## EXTERNAL STRESS AND THE INTERNATIONAL INVESTMENT POSITION — Online Annex 2.1. Additional Details on Empirical Analysis

## Source details

**Canonical URL:** [EXTERNAL STRESS AND THE INTERNATIONAL INVESTMENT POSITION — Online Annex 2.1. Additional Details on Empirical Analysis](https://www.imf.org/-/media/files/publications/esr/2020/english/onlineannex21.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/esr/2020/english/onlineannex21.pdf.md)
- [Structured JSON version](/-/media/files/publications/esr/2020/english/onlineannex21.pdf.json)

---

### Data Sources
- Sample: 73 advanced and emerging market and developing economies; period considered 1991–2018.
- Primary datasets and measures:
  - International Investment Position (IIP), current account, and nominal GDP: External Wealth of Nations database (Lane and Milesi-Ferretti, 2007).
  - Foreign currency share of external debt: Bénétrix and others (2019); for countries not available in that database, Bénétrix, Lane, and Shambaugh (2015) with constant weights to extend the sample to 2017.
  - Debt and equity definitions: “Net” = asset minus liability positions; “equity” = FDI equity + portfolio equity; “debt” = portfolio debt securities + other investment + FDI debt. FDI debt/equity breakdown supplemented with IMF International Financial Statistics where available.
  - External stress episodes: sovereign debt default or restructuring or an IMF arrangement. Private creditor defaults/restructurings from Asonuma and Trebesch (2016); official restructurings from Paris Club and Das, Papaioannou, and Trebesch (2011).
  - Other variables and sources (as listed): Real effective exchange rate (2010=100) from IMF Information Notice Systems; Income per capita and Fiscal balance from IMF WEO; Credit from BIS and World Development Indicators; Financial development index from Svirydzenka (2016); VXO from Haver Analytics.

- Identified external stress cases and composition:
  - Total cases: 128 external stress episodes.
  - Composition: 111 IMF arrangements, 6 defaults with private creditors, 35 sovereign debt restructurings (some overlap).

### Event Study Analysis (Unconditional Specification)
- Regression specification used (country and time fixed effects included):
  - y_i,t+s = α_i + δ_t + Σ_{s=-5}^{5} β_s D_{i,t+s} + ε_i,t
  - D_{i,t+s} are 11 dummy variables taking value 1 at year t (event year).
- Purpose: track evolution of IIP and components around stress episodes to assess exposures relative to each country’s historical mean.
- Coefficients β_s plotted to capture association between variable movements and external stress episodes.

### Probit Estimates: Results and Robustness
- Model:
  - Pooled probit model to study likelihood of external stress episodes; setup similar to Catão and Milesi-Ferretti (2014).
- Independent variable groups:
  1. IIP components for each country.
  2. Macroeconomic variables (lagged one year): current account balance; real exchange rate gap (deviations from average of previous five years); credit gap (deviations of credit-to-GDP ratio from average of previous five years); level of financial development; income per capita relative to United States (including interaction term).
  3. Global risk variable: VXO (analogous to VIX with longer time series).
- Robustness checks and alternative specifications:
  - Changes in country sample (e.g., exclude oil exporters).
  - Changes in stress episode definition (exclude consecutive years; different IMF arrangement types; include Spain’s European Stability Mechanism program).
  - Inclusion of additional controls (exchange rate regime, capital account openness, terms-of-trade changes).
  - Alternative estimation technique: logit model.
  - Winsorizing independent variables at 1 percent from both tails.
  - Restricting to large crises: IMF loans greater than 200 percent of quota (results robust; reference Catão and Milesi-Ferretti, 2014).
  - Excluding large financial centers (Iceland, Ireland): exclusion of Iceland strengthens coefficients for foreign-currency-denominated debt liabilities relative to entire sample.
  - Market-access considerations:
    - Regressions for periods of market access using Asonuma and Trebesch (2016) data for 50 countries (including 40 emerging market and developing economies); results robust.
    - Removing one, two, three, and four years after the stress event (to address potential postcrisis bias): results robust (and often stronger) for full sample and emerging market and developing economy sample in most cases; some exceptions when three or four years removed for emerging markets and developing economies and when four years removed for full sample (coefficients for foreign-currency-denominated debt liabilities smaller in those specific specifications).
    - Removing consecutive stress-event years: results robust.
  - Overall: main findings generally robust across these alternative specifications; when incorporating methodologies to address “potential postcrisis” bias, results remain robust and often highly statistically significant; in one case estimations show significant but weaker results.

### Measuring the Stress Episode Impact on Output, Current Account, and Exchange Rate Effects: Local Projection Method (LPM)
- LPM specification to estimate dynamic responses of output, REER, and current-account-to-GDP ratio to external stress events:
  - y_{i,t+k} - y_{i,t-1} = α_i + γ_t + β_H D^H_{i,t} Stress_{i,t} + β_O (1 - D^H_{i,t}) Stress_{i,t} + ν X_{i,t} + ε_{i,t}
- Variables and definitions:
  - y_{i,t+k}: natural logarithm of output, natural logarithm of REER, or current-account-to-GDP ratio of country i at time t+k.
  - α_i and γ_t: country and time fixed effects.
  - D^H_{i,t}: dummy = 1 for countries with high foreign currency debt (above sample median), a high current account deficit (below sample median for current account balance), and low foreign currency reserves (below sample median); = 0 otherwise.
  - Stress_{i,t}: dummy = 1 for stress episode in country i at time t; = 0 otherwise.
  - X_{i,t}: additional regressors — two-year lags of change in output, REER, and current account for country i at time t.
  - ε_{i,t}: residual.
- Purpose: estimate heterogeneous dynamic responses depending on preexisting vulnerabilities (high foreign-currency debt, large current-account deficits, low reserves).

### Key Quantitative Values and Sample Notes
- Number of economies in sample: 73.
- Period: 1991–2018.
- Total external stress episodes identified: 128.
  - IMF arrangements: 111.
  - Defaults with private creditors: 6.
  - Sovereign debt restructurings: 35.
- Event-study dummy span: 11 (s = −5 to +5).
- Winsorization in robustness: 1 percent from both tails.
- Large-crisis cutoff: IMF loans greater than 200 percent of quota.
- Sample starting year referenced in robustness discussion: 1990.
- Robustness checks include removing one, two, three, and four years following stress events in some specifications.

*International Monetary Fund | July 2020*

---


_Source: https://www.imf.org/-/media/files/publications/esr/2020/english/onlineannex21.pdf_