2.1 Debt Buildups Amplify Recessions in Both EMEs and AEs

Our first question asks whether rapid private and public debt buildups prior to recessions predict deeper contractions and slower recoveries. To address this issue, we use IFS data on bank credit to the private sector and IMF data on public debt covering a large set of EMEs and AEs (see Appendix A) and the Local Projection Method (LPM) pioneered by Jordà (2005) and applied in a similar context in Jordà, Schularick and Taylor (2016). Mian, Sufi and Verner (2015) use a similar approach to assess the effect of private debt, specifically household debt, on growth.

Our baseline model regresses the cumulative change in the variables of interest y (e.g. real per capita GDP) in bad times on a predetermined information set. The LPM facilitates the identification of the effect of a starting condition, that is, the accumulation of debt prior to a recession, on the dynamics during the ensuing recession, by running a sequence of regressions for different horizons. Specifically, we estimate the following local projection model:

for h = 1,…, 5, where Δ_hy_i,p+h/y_ip is the cumulated percentage change in real percapita GDP h years after a peak p in the business cycle, θ_h is the intercept, ${\mathrm{\Delta }}_{\overline{5}}PR{Y}_{i,p}$ and ${\mathrm{\Delta }}_{\overline{5}}PU{Y}_{i,p}$ measure the average annual change in the private and public debt ratios over the five years preceding a downturn, Y_ip−l are a set predetermined control variables, α_hi are fixed country effects, and u_i,p+h is the residual.⁵ All variables are expressed in real per-capita terms. Controls include annual growth rates of GDP, private debt, public debt, and government expenditures. In our baseline specification we use three lags, i.e. we set L = 2. Following Jordà, Schularick and Taylor (2016), standard errors are clustered at the country-level as a conservative fix for the leftover serial correlation typical of local projections. The business cycle peaks are identified using the Bry and Boschan (1971) algorithm. Formally, a generic year t is identified as a “peak year” p if the level of real per-capita GDP grows in year t and drops in year t + 1.

To allow for a meaningful interpretation of the intercept θ_h, all the right hand side (RHS) variables are expressed in difference from their pooled means and the set of country-dummies is normalized to satisfy the condition ${\mathrm{\Sigma }}_{i=1}^N{\alpha }_{hi}=0$.⁶ In this way, θ_h measures the GDP path in “average-debt” recessions, that is, slumps associated with private and public debt accumulation at their respective sample averages (along with the remaining set of regressors). ${\beta }_h^{PR}\left({\beta }_h^{PU}\right)$, instead, measures how the path deviates when a country enters a recession with larger-than-average private (public) debt accumulation and public (private) debt accumulation at its sample average. Finally, ${\beta }_h^{PRPU}$ measures the marginal (additional) effect of entering a recession with both debt buildups at the same time. In the rest of the discussion, we simulate illustrative scenarios in which debt buildups amount to five percentage points. We use the same five percentage points buildups across models, types of debt, country-groups, and types of recessions in order to ease comparisons.⁷

Figure 2 shows the path of real per capita GDP during downturns under different scenarios in EMEs and AEs. The estimation uses information from a set of more than 300 recessions in 80 countries. First, we focus our discussion on the solid lines, which denotes our baseline. In AEs, an average recession lasts for one year and is associated with a 2 percent decrease in real per capita GDP. Between years 2 and 3, output fully recovers; at year 5, it is 5.5 percent higher than at the peak. Although the duration of the recession is the same, it is slightly deeper in EMEs, reaching a negative of -3 percent. Also, the recovery takes longer and it is only completed at year 4. Finally, five years from the start of the recession, output is only 2.2 percent higher than its pre-peak level.

View Full Size

Debt Buildups and Recessions in Advanced and Emerging Economies

Citation: IMF Working Papers 2017, 061; 10.5089/9781475588170.001.A001

Note. Top panel. Dynamics of real per capita GDP in advanced economies and emerging market economies, starting from the year preceding a recession (peak). The solid line shows the average path in recessions while the dashed line shows how the path deviates when the recession is preceded by a rapid debt buildup. Bottom panel. Robust standard errors, clustered by country, are reported in parentheses.

• Download Figure
• Download figure as PowerPoint slide

How does this pattern change when a country enters a recession after rapid private or public debt buildups? The dotted lines in the figure show three alternative scenarios. When a country enters a recession after a rapid private debt buildup (left panel), while AEs recover at year 4, EMEs still have not completed the recovery at year 5. A similar scenario occurs in the event of a public debt buildup (middle panel). The effect is especially marked in EMEs, where GDP drops and stays below the prerecession level five years from its start. Not surprisingly, prospects deteriorate further when both private and public debt buildups are present at the start of a recession (right panel). While AEs recover around year 5, GDP remains persistently lower than the pre-crisis levels in EMEs.

Overall, we find that rapid debt accumulation prior to recessions, both private and public, predicts deeper recessions and slower recoveries also in EMEs. Thus, it appears that the interaction between debt dynamics and business cycles is a global phenomenon. Moreover, such interaction appears to be particularly strong in EMEs. The next section looks deeper into this issue by focusing on a particularly severe type of recession in which private debt plays an important role-the banking crisis or financial recession.

3.1 Public Debt Buildups Do Not Predict Banking Crises…

To analyze the role of debt in predicting banking crises, we follow Jordà, Schularick and Taylor (2016) and estimate the following probabilistic model:

where $\ln \frac{P(B_{i,t}=1)}{1-P(B_{i,t}=1)}$ is the log-odds ratio of a banking crisis for a country i at year t, ${\mathrm{\Delta }}_{\overline{5}}PR{Y}_{i,t-1}$ and ${\mathrm{\Delta }}_{\overline{5}}PR{Y}_{i,t-1}$ measure the average annual change in the private and public debt ratios over the five years before year t, α and α_i are the intercept and a set of fixed country effects, and u_i,t is the residual. Financial crises episodes B_i,t are taken from Reinhart and Rogoff (2009), Jordà, Schularick and Taylor (2016), and Laeven and Valencia (2013).⁹

Table 1 shows the results for different specifications of the general model in equation (2). In particular, we report the marginal effects on the probability of a banking crisis when all predictors are at the mean. The main finding is that while private debt buildups are likely to trigger banking crises in both EMEs and AEs, public debt buildups do not predict such events.

Table 1.

Marginal Effects of Debt Buildups on the Probability of a Banking Crisis

Note. The table shows the marginal effects of private and public debt buildups on the probability of a banking crisis (percent), for every percentage point increase in the average annual change of private credit-to-GDP and public debt-to-GDP over five years. Predicted probabilities are calculated with the assumption that all predictors in equation (2) are at their mean values. Robust standard errors are reported in parentheses.

Table 1.

Marginal Effects of Debt Buildups on the Probability of a Banking Crisis

	Advanced Economies P (Bi,t = 1)%			Emerging Market Economies P (Bi, t = 1)%
β PR	0.35		0.35	0.39		0.72
	(0.09)		(0.09)	(0.22)		(0.19)
β PU		−0.08	−0.04		−0.25	−0.18
		(0.15)	(0.17)		(0.12)	(0.13)
Obs	1239	1417	1178	1548	1470	1259

Note. The table shows the marginal effects of private and public debt buildups on the probability of a banking crisis (percent), for every percentage point increase in the average annual change of private credit-to-GDP and public debt-to-GDP over five years. Predicted probabilities are calculated with the assumption that all predictors in equation (2) are at their mean values. Robust standard errors are reported in parentheses.

View Table

Table 1.

Marginal Effects of Debt Buildups on the Probability of a Banking Crisis

	Advanced Economies P (Bi,t = 1)%			Emerging Market Economies P (Bi, t = 1)%
β PR	0.35		0.35	0.39		0.72
	(0.09)		(0.09)	(0.22)		(0.19)
β PU		−0.08	−0.04		−0.25	−0.18
		(0.15)	(0.17)		(0.12)	(0.13)
Obs	1239	1417	1178	1548	1470	1259

Note. The table shows the marginal effects of private and public debt buildups on the probability of a banking crisis (percent), for every percentage point increase in the average annual change of private credit-to-GDP and public debt-to-GDP over five years. Predicted probabilities are calculated with the assumption that all predictors in equation (2) are at their mean values. Robust standard errors are reported in parentheses.

In particular, we find that for every percentage point increase in the average annual change of private debt-to-GDP, the predicted probability of a banking crises goes up by about 0.35 – 0.72 percentage points. This effect is estimated with high precision, that is, it passes the 3-standard deviations threshold. The effect associated to an increase in the average annual change of public debt-to-GDP, instead, is slightly negative around -0.25 – -0.04 percentage points and is significantly different from zero at the one-standard deviation level. This result is rather surprising although in line with Jordà, Schularick and Taylor (2016). In principle, our estimate might be capturing the fact that public debt tends to rise sharply right after crises and therefore one might erroneously conclude that relatively low public debt levels make crises more likely, as discussed in Gourinchas and Obstfeld (2012). However, we obtain qualitatively similar results if we drop the first 4-8 years after the start of each financial crisis. An alternative possible explanation for the negative coefficient on public debt is that countries might pre-empt banking crises via recapitalizations that, despite having an impact on the public balance sheet, reduce the probability of hitting a banking crisis.

3.2 … But Exacerbate Non-financial and Financial Recessions

To isolate the effects of private and public debt buildups during banking crises, we slightly change the benchmark local projection model as follows:

The structure of the model in equation (3) is similar to the one in equation (1). The only difference is that now each parameter of interest is interacted with a 0-1 financial peak indicator F_i,p and its complement (1 — F_i,p).¹⁰ Accordingly, we now center all of the RHS variables with respect to their means in the non-financial and financial recession bins, respectively.

Figure 3 reports the evolution of real per capita GDP from the year preceding non-financial and financial recessions in EMEs and AEs. The solid line shows the dynamics in banking crisis recessions (red lines) and in other recessions (blue lines). The dashed lines, as before, show how GDP deviates from the respective average paths when the preceding expansion is accompanied by a private credit boom (left panel), a public debt boom (middle panel), or both (right panel). As expected, financial recessions are considerably more painful than non-financial recessions in general, both in EMEs and AEs. This points to the importance of looking at financial and non-financial recessions separately.

View Full Size

Non-financial and Financial Recessions

Citation: IMF Working Papers 2017, 061; 10.5089/9781475588170.001.A001

Note. Top panel. Dynamics of real per capita GDP in AEs and EMEs, starting from the year preceding non-financial (blue lines) and financial (red lines) recessions. The solid line shows the average path in recessions, while the dashed line shows how the path deviates when the recession is preceded by a rapid debt buildup. Bottom panel. Robust standard errors, clustered by country, are reported in parentheses.

• Download Figure
• Download figure as PowerPoint slide

In line with Jordà, Schularick and Taylor (2016), pre-recession buildups in private debt are found to be more toxic in financial crises. This finding suggests that when private debt is the underlying problem driving the recession, it plays a more prominent role in slowing down the recovery. Nevertheless, the most striking results stem from the marginal effects of debt. When a country enters a banking crisis recession with rapid debt accumulations in both sectors, its economic performance tends to deteriorate further, slowing down the recovery. This result can be appreciated by looking at how the dashed lines lie below the solid lines in the right-hand-side panels in Figure 3. Moreover, the marginal effects of debt appear especially strong in EMEs. The case of a country entering a financial recession with both debt buildups is striking. Three years after the recession is started, the drop in GDP is twice as large in EMEs than in AEs, -16 percent and -8 percent, respectively.

Interestingly, our results point to a significant effect of public debt accumulations also in non-financial recessions. The blue dashed lines are consistently below the blue solid line in both EMEs and AEs. This suggests that the constraints that a pre-crisis loose fiscal stance might impose on the policymakers after the start of the crisis can be at times significant. Private debt accumulations, on the contrary, appears not to have any significant effect in amplifying the GDP fall during non-financial recessions. To summarize, we have shown that when we control for the same type of recessions, focusing on rare events typically associated with extremely acute downturns, we still find that rapid debt buildups can add further slack to economies and that this effect is especially amplified in EMEs. In the next section, we look into the determinants of these different results.

4.1 Limited Fiscal Support in EMEs…

Economic history and empirical studies have suggested that fiscal policy often tends to be pro-cyclical in EMEs (Gavin and Perotti, 1997; Lane, 2003; Kaminsky, Reinhart and Végh, 2004). This differs from the experience of AEs, where fiscal policy is usually a-cyclical or countercyclical. At the same time, the literature on EMEs suggests that monetary policy in these countries is often constrained (for example, by an exchange rate policy) or limited in its effectiveness (either because the financial systems are small or because the transmission mechanism of interest rates is poor). Accordingly, the main policy tool to address cyclical considerations in many EMEs has been fiscal policy. A natural question therefore arises. Can the large and significant effects of public and private debt buildups in EMEs that we uncovered be partly due to a different response of fiscal policy during crises? To explore this question, we look at the evolution of fiscal variables during non-financial and financial recessions in EMEs and AEs. Our analysis contributes to the literature on fiscal pro-cyclicality in EMEs by linking pro-cyclicality during crisis to pre-crisis debt accumulation.

The path of fiscal variables during non-financial and financial recessions is estimated using the regression model (3). We look first at the evolution of public debt, and we then try to identify the fiscal policy stance by looking at per capita real government spending. The literature on EMEs fiscal policy has identified real per capita government spending as the best available indicator to assess the stance of fiscal policy (Kaminsky, Reinhart and Végh, 2004). The reason is that the cyclically adjusted primary balance (CAPB) is not available for a large set of emerging economies going back in time. Moreover, EMEs tend to have a rather volatile growth, and more so during financial crisis, therefore output gap estimates necessary to compute the CAPB are subject to high uncertainty. At the same time, most of the impact to the fiscal accounts of recessions in EMEs works through the revenues, as these countries tend to have limited automatic stabilizers on the spending side (as for example large unemployment schemes). For all these reasons, real per-capita government spending seems to be the best measure to assess the discretionary fiscal stance, as it is the fiscal measure which is more independent from GDP fluctuations.¹¹

To allow for a natural comparison across variables and country groups, prior to the estimation we scale the LHS variables using y_i,p, the level of real per capita GDP at the peak (Hall, 2009; Barro and Redlick, 2011). Formally, by denoting with v real per-capita government spending or real per capita public debt, the LHS variable is now given by the following formula:

Figure 4 shows how public debt, scaled by the pre-downturn level of GDP, evolves during non-financial and financial recessions, with and without the presence of larger-than-average debt buildups. Three main results emerge from this exercise. First, recessions are typically associated with increases in public debt, which tend to be larger during banking crises. Second, the increase in public debt is found to be more marked in AEs than in EMEs. This is not only the case for banking crises (Laeven and Valencia, 2013); it applies to non-financial recessions as well. Finally, if a country enters a non-financial or financial recession with higher private debt buildups, the increase in public debt tends to be larger. This suggests that, especially for financial recessions, private sector support generally weights on the government balance sheet, determining a relocation of debt from private to public debt.

View Full Size

Public Debt Dynamics

Citation: IMF Working Papers 2017, 061; 10.5089/9781475588170.001.A001

Note. Top panel. Dynamics of real per capita public debt in AEs and EMEs, starting from the year preceding non-financial (blue lines) and financial (red lines) recessions. The solid line shows the average path in recessions, while the dashed line shows how the path deviates when the recession is preceded by a rapid debt buildup. Bottom panel. Robust standard errors, clustered by country, are reported in parentheses.

• Download Figure
• Download figure as PowerPoint slide

The analysis in Figure 4 gives a first indication that fiscal support in AEs, measured as the change in public debt with respect to pre-slump GDP levels, could be more intense than in EMEs, as the profile of public debt grows less in EMEs than in AEs after both non-financial and financial recessions. However, despite providing interesting insights on the relationship of debt, downturns, and fiscal policy, the evolution of public debt during recessions is not a clear indicator of the fiscal stance. The public debt dynamics is largely affected by the pre-crisis fiscal position (i.e. if a country enters a financial crisis with a large fiscal deficit, the public debt will keep increasing even if the country embarks in fiscal consolidation). Accordingly, in Figure 5 we also look at the dynamics of real per capita government spending.

View Full Size

Government Spending Dynamics

Citation: IMF Working Papers 2017, 061; 10.5089/9781475588170.001.A001

Note. Top panel. Dynamics of real per capita government spending in AEs and EMEs, starting from the year preceding non-financial (blue lines) and financial (red lines) recessions. The solid line shows the average path in recessions, while the dashed line shows how the path deviates when the recession is preceded by a rapid debt buildup. Bottom panel. Robust standard errors, clustered by country, are reported in parentheses.

• Download Figure
• Download figure as PowerPoint slide

From looking at Figure 5, we see that real per-capita government spending appears to be an important source of heterogeneity between EMEs and AEs. In particular, while non-financial and financial recessions in AEs are typically associated with relatively large increases in real government spending, the corresponding dynamic is more muted in EMEs. Interestingly, large buildups of both private and public debt are associated with a significant negative growth of real government spending after the start of the crisis in EMEs. During financial crises characterized by public and private debt overhangs, we find that fiscal policy tends to react asymmetrically between country-groups: while AEs increase government spending, EMEs decrease it pro-cyclically.

Overall, we find that the fiscal policy stance can act as an amplifier of debt-recession cycles, particularly in EMEs. Our analysis uses government spending as a proxy for discretionary fiscal policy, which admittedly is not completely exogenous from the GDP dynamics. Therefore, our results regarding the effect of public spending on GDP should be considered more as a finding that the two variables are correlated more than causally linked. Still, in our setup the GDP dynamics is initially mainly affected by the outbreak of a banking crisis. Therefore, the chain of causality between larger debt buildups, deeper recessions and larger reductions in real public spending is rather clear in our setup. The start of the recession is triggered by a banking crisis which is exogenous from the impact of public spending on the economy. Reductions in public spending therefore are not the driving force of the recessions, although they certainly contribute to propagate the initial shock.

4.2 … Driven By Tighter Financial Constraints

In this section we present evidence regarding the transmission mechanism between debt accumulation and fiscal policy stance during crises in EMEs. Our goal is to assess why countries that enter a financial recession with larger debt buildups tend to cut real public spending more. In particular, we will show evidence suggesting that countries entering a financial recession with a larger accumulation of private and public debts tend to be more constrained in financing the government deficit once the crisis starts and are therefore forced to contain public spending more. This evidence is particularly in line with our analysis, as financing constraints should indeed be tighter if the country enters a crisis with larger private and/or public debt buildups.

The link that we highlight between large debt buildups on one side and, on the other, tighter financing constraints and larger reduction of public spending during crises contributes to our understanding of the nature of pro-cyclicality of fiscal policy in EMEs. But some clarifications are in order. First, our result point to a pro-cyclical fiscal stance during financial crises and not in general. Gavin and Perotti (1997) for example focus on the experience of Latin America and argue that “the pro-cyclicality of fiscal policy in Latin America has to do with a loss during macroeconomic bad times of the market access that would be required to support a more countercyclical fiscal policy”. This interpretation is not inconsistent with alternative explanations of fiscal pro-cyclicality in EMEs. For example, Ilzetzki (2011) shows how the presence of polarized political systems can account for the pro-cyclical stance of fiscal policy, while Frankel, Vegh and Vuletin (2013) find evidence of a causal link running from institutional quality to fiscal stance. It is not unconceivable that countries with highly polarized political systems and low institutional quality are more prone to debt accumulations and might therefore end up with tighter access to financial markets in bad times.

But how do we assess whether financing constraints were indeed tighter for financial recessions burdened by excessive pre-crisis accumulation of debt? To provide some evidence on this issue, we follow Gavin and Perotti (1997) and look at the access to official emergency credit during financial crisis. The main source of official credit of last resort for countries is IMF loans. We therefore re-run our regressions in equation (3) using as a dependent variable the use of IMF credit.¹² IMF credit measures the net outstanding debt of a country toward the IMF and includes normal and exceptional financing under all IMF facilities.¹³ Figure 6 shows that the use of IMF credit, scaled by the pre-downturn level of public debt pub_i,P, is higher in EMEs, and more so during financial recessions.¹⁴ It is particularly high during financial recessions with high private and public debt buildups. This suggests that during these types of recessions, government financing needs could not be possibly met by borrowing on the market. Resort to the IMF was the only way to secure financing. Clearly, with rising public debt (Figure 4) and no access to financial markets, countries were not able to run counter-cyclical fiscal policy and had to restrain spending.¹⁵

View Full Size

IMF Credit Dynamics

Citation: IMF Working Papers 2017, 061; 10.5089/9781475588170.001.A001

Note. Top panel. Dynamics of real per-capita IMF credit in AEs and EMEs, starting from the year preceding non-financial (blue lines) and financial (red lines) recessions. The solid line shows the average path in recessions, while the dashed line shows how the path deviates when the recession is preceded by a rapid debt buildup. Bottom panel. Robust standard errors, clustered by country, are reported in parentheses.

• Download Figure
• Download figure as PowerPoint slide

Our evidence so far underscores an important difference between EMEs and AEs, i.e. their capacity to run counter-cyclical fiscal policy during recessions and crises. This might be related to institutional differences that make EMEs more prone to public debt accumulation and that constraint their fiscal response when crises explode. In the case of AEs, Jordà, Schularick and Taylor (2016) have highlighted a rather different transmission mechanism, one that works mainly through the reduction of the availability of credit during a banking crisis causing a fall in investment. A reduction in credit and investment is not inconsistent with the fiscal pro-cyclicality interpretation that we have highlighted. Indeed, in the following we report evidence showing that the fall in investment is a characteristic also of EMEs financial crisis.

As a proxy for investment we look at the National Account series for gross fixed capital formation, which includes both private and public investment. Given the higher volatility of private as opposed to public investment, gross fixed capital formation is mainly driven by private investment, whose dynamic substantially reflects credit market conditions as the financing of new capital usually requires the use of credit. Figure 7 shows that indeed countries with the most rapid debt accumulations before recessions are the ones with the highest reductions in gross fixed capital formation once the crisis starts. This effect is particularly significant after large private debt accumulations. This confirms that the fall in GDP after pre-crisis buildups in debt is accompanied by a severe decline in investment spending, particularly in EMEs.

View Full Size

Capital Formation Dynamics

Citation: IMF Working Papers 2017, 061; 10.5089/9781475588170.001.A001

Note. Top panel. Dynamics of real per-capita capital formation in AEs and EMEs, starting from the year preceding non-financial (blue lines) and financial (red lines) recessions. The solid line shows the average path in recessions, while the dashed line shows how the path deviates when the recession is preceded by a rapid debt buildup. Bottom panel. Robust standard errors, clustered by country, are reported in parentheses.

• Download Figure
• Download figure as PowerPoint slide

Overall, our evidence suggests that the transmission mechanisms through which large pre-crisis debt buildups amplify recessions in EMEs is somewhat similar to that of AEs and somewhat different. It is similar to the extent that investment falls considerably. It is different to the extent that the response of fiscal policy tends to be pro-cyclical or, more generally, less supportive than in AEs.

A Data

As a proxy for private debt, we use International Financial Statistics (IFS) data on banks’ claims on the private sector (line 22d in IFS) for 36 AEs and 77 EMEs, starting as early as 1950 and, on average, from 1964. This coverage is much larger than the widely used Bank for International Settlements (BIS) data, which cover 27 AEs and 14 EMEs. Claims include relatively minor items other than loans, such as securities and shares, financial derivatives, and trade credit advances. With the exception of shares, these other items are part of private sector debt. However, the difference between claims and loans is generally small. Loans accounted, on average, for 98 percent of bank claims in countries reporting to the IFS Standardized Reporting Forms in 2015 (for which the decomposition between loans and other item is available). Data on public debt come from an updated version of the IMF historical debt dataset (Abbas et al., 2010). The original historical dataset has been updated with the April 2015 World Economic Outlook (WEO) data on general government debt. The data cover 37 AEs and 87 EMEs, starting, on average, in 1940. Figure 1 summarizes the data on private and public debt as a share of GDP. It takes simple averages across EMEs and AEs of bank credit and public debt starting in 2001. It shows very different patterns in the two groups of countries. In AEs, private debt started to decline after the global financial crisis; in EMEs, it continued to increase at a fast pace. In EMEs, public debt also increased rapidly after the global financial crisis, inverting the previous trend. These data on private and public debt are combined with data on real per capita GDP from the WEO database complemented by data from the Penn Word Tables. Data on government expenditures are taken from Mauro, Romeu, Binder and Zaman (2015). Data on IMF credit are from the IMF Financial Flows Analytics database. Finally, data on gross fixed capital formation are from the IMF IFS database, complemented in the case of a few countries by data from the WEO database. The map in Figure A1 shows the country coverage of our dataset.

View Full Size

Worldwide Data Coverage

Citation: IMF Working Papers 2017, 061; 10.5089/9781475588170.001.A001

Note. The figure shows the worldwide country coverage of our dataset, which is exploited either in the local projections analysis or for the estimation of the probabilistic models.

• Download Figure
• Download figure as PowerPoint slide

B Additional Information

In this section we report additional information on some key variables used throughout the analysis.

Table A2.

List of Business Cycle Peaks (t = p) Used in the Local Projections

Note. The list shows the peaks (t = p) that are used, at least once, for the estimation of the local projection models. A generic country i at time p is included only if (a) all the corresponding LHS and RHS observations are available, and (b) the country in question has experienced at least two recessions in the available sample. The years in blue are non-financial peaks. The boldfaced years in red are financial peaks.

Table A2.

List of Business Cycle Peaks (t = p) Used in the Local Projections

	Advanced Economies		Emerging Market Economies
Australia	1973, 1981, 1990, 2008	Algeria	1996, 2008
Austria	1974, 1977, 1980, 1992, 2008, 2012	Argentina	1969, 1974, 1977, 1979, 1984, 1987,
Belgium	1974, 1980, 1992, 2007, 2011		1994, 1998, 2008, 2011
Canada	1956, 1981, 1989, 2007	Barbados	2000, 2008
Czech Republic	2008, 2011	Belize	2004, 2006, 2010, 2012
Denmark	1973, 1979, 1987, 1992, 2007, 2011	Bosnia and Herzegovina	2008, 2011
Finland	1975, 1989, 2008, 2011	Brazil	1969, 1980, 2000, 2001, 2008
France	1974, 1992, 2007, 2011	Brunei	2006, 2011
Germany	1966, 1974, 1980, 1992, 2001, 2008	Chile	1971, 1981, 1998, 2008
Greece	1973, 1979, 1986, 1989, 1991, 2007	Colombia	1974, 1979, 1981, 1997
Iceland	1960, 1966, 1982, 1987, 1990, 1994,	Costa Rica	1974, 1979, 1984, 1990, 1995, 2000,
	2001, 2007		2008
Ireland	1974, 1982, 2007, 2011	Dominican Republic	1977, 1981, 1983, 1987, 1989, 2001,
Israel	1988, 1990, 2000, 2008		2008
Italy	1974, 1992, 2001, 2007, 2011	Ecuador	1998, 2008
Japan	1973, 2007, 2010	Equatorial Guinea	2005, 2008, 2012
Korea	1979, 1997	Gabon	1998, 2008
Luxembourg	2007, 2011	Hungary	2008, 2011
Netherlands	2001, 2008, 2011	India	1970, 1973, 1975, 1978
New Zealand	1975, 1978, 1982, 2007	Iran	2007, 2011
Norway	1977, 1981, 1987	Jamaica	1999, 2001, 2007, 2011
Portugal	1974, 1983, 1992, 2001, 2008, 2010	Kuwait	1994, 1998, 2000, 2006, 2008, 2012
Singapore	1997, 2000, 2007	Lebanon	1998, 2001, 2004, 2010
Slovak Republic	1998, 2008	Libya	1996, 2000
Slovenia	2008, 2011	Macedonia	2008, 2011
Spain	1978, 1980, 1992, 2007	Malaysia	1997, 2000, 2008
Sweden	1976, 1990, 2007, 2011	Mexico	1981, 1985, 1994, 2000, 2008
Switzerland	1974, 1981, 1990, 1994, 2001, 2008	Morocco	1996, 1998
United Kingdom	1957, 1973, 1979, 1990, 2007	Pakistan	2000, 2008
United States	1956, 1969, 1973, 1979, 1981, 1990,	Panama	1957, 1973, 1979, 2000, 2008
	2000, 2007	Paraguay	1995, 1997, 2008, 2011
		Peru	1975, 1981, 1984, 1987, 1991, 1997,
			2000, 2008
		Serbia	2008, 2011
		Seychelles	1993, 1998, 2000, 2007
		South Africa	1971, 1974, 1981, 1984, 1989, 1997,
			2008
		Syria	1996, 1998, 2001
		Thailand	1996,2008
		Trinidad and Tobago	1991, 2008
		Turkey	1977, 1979, 1982, 1988, 1990,
			1993, 1997, 2000, 2007
		Uruguay	1981, 1989, 1994, 1998
		Venezuela	1970, 1974, 1977, 1984, 1988,
			1992, 2008, 2012

Note. The list shows the peaks (t = p) that are used, at least once, for the estimation of the local projection models. A generic country i at time p is included only if (a) all the corresponding LHS and RHS observations are available, and (b) the country in question has experienced at least two recessions in the available sample. The years in blue are non-financial peaks. The boldfaced years in red are financial peaks.

View Table

Table A2.

List of Business Cycle Peaks (t = p) Used in the Local Projections

	Advanced Economies		Emerging Market Economies
Australia	1973, 1981, 1990, 2008	Algeria	1996, 2008
Austria	1974, 1977, 1980, 1992, 2008, 2012	Argentina	1969, 1974, 1977, 1979, 1984, 1987,
Belgium	1974, 1980, 1992, 2007, 2011		1994, 1998, 2008, 2011
Canada	1956, 1981, 1989, 2007	Barbados	2000, 2008
Czech Republic	2008, 2011	Belize	2004, 2006, 2010, 2012
Denmark	1973, 1979, 1987, 1992, 2007, 2011	Bosnia and Herzegovina	2008, 2011
Finland	1975, 1989, 2008, 2011	Brazil	1969, 1980, 2000, 2001, 2008
France	1974, 1992, 2007, 2011	Brunei	2006, 2011
Germany	1966, 1974, 1980, 1992, 2001, 2008	Chile	1971, 1981, 1998, 2008
Greece	1973, 1979, 1986, 1989, 1991, 2007	Colombia	1974, 1979, 1981, 1997
Iceland	1960, 1966, 1982, 1987, 1990, 1994,	Costa Rica	1974, 1979, 1984, 1990, 1995, 2000,
	2001, 2007		2008
Ireland	1974, 1982, 2007, 2011	Dominican Republic	1977, 1981, 1983, 1987, 1989, 2001,
Israel	1988, 1990, 2000, 2008		2008
Italy	1974, 1992, 2001, 2007, 2011	Ecuador	1998, 2008
Japan	1973, 2007, 2010	Equatorial Guinea	2005, 2008, 2012
Korea	1979, 1997	Gabon	1998, 2008
Luxembourg	2007, 2011	Hungary	2008, 2011
Netherlands	2001, 2008, 2011	India	1970, 1973, 1975, 1978
New Zealand	1975, 1978, 1982, 2007	Iran	2007, 2011
Norway	1977, 1981, 1987	Jamaica	1999, 2001, 2007, 2011
Portugal	1974, 1983, 1992, 2001, 2008, 2010	Kuwait	1994, 1998, 2000, 2006, 2008, 2012
Singapore	1997, 2000, 2007	Lebanon	1998, 2001, 2004, 2010
Slovak Republic	1998, 2008	Libya	1996, 2000
Slovenia	2008, 2011	Macedonia	2008, 2011
Spain	1978, 1980, 1992, 2007	Malaysia	1997, 2000, 2008
Sweden	1976, 1990, 2007, 2011	Mexico	1981, 1985, 1994, 2000, 2008
Switzerland	1974, 1981, 1990, 1994, 2001, 2008	Morocco	1996, 1998
United Kingdom	1957, 1973, 1979, 1990, 2007	Pakistan	2000, 2008
United States	1956, 1969, 1973, 1979, 1981, 1990,	Panama	1957, 1973, 1979, 2000, 2008
	2000, 2007	Paraguay	1995, 1997, 2008, 2011
		Peru	1975, 1981, 1984, 1987, 1991, 1997,
			2000, 2008
		Serbia	2008, 2011
		Seychelles	1993, 1998, 2000, 2007
		South Africa	1971, 1974, 1981, 1984, 1989, 1997,
			2008
		Syria	1996, 1998, 2001
		Thailand	1996,2008
		Trinidad and Tobago	1991, 2008
		Turkey	1977, 1979, 1982, 1988, 1990,
			1993, 1997, 2000, 2007
		Uruguay	1981, 1989, 1994, 1998
		Venezuela	1970, 1974, 1977, 1984, 1988,
			1992, 2008, 2012

Note. The list shows the peaks (t = p) that are used, at least once, for the estimation of the local projection models. A generic country i at time p is included only if (a) all the corresponding LHS and RHS observations are available, and (b) the country in question has experienced at least two recessions in the available sample. The years in blue are non-financial peaks. The boldfaced years in red are financial peaks.

Table A3.

List of Banking Crises Episodes (B_i,t = 1) Used in the Logistic Models

Note. The list shows the banking crises episodes (B_i,t = 1) that are used, at least once, for the estimation of the probabilistic models. A banking crisis episode is included only if all the corresponding LHS and RHS observations are available.

Table A3.

List of Banking Crises Episodes (B_i,t = 1) Used in the Logistic Models

	Advanced Economies		Emerging Market Economies
Australia	1989	Algeria	1990
Austria	2008	Argentina	1980, 1989, 1995, 2001
Belgium	2008	Brazil	1963, 1985, 1990, 1994
Canada	1983	Chile	1976, 1981
Denmark	1987, 2008	China	1992, 1998
Finland	1991	Colombia	1982, 1998
France	1994, 2008	Costa Rica	1987, 1994
Germany	1977, 2008	Dominican Republic	1996, 2003
Greece	1991, 2008	Ecuador	1981, 1998
Iceland	1985, 1993, 2007	Egypt	1980, 1990
Ireland	2007	El Salvador	1989
Israel	1977	Guatemala	1990, 2001, 2006
Italy	1990, 2008	Hungary	1991, 2008
Japan	1992, 1997	India	1993
Korea	1983, 1985, 1997	Indonesia	1992, 1994, 1997
Latvia	2008	Jamaica	1996
Luxembourg	2008	Jordan	1989
Netherlands	2008	Kazakhstan	2008
New Zealand	1987	Kuwait	1982
Norway	1987, 1991	Lebanon	1990
Portugal	2008	Malaysia	1985, 1997
Singapore	1982	Mexico	1981, 1994
Slovak Republic	1998	Morocco	1980, 1983
Slovenia	2008	Panama	1988
Spain	1977, 2008	Paraguay	1995, 2001
Sweden	1991, 2008	Peru	1983, 1999
Switzerland	1991, 2008	Philippines	1981, 1983, 1997
United Kingdom	1974, 1984, 1991, 1995, 2007	Poland	1991
United States	1984, 1988, 2007	Romania	1990
		Russia	1998, 2008
		South Africa	1977, 1989
		Sri Lanka	1989
		Swaziland	1995
		Thailand	1980, 1983, 1996
		Tunisia	1991
		Turkey	1982, 1991, 1994, 2000
		Ukraine	1998, 2008
		Uruguay	1971, 1981, 2001
		Venezuela	1978, 1993

Note. The list shows the banking crises episodes (B_i,t = 1) that are used, at least once, for the estimation of the probabilistic models. A banking crisis episode is included only if all the corresponding LHS and RHS observations are available.

View Table

Table A3.

List of Banking Crises Episodes (B_i,t = 1) Used in the Logistic Models

	Advanced Economies		Emerging Market Economies
Australia	1989	Algeria	1990
Austria	2008	Argentina	1980, 1989, 1995, 2001
Belgium	2008	Brazil	1963, 1985, 1990, 1994
Canada	1983	Chile	1976, 1981
Denmark	1987, 2008	China	1992, 1998
Finland	1991	Colombia	1982, 1998
France	1994, 2008	Costa Rica	1987, 1994
Germany	1977, 2008	Dominican Republic	1996, 2003
Greece	1991, 2008	Ecuador	1981, 1998
Iceland	1985, 1993, 2007	Egypt	1980, 1990
Ireland	2007	El Salvador	1989
Israel	1977	Guatemala	1990, 2001, 2006
Italy	1990, 2008	Hungary	1991, 2008
Japan	1992, 1997	India	1993
Korea	1983, 1985, 1997	Indonesia	1992, 1994, 1997
Latvia	2008	Jamaica	1996
Luxembourg	2008	Jordan	1989
Netherlands	2008	Kazakhstan	2008
New Zealand	1987	Kuwait	1982
Norway	1987, 1991	Lebanon	1990
Portugal	2008	Malaysia	1985, 1997
Singapore	1982	Mexico	1981, 1994
Slovak Republic	1998	Morocco	1980, 1983
Slovenia	2008	Panama	1988
Spain	1977, 2008	Paraguay	1995, 2001
Sweden	1991, 2008	Peru	1983, 1999
Switzerland	1991, 2008	Philippines	1981, 1983, 1997
United Kingdom	1974, 1984, 1991, 1995, 2007	Poland	1991
United States	1984, 1988, 2007	Romania	1990
		Russia	1998, 2008
		South Africa	1977, 1989
		Sri Lanka	1989
		Swaziland	1995
		Thailand	1980, 1983, 1996
		Tunisia	1991
		Turkey	1982, 1991, 1994, 2000
		Ukraine	1998, 2008
		Uruguay	1971, 1981, 2001
		Venezuela	1978, 1993

Note. The list shows the banking crises episodes (B_i,t = 1) that are used, at least once, for the estimation of the probabilistic models. A banking crisis episode is included only if all the corresponding LHS and RHS observations are available.