexisting vulnerabilities (foreign-currency debt, current account deficits, foreign exchange reserves).
- Output loss dynamics (EMDEs):
  - Vulnerable economies: output loss within first two years ~4.1 percent.
  - Less vulnerable economies: estimated loss within first two years ~1 percent.
  - Five years after episode: vulnerable economies output loss ~2.6 percent; less vulnerable economies recover to pre-episode GDP levels within five years.
- Exchange rate and current account responses for high-vulnerability countries:
  - Real effective exchange rate: depreciates by about 10 percent within the first year.
  - Current account balance: rises by more than 2.5 percent of GDP within the first year.
  - Movements are much smaller for less vulnerable economies.

### Consequences for creditor economies
- Creditor losses channels: adverse exchange rate movements, lower asset and bond prices, valuation changes, debt restructuring and write-offs.
- Historical evidence: Creditor advanced economies (for example, Belgium, Denmark, Germany, Sweden, Switzerland) suffered banking crises in 2008 partly due to exposures to distressed assets in debtor economies.
- Aggregate valuation analysis: Valuation effects in net IIP estimated as difference between annual change in net IIP and financial account flows (residual approach). Results indicate sustained valuation losses for countries with persistent current account surpluses in the aftermath of the global financial crisis that were not present in the precrisis period.

### Ratio of Current Account Balance to GDP (valuation changes)
- Precrisis period: no systematic pattern between sustained current account balances and valuation gains or losses.
- Post-global-financial-crisis period: relationship is negative and statistically significant.
  - Estimated slope coefficient: –0.5. Interpretation: a sustained current account surplus of 2 percent of GDP led, on average, to a valuation loss of 1 percent of GDP a year.
- Implication: in countries with sustained current account surpluses, the net IIP increases by less than would be expected from the cumulative current account balances.
- NIIP valuation change definition: {(change of total asset – net acquisition of asset) – (change of total liabilities – net incurrence of liabilities)} / GDP.
- FX-related NIIP valuation change formula: –(net foreign exchange share in GDP × percent change in real effective exchange rate).
- Sample note: includes all External Sector Report economies excluding Hong Kong SAR, Saudi Arabia, and Singapore.

### Implications for the outlook and policies
- For debtor economies:
  - Ongoing global financial stress has increased the probability of experiencing external stress with either a debt default, debt restructuring, or the need for IMF financial support.
  - Economies most at risk are likely to be emerging market and developing economies with preexisting vulnerabilities: relatively high levels of foreign currency external debt, large current account deficits, and relatively low levels of international official reserves.
  - During spikes in global risk aversion, the overall risk of an external stress episode for such economies is several times greater than for EMDEs with limited preexisting vulnerabilities.
  - Macroeconomic consequences (lost real GDP, sharpness of current account and REER adjustment) are likely to be significantly greater for economies with greater preexisting vulnerabilities.
  - The rise in debt ratios and fall in foreign exchange reserves in a number of EMDEs could increase the near-term likelihood of external stress episodes.
  - COVID-19 crisis adds unique additional risk factors: evolution of the pandemic; sharp terms-of-trade movements; disruptions to economic activity, trade, travel, and remittances; and implications for net exporters of commodities and tourism.
- For creditor economies:
  - Running large and persistent current account surpluses can come with potential valuation losses in the aftermath of large systemic crises.
  - Countries with large current account surpluses entering a crisis may be at negligible risk of an external crisis themselves but may experience IIP valuation losses from exposures to distressed assets or markets.

### Policy recommendations and monitoring
- Limit buildup of external vulnerabilities by monitoring various components of external flows and the IIP.
- When financing priority investment through external public and private sector debt, limit the foreign-currency-denominated component and currency mismatches by maintaining adequate buffers in the form of official and private sector reserves, even when accumulation of foreign assets may carry the risk of valuation losses.
- Increased foreign ownership of domestic currency debt can reduce borrowing costs but may increase price volatility where domestic markets lack depth.
- Monitoring currency mismatches requires timely data on the currency composition of external assets and liabilities; further efforts are needed to compile official data on currency composition.
- IMF staff already factor excessive IIP and financing risk considerations when assessing external positions; the chapter results can further inform the external sector assessment process.
- The potential risks and costs associated with both large creditor and debtor positions provide additional reason to avoid excessive and persistent current account imbalances over the medium term.

*Source: ch2 - Introduction (IMF 2020).*

### Introduction

### ch2 - Introduction

### Background and motivation
- External assets and liabilities more than tripled as a share of GDP from the early 1990s to the years preceding the Great Lockdown.
- The sharp increase in gross and net external positions—referred to as the rise of “stock imbalances”—raises sustainability questions for debtor economies and macrofinancial vulnerabilities to domestic and global shocks.
- The initial tightening in global financial conditions and large terms-of-trade fluctuations caused by the coronavirus (COVID-19) outbreak and the Great Lockdown led to sharp currency and current account movements; in most cases the exchange rate acted as a shock absorber, while a few countries resorted to foreign exchange intervention and capital flow management measures.

### Research questions
- How do the size and composition of the various types of external assets and liabilities relate to the risk of external stress episodes? Is the relationship for emerging market and developing economies different from that for other (advanced) economies?
- What is the role of other well-studied variables, such as the level of global financial risk aversion and external current account balances, in explaining the likelihood of external stress episodes? How do these factors combine with the structure of the IIP in amplifying or mitigating risks?
- When an external stress event occurs, how does the size and composition of the IIP relate to the impact on output, the current account, and the exchange rate? How do external stress events impact creditor economies?

### Data, sample, and definitions
- Sample: 73 advanced and emerging market and developing economies during 1991–2018.
- External stress episode definition: an event that involves either a sovereign external debt default, debt restructuring, or recourse to an IMF arrangement. Sources used include updated Das, Papaioannou, and Trebesch (2011); Asonuma and Trebesch (2016); and Paris Club reports.
- Currency-denomination data for external debt assets and liabilities available starting in 1991.
- Large external crises subset: cases of IMF financial assistance exceeding 200 percent of quota.

### Main findings (high-level)
- Not all IIP components relate equally to the likelihood of external stress episodes.
- The net IIP declines in the run-up to an external stress episode; the more negative it becomes, the greater the likelihood of external stress materializing.
- Gross external debt liabilities are stronger predictors of external stress than equity liabilities or private external debt assets.
- Larger stock of foreign official reserves mitigates the likelihood of an external stress episode, although with diminishing effects.
- For the full sample, external debt liabilities are strong predictors of stress irrespective of currency denomination; foreign-currency-denominated debt liabilities are particularly relevant predictors for emerging market and developing economies.
- Traditional predictors remain important: large current account deficits and higher levels of global risk aversion increase external financing risks.
- The nature of external vulnerabilities has rotated over time: before the Asian financial crisis low reserves were central; ahead of the global financial crisis large current account deficits were central; in the years preceding the Great Lockdown elevated gross external debt liabilities and their foreign-currency-denominated component were central vulnerabilities for emerging market and developing economies.
- Preexisting external vulnerabilities amplify macroeconomic costs of external stress episodes.

### Key quantitative impacts of external stress episodes
- Identified external stress cases: 128.
- Real GDP: For countries with large current account deficits, elevated foreign-currency-denominated debt, and low levels of reserves, real GDP falls by about 4.1 percent within two years of an external stress episode; for countries with more limited external vulnerabilities, the decline in real GDP levels is typically about 1 percent.
- Real effective exchange rate: depreciates by about 10 percent within the first year of an external stress episode in countries with high preexisting vulnerabilities.
- Current account balance: rises by more than 2 percent of GDP within the first year of an external stress episode in countries with high preexisting vulnerabilities.
- Creditor economies: following large global crises, on average in the decade after the global financial crisis, a 1 percent of GDP rise in the current account surplus has been associated with a 0.5 percent of GDP valuation loss in creditor economies.

### International Investment Position (IIP) dynamics around stress episodes
- Event-study controls: country and time fixed effects to capture differences in country average IIP levels and common shocks.
- Run-up to stress episode:
  - Net IIP declines, driven predominantly by a sharp rise in foreign-currency-denominated external debt liabilities as a share of GDP (partly reflecting currency depreciation dynamics).
  - Domestic-currency-denominated debt liabilities also increase ahead of the stress episode, but by a smaller magnitude.
  - Equity assets and liabilities decline gradually.
  - Foreign-currency-denominated external debt assets increase; private foreign-currency-denominated external debt assets increase ahead of the stress episode (likely reflecting private capital flight and valuation effects).
  - Official foreign exchange reserves decline sharply just ahead of the stress episode.
- After onset of stress:
  - Net IIP typically rises, driven primarily by a significant drop in foreign-currency-denominated external debt liabilities (deleveraging and restructuring).
  - Official foreign exchange reserves typically decline in the aftermath and then bounce back.
- For large external crises, these dynamics are starker: larger drops in net IIP ahead of crisis, larger rises in foreign-currency-denominated debt liabilities, and sharper declines in gross equity and official reserve assets that rebound to levels well below precrisis peaks.

### Estimating external stress probabilities: methodology
- Approach: pooled probit model where the dependent variable equals 1 for a stress episode in a given country-year and 0 otherwise.
- Explanatory variables include:
  - IIP components (gross and net), including currency denomination of assets and liabilities.
  - Current account balance.
  - Global risk aversion.
  - Real effective exchange rate gap (deviations from average of previous five years).
  - Income per capita relative to the United States.
  - Credit gap (constructed analogously to the real exchange rate gap).
  - Degree of financial development (index including market depth, access, and efficiency).
- Sample and specification broadly similar to Catão and Milesi-Ferretti (2014) with extensions for currency denomination.

### Estimation results (summary)
- Lower net IIP (larger net debtor position) is associated with higher probability of external stress.
- Disaggregated IIP findings:
  - Both higher foreign-currency and domestic-currency external debt liabilities increase the probability of external stress events.
  - Gross positions (assets versus liabilities) provide useful information beyond net positions; coefficients for symmetric external debt categories differ for assets and liabilities.
  - Higher levels of foreign exchange reserves reduce the occurrence of stress episodes.
  - Private external debt assets do not appear to play a mitigating role, possibly reflecting capital flight dynamics that often accompany deteriorating conditions.

### Chapter organization and further analysis
- The chapter proceeds with:
  1. Empirical patterns of the main IIP components around external stress episodes.
  2. Results from estimating an external stress probability model, focusing on IIP components and combinations of vulnerabilities.
  3. Computation of costs for debtor and creditor economies after external stress episodes.
  4. Conclusions summarizing implications for outlook and risks.

*Source: ch2 - Introduction (IMF 2020).*

### Box 2.1 presents work by IMF staff on predicting external crises

### Box 2.1 presents work by IMF staff on predicting external crises

### Overview and methodology
- Objective: Predict external crises using alternative definitions, including sudden stop episodes with high growth impact and exchange rate market pressure episodes.
- Techniques: Signal extraction and machine-learning techniques; probit estimation for external stress events; VXO (precursor of the VIX) used to maximize data coverage as a proxy for global risk aversion.
- Sample and period: Estimation period: 1991–2018. Analysis covers a full sample (advanced + emerging market and developing economies) and an EMDE (emerging market and developing economies) subsample.
- Controls and data notes: Country-specific variables are lagged by one year. The current account/GDP is included as a two-year moving average. Additional controls include the credit gap, the real effective exchange rate gap, income per capita relative to the United States, and a financial development index. Data limitations preclude inclusion of additional countries and limit breakdowns (for example, currency composition of short-term external debt).

### Key predictors and statistical findings
- Central finding: External debt is a strong predictor of external stress episodes; stock vulnerabilities such as external debt measures are reliable predictors across crisis types.
- Other significant predictors:
  - Larger current account deficits are associated with higher external stress.
  - Global risk aversion increases the likelihood of external stress, indicating global “push” factors matter.
  - Higher levels of foreign exchange reserves play a mitigating role.
- Differences between samples:
  - Foreign-currency-denominated debt liabilities have a statistically significant relationship with external stress risk for EMDEs, whereas domestic-currency-denominated debt liabilities do not (in EMDEs).
  - Private external debt assets denominated in foreign currency reduce the probability of a stress episode in EMDEs.
  - Equity assets and liabilities and external debt assets denominated in domestic currency do not play a statistically significant role in EMDEs.

### Probit estimates (Estimation period: 1991–2018)
- Dependent variable: probability of external stress event (0/1; probit). Probit coefficients presented below. Significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1.
- Number of Observations: Full Sample columns: 1,838 and 1,828; EMDE Sample columns: 1,014 and 1,004.
- Coefficients reported (probit coefficients):
  - NIIP/GDP: Full Sample: –0.27* ; EMDE Sample: –0.58**
  - Debt Assets: Foreign Currency/GDP: Full Sample: 0.40 ; EMDE Sample: –0.13
  - Debt Assets: Domestic Currency/GDP: Full Sample: –0.27 . . .
  - Debt Liabilities: Foreign Currency/GDP: Full Sample: 0.44*** ; EMDE Sample: 1.78***
  - Debt Liabilities: Domestic Currency/GDP: Full Sample: 0.75** ; EMDE Sample: 1.32
  - Equity Assets/GDP: Full Sample: 0.34 ; EMDE Sample: –0.52
  - Equity Liabilities/GDP: Full Sample: –0.66*** ; EMDE Sample: –0.56
  - FX Reserves/GDP: Full Sample: –5.22*** ; EMDE Sample: –5.47***
  - Current Account/GDP: Full Sample: –5.45***–4.61*** ; EMDE Sample: –6.89***–5.10***
  - Global Risk Aversion (VXO): Full Sample: 0.02**0.02** ; EMDE Sample: 0.02***0.02***
  - Constant: Full Sample: –0.11–0.61** ; EMDE Sample: –0.67**–1.24***
- Note: Country-specific variables are lagged by one year. EMDE = emerging market and developing economies; FX = foreign exchange; NIIP = net international investment position; VXO = Chicago Board Options Exchange Volatility Index.

### Predicted probabilities and economic significance
- Approach: Compute predicted probabilities by holding all variables at sample means and changing a variable of interest in specified increments to reveal nonlinear effects.
- Selected marginal effects for EMDE sample (economic significance):
  - Foreign-currency-denominated debt liabilities:
    - Increase from 40 percent of GDP to 60 percent of GDP → predicted probability of external stress increases by 5 percentage points (EMDE sample).
    - Same rise in full sample → predicted probability increases by 0.2 percentage points.
  - Current account balance:
    - Decline from a surplus of 5 percent of GDP to a deficit of 5 percent of GDP → predicted probability of external stress increases by 5.3 percentage points (EMDE sample).
    - Same change in full sample → probability rises by 1.1 percentage points.
  - Official foreign exchange reserves (nonlinear effects):
    - Predicted external stress probability near zero when reserves are above 40 percent of GDP.
    - Decline in reserves from 20 percent to 10 percent of GDP → predicted external stress probability increases by 6.5 percentage points (EMDE sample).
    - Further decline from 10 percent to 0 percent of GDP → predicted external stress probability increases by an additional 12.6 percentage points (EMDE sample).
    - Corresponding values for entire sample: 0.7 percent and 2.1 percent, respectively.
- Interaction effects and heterogeneity:
  - A combination of two or more external vulnerabilities greatly increases the probability of external stress for EMDEs.
  - Example: With foreign-currency-denominated debt at 40 percent of GDP, predicted probability ranges from 2–12 percent depending on whether reserves and current account are at the 75th percentile or 25th percentile of the sample.
- Implication under heightened global risk aversion:
  - When global risk aversion reaches peak values seen during the global financial crisis or the Great Lockdown, the predicted external stress episode probability for an EMDE with an average level of preexisting vulnerabilities rises to about 40 percent—more than double the estimated probability for less vulnerable EMDEs.

### External stress drivers over time (configuration in EMDEs)
- Before the Asian financial crisis (1996): external risks mostly associated with low foreign exchange reserves and, to a lesser extent, large current account deficits.
- At the onset of the global financial crisis (2007): external risks mainly reflected current account deficits and, to a lesser extent, foreign-currency-denominated debt liabilities; low reserves were less of a vulnerability.
- Latest period (2018): elevated foreign-currency-denominated debt liabilities became a central vulnerability for EMDEs but were often mitigated by relatively small current account deficits and relatively high foreign exchange reserves.
- Venn-diagram summary (percent of sample in vulnerable zone vs not):
  - 1996: Not in vulnerable zone: 32 (percent of sample)
  - 2007: Not in vulnerable zone: 52
  - 2018: Not in vulnerable zone: 50

### Consequences for debtor economies
- Method: Local projections following Jordà (2005) to estimate dynamic responses of real GDP, real effective exchange rate (REER), and current account balance; countries classified by preexisting vulnerabilities (foreign-currency debt, current account deficits, foreign exchange reserves).
- Output loss dynamics (EMDEs):
  - Vulnerable economies: output loss within first two years ~4.1 percent.
  - Less vulnerable economies: estimated loss within first two years ~1 percent.
  - Five years after episode: vulnerable economies output loss ~2.6 percent; less vulnerable economies experience recovery to pre-episode GDP levels within five years.
- Exchange rate and current account responses (vulnerable vs less vulnerable):
  - Real effective exchange rate: depreciates by about 10 percent within the first year for high-vulnerability countries.
  - Current account balance: rises by more than 2.5 percent of GDP within the first year for high-vulnerability countries.
  - Movements are much smaller for less vulnerable economies.

### Consequences for creditor economies
- Creditor losses channels: adverse exchange rate movements, lower asset and bond prices, valuation changes, debt restructuring and write-offs.
- Historical evidence: Creditor advanced economies (for example, Belgium, Denmark, Germany, Sweden, Switzerland) suffered banking crises in 2008 partly due to exposures to distressed assets in debtor economies.
- Aggregate valuation analysis: Valuation effects in net IIP estimated as difference between annual change in net IIP and financial account flows (residual approach). Results indicate sustained valuation losses for countries with persistent current account surpluses in the aftermath of the global financial crisis that were not present in the precrisis period.

*Source: IMF staff estimates.*

### 3. Ratio of Current Account Balance to GDP

### 3. Ratio of Current Account Balance to GDP

### Relationship between current account balances and valuation changes
- In the precrisis period, there is no systematic pattern: sustained valuation gains or losses were not related to average current account balances.
- In the post-global-financial-crisis period (which also includes the euro area sovereign debt crisis of 2010), the relationship is negative and statistically significant.
  - Estimated slope coefficient: –0.5. Interpretation given: a sustained current account surplus of 2 percent of GDP led, on average, to a valuation loss of 1 percent of GDP a year.
- Implication: in countries with sustained current account surpluses, the net IIP increases by less than would be expected from the cumulative current account balances.
- For the pre-global-financial-crisis period, the coefficient is near zero and not statistically significant.

### Role of valuation effects and drivers
- The stabilizing role of valuation effects in the net IIP is especially strong after large systemic crises.
- Valuation gains can reflect adverse macroeconomic and financial factors (example: euro area debtor economies, including Italy and Spain, generally experienced valuation gains following the global financial crisis; Greece and Portugal experienced large valuation gains that intensified after the euro area sovereign debt crisis).
- Valuation losses can result from relatively strong underlying fundamentals (example: since 2008 the United States has seen valuation losses despite current account deficits, driven by (1) an appreciation of the US dollar reducing the value of US external assets denominated in foreign currency, and (2) better performance of equity valuations compared with peers that raises the value of US foreign equity liabilities and lowers the net IIP).
- Figure 2.7 finds no systematic relationship between current account balances and valuation changes resulting from exchange rates across the two subperiods.
  - This suggests that factors linked to bond and asset price differentials, debt restructuring, and debt write-offs are driving the valuation effects more than exchange rate fluctuations, when averaged over long periods.

### Empirical details and data notes
- NIIP valuation change definition: {(change of total asset – net acquisition of asset) – (change of total liabilities – net incurrence of liabilities)} / GDP.
- FX-related NIIP valuation change formula given: –(net foreign exchange share in GDP × percent change in real effective exchange rate).
- The ratio of valuation changes to nominal GDP is estimated by converting both measures to US dollars, following the literature. Results are noted to be quite similar when both measures are converted to domestic currency, particularly for economies with sustained current account surpluses.
- Sample note: includes all External Sector Report economies excluding Hong Kong SAR, Saudi Arabia, and Singapore.

### Implications for the outlook and policies
- For debtor economies:
  - Ongoing global financial stress has increased the probability of experiencing external stress with either a debt default, debt restructuring, or the need for IMF financial support.
  - Economies most at risk are likely to be emerging market and developing economies with preexisting vulnerabilities: relatively high levels of foreign currency external debt, large current account deficits, and relatively low levels of international official reserves.
  - During spikes in global risk aversion, the overall risk of an external stress episode for such economies is several times greater than for EMDEs with limited preexisting vulnerabilities.
  - Macroeconomic consequences (lost real GDP, sharpness of current account and REER adjustment) are likely to be significantly greater for economies with greater preexisting vulnerabilities.
  - The rise in debt ratios and fall in foreign exchange reserves in a number of EMDEs could increase the near-term likelihood of external stress episodes.
  - COVID-19 crisis adds unique additional risk factors: evolution of the pandemic; sharp terms-of-trade movements; disruptions to economic activity, trade, travel, and remittances; and implications for net exporters of commodities and tourism.
- For creditor economies:
  - Running large and persistent current account surpluses can come with potential valuation losses in the aftermath of large systemic crises.
  - Countries with large current account surpluses entering a crisis may be at negligible risk of an external crisis themselves but may experience IIP valuation losses from exposures to distressed assets or markets.
- Policy recommendations and monitoring:
  - Limit buildup of external vulnerabilities by monitoring various components of external flows and the IIP.
  - When financing priority investment through external public and private sector debt, limit the foreign-currency-denominated component and currency mismatches by maintaining adequate buffers in the form of official and private sector reserves, even when accumulation of foreign assets may carry the risk of valuation losses.
  - Increased foreign ownership of domestic currency debt can reduce borrowing costs but may increase price volatility where domestic markets lack depth.
  - Monitoring currency mismatches requires timely data on the currency composition of external assets and liabilities; further efforts are needed to compile official data on currency composition.
  - IMF staff already factor excessive IIP and financing risk considerations when assessing external positions; the chapter results can further inform the external sector assessment process.
  - The potential risks and costs associated with both large creditor and debtor positions provide additional reason to avoid excessive and persistent current account imbalances over the medium term.

### Box 2.1 — Robustness and drivers of external crisis predictions
- Complementary episode definitions studied: 
  - Sudden stops with growth impact (SSGIs): episodes where net private capital inflow as a percentage of GDP is at least 2 percentage points lower than in the two previous years with large multilateral support.
  - Exchange market pressure events (EMPEs): episodes where the weighted average of the annual percentage depreciation in the nominal exchange rate and the annual decline in reserves as a percentage of the previous year’s GDP is below the 15th percentile of the worldwide pooled sample, with large multilateral support.
- Methods:
  - Signal extraction methods used as a starting point for predicting external crises; they calculate a threshold for each variable separately.
  - Machine-learning techniques (tree-based models, random forest, RUSBoost) explored to uncover nonlinearities and interactions.
- Data and variables:
  - About 80 predictive indicators considered, covering (1) policy regimes, (2) imbalances and mismatches, (3) asset price booms and busts, (4) global liquidity and contagion, and (5) political shocks.
  - Table 2.1.1 lists extensive variables, including fiscal balance/GDP; Reserves/M2; Real GDP growth; Current account balance/GDP; Amortization/exports; FX share of external debt; External debt/GDP; Private external debt/GDP; Reserves/imports; VIX; TED spread; Federal funds rate; Real stock price growth; REER growth; and many more.
- Main results on predictors:
  - Stock vulnerabilities are generally reliable predictors of external crises; ranking and importance of interactions vary across crisis categories and country groupings.
  - SSGIs in emerging market economies are well predicted by signal extraction methods. Top predictors include debt liabilities and asset price and credit bubbles, plus global factors (including the TED spread, the incidence of financial crisis in advanced economies, and interbank liabilities to banks in these advanced economies), medium-term bubbles (stock prices, house prices, REER), and external debt measures (scheduled amortization, cross-border interbank debt).
  - EMPEs in emerging market economies are better predicted by machine learning techniques; best predictors span several crisis generations, including reserve adequacy metrics, measures of equity outflows, fiscal vulnerabilities (EMBI sovereign spread, change in public debt), and competitiveness indicators (cumulative inflation).
  - EMPEs in advanced economies are well predicted by signal extraction techniques; most important predictors are indicators of external debt (private external debt, amortization, foreign currency and external shares of public debt).
  - EMPEs in low-income countries may be better predicted either by signal extraction or by machine learning, depending on context.
- Estimation notes and caveats:
  - Models estimated with data from 1990 onward; best-performing model presented with out-of-sample testing between 2008–17.
  - Missing variables imputed using a machine-learning-based surrogate technique.
  - Variable importance rankings are subject to caveats: random seed effects in machine learning, sensitivity to variable subsets, and differences between in-sample and out-of-sample rankings.

*Source: Chapter 2, “External Stress and the International Investment Position,” 2020 External Sector Report, International Monetary Fund.*

### Box 2.1 (continued)

### Box 2.1 (continued)

### Key findings on predictors of external and financial crises
- When foreign currency share data are included:
  - Net open foreign currency share measures are important.
  - Other important predictors include indicators of first-generation currency crises: cumulative inflation, fiscal vulnerabilities, exchange rate regime.
  - Banking system health indicators matter: share of non-investment-grade debt, capital-to-assets ratio.
  - For countries where it is available, stock market overvaluation (price-to-earnings ratio) is important.
- When foreign currency share data are not available:
  - Machine-learning methods deliver superior performance.
  - In addition to the above variables, global factors are identified as important, including TED spread and US term premium.

### Top predictive variables (Figure 2.1.1)
- The figure presents top predictive variables for four crisis categories:
  1. EMPEs in EMs
  2. EMPEs in LICs
  3. EMPEs in AEs
  4. SSGIs in EMs
- Selected variables listed in the figure (preserved as in source):
  - EMBI sovereign spread
  - Five-year inflation
  - Inflation
  - Change in reserves to GDP
  - Change in public debt to GDP
  - Reserves to GDP
  - Change in external equity liabilities to GDP
  - Interest coverage ratio
  - Change in foreign liabilities to domestic credit
  - Export growth
  - External debt to GDP
  - Reserves to imports
  - Five-year change in private credit to GDP
  - Five-year change in broad money to GDP
  - GDP growth
  - Price-earnings ratio
  - Share of non-investment grade debt
  - Net open FX position to GDP
  - Fiscal balance to GDP
  - Net open debt FX position to GDP
  - Change in public debt to GDP
  - Hard peg indicator
  - Banks’ capital-asset ratio
  - Change in external debt to GDP
  - Primary gap to GDP
  - Change in nonbank pri. ext. debt to GDP
  - FX share of external debt
  - Change in private external debt to GDP
  - Private external debt to GDP
  - FX share of government Debt
  - Change in public external debt to GDP
  - Amortization to exports
  - Nonbank private external debt to GDP
  - CA balance to GDP
  - Public external debt to GDP
  - Five-year change in inter-bank ext. debt to GDP
  - Loan-deposit ratio
  - Five-year change in construction contribution to GDP
  - Five-year stock price growth
  - TED spread
  - Interbank liabilities to AEs in fin. crisis
  - House price acceleration
  - Five-year house price growth
  - AEs’ banking crisis frequency
  - Debt service to exports
  - Interbank external debt to GDP
  - Foreign liabilities to domestic credit
  - Five-year change in interbank external debt to GDP
  - Change in private credit to GDP
  - Five-year REER appreciation
  - Five-year change in private credit to GDP

### Notes on variable importance metrics
- For the signal extraction model, the metric is the weight of the variable.
- For the machine-learning model, the metric is the percentage of in-sample variation in the sum of errors explained by removal of the variable from the model-generated trees.

*Source: IMF staff calculations.*

---


_Source: https://www.imf.org/-/media/files/publications/esr/2020/english/ch2.pdf_