A. Event Analysis: Pre- and Post-crisis IIP Dynamics

We start our analysis by studying the evolution of the main IIP components around external crisis episodes. The sample comprises 73 advanced and emerging market economies during 1991–2018.¹⁰ This event-study analysis controls for country and time fixed effects to capture differences in countries’ average IIP levels as well as the influence of common shocks. In line with the analysis in Gourinchas and Obstfeld (2012) and Catão and Milesi-Ferretti (2014), the following specification is used to perform a standard unconditional event analysis:

where α_t and β_t are country and time fixed effects, respectively; D_t+s are dummy variables to identify when the crisis event occurs; and y_it is the IIP component under consideration. The coefficient β_s thus captures the extent to which the IIP component affects the external stress event. ¹¹

As shown in Figure 4, which presents estimates for the evolution of key IIP components (together with 90 percent confidence bands), the deterioration of the net IIP as a share of GDP ahead of an external crisis seems to be driven predominantly by a sharp increase in foreign currency-denominated external debt liabilities, (which in part reflects valuation effects from currency depreciation, and lower real GDP growth in the years preceding the crisis). Meanwhile, foreign-currency external debt assets also rise, reflecting a combination of private capital flight and valuation effects, while foreign exchange reserves decline sharply just ahead of the crisis. ¹² Equities play a mitigating role (in particular the value of equity liabilities falls ahead of crisis) while the rise in domestic-currency external indebtedness is much less pronounced. Following the crisis, net IIP typically rises, driven primarily by a significant drop in foreign-currency-denominated external debt liabilities, which likely reflects the necessary deleveraging and restructuring as well as some positive valuation effects from currency appreciation in the years after the crisis. The decline in foreign official reserves precedes the crisis but continues in its aftermath. Other components exhibit smaller fluctuations or remain essentially flat.

View Full Size

Conditional Mean of Key IIP Components, the REER and Growth Around External Crisis Episodes

Citation: IMF Working Papers 2021, 020; 10.5089/9781513568102.001.A001

Source: Authors’ calculations.Note: The solid blue line corresponds to the mean of all external crisis episodes, while the blue shaded area corresponds to the 90 percent confidence interval. The red line corresponds to the mean of large crisis episodes. An increase (decrease) in the Real Effective Exchange Rate denotes appreciation (depreciation).

• Download Figure
• Download figure as PowerPoint slide

Similar, yet starker, dynamics of IIP components occur for a subsample of large external crises, which involve cases of IMF financial assistance exceeding 200 percent of quota. In these cases, the drop in net IIP is far more pronounced, driven even more important rise in foreign currency-denominated debt liabilities and decline in official reserve assets. That said, while IIP components also rebound more sharply following the crisis, they tend to end well below pre-crisis peaks.

B. Probit Estimation Results

A pooled probit model is used to study how the IIP and its components relate to the likelihood of external crisis events. The empirical approach is similar to Catão and Milesi-Ferretti (2014) and Turrini and Zeugner (2019), and is extended to include the currency denomination of external assets and liabilities. The dependent variable is the occurrence of an external crisis (a value of 1 indicates a crisis episode in a given country and year), and explanatory variables include various IIP components and standard macroeconomic variables, including the current account balance, the real effective exchange rate gap (measured as deviations of the real exchange rate from the average of the previous five years), relative income per capita, and a global risk indicator.¹³ The estimated specification includes additional controls, including (i) the credit gap (constructed analogously to the real exchange rate gap), to measure the role of financial excess; and (ii) the degree of financial development (an index measuring market depth, access, and efficiency, which is also interacted with relative income per capita). to account for differences in the ability to respond to external shocks (see Svirydzenka 2016). The country and time coverages and crisis definition are same as in the event study analysis. All variables, except global risk aversion, are lagged one year to address endogeneity issues.¹⁴ Regression results also report log pseudolikelihood and AUROC estimates to give a sense of goodness of fit and compare across specifications.¹⁵

In line with the literature, a negative net IIP position is associated with a higher likelihood of an external crisis, a result that is robust to the inclusion of additional controls (see Table 1, columns 1 and 5). When disaggregating net IIP into its main net components, we find that while net external debt and net foreign official reserves affect the probability of external crises, the net equity position does not (Table 1, columns 2 and 6). Moreover, when further disaggregating the net external debt components by currency, only the net foreign-currency-denominated debt component becomes significant (Table 1, columns 3 and 7). Finally, when focusing on the gross components of the IIP, foreign-currency denominated debt liabilities and foreign official reserves become the most robust and significant determinants (Table 1, columns 4 and 8).¹⁶ The log-likelihood and AUROC criteria suggest that the most disaggregated model with gross positions and additional controls is the one that fits the data best (Table 1).

^{Table 1.}

Probit Estimates for External Crisis Events

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 All country-specific variables are lagged by one year. Net refers to assets minus liabilities. Net debt, net equity, net foreign debt, and net domestic debt are computed using the underlying gross numbers.

^1/The main regressions use stata command “xtprobit”. Approximate AUROC estimates are calculated using stata command “probit” and then applying “lroc”.

^{Table 1.}

Probit Estimates for External Crisis Events

Variables	1	2	3	4	5	6	7	8
Net IIP, share of GDP	-0.34* (0.19)				-0.27* (0.15)
Debt, share of GDP
Net debt		-0.55*** (0.21)				-0.50*** (0.16)
Net foreign currency debt			-0.61** (0.25)				-0.52*** (0.17)
Net domestic currency debt			0.13 (0.40)				-0.29 (0.36)
Foreign currency debt assets				0.22 (0.33)				0.40 (0.29)
Domestic currency debt assets				-0.24 (0.49)				-0.27 (0.48)
Foreign currency debt liabilities				0.62** (0.28)				0.44** (0.21)
Domestic currency debt liabilities				0.12 (0.39)				0.75* (0.39)
Equity, share of GDP
Net equity		-0.31 (0.23)	-0.29 (0.24)			-0.34 (0.40)	-0.32 (0.39)
Equity assets				-1.99* (1.18)				0.34 (0.76)
Equity liabilities				0.39 (0.40)				-0.66** (0.26)
Foreign exchange reserves, share of GDP		-7.22*** (1.43)	-7.21*** (1.44)	-7.02*** (1.65)		-6.13*** (1.65)	-6.20*** (1.70)	-5.22*** (1.81)
Other variables
Income per capita, relative to US					-3.92*** (1.00)	-2.90** (1.18)	-2.88** (1.18)	-1.28 (1.34)
Current account/GDP, 2yr MA					-5.32*** (1.52)	-5.71*** (1.54)	-5.73*** (1.53)	-6.89*** (1.50)
REER gap, relative to 5yr MA					-0.29 (0.44)	-0.34 (0.42)	-0.34 (0.42)	-0.19 (0.43)
Credit gap, credit/GDP compared to 5yr MA					0.90 (0.70)	1.14 (0.69)	1.12 (0.70)	0.59 (0.59)
Financial development index					-4.96*** (1.19)	-2.87** (1.22)	-2.72** (1.37)	-0.42 (1.34)
Fin. develop. * income per capita relative to US					5.76*** (1.70)	1.80 (2.16)	1.70 (2.21)	-5.03* (3.00)
VXO					0.02** (0.01)	0.02** (0.01)	0.02** (0.01)	0.02** (0.01)
Constant	-1.95*** (0.14)	-1.41*** (0.20)	-1.35*** (0.20)	-1.24*** (0.18)	-0.08 (0.34)	-0.08 (0.36)	-0.10 (0.37)	-0.67** (0.33)
Observations	1,828	1,828	1,828	1,828	1,828	1,828	1,828	1,828
Number of ifs_code	73	73	73	73	73	73	73	73
Log pseudolikelihood	-410.38	-385.51	-384.42	-377.31	-358.33	-342.39	-342.32	-333.29
AUROC estimates 1/	0.66	0.67	0.72	0.82	0.83	0.84	0.84	0.87

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 All country-specific variables are lagged by one year. Net refers to assets minus liabilities. Net debt, net equity, net foreign debt, and net domestic debt are computed using the underlying gross numbers.

^1/The main regressions use stata command “xtprobit”. Approximate AUROC estimates are calculated using stata command “probit” and then applying “lroc”.

View Table

^{Table 1.}

Probit Estimates for External Crisis Events

Variables	1	2	3	4	5	6	7	8
Net IIP, share of GDP	-0.34* (0.19)				-0.27* (0.15)
Debt, share of GDP
Net debt		-0.55*** (0.21)				-0.50*** (0.16)
Net foreign currency debt			-0.61** (0.25)				-0.52*** (0.17)
Net domestic currency debt			0.13 (0.40)				-0.29 (0.36)
Foreign currency debt assets				0.22 (0.33)				0.40 (0.29)
Domestic currency debt assets				-0.24 (0.49)				-0.27 (0.48)
Foreign currency debt liabilities				0.62** (0.28)				0.44** (0.21)
Domestic currency debt liabilities				0.12 (0.39)				0.75* (0.39)
Equity, share of GDP
Net equity		-0.31 (0.23)	-0.29 (0.24)			-0.34 (0.40)	-0.32 (0.39)
Equity assets				-1.99* (1.18)				0.34 (0.76)
Equity liabilities				0.39 (0.40)				-0.66** (0.26)
Foreign exchange reserves, share of GDP		-7.22*** (1.43)	-7.21*** (1.44)	-7.02*** (1.65)		-6.13*** (1.65)	-6.20*** (1.70)	-5.22*** (1.81)
Other variables
Income per capita, relative to US					-3.92*** (1.00)	-2.90** (1.18)	-2.88** (1.18)	-1.28 (1.34)
Current account/GDP, 2yr MA					-5.32*** (1.52)	-5.71*** (1.54)	-5.73*** (1.53)	-6.89*** (1.50)
REER gap, relative to 5yr MA					-0.29 (0.44)	-0.34 (0.42)	-0.34 (0.42)	-0.19 (0.43)
Credit gap, credit/GDP compared to 5yr MA					0.90 (0.70)	1.14 (0.69)	1.12 (0.70)	0.59 (0.59)
Financial development index					-4.96*** (1.19)	-2.87** (1.22)	-2.72** (1.37)	-0.42 (1.34)
Fin. develop. * income per capita relative to US					5.76*** (1.70)	1.80 (2.16)	1.70 (2.21)	-5.03* (3.00)
VXO					0.02** (0.01)	0.02** (0.01)	0.02** (0.01)	0.02** (0.01)
Constant	-1.95*** (0.14)	-1.41*** (0.20)	-1.35*** (0.20)	-1.24*** (0.18)	-0.08 (0.34)	-0.08 (0.36)	-0.10 (0.37)	-0.67** (0.33)
Observations	1,828	1,828	1,828	1,828	1,828	1,828	1,828	1,828
Number of ifs_code	73	73	73	73	73	73	73	73
Log pseudolikelihood	-410.38	-385.51	-384.42	-377.31	-358.33	-342.39	-342.32	-333.29
AUROC estimates 1/	0.66	0.67	0.72	0.82	0.83	0.84	0.84	0.87

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 All country-specific variables are lagged by one year. Net refers to assets minus liabilities. Net debt, net equity, net foreign debt, and net domestic debt are computed using the underlying gross numbers.

^1/The main regressions use stata command “xtprobit”. Approximate AUROC estimates are calculated using stata command “probit” and then applying “lroc”.

Among standard macroeconomic variables, the current account balance (included in the regression as the lagged two year average) and global risk aversion (proxied by the VXO) are the most robust determinants of external crises and are always significant when included in the probit regressions (Table 1, columns 5 to 8). The financial development index is significant when the net IIP components are considered, as well as when the gross IIP components are considered, although it requires interacting financial development with relative income per capita. The real exchange rate and credit gaps appear with the correct sign in the regressions but are not significant in any of the specifications.

The main findings in Table 1, namely the importance of foreign-currency-denominated debt liabilities, foreign official reserves, current account balances and global risk aversion are robust to alternative specifications of the probit baseline regression. When limiting the sample to EMDEs (Table 2, column 2), the impact of foreign-currency denominated debt liabilities increases (consistent with the probability margin exercise in the next subsection), while other components like external debt in domestic currency and equity positions drop in significance. ¹⁷ Meanwhile, foreign official reserves serve a more important mitigating factor in crisis, yet crisis risks rise even more with higher current account deficits and levels of global risk aversion. Finally, the level of financial development and income per capita are not significant in the EMDE sample, likely reflecting the more homogeneous nature of this group. External assets held by the private sector play a limited mitigating role, suggesting these reflect resident capital flight that respond to weak policy frameworks and underlying vulnerabilities. ¹⁸

^{Table 2.}

Robustness Checks for External Crisis Events

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 All country-specific variables are lagged by one year.

^1/The main regressions use stata command “xtprobit”. Approximate AUROC estimates are calculated using stata command “probit” and then applying “lroc”.

^{Table 2.}

Robustness Checks for External Crisis Events

	Sample Variation						Episode Variation						Additional Controls				Estimation method
							Full Sample			EMDE sample
Variables	Baseline	EMDEs only	Exclude oil exporters	Exclude Iceland and Ireland	Exclude GFC (2008)	Include Spain ESM	Large crises only	Crisis over next two years	Exclude consecutive years	Large crises only	Crisis over next two years	Exclude consecutive years	Exchange Rate Regime	Capital Account Openness	Commodity Prices	Fiscal Balance	Winsorize (1 percent both tails)	Log it
Debt, share of GDP
Foreign currency debt assets	0.40 (0.29)	-0.13 (0.47)	0.52 (0.34)	-0.51 (0.42)	0.56** (0.28)	0.37 (0.28)	0.51 (0.38)	0.28 (0.44)	0.50 (0.31)	0.59 (0.56)	-1.06* (0.56)	0.06 (0.45)	0.42 (0.29)	0.30 (0.32)	0.42 (0.30)	0.37 (0.36)	-0.01 (0.35)	0.95 (0.58)
Domestic currency debt assets	-0.27 (0.48)	… …	-0.41 (0.52)	0.37 (0.69)	1.01 (1.11)	-0.73 (0.63)	-1.08* (0.63)	0.25 (0.52)	-0.20 (0.49)	… …	… …	… …	-0.22 (0.51)	-0.50 (0.51)	-0.70 (0.50)	-0.57 (0.60)	0.03 (0.63)	-0.64 (0.97)
Foreign currency debt liabilities	0.44** (0.21)	1.78*** (0.33)	0.44** (0.21)	1.55*** (0.27)	0.63*** (0.22)	0.42** (0.20)	0.41** (0.20)	0.58** (0.29)	0.32* (0.18)	1.09** (0.47)	2.55*** (0.51)	1.52*** (0.31)	0.43** (0.20)	0.49** (0.22)	0.41** (0.20)	0.50** (0.20)	0.95*** (0.26)	0.86** (0.38)
Domestic currency debt liabilities	0.75* (0.39)	1.32 (2.04)	0.81* (0.42)	0.45 (0.44)	0.39 (0.61)	1.10** (0.52)	1.09** (0.46)	0.64* (0.38)	0.72* (0.39)	-0.21 (2.08)	2.22 (2.90)	1.35 (1.97)	0.73* (0.42)	0.94** (0.45)	1.08*** (0.39)	0.93** (0.45)	0.57 (0.40)	1.52** (0.76)
Equity, share of GDP
Equity assets	0.34 (0.76)	-0.52 (0.86)	0.30 (0.90)	-0.22 (1.38)	-2.49 (1.68)	0.16 (0.69)	-0.19 (0.82)	0.01 (0.98)	0.21 (0.82)	-0.92 (0.62)	-0.60 (0.80)	-0.54 (0.80)	0.40 (0.72)	0.29 (0.83)	0.25 (0.73)	0.36 (0.68)	0.02 (0.46)	0.59 (1.61)
Equity liabilities	-0.66** (0.26)	-0.56 (0.49)	-0.68** (0.29)	-0.60 (0.43)	-0.30 (0.35)	-0.51* (0.26)	-0.15 (0.38)	-0.88* (0.47)	-0.66*** (0.24)	0.42 (0.36)	-0.63 (0.58)	-0.67* (0.37)	-0.73*** (0.27)	-0.56* (0.30)	-0.46* (0.26)	-0.56 (0.34)	-0.50* (0.29)	-1.34** (0.58)
Foreign exchange reserves, share of GDP	-5.22*** (1.81)	-5.47*** (1.81)	-5.64*** (1.87)	-5.36*** (1.84)	-5.74*** (1.85)	-5.29*** (1.80)	-5.55*** (1.77)	-5.66*** (1.98)	-3.40** (1.53)	-6.46*** (1.79)	-6.30*** (2.19)	-3.69** (1.46)	-4.85*** (1.74)	-5.34** (2.17)	-4.44** (1.77)	-5.28** (2.25)	-4.02*** (1.24)	-10.57*** (3.68)
Other variables
Income per capita, relative to US	-1.28 (1.34)	-1.05 (0.68)	-1.30 (1.54)	-0.74 (1.46)	-0.78 (1.72)	-2.12 (1.33)	-0.34 (1.41)	-2.95* (1.65)	-1.28 (1.19)	-1.02 (0.79)	-1.06 (1.01)	-0.98* (0.55)	-0.84 (1.36)	-1.00 (1.59)	-1.22 (1.25)	-0.71 (1.52)	-1.17 (1.33)	-1.80 (2.65)
Current account/GDP, 2yr MA	-6.89*** (1.50)	-5.10*** (1.33)	-7.31*** (1.68)	-6.06*** (1.49)	-5.74*** (1.50)	-6.60*** (1.45)	-7.08*** (1.86)	-8.48*** (2.66)	-7.55*** (1.60)	-4.69*** (1.73)	-7.38*** (2.45)	-5.71*** (1.42)	-6.81*** (1.50)	-7.22*** (1.85)	-7.42*** (1.50)	-5.78*** (1.72)	-6.93*** (1.51)	-12.75*** (2.89)
REER gap, relative to 5yr MA	-0.19 (0.43)	0.34 (0.43)	0.16 (0.60)	0.17 (0.42)	-0.20 (0.41)	-0.17 (0.43)	0.02 (0.59)	-0.33 (0.40)	0.19 (0.47)	0.25 (0.57)	0.35 (0.44)	0.68 (0.48)	-0.14 (0.39)	-0.13 (0.53)	-0.10 (0.44)	-0.55 (0.46)	0.25 (0.59)	-0.28 (0.78)
Credit gap, credit/GDP compared to 5yr MA	0.59 (0.59)	-0.68 (0.82)	0.61 (0.58)	-0.48 (0.69)	-0.38 (0.66)	0.44 (0.56)	1.12* (0.58)	1.80** (0.86)	0.14 (0.65)	0.36 (0.90)	-0.20 (1.04)	-1.31 (0.94)	0.95* (0.56)	1.56** (0.69)	0.49 (0.58)	1.57** (0.79)	0.15 (0.71)	1.38 (1.18)
Financial development index	-0.42 (1.34)	-1.17 (1.20)	-0.10 (1.39)	-0.05 (1.40)	0.30 (1.57)	-0.98 (1.34)	2.36* (1.35)	-2.03 (1.42)	-0.13 (1.22)	0.60 (1.01)	-2.79** (1.25)	-0.35 (1.05)	-0.16 (1.38)	-0.21 (1.52)	-0.16 (1.27)	1.34 (1.30)	-0.04 (1.38)	0.16 (2.56)
Fin. develop. * income per capita relative to US	-5.03* (3.00)	… …	-5.56* (3.35)	-4.97 (3.66)	-5.85 (4.62)	-2.60 (2.92)	-7.32** (3.33)	-1.97 (3.47)	-4.28 (2.67)	… …	… …	… …	-6.00** (3.01)	-5.95* (3.20)	-4.98* (2.86)	-7.76** (3.14)	-5.35* (2.93)	-11.98** (5.90)
VXO	0.02** (0.01)	0.02*** (0.01)	0.02** (0.01)	0.02*** (0.01)	0.02** (0.01)	0.02** (0.01)	0.02** (0.01)	0.02** (0.01)	0.01** (0.01)	0.03*** (0.01)	0.03*** (0.01)	0.02*** (0.01)	0.02** (0.01)	0.02** (0.01)	0.01* (0.01)	0.02** (0.01)	0.02** (0.01)	0.03** (0.01)
Exchange rate regime													0.01 (0.04)
Capital account openness														0.02 (0.27)
Commodity prices															-0.11*** (0.03)
Fiscal																0.14 (3.04)
Constant	-0.67** (0.33)	-1 24*** (0.27)	-0.72** (0.35)	-1.24*** (0.36)	-0.91** (0.44)	-0.50 (0.34)	-2.20*** (0.41)	0.16 (0.43)	-1.00*** (0.31)	-2.22*** (0.31)	-0.76* (0.44)	-1.58*** (0.24)	-0.90*** (0.32)	-0.85** (0.42)	-0.48 (0.31)	-1.39*** (0.38)	-0.94** (0.37)	-1.33** (0.61)
Observations	1,828	1,004	1,706	1,772	1,756	1,828	1,828	1,828	1,802	1,004	1,004	978	1781	1,509	1,828	1,454	1,828	1,828
Number of ifs_code	73	41	68	71	73	73	73	73	73	41	41	41	73	65	73	72	73	73
Log pseudolikelihood	-333.29	-294.38	-313.80	-322.31	-31 5.55	-339.01	-213.19	-339.01	-298.35	-185.67	-366.14	-259.70	-318.83	-264.67	-329.62	-206.28	-330.72	-333.71
AUROC estimates 1/	0.87	0.81	0.87	0.88	0.88	0.86	0.85	0.88	0.83	0.79	0.82	0.79	0.87	0.87	0.88	0.87	0.87	0.87

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 All country-specific variables are lagged by one year.

^1/The main regressions use stata command “xtprobit”. Approximate AUROC estimates are calculated using stata command “probit” and then applying “lroc”.

View Table

^{Table 2.}

Robustness Checks for External Crisis Events

	Sample Variation						Episode Variation						Additional Controls				Estimation method
							Full Sample			EMDE sample
Variables	Baseline	EMDEs only	Exclude oil exporters	Exclude Iceland and Ireland	Exclude GFC (2008)	Include Spain ESM	Large crises only	Crisis over next two years	Exclude consecutive years	Large crises only	Crisis over next two years	Exclude consecutive years	Exchange Rate Regime	Capital Account Openness	Commodity Prices	Fiscal Balance	Winsorize (1 percent both tails)	Log it
Debt, share of GDP
Foreign currency debt assets	0.40 (0.29)	-0.13 (0.47)	0.52 (0.34)	-0.51 (0.42)	0.56** (0.28)	0.37 (0.28)	0.51 (0.38)	0.28 (0.44)	0.50 (0.31)	0.59 (0.56)	-1.06* (0.56)	0.06 (0.45)	0.42 (0.29)	0.30 (0.32)	0.42 (0.30)	0.37 (0.36)	-0.01 (0.35)	0.95 (0.58)
Domestic currency debt assets	-0.27 (0.48)	… …	-0.41 (0.52)	0.37 (0.69)	1.01 (1.11)	-0.73 (0.63)	-1.08* (0.63)	0.25 (0.52)	-0.20 (0.49)	… …	… …	… …	-0.22 (0.51)	-0.50 (0.51)	-0.70 (0.50)	-0.57 (0.60)	0.03 (0.63)	-0.64 (0.97)
Foreign currency debt liabilities	0.44** (0.21)	1.78*** (0.33)	0.44** (0.21)	1.55*** (0.27)	0.63*** (0.22)	0.42** (0.20)	0.41** (0.20)	0.58** (0.29)	0.32* (0.18)	1.09** (0.47)	2.55*** (0.51)	1.52*** (0.31)	0.43** (0.20)	0.49** (0.22)	0.41** (0.20)	0.50** (0.20)	0.95*** (0.26)	0.86** (0.38)
Domestic currency debt liabilities	0.75* (0.39)	1.32 (2.04)	0.81* (0.42)	0.45 (0.44)	0.39 (0.61)	1.10** (0.52)	1.09** (0.46)	0.64* (0.38)	0.72* (0.39)	-0.21 (2.08)	2.22 (2.90)	1.35 (1.97)	0.73* (0.42)	0.94** (0.45)	1.08*** (0.39)	0.93** (0.45)	0.57 (0.40)	1.52** (0.76)
Equity, share of GDP
Equity assets	0.34 (0.76)	-0.52 (0.86)	0.30 (0.90)	-0.22 (1.38)	-2.49 (1.68)	0.16 (0.69)	-0.19 (0.82)	0.01 (0.98)	0.21 (0.82)	-0.92 (0.62)	-0.60 (0.80)	-0.54 (0.80)	0.40 (0.72)	0.29 (0.83)	0.25 (0.73)	0.36 (0.68)	0.02 (0.46)	0.59 (1.61)
Equity liabilities	-0.66** (0.26)	-0.56 (0.49)	-0.68** (0.29)	-0.60 (0.43)	-0.30 (0.35)	-0.51* (0.26)	-0.15 (0.38)	-0.88* (0.47)	-0.66*** (0.24)	0.42 (0.36)	-0.63 (0.58)	-0.67* (0.37)	-0.73*** (0.27)	-0.56* (0.30)	-0.46* (0.26)	-0.56 (0.34)	-0.50* (0.29)	-1.34** (0.58)
Foreign exchange reserves, share of GDP	-5.22*** (1.81)	-5.47*** (1.81)	-5.64*** (1.87)	-5.36*** (1.84)	-5.74*** (1.85)	-5.29*** (1.80)	-5.55*** (1.77)	-5.66*** (1.98)	-3.40** (1.53)	-6.46*** (1.79)	-6.30*** (2.19)	-3.69** (1.46)	-4.85*** (1.74)	-5.34** (2.17)	-4.44** (1.77)	-5.28** (2.25)	-4.02*** (1.24)	-10.57*** (3.68)
Other variables
Income per capita, relative to US	-1.28 (1.34)	-1.05 (0.68)	-1.30 (1.54)	-0.74 (1.46)	-0.78 (1.72)	-2.12 (1.33)	-0.34 (1.41)	-2.95* (1.65)	-1.28 (1.19)	-1.02 (0.79)	-1.06 (1.01)	-0.98* (0.55)	-0.84 (1.36)	-1.00 (1.59)	-1.22 (1.25)	-0.71 (1.52)	-1.17 (1.33)	-1.80 (2.65)
Current account/GDP, 2yr MA	-6.89*** (1.50)	-5.10*** (1.33)	-7.31*** (1.68)	-6.06*** (1.49)	-5.74*** (1.50)	-6.60*** (1.45)	-7.08*** (1.86)	-8.48*** (2.66)	-7.55*** (1.60)	-4.69*** (1.73)	-7.38*** (2.45)	-5.71*** (1.42)	-6.81*** (1.50)	-7.22*** (1.85)	-7.42*** (1.50)	-5.78*** (1.72)	-6.93*** (1.51)	-12.75*** (2.89)
REER gap, relative to 5yr MA	-0.19 (0.43)	0.34 (0.43)	0.16 (0.60)	0.17 (0.42)	-0.20 (0.41)	-0.17 (0.43)	0.02 (0.59)	-0.33 (0.40)	0.19 (0.47)	0.25 (0.57)	0.35 (0.44)	0.68 (0.48)	-0.14 (0.39)	-0.13 (0.53)	-0.10 (0.44)	-0.55 (0.46)	0.25 (0.59)	-0.28 (0.78)
Credit gap, credit/GDP compared to 5yr MA	0.59 (0.59)	-0.68 (0.82)	0.61 (0.58)	-0.48 (0.69)	-0.38 (0.66)	0.44 (0.56)	1.12* (0.58)	1.80** (0.86)	0.14 (0.65)	0.36 (0.90)	-0.20 (1.04)	-1.31 (0.94)	0.95* (0.56)	1.56** (0.69)	0.49 (0.58)	1.57** (0.79)	0.15 (0.71)	1.38 (1.18)
Financial development index	-0.42 (1.34)	-1.17 (1.20)	-0.10 (1.39)	-0.05 (1.40)	0.30 (1.57)	-0.98 (1.34)	2.36* (1.35)	-2.03 (1.42)	-0.13 (1.22)	0.60 (1.01)	-2.79** (1.25)	-0.35 (1.05)	-0.16 (1.38)	-0.21 (1.52)	-0.16 (1.27)	1.34 (1.30)	-0.04 (1.38)	0.16 (2.56)
Fin. develop. * income per capita relative to US	-5.03* (3.00)	… …	-5.56* (3.35)	-4.97 (3.66)	-5.85 (4.62)	-2.60 (2.92)	-7.32** (3.33)	-1.97 (3.47)	-4.28 (2.67)	… …	… …	… …	-6.00** (3.01)	-5.95* (3.20)	-4.98* (2.86)	-7.76** (3.14)	-5.35* (2.93)	-11.98** (5.90)
VXO	0.02** (0.01)	0.02*** (0.01)	0.02** (0.01)	0.02*** (0.01)	0.02** (0.01)	0.02** (0.01)	0.02** (0.01)	0.02** (0.01)	0.01** (0.01)	0.03*** (0.01)	0.03*** (0.01)	0.02*** (0.01)	0.02** (0.01)	0.02** (0.01)	0.01* (0.01)	0.02** (0.01)	0.02** (0.01)	0.03** (0.01)
Exchange rate regime													0.01 (0.04)
Capital account openness														0.02 (0.27)
Commodity prices															-0.11*** (0.03)
Fiscal																0.14 (3.04)
Constant	-0.67** (0.33)	-1 24*** (0.27)	-0.72** (0.35)	-1.24*** (0.36)	-0.91** (0.44)	-0.50 (0.34)	-2.20*** (0.41)	0.16 (0.43)	-1.00*** (0.31)	-2.22*** (0.31)	-0.76* (0.44)	-1.58*** (0.24)	-0.90*** (0.32)	-0.85** (0.42)	-0.48 (0.31)	-1.39*** (0.38)	-0.94** (0.37)	-1.33** (0.61)
Observations	1,828	1,004	1,706	1,772	1,756	1,828	1,828	1,828	1,802	1,004	1,004	978	1781	1,509	1,828	1,454	1,828	1,828
Number of ifs_code	73	41	68	71	73	73	73	73	73	41	41	41	73	65	73	72	73	73
Log pseudolikelihood	-333.29	-294.38	-313.80	-322.31	-31 5.55	-339.01	-213.19	-339.01	-298.35	-185.67	-366.14	-259.70	-318.83	-264.67	-329.62	-206.28	-330.72	-333.71
AUROC estimates 1/	0.87	0.81	0.87	0.88	0.88	0.86	0.85	0.88	0.83	0.79	0.82	0.79	0.87	0.87	0.88	0.87	0.87	0.87

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 All country-specific variables are lagged by one year.

^1/The main regressions use stata command “xtprobit”. Approximate AUROC estimates are calculated using stata command “probit” and then applying “lroc”.

The main results are also robust to a battery of alternative specifications. In particular, foreign currency-denominated debt liabilities are always statistically significant, and the magnitude is remarkably stable across alternative specifications.¹⁹The first set of robustness checks involves changes in country sample, to exclude oil exporters and financial centers, and consider Spain’s European Stability Mechanism (ESM) program (Table 2, columns 2, 3 and 4). The second set involves alternative definitions of the left-hand-side variable, such as: (i) focusing on large crises (IMF programs greater than 200 percent of quota); (ii) using a two-year window in defining a crisis, and (iii) excluding consecutive crisis years, both for the full sample and for the EMDE sample. The third set of alternative specifications involves having additional controls to consider the exchange rate regime, capital account openness, commodity price changes and the fiscal balance.,²⁰, Only commodity prices are significant, and the inclusion of additional controls does not alter key results. Finally, windsorizing the data prior to estimation (1 percent for both tails), and estimating a logit regression delivers similar results. ²¹

C. Probit Margins

Predicted probabilities are computed by keeping all the variables in the estimated model fixed at their sample means, while changing the variable of interest in specified increments (for other applications of this approach, see, for example, Gourinchas and Obstfeld 2012). This estimation is useful in uncovering the nonlinear effects of certain variables on the likelihood of external crises. As shown in Figure 5. the estimated effects are economically more meaningful in the model estimated for EMDEs. Specifically:

• In the EMDE sample, an increase in foreign-currency-denominated debt liabilities from 40 percent of GDP (near the EMDE median) to 60 percent of GDP is associated with a 5-percentage point increase in the predicted probability of an external crisis. Meanwhile, in the full sample, this rise in debt would result increases the crisis probability by only 0.2 percentage points.

• A decline in the current account balance from a surplus of 5 percent of GDP to a deficit of 5 percent of GDP is associated with an increase in the predicted probability of an external crisis by 5.3 percentage points for EMDEs. For the full sample, the probability rises by only 1.1 percentage points.

• The relationship between foreign official reserves and external crises is markedly nonlinear. The predicted external stress probability is near zero when reserves are above 30 percent of GDP. As reserves decline, the predicted external crisis probability increases. In EMDEs, a decline in foreign exchange reserves from 20 percent to 10 percent of GDP is associated with a 6.5 percentage points increase in the likelihood of an external crisis, while a further decline from 10 percent to 0 percent of GDP increases the predicted probability by an additional 12.6 percentage points. For the full sample, corresponding values for the entire sample are much lower (0.7 and 2.1 percentage points, respectively).

• Global risk aversion also has a differentiated and nonlinear impact depending on the sample. In EMDEs, a doubling of global risk aversion from its historical mean increases crisis predicted probabilities by 6.5 percentage points, with a further doubling to levels witnessed in the early phase of the Global Financial Crisis and the Covid-19 pandemic increases predicted crisis probabilities by 23.5 percentage points. For the full sample, the probability rises by 0.7 and 4 percentage points, respectively.

View Full Size

External Crises: Model Predicted Probabilities (Margins)

Citation: IMF Working Papers 2021, 020; 10.5089/9781513568102.001.A001

Note: All panels display the predicted probabilities of an external crisis episode, keeping all covariates except foreign currency debt, foreign exchange reserves, the current account, and global risk aversion at their sample mean. The solid blue line corresponds to the mean estimate, the dotted lines correspond to the 90 percent confidence interval.

• Download Figure
• Download figure as PowerPoint slide

The finding that external vulnerabilities are more strongly related to risks of external stress for EMDEs reflects differences in the estimated coefficients and mean of control variables across the full and EMDE sample. For instance, the estimated coefficient on the effect of foreign-currency-denominated debt on the probability of an external crisis is about four times larger than for the full sample. In addition, the EMDE sample has a lower average financial development index, which likely constrains a country’s ability to weather external shocks.

The results also imply that a combination of two or more external vulnerabilities greatly increases the probability of external crisis in EMDEs (Figure 6). The same level for foreign-currency-denominated debt liabilities could signal very different risks of an external crisis episode, depending on other vulnerabilities. When foreign currency debt is 40 percent of GDP, the predicted probability ranges from 2–13 percent, depending on whether foreign official reserves and the current account balance are at high levels (75th percentile of the sample) or at low levels (25th percentile). Similarly, the vulnerabilities associated with large current account deficits depend on the levels of foreign official reserves and foreign-currency-denominated debt. Predicted crisis probabilities associated with a current account deficit of 5 percent range from 2.5 to 21 percent depending on whether the economy is less vulnerable (with official reserves above the 75^th percentile and foreign-currency-denominated debt below the 25^th percentile) or more vulnerable (with official reserves below the 25^th percentile and foreign-currency-denominated debt above the 75^th percentile). A similar logic applies to the risks associated with low foreign official reserves, which are higher when current account deficits and foreign-currency-denominated debt are at high levels as well.

View Full Size

Selected Predictors of External Crises in the Emerging Market and Developing Economies Sample

Citation: IMF Working Papers 2021, 020; 10.5089/9781513568102.001.A001

Sources: External Wealth of Nations database (Lane and Milesi-Ferretti 2007); Haver Analytics; IMF, World Economic Outlook database; and authors’ calculations.Note: All panels display the predicted probabilities of an external crisis episode, keeping all covariates except foreign currency debt, foreign exchange reserves, the current account, and global risk aversion at their sample mean. More vulnerable countries are defined as those with foreign currency debt at the 75th percentile and foreign exchange reserves and current account balance at the 25th percentile of the sample. Less vulnerable countries are defined as those with foreign currency debt at the 25th percentile and foreign exchange reserves and current account balance at the 75th percentile. Median countries are defined as those with foreign currency debt, foreign exchange reserves, and current account balance at the median.

• Download Figure
• Download figure as PowerPoint slide

Finally, the estimated model has important implications for the global risks currently facing EMDEs. When global risk aversion reaches the peak values as seen in the initial phase of the Covid-19 pandemic (about 80, a value also reached during the Global Financial Crisis), the external crisis probability for an EMDE with median levels of preexisting vulnerabilities rises to about 40 percent— more than double the estimated probability for less vulnerable EMDEs (as defined previously). Results reaffirm the importance of external buffers to deal with global risk shocks.

D. Output Costs and Macroeconomic Effects of Crises

In addition to affecting the likelihood of an external crisis, high levels of foreign currency-denominated debt liabilities, high current account deficits, and low foreign official reserves may affect post-crisis macroeconomic outcomes and the speed of the subsequent recovery. To investigate this hypothesis, the asymmetry in post-crisis outcomes related to these indicators is analyzed using the local projections method (Jordà, 2005), which helps trace out the dynamic response of output, exchange rates and the current account after a crisis materializes, for the EMDE sample. ²²

For the purposes of the analysis, countries are again classified as having higher or lower vulnerabilities based on the preexisting level of foreign-currency-denominated debt liabilities, current account deficits, and foreign official reserves. In particular, the following specification is specified:

for horizons k=0,1,...5 years and where:

• y_i,t+k is the log of real GDP, the log of the REER, or the CA/GDP ratio of country i at time t+k,

• α_i and γ_t are country and time fixed effects,

• D^H_i,t is a dummy variable = 1 for countries with high foreign currency denominated debt (above median of sample), high current account deficit (below median of current account balance), and low foreign official reserves (below median); and 0 for the rest of the sample,

• Crisis_i,t is a dummy variable (1 for crisis and 0 otherwise),

• X_i,t contains additional regressors: 2-year lags of the first difference of the log of real GDP, the log of the REER, and the CA/GDP ratio. Γ is the associated vector of estimated coefficients.

• ε is the regression residual.

The results suggest that EMDEs with a high vulnerabilities (a combination of high foreign-currency debt liabilities, high current account deficit, and low foreign official reserves) tend to experience higher and more proacted output losses following the crisis (Figure 7, top panels). Specifically, in vulnerable EMDEs the initial output loss is over 4 percent, and output remains below pre-crisis levels after five years. In the rest of the EMDE sample. initial losses are smaller (1 percent) and output rises to above pre-crisis level after four years. In addition (Figure 7, mid and bottom panels), more vulnerable EMDEs observe larger initial exchange rate depreciations (averaging 10 percent) and current account adjustment (averaging 2 percent of GDP). Overall, these results suggest that the post-crisis adjustment is larger when countries have multiple pre-existing external vulnerabilities, and the recovery is slower.

View Full Size

Dynamic Evolution of Output, Real Exchange Rates, and Current Account Balance Following External Crisis Episodes

Citation: IMF Working Papers 2021, 020; 10.5089/9781513568102.001.A001

Note: Shaded area corresponds to the 90 percent confidence interval. The horizontal axis denotes time in years, and 0 is the year of the external crisis episode. EMDEs = emerging market and developing economies; REER = real effective exchange rate.1/ Vulnerable EMDEs are defined as those with foreign currency debt above the EMDE median, and current account balance and foreign exchange reserves below the EMDE median.

• Download Figure
• Download figure as PowerPoint slide

A. Event Analysis

The qualitative results (Figure 8) are similar to the case of external crisis. Sudden stops are preceded by (i) net IIP declines, which are mostly driven by increases in external debt liabilities denominated in foreign currency as percent of GDP (with currency depreciation and negative real growth also playing a role); and (ii) foreign official reserves declines, which continue in its aftermath, suggesting reserve sales are often used to buffer the impact of sudden stops. Meanwhile, external debt liabilities denominated in domestic currency are flat around sudden stop episodes, with equities following a similar path as in crisis episodes.

View Full Size

Conditional Mean of Key IIP Components, the REER and Growth Around Sudden Stop Episodes

Citation: IMF Working Papers 2021, 020; 10.5089/9781513568102.001.A001

Note: The solid blue line corresponds to the mean of all sudden stop episodes, while the blue shaded area corresponds to the 90 percent confidence interval. IIP components are expressed as percent of GDP. An increase (decrease) in the Real Effective Exchange Rate denotes appreciation (depreciation).

• Download Figure
• Download figure as PowerPoint slide

In quantitative terms, the magnitudes involved are much larger, suggesting that the identified sudden stop events are more severe than the identified external crises (Section IV A). The net IIP declines by more than 40 percentage points around a sudden stop episode, while it declines by about 3 percentage points for all external crises and by about 14 percentage points for large external crises, which involve cases of IMF financial assistance exceeding 200 percent of quota. Foreign-currency-denominated external debt liabilities increase by almost 80 percentage points for sudden stops, compared with 12 percentage points in the case of external crises and 22 percentage points in the case of large external crises.

B. Probit Estimation Results and Probability Margins

As in the case of external crisis events, deterioration in the net IIP position increases the probability of sudden stops, with statistically significant effects (column 1), driven both by an increase in net debt and net equity positions (Table 3). However, once foreign and domestic currency components of net debt are considered, the net equity position loses statistical significance (columns 3, 4, 7, and 8), suggesting that sudden stops are driven by debt and its foreign-currency component. Higher net foreign-currency denominated assets reduce the likelihood of sudden stops, although the effect is mostly driven by lower gross foreign currency debt liabilities (columns 4 and 8). This finding is in line with the results of the event analysis, confirming that gross external debt liabilities are the main driver of IIP risks. Interestingly, domestic-currency denominated debt liabilities appear to mitigate such risks, likely reflecting (as discussed earlier) stronger policy frameworks and deeper capital markets, which in turn increase the ability to issue external debt in domestic currency.²³

^{Table 3.}

Probit Estimates for Sudden Stops

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 All country-specific variables are lagged by one year. Net refers to assets minus liabilities. Net debt, net equity, net foreign debt, and net domestic debt are computed using the underlying gross numbers.

^1/The main regressions use stata command “xtprobit”. Approximate AUROC estimates are calculated using stata command “probit” and then applying “lroc”.

^{Table 3.}

Probit Estimates for Sudden Stops

Variables	1	2	3	4	5	6	7	8
Net IIP, share of GDP	-0.26*** (0.07)				-0.50 (0.35)
Debt, share of GDP
Net debt		-0.26*** (0.07)				-0.50* (0.27)
Net foreign currency debt			-0.33*** (0.09)				-0.62*** (0.21)
Net domestic currency debt			1.05*** (0.35)				1.86** (0.77)
Foreign currency debt assets				0.09 (0.19)				-0.15 (0.37)
Domestic currency debt assets				0.84** (0.34)				1.88*** (0.65)
Foreign currency debt liabilities				0.34*** (0.08)				0.59*** (0.21)
Domestic currency debt liabilities				-1.40*** (0.51)				-2.50*** (0.77)
Equity, share of GDP
Net equity		-0.26*** (0.09)	-0.06 (0.16)			-0.26 (0.44)	0.36 (0.50)
Equity assets				-0.62 (0.43)				0.13 (1.03)
Equity liabilities				0.24 (0.23)				-0.27 (0.49)
Foreign exchange reserves, share of GDP		-0.42 (0.75)	-0.37 (0.84)	-1.45 (0.95)		0.37 (1.71)	0.85 (1.99)	-0.15 (1.88)
Other variables
Income per capita, relative to US					4.46*** (1.39)	4.36*** (1.47)	5.23*** (1.61)	5.52*** (1.58)
Current account/GDP, 2yr MA					-2.81 (2.18)	-2.89 (1.94)	-3.15 (1.92)	-2.81 (1.89)
REER gap, relative to 5yr MA					2.89*** (0.67)	2.95*** (0.66)	3.03*** (0.64)	2.86*** (0.63)
Credit gap, credit/GDP compared to 5yr MA					4.31*** (1.23)	4.45*** (1.30)	4.68*** (1.29)	4.35*** (1.33)
Financial development index					0.05 (1.66)	0.01 (1.64)	1.60 (2.04)	2.59 (2.29)
Fin. develop. * income per capita relative to US					-6.79** (3.27)	-6.70** (3.36)	-9.60** (4.05)	-10.92** (4.49)
VXO					0.09*** (0.01)	0.09*** (0.01)	0.09*** (0.01)	0.09*** (0.01)
Constant	-2.38*** (0.14)	-2.36*** (0.1 6)	-2.31*** (0.16)	-2.16*** (0.17)	-5.61*** (0.76)	-5.66*** (0.71)	-6.12*** (0.75)	-6.17*** (0.72)
Observations	1,808	1,808	1,808	1,808	1,808	1,808	1,808	1,808
Number of ifs_code	73	73	73	73	73	73	73	73
Log pseudolikelihood	-156.03	-156.02	-152.90	-149.78	-105.47	-104.81	-101.69	-100.55
AUROC estimates 1/	0.60	0.61	0.64	0.73	0.92	0.92	0.93	0.93

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 All country-specific variables are lagged by one year. Net refers to assets minus liabilities. Net debt, net equity, net foreign debt, and net domestic debt are computed using the underlying gross numbers.

^1/The main regressions use stata command “xtprobit”. Approximate AUROC estimates are calculated using stata command “probit” and then applying “lroc”.

View Table

^{Table 3.}

Probit Estimates for Sudden Stops

Variables	1	2	3	4	5	6	7	8
Net IIP, share of GDP	-0.26*** (0.07)				-0.50 (0.35)
Debt, share of GDP
Net debt		-0.26*** (0.07)				-0.50* (0.27)
Net foreign currency debt			-0.33*** (0.09)				-0.62*** (0.21)
Net domestic currency debt			1.05*** (0.35)				1.86** (0.77)
Foreign currency debt assets				0.09 (0.19)				-0.15 (0.37)
Domestic currency debt assets				0.84** (0.34)				1.88*** (0.65)
Foreign currency debt liabilities				0.34*** (0.08)				0.59*** (0.21)
Domestic currency debt liabilities				-1.40*** (0.51)				-2.50*** (0.77)
Equity, share of GDP
Net equity		-0.26*** (0.09)	-0.06 (0.16)			-0.26 (0.44)	0.36 (0.50)
Equity assets				-0.62 (0.43)				0.13 (1.03)
Equity liabilities				0.24 (0.23)				-0.27 (0.49)
Foreign exchange reserves, share of GDP		-0.42 (0.75)	-0.37 (0.84)	-1.45 (0.95)		0.37 (1.71)	0.85 (1.99)	-0.15 (1.88)
Other variables
Income per capita, relative to US					4.46*** (1.39)	4.36*** (1.47)	5.23*** (1.61)	5.52*** (1.58)
Current account/GDP, 2yr MA					-2.81 (2.18)	-2.89 (1.94)	-3.15 (1.92)	-2.81 (1.89)
REER gap, relative to 5yr MA					2.89*** (0.67)	2.95*** (0.66)	3.03*** (0.64)	2.86*** (0.63)
Credit gap, credit/GDP compared to 5yr MA					4.31*** (1.23)	4.45*** (1.30)	4.68*** (1.29)	4.35*** (1.33)
Financial development index					0.05 (1.66)	0.01 (1.64)	1.60 (2.04)	2.59 (2.29)
Fin. develop. * income per capita relative to US					-6.79** (3.27)	-6.70** (3.36)	-9.60** (4.05)	-10.92** (4.49)
VXO					0.09*** (0.01)	0.09*** (0.01)	0.09*** (0.01)	0.09*** (0.01)
Constant	-2.38*** (0.14)	-2.36*** (0.1 6)	-2.31*** (0.16)	-2.16*** (0.17)	-5.61*** (0.76)	-5.66*** (0.71)	-6.12*** (0.75)	-6.17*** (0.72)
Observations	1,808	1,808	1,808	1,808	1,808	1,808	1,808	1,808
Number of ifs_code	73	73	73	73	73	73	73	73
Log pseudolikelihood	-156.03	-156.02	-152.90	-149.78	-105.47	-104.81	-101.69	-100.55
AUROC estimates 1/	0.60	0.61	0.64	0.73	0.92	0.92	0.93	0.93

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 All country-specific variables are lagged by one year. Net refers to assets minus liabilities. Net debt, net equity, net foreign debt, and net domestic debt are computed using the underlying gross numbers.

^1/The main regressions use stata command “xtprobit”. Approximate AUROC estimates are calculated using stata command “probit” and then applying “lroc”.