2013年-IMF国际货币组织全球_External_Liabilities_and_Crises_37页_1mb

Summary of "External Liabilities and Crises" by Luis A. V. Catão and Gian Maria Milesi-Ferretti

Core Content

This working paper by Luis A. V. Catão and Gian Maria Milesi-Ferretti investigates the determinants of external crises, with a focus on the role of net foreign liabilities (NFL) and their composition. The authors analyze a dataset spanning 1970–2011 for 70 countries, including 41 emerging markets, to determine how various financial indicators predict the likelihood of external crises.

Main Findings

• Net Foreign Liabilities (NFL) as a Crisis Predictor: The ratio of NFL to GDP is a significant indicator of external crisis risk. The risk increases sharply when this ratio exceeds 50 percent of GDP or rises by 20 percentage points above the country-specific historical mean.

• Composition of NFL Matters: Among the components of NFL, net external debt is the most critical factor in raising crisis risk. The effects of portfolio equity liabilities are more mixed and generally weaker, while foreign direct investment (FDI) liabilities seem to have a mitigating effect on crisis risk.

• Current Account as a Strong Predictor: The current account is a powerful predictor of external crises. Its unconditional levels have greater predictive power than deviations from a "norm" in most specifications. Crisis-stricken countries typically have current account deficits around 4 percent of GDP, which worsen before the crisis and then sharply reverse.

• Foreign Exchange Reserves as a Protective Factor: Higher foreign exchange reserves reduce the likelihood of external crises more than other foreign asset holdings, consistent with the idea of reserves as a precautionary tool.

• Predictive Power of a Multivariate Model: A parsimonious multivariate probit model incorporating NFL, current account, FX reserves, and other controls has strong in-sample and out-of-sample predictive performance, particularly for the 2008–2011 crises.

• Robustness to Variable Omission and Sample Definition: The model's predictive power remains robust even when excluding certain variables or when using different crisis definitions. The focus on external crises stricto sensu (including defaults, rescheduling, and large IMF support) is crucial for the model's effectiveness.

• Limitations of Earlier Models: The paper finds that many variables previously used in crisis early warning systems (EWS) do not add significant explanatory or predictive power, suggesting the need for a more focused approach.

Key Variables and Their Impact

• Net Foreign Liabilities (NFL): A key determinant of crisis risk, especially when it exceeds 50% of GDP.
• Net External Debt: Strongly associated with increased crisis risk.
• Portfolio Equity and FDI Liabilities: Mixed or mitigating effects on crisis risk.
• Foreign Exchange Reserves: Strongly protective against external crises.
• Current Account Deficits: A consistent and powerful predictor of external crises.
• Real Exchange Rate Gaps: Significant precursors to external crises, showing a pattern of appreciation followed by depreciation.
• Global Financial Conditions: Indicators such as the VIX index and interest rate spreads between AAA and BAA-rated U.S. corporates are relevant to crisis prediction.

Methodology and Data

• Data Sample: 70 countries (41 emerging) from 1970–2011.
• Crisis Definition: Includes defaults, rescheduling, and large IMF support. The authors exclude post-crisis periods up to the year before market re-entry to avoid feedback effects.
• Model Selection: The authors use the ROC curve to select the best model, focusing on the trade-off between true and false positives.
• Event Analysis: The paper uses treatment effect regressions to compare pre- and post-crisis dynamics of variables, controlling for fixed and time effects.

Implications

• The findings have important implications for fiscal and macro-prudential policies, as they highlight the role of external liabilities in crisis risk.
• The results also inform country risk assessment and international financial architecture discussions, emphasizing the need for a broader set of controls and a more precise crisis definition.
• The focus on external crises stricto sensu helps address the limitations of earlier models, particularly their poor out-of-sample performance.

Conclusion

The paper contributes to the literature on crisis early warning systems by emphasizing the importance of NFL composition, current account imbalances, and foreign exchange reserves in predicting external crises. It underscores the value of a multivariate probit model that incorporates these variables for improved predictive accuracy, especially in the context of recent global financial crises.