Ricardian Equivalence and Sovereign Default Risk

Stefan Eichler†

Ju Hyun Pyun‡ Korea University Business School

Technische Universität Dresden and Halle Institute for Economic Research

This version: January 2017

Abstract We study the impact of sovereign default risk on the private–public savings offset. Using data on 80 countries for the period 1989–2010, we find robust evidence for a U-shaped pattern in the private–public savings offset in foreign currency sovereign credit ratings. While Ricardian Equivalence holds approximately at intermediate levels of sovereign solvency, it breaks down at very low and very high levels of sovereign default risk. In particular, the U-shaped pattern is an emerging market phenomenon as well as confirmed by external public debt, but not domestic public debt. A key result is that in the presence of foreign ownership of sovereign bonds, sovereign default constitutes a net wealth gain for domestic consumers as the present value of saved future taxes outweighs their wealth loss on bond holding. Thus, in times of high default risk, consumers appear to anticipate that the government would rather dilute bondholders than repay sovereign debt using higher taxes.

JEL classification: E21, E62, F40, G01, H60 Keywords: Fiscal Policy, Sovereign Default Risk, Ricardian Equivalence, Private Saving, External Public Debt, Emerging Markets.



We are grateful to Jongseok Han, Soyoung Kim, Yun Jung Kim, Eric Leeper, Dominik Maltritz, Morten Ravn, and Martin Uribe for their valuable comments. All remaining errors are our own. † Technische Universität Dresden, Chair of International Monetary Economics, Dresden Germany; Halle Institute for Economic Research, Department of Financial Markets; E-mail: [email protected]. ‡ Corresponding author; Business School, Korea University, 145 Anam-Ro, Seongbuk-Gu, Seoul 02841, Korea, Email: [email protected], Tel: 82-2-3290-2610.

1

1. Introduction We investigate the role of sovereign default risk in the relationship between private and public savings across countries. According to Barro (1989), the modification to Ricardian Equivalence (RE)1 suggests that rational consumers will increase their private savings when the government increases the current fiscal deficits. This is because they expect higher public debt levels to be equivalent to a higher future tax burden, so that a decrease in public savings is offset by an increase in private savings. Previous research has discussed breakdowns of RE in a Keynesian state of the economy, where free fiscal resources allow the government to increase public debt without the need to increase taxes in the future. Here, we focus on sovereign default risk as an alternative explanation as to why RE may break down. In a state with significant sovereign default risk, consumers may expect that fiscal deficits will be accompanied by debt default (especially external debt) rather than by higher future taxes. Consequently, rational consumers will not increase private savings in a one-to-one ratio as a response to higher fiscal deficits. Using panel data on 80 economies for the period 1989–2010, we find robust evidence of a U-shaped pattern in the private–public savings offset in foreign currency sovereign credit ratings. While consumers’ responses to fiscal deficit get closer to RE for intermediate levels of ratings (where fiscal space is limited, but sovereign default risk is still moderate), RE breaks down not only in the Keynesian state with good ratings, but also in the sovereign default risk state. These findings suggest that consumers consider sovereign default risk (as indicated by credit ratings) when deciding on their optimal consumption in response to fiscal deficits.

“Ricardian equivalence (RE)” means that changes in the composition of government expenditure finance have no real effect on consumption. This is why an increase in today’s fiscal deficit (via a cut in taxes) should be matched by a corresponding increase in the present value of future taxes, holding fixed the path of government expenditure (Barro, 1989). Theoretically, the validity of RE proposition depends upon the length of consumer’s planning horizons and the existence of intergenerational transfer (Bernheim, 1987). This RE view is also skeptical about the Keynesian view of fiscal deficit and argues that fiscal policy cannot be effective even in the short run (see Ricciuti (2003) for a good summary of RE). 1

2

We present several empirical indications that the U-shaped behavior of the private-public savings offset in sovereign credit ratings is only prevalent in the setting where foreign ownership plays a role. For example, we find that the U-shaped pattern only prevails when foreign currency ratings are considered, while insignificant results are obtained for domestic currency ratings which are only indicative for the risk faced by domestic creditors. Second, we find that the Ushaped pattern only prevails for emerging markets countries (and not for developed countries) where typically a large fraction of sovereign bonds are held by foreign creditors who bear the lion’s share of wealth losses associated with sovereign default. Moreover, investigating different types of sovereign debt ratios, only the external sovereign debt ratios reveal the U-shaped pattern. Furthermore, the U-shaped pattern is more pronounced in countries with a sovereign default history. The pattern only occurs in countries with relatively small and closed financial markets, suggesting that sovereign refinancing capacities shape consumers’ expectations of sovereign default. The existing literature has so far focused on only two states when examining the nonlinearity of fiscal policy: free fiscal space, with a Keynesian state of the economy; and tight/no fiscal space, where RE holds.2 However, there is no theoretical or empirical research that investigates the nonlinearity of private–public savings offsets when government is on the verge of default. We focus on states where considerable sovereign risk is anticipated by consumers, which gives rise to a U-shaped response of private saving to a fiscal deficit, with respect to fiscal sustainability. Our empirical findings confirm the above predictions and provide important

2

Given the theoretical predictions of previous studies, the response of private saving to public saving, which measures the Ricardian offset, varies with fiscal sustainability. In a Keynesian state with low sovereign default risk and good ratings, the response of private saving to a fiscal deficit is far from RE, because the government has enough fiscal space to finance the fiscal deficit without raising taxes in the near future. For intermediate levels of fiscal solvency, where fiscal space is limited, but sovereign default risk is still moderate, a higher fiscal deficit will probably be refinanced by higher future taxes, leading to a one-to-one private–public savings offset.

3

insights for current fiscal policy research. Previous theoretical contributions considering intertemporal budget constraints argue that the effect of fiscal policy is nonlinear in terms of public debt levels. Sutherland (1997) and Perotti (1999) show that at moderate levels of debt, fiscal policy has traditional Keynesian effects. However, when public debt reaches high values, a fiscal deficit can have a contractionary effect on consumption. According to their arguments, RE appears when the debt–income ratio increases. Bertola and Drazen (1993) consider the nonlinear effects of fiscal policy in terms of the size of government spending, rather than debt (see Beetsma, 2008 or Giavazzi et al., 2000, for an excellent discussion on the competing theoretical hypotheses). Thus, the public debt level or the size of fiscal deficit may determine the private–public savings offset. Nevertheless, the aforementioned theoretical studies focus on public (domestic) debt in a closed economy setting and do not consider foreign creditors and the possibility of sovereign default. This study, by introducing sovereign default risk in foreign currency, argues that breakdown of RE under heightened default risk would be more pronounced when government has a larger share of external debts. Sovereign default has non-trivial net wealth effects for domestic consumers depending on the share of foreign creditors of domestic sovereign debt. If domestic sovereign debt was entirely held by domestic consumers, their wealth losses associated with sovereign default would equal the present value of saved future taxes. Thus, in such situations with no foreign ownership of domestic sovereign debt, RE would hold. On the contrary, if some fraction of domestic sovereign debt is owned by foreign creditors, the wealth losses associated with sovereign default would not entirely be made by domestic consumers and thereby their net wealth effect after considering their savings on future taxes is positive. Thus, the presence of foreign creditors of sovereign debt, constitutes a deviation from 4

RE in the sovereign default risk state, that is, fiscal deficit under heightened sovereign default risk allows domestic households to consume more (and save less) and breaks RE. The empirical specification of our study is built upon the work of Corbo and SchmidtHebbel (1991) and Loayza et al. (2000), which uses a private saving function to investigate the effect of fiscal policy. The advantage of examining the saving function and fiscal policy is that we can compare the results directly with the RE proposition (Giavazzi et al., 2000). 3 Giavazzi et al. (2000) examine the effect of fiscal policy conditional on the size of the fiscal impulse and public debt-to-GDP ratio using the specification of the national saving rate. Our empirical study extends previous theoretical predictions on the nonlinear fiscal policy effect in terms of fiscal status (Sutherland 1997; Perotti 1999) by adding another possibility for nonlinearity. Our novel finding is that RE breaks down again with extremely poor sovereign ratings, which suggests that the private–public saving offset indeed exhibits a U-shape. While previous studies have focused on the effect of the fiscal policy of developed countries with respect to public debt level―where the risk of sovereign default is typically a minor issue (e.g., Giavazzi et al., 2000), few studies have examined the effectiveness of the fiscal policy in countries with (perceived) very low fiscal solvency, or even fiscal insolvency.4 This study contributes to a growing body of literature on fiscal sustainability. Previous theoretical studies distinguish between the fiscal limit and the actual debt-to-GDP ratio of countries, and then examine the relationship between sovereign default risk and the level of 3

Barro (1989) states that when RE holds (i.e., fiscal policy is ineffective), a reduction in public saving is offset, onefor-one, by an increase in private saving, with national saving being constant. In this setting, taxes are lump sum or otherwise completely non-distorting and government debt is real (indexed to inflation) and not defaultable. Many previous studies have made cross-country comparisons of the private–public saving offset in the context of RE. Seater (1993) provides an extensive survey of the empirical literature on RE, and concludes that RE is valid whereas Bernheim (1987) refutes the validity of RE. A subsequent empirical study by Röhn (2010) confirms the nonlinearity of RE using cross-country data. 4 In this regard, our work is closely related to those previous studies that examine the heterogeneous effects of fiscal policy on a broader set of macro variables, such as output, consumption, and the current account, using cross-country data (Corsetti et al., 2012; Ilzetzki et al., 2013; Kim, 2015; Nickel and Tudyka, 2014; Pyun and Rhee, 2015).

5

government debt5 (e.g., Bi, 2012). Subsequent empirical contributions, such as Ghosh et al. (2013) and Ostry et al. (2010), estimate the fiscal space (the distance between fiscal limits and the current debt level) in developed countries based on historical fiscal reactions to debt.6 Our work is in line with these previous works in that we distinguish between fiscal sustainability, measured as sovereign ratings, and the public debt-to-GDP ratio. However, our study is differentiated from these works because we focus on the relevance of fiscal sustainability for the private–public savings offset by using sovereign credit ratings. As opposed to the fiscal measures used in the studies above, ratings can be easily observed by domestic consumers in real time and may thus shape the private–public savings offset. Moreover, we further distinguish between foreign currency and local currency ratings in order to study if a possible dilution of foreign creditors affects the consumption of domestic consumers. Lastly, this study contributes to a strand of literature focusing on the implications of sovereign default risk. The negative effects of sovereign default include diminished access of private firms to international credit during sovereign default episodes (Arteta and Hale, 2008), higher corporate borrowing costs (Agca and Celasun, 2012), a decrease in economic growth (Borensztein and Panizza, 2009; Levy-Yeyati and Panizza, 2011; Furceri and Zdzienicka, 2012), a reduction in international trade (Rose, 2005), and a higher probability of banking crises (Reinhart and Rogoff, 2011, 2013; Demirgüç-Kunt and Huizinga, 2013; Acharya et al., 2014; Gennaioli et al., 2014). We focus on the role of sovereign default risk for the Ricardian offset, which, to the best of our knowledge, has not been studied before.

5

Many theoretical studies assess fiscal sustainability in a general equilibrium model. Mendoza and Oviedo (2009) introduce the “natural debt limit,” capturing the maximum debt level that a government remains able to service fully. Corsetti et al. (2013) highlight the direct role of sovereign risk in determining fiscal policy effectiveness (when monetary policy is constrained). 6 They compute a debt limit for each country, which is determined completely by the risk-free interest rate, the recovery rate, and the support of the shock to the primary balances.

6

The remainder of the paper is organized as follows. In Section 2, introduces the data set and formulates an empirical specification for private saving rates. Section 3 presents the main empirical results on the private–public saving offset. In Section 4, we provide several channels that constitute the U-shaped pattern of the savings offset. Concluding remarks follow in Section 5.

2. Empirical model and data 2.1. Data Our data set covers 80 developed and emerging/developing countries for the period 1989–2010 (the country list is provided in the appendix, Table A1). Table A2 in the appendix presents the definitions and sources of the variables used. Then, Table A3 in the appendix reports the summary statistics. We set up a reduced form, cross-country private saving function following the life cycle model (e.g., Modigliani and Brumberg, 1955), which explains saving patterns in terms of the income and demographic structure of the economy. First, our analysis focuses on fiscal policy (i.e., the response of private savings to a fiscal deficit). Here, we include gross public savings (current government revenue – current government expenditure) over GDP to measure the extent to which fiscal policy affects economic agents’ savings. Barro (1989) also suggests that RE theorem can be tested by examining whether a reduction in public saving (or an increase in the current fiscal deficit) is offset by an increase in private saving. In order to estimate a U-shaped pattern of the private savings–public savings offset, we relate a country’s public savings to its sovereign credit rating in an interaction framework. Sovereign ratings constitute the rating agency’s assessment of the ability of the sovereign issuer’s

7

ability to repay its debt. We use Standard & Poor’s sovereign credit ratings.7 We consider foreign currency sovereign credit ratings in order to prevent currency risk affecting our sovereign risk measure. In order to convert the ratings (AAA to D) into numerical values, we use the rating migration table proposed by Correa et al. (2014), which is similar to other migration tables used in the literature (e.g., see Brooks et al., 2004; Gande and Parsley, 2005; and Afonso et al., 2012). The rating variable ranges from 0 (D, SD – sovereign default) to 21 (AAA). The rating variable also accounts for a rating outlook, where 0.2 is added for “Positive outlook,” 0.1 for “Watchlist positive,” -0.1 for “Watchlist negative,” and -0.2 for “Negative outlook.” Table A4 in the appendix reports the numerical values assigned to each rating category. Previous studies typically use the public debt-to-GDP ratio as a measure of fiscal sustainability (Sutherland, 1997; Perotti, 1999; Giavazzi et al., 2000; Nickel and Tudyka, 2014).8 These studies distinguish between a Ricardian state with high public debt levels and a Keynesian state with low public debt levels. We argue that states with high public debt comprise two states of sovereign creditworthiness: a state of moderate sovereign default risk, where RE approximately holds; and a high sovereign default risk state, where RE breaks down. For example, for some developed countries such as Japan or the United States, large public debt-toGDP ratios (i.e., even above 100%) are considered fiscally sustainable. However, for most emerging market countries, such public debt ratios are out of reach, and most of these countries would experience a sovereign debt crisis when breaking much lower public debt ratios. Sovereign The results are robust when ratings of the other two major rating agencies (Moody’s and Fitch) are used. However, we opt to use the ratings issued by Standard & Poor’s, because several papers find that their rating adjustments are faster, on average, than those of their competitors (e.g., see Brooks et al., 2004; Gande and Parsley, 2005). 8 Other papers use financial crisis dummies. The definition of these financial crises are often based on the occurrence of banking crises (Laeven and Valencia, 2012), for example. However, such banking crisis dummies are not sufficient to conclude fiscal sustainability. Moreover, even if sovereign debt crisis dummies could be used, the information value of the dummy would be limited because we aim to test for a possible U-shaped effect of public spending shocks on saving, conditional on sovereign solvency. Thus, using a sovereign rating allows us to measure sovereign solvency in a relatively precise manner, and enables us to test for a continuous U-shaped impact of public spending shocks on the savings rate. 7

8

ratings capture the country- and time-specific features that drive fiscal sustainability more effectively than public debt ratios do. In addition, consumers in a country may monitor sovereign rating changes to assess the sustainability of public finances in their country. In addition to these informational advantages, sovereign rating agencies, to some extent, govern the fiscal sustainability of countries, because they can potentially shut down a country’s access to capital markets after severe rating downgrades. In the private saving function, we also include several control variables such as (log) GDP per capita and GDP per capita growth. While a higher level of GDP per capita or income is expected to increase the saving rate proportionally, the effect of income growth on the private saving rate is ambiguous. Loayza et al. (2000) confirm the positive effect of income on private saving rates. A high income growth can raise the aggregate saving if this income growth is temporary (Masson et al., 1998). However, if the income growth is expected to be “permanently” high, then the income growth causes an increase in consumption, not saving, because private agents expect that their permanent income will increase too. We control for demographic variables such as young and old age dependency ratios and life expectancy. The young dependency ratio is the ratio of the population aged 0–14 to the population aged 15–64. The old dependency ratio is the ratio of the population aged 65+ to the population aged 15–64. A large youth and old population leads to an increase in consumption at the expense of saving. Previous studies have shown that higher youth and old dependency ratios are associated with lower private saving rates. Life expectancy is total years of life expectancy at birth, which is expected to be positively associated with the need for savings. Financial depth (or the size of the financial system) and overall liquidity condition affect private savings. We include M2/GDP to measure financial depth. If agents suffer from liquidity 9

constraints or less available domestic credit, they would save more (Jappelli and Pagano, 1994). We also include terms of trade shocks. Changes in the terms of trade in goods and services can have ambiguous consequences for savings.9 Banking, currency and debt crisis dummies are also included to control for the effects of the external shocks on private saving rate.

2.2. Empirical specification To test for a possible U-shaped pattern of RE in terms of the sovereign rating variable, we set up a private saving equation, as follows:

𝑃𝑟𝑖𝑣𝑎𝑡𝑒_𝑠𝑎𝑣𝑖𝑡 = 𝛼 + 𝛽1 𝑃𝑢𝑏𝑙𝑖𝑐_𝑠𝑎𝑣𝑖𝑡 +𝛽2 𝑃𝑢𝑏𝑙𝑖𝑐_𝑠𝑎𝑣𝑖𝑡 × 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 2 +𝛽3 𝑃𝑢𝑏𝑙𝑖𝑐_𝑠𝑎𝑣𝑖𝑡 × 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 + 𝑋𝑖𝑡 𝛿 + 𝜑𝑖 + 𝑡𝑡 + 𝑒𝑖𝑡 , (1)

where the dependent variable, 𝑃𝑟𝑖𝑣𝑎𝑡𝑒_𝑠𝑎𝑣𝑖𝑡 , is the private savings-to-GDP ratio of country i at time t. We also include the public savings-to-GDP ratio (𝑃𝑢𝑏𝑙𝑖𝑐_𝑠𝑎𝑣𝑖𝑡 ). For robustness, we use an alternative saving rate measure, namely the GNI (instead of GDP). Then 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 and 2 𝑅𝑎𝑡𝑖𝑛𝑔𝑖𝑡 denote the rating measure and its quadratic term, respectively. The latter two variables

interact with the public savings-to-GDP ratio, after which we test the nonlinearity of the RE. In addition to our main regressors, 𝑋𝑖𝑡 is a vector of controls that affect the private saving rate, as discussed above. Then, 𝑋𝑖𝑡 includes GDP per capita and its growth, as well as the demographic variables—the young and old dependency ratios, life expectancy ratio, M2/GDP, a change in the terms of trade and financial crisis dummies. Lastly, 𝜑𝑖 and 𝑡𝑡 denote country fixed effects and

9

A temporary improvement in the terms of trade leads to an increase in savings because it causes real income to increase temporarily (the Harberger–Laursen–Metzler effect). However, a permanent positive shock to the terms of trade increases permanent income, which leads to an increase in consumption (or a decrease in saving).

10

year fixed effects, respectively, and 𝑒𝑖𝑡 is an error term. The country fixed effects estimator, in general, is consistent; however, its estimates may be biased when the dependent variable depends on its lagged dependent variable in the panel structure (Nickell, 1981) 10 . For instance, previous studies such as Loyaza et al. (2000) and Giavazzi et al. (2000) employ dynamic panel estimation to address the inertia in saving rates that can arise from lagged effects of the explanatory variables on saving. Thus, in addition to our baseline specification (1), we consider another specification including the lagged dependent variable, 𝑃𝑟𝑖𝑣𝑎𝑡𝑒_𝑠𝑎𝑣𝑖𝑡−1 as a regressor and employ system generalized-method-of-moments (GMM) estimators applied to dynamic panel models (Arellano and Bover 1995; Blundell and Bond 1998). 11 These estimators allow us to control not only for unobserved country-specific effects and but also potential endogeneity of the explanatory variables. As the reliability of the GMM estimator depends on whether the explanatory variables’ lagged values are valid instruments, we conduct weak instrument test (Bazzi and Clemens, 2013) and over-identification restriction test―where failure to reject the null hypothesis indicates the instruments’ validity. Lastly, it is necessary to check whether the error term, ei,t, is serially correlated; if it is not, then the first-order differenced error terms (ei,t and ei,t-1) are expected to have serial correlation (AR(1) test). As a result, it is expected that the second order differenced error terms (ei,t and ei,t-2) will have no serial autocorrelation (AR(2) test). As such, test results for first and second order autocorrelation in the differenced error terms are reported. 10

The first difference of equation (1) will eliminate the country-specific effects, 𝜑𝑖 , but generate an additional correlation between the first difference in dependent variables and that in error terms, which causes a “Nickell” bias in the estimation 11 Alonso-Borrego and Arellano (1996) and Blundell and Bond (1998) point out that the difference-GMM estimator proposed by Arellano and Bond (1991) cannot account for cross-country variations and that the regressors’ lagged levels might be weak instruments for the first-differences if the regressors are stable (close to a random walk process) over time. Thus, the difference-GMM performs poorly because the past levels convey little information about future changes. To overcome this obstacle, Arellano and Bover (1995) propose the system-GMM estimator, which combines the first-differenced regression with the level regression in (1). Using equation (1), level variables are augmented with suitable lags of their own first differences.

11

3. Empirical results 3.1. Main results I Table 1 shows our baseline results for the private–public savings offset. We include our main variables of interest in column (1), and add further controls, step by step, from column (2) to column (3). In all specifications, country and year fixed effects are included to control for unobserved country- and time-specific components. In each specification, the coefficient of public saving to GDP is negative, ranging between -0.33 and -0.46. Note that the results in columns (2) and (3) show statistical significance at the 5 percent level. These findings indicate that full RE does not hold, but a decrease in public savings by 1 percentage point (either a 1 percentage point increase in government spending or a decrease in tax) leads to an increase in private savings of about 0.4 percentage points. The coefficient of the interaction term between the public saving rate and the sovereign credit rating shows a significant and negative sign. More interestingly, the interaction term with the squared rating variable is significant and positive. This implies that the savings offset coefficient is a U-shaped function in the rating variable. In columns (4) and (5), we include lagged dependent variable as a regressor and estimate the augmented specification using the two-step system-GMM. In column (5), we introduce an alternative saving rate using GNI instead of GDP. We consider the lagged private saving rate, public saving rate and its interaction terms with rating variables as endogenous variables to control for simultaneity between private and public saving rate, but include other variables as predetermined or exogenous variables. The estimated coefficient on the lagged private saving rate is significant and positive in both columns (4) and (5). More importantly, the coefficients on the interaction terms of public saving rate and the sovereign credit rating and the squared rating are 12

consistent with those in column (3). Note that coefficient on public saving rate turns out to be statistically insignificant but the negative and significant coefficient on the interaction term of public saving rate and the credit rating becomes about two and a half times greater. Thus, these three estimated coefficients in both columns (4) and (5) still support U-shaped marginal effect of public saving rate on private saving with respect to the rating variable. For consistent estimation in the dynamic panel, the error 𝑒𝑖𝑡 is required to be serially uncorrelated. AR(1) and AR(2) tests that we report support the validity of dynamic specification. The Cragg–Donald statistic for testing the null hypothesis—such that the instruments are weak— is above 10, which is a rule of thumb critical value (Staiger and Stock, 1997).12 The Hansen’s over-identifying restriction cannot be rejected and support the validity of instruments. However, in our system GMM estimation, the number of instruments is greater than the number of countries (due to the relatively longer T compared to N (N=80, T=22)), and the reported p-value of the Hansen test is turns out to be close to 1 even though Roodman’s (2009a) “collapse” method is applied to control for an excess of instruments. Bowsher (2002) argues, though, that instrument proliferation vitiates the Hansen test of over-identification, and the test may implausibly return a perfect p-value of 1. However, we find that the results are often sensitive in terms of statistical significance to the choice of lags of instruments so we report the most conservative results without restricting the lags of our instruments arbitrarily. 13 The system GMM results do not alter qualitatively and even we find the better results in terms of statistical significance when restricting the lags of the instruments.

12

Stock and Yogo (2005) provide the weak IV critical values when there are multiple endogenous regressors. However, our results fail to compute Stock-Yogo weak IV test critical values. 13 In practice, Roodman (2009a) suggests restricting the number of instruments not only by using the “collapse” command but also by limiting the maximum lag in order to improve the Hansen test of the instruments’ joint validity. Certainly, when restricting the lags of instruments, the p-value of over-identifying restriction becomes smaller than 1 but still cannot be rejected at the 1 percent level.

13

[Insert Table 1] [Insert Figure 1] Figure 1 plots the marginal effects of public saving rates on the private saving rates for the range of sovereign credit ratings observed in our data set, based on results in columns (3), (4) and (5). The marginal effects of public saving rates on private saving rates clearly show a U-shaped pattern in sovereign credit ratings in all panels. For instance, in a top panel, the marginal effect characterizing the private–public savings offset reaches a minimum value of around -0.8 at the medium sovereign credit rating value (i.e., 10, which is equal to a rating of BB), which is near the value of -1 predicted by RE. Thus, for a medium sovereign credit rating value, which indicates strained fiscal resources, but still low levels of sovereign default risk, RE approximately holds. In other words, consumers expect that higher public spending (and, thus, higher fiscal deficits) in the present period will be associated with higher future taxes, which prompts them to increase their current private savings. However, the U-shaped pattern of the savings offset coefficient reveals a breakdown of RE for the extreme value of sovereign credit ratings. The Keynesian state of the economy is located at the right end of the U-shaped function, where good ratings signal high levels of sovereign solvency. The middle range of the U-shaped function represents the Ricardian state of the economy (i.e., the private–public savings offset is observed). More importantly, we reveal the sovereign default risk state of the economy located at the left range of the U-function, where the extent of the private-public savings offset is reduced again (statistically different from -1). That is, for poor sovereign credit ratings (i.e., for all S&P ratings below the investment grade rating BBB-) the savings offset is substantially below the value of -1 predicted by RE. This result suggests that consumers anticipate a net wealth gain of a potential sovereign default (with a present value of 14

saved taxes being lager than the wealth loss associated with default) and thus increase their private savings by less than the expansion of fiscal deficits.

3.2. Main results II: Alternative identification In empirical fiscal policy works, different identifications of fiscal policy (shocks) often lead to different conclusions, thus we first test the robustness of our results using an alternative identification strategy, which focuses on discretionary government spending. Here, we use a twostep approach, as proposed by Perotti (1999) and Corsetti et al. (2012). 14 By isolating past information, we can extract discretionary public spending shocks that are uncorrelated with contemporaneous economic variables (such as our main variable, the private savings rate, or current GDP). In the first step, we estimate a country-specific fiscal policy function in a time series framework to allow for country-specific heterogeneity in the slope parameters. In this specification, we assume that public spending is based on past information on public spending and GDP. Thus, the innovations of the fiscal policy function represent public spending shocks, which are orthogonal to contemporaneous economic fundamentals. We estimate the countryspecific public spending function as follows:

log(𝐺𝑖𝑡 ) = 𝛼 + 𝛽𝑖1 log(𝐺𝑖𝑡−1 ) + 𝛽𝑖2 log(𝐺𝑖𝑡−2 ) + 𝛾𝑖1 log(𝑌𝑖𝑡−1 ) +𝛾𝑖2 log(𝑌𝑖𝑡−2 ) +𝛿𝑖1 (𝐷𝑒𝑏𝑡/𝑌)𝑖𝑡−1 + 𝛿𝑖2 𝑃𝑒𝑔𝑖𝑡−1 + 𝛿𝑖3 𝐷𝑒𝑏𝑡_𝐶𝑟𝑖𝑠𝑖𝑠𝑖𝑡−1 + 𝛿𝑖4 𝑂𝑡ℎ𝑒𝑟_𝐶𝑟𝑖𝑠𝑖𝑠𝑖𝑡−1 + 𝛿𝑖5 𝑡𝑟𝑒𝑛𝑑 + 𝜀𝑖𝑡 (2)

14

Although this identification strategy is inspired by previous VAR studies, we follow Perotti (1999) and Corsetti et al. (2012), and do not use a VAR estimation, because this linear model would not be flexible enough to capture the conditional effects of government spending shocks on the national savings rate we are interested in. Other studies, such as Ilzetzki et al. (2013), investigate the conditional effects of government spending shocks using sample splits. However, we are interested in the conditional effects of government spending on savings, conditional on the level of sovereign default risk, which can best be assessed using a panel interaction model.

15

where the public spending of country i at time t, Git, is assumed to depend on the lagged levels of public spending (G), GDP (Y), the public debt-to-GDP ratio (Debt/Y), a peg dummy, a debt crisis dummy, and a dummy indicating the occurrence of another type of financial crisis (banking and currency crises). 15 The residual, 𝜀𝑖𝑡 , captures discretionary public spending shocks. This identification strategy rests on the assumption that public spending – due to budgetary planning and approval – is based on past economic fundamentals, and that the residual is orthogonal to current economic fundamentals (such as the savings rate). A positive residual 𝜀𝑖𝑡 measures positive discretionary spending shocks, where public spending is higher than that implied by the fiscal policy function, while negative values of 𝜀𝑖𝑡 measure negative discretionary policy shocks. Our identification assumption that the spending shock is orthogonal to current economic fundamentals can be justified by several empirical observations. First, automatic stabilizers (e.g., in the tax system) are rare in the context of public spending, ruling out the contemporaneous effect of economic fundamentals. Second, in practical policymaking, there is a considerable time lag between the planning and the implementation of the government budget, owing to a lag in real-time information on economic fundamentals and a potentially lengthy budget planning and approval process.16 In the second step of our identification design, we use the calculated discretionary public 15

The peg dummy is drawn from Klein and Shambaugh (2006). This variable is coded as 1 when a currency either stays within 2% bands against the base currency or has zero volatility in all months, except for a one-off devaluation. The debt crisis and other banking and currency crisis dummies are collected from Laeven and Valencia (2012). 16 Identifying exogenous fiscal policy shocks is difficult, because fiscal variables can be affected endogenously by the business cycle. Government expenditure and income tax revenue can be seen as automatic stabilizers, because they decrease during economic downturns. Therefore, the endogenous responses of fiscal variables should be disentangled from the exogenous fiscal shocks. To identify the exogenous process of fiscal shocks, Ramey and Shapiro (1998) focus on defense multipliers, which are approximately orthogonal to output. On the other hand, Blanchard and Perotti (2002) calculate the output elasticity of fiscal variables using institutional information and imposing short-run restrictions on the VAR system. Lastly, Mountford and Uhlig (2009) devise a sign restriction scheme to identify fiscal shocks.

16

spending shock as an explanatory variable in a panel regression for national savings rates (instead of the private saving rates), as is in Giavazzi et al. (2000).17 The discretionary spending shock ( 𝜀𝑖𝑡 ) is measured in a log term in a log-linear government spending equation. Thus, the discretionary spending shock (𝜀𝑖𝑡 ) can be interpreted as a percentage change in discretionary government spending. To make it comparable to the national saving rate (in terms of GDP), we standardize it by subtracting it from GDP growth. Furthermore, we include a percentage change in the national saving rate as an alternative dependent variable. [Insert Table 2] [Insert Figure 2] The results reported in Table 2 and Figure 2 support our baseline regressions. In line with our hypotheses, we find an inverted U-shaped pattern of the impact of discretionary spending shocks on the national savings rate, with a minimum of zero. In the medium range of sovereign credit ratings (i.e., in a Ricardian state of the economy), the effect of public spending on national savings is statistically insignificant (not different from zero). Higher discretionary public spending, which reduces current public savings, is associated with a one-to-one increase in private savings because consumers expect higher future taxes in order to cover the current budget deficit. This leaves the overall national savings rate unchanged. However, in the Keynesian state (the right tail) and the sovereign default risk state (the left tail), higher discretionary public 17

Contrary to our regression models presented above, as well as previous works on RE, which regress the private saving rate on the public saving rate (such as Loayza et al. (2000)), this section tests for robustness by tackling the potential endogeneity issue for public savings using our identification procedure and the discretionary public spending shock. Using the private saving rate would not be appropriate in our context because a positive spending shock (𝜀𝑖𝑡 > 0) could be associated with a positive revenue shock (which, because of endogeneity concerns and a lack of data, we do not include in the model). Thus, private savings (=Y – T – C) could respond to such a Havelmotype budget-neutral increase in public expenditure owing to a higher tax burden, T, or a drop in consumption expenditure, C. Thus, in order to isolate the effect of unobserved public revenue variations, we use the total national saving rate, which, by definition, is unaffected by public revenues (because public revenues enter positively into the public savings function and negatively in the private savings function).

17

spending and the associated drop in public savings are not entirely compensated for by private savings. In other words, consumers increase their savings by less than the public spending, leading to an overall reduction in total national savings.

3.3. Foreign vs. domestic ownership of sovereign debt To strengthen the robustness of the U-shaped pattern of the savings offset, we perform three sensitivity analyses. First, we check the robustness of our results using alternative indicators for fiscal sustainability instead of the foreign currency sovereign credit ratings used in the baseline regressions. We use Standard and Poor’s sovereign credit ratings on domestic currency. As domestic investors typically prefer to hold domestic currency sovereign bonds, they should closely monitor domestic currency ratings. Foreign investors, on the contrary, may prefer to invest in foreign currency bonds. Thus, time variation in domestic currency ratings represents the wealth losses of domestic investors following sovereign default, while foreign currency ratings more precisely represent wealth losses for foreign investors. Accounting for the foreign ownership issue, time variation of domestic currency ratings should therefore reproduce the RE result, but not the U-shaped pattern since default on domestic currency sovereign bonds would not imply a net wealth gain for domestic investors. Furthermore, we examine whether the U-shaped impact remains robust when we consider debt-to-income ratios used frequently in previous studies as a proxy of fiscal sustainability. Here, we introduce the public debt-to-GDP ratio.18 In addition, we collect domestic public debt data and external public debt data separately from the World Bank’s World Development Indicators (WDI), to check whether they have distinct effects on the savings offset. Table 3 reports the 18

Data on total public debt-to-GDP data is collected from the IMF public debt database, constructed by Abbas et al. (2010). However, this data set does not distinguish between domestic public debt and external public debt. However, this data set does not distinguish between domestic public debt and external public debt.

18

results with the alternative fiscal measurements, and Figure 3 plots the public–private savings offset coefficients. As expected, the results for the domestic sovereign solvency indicators such as local currency sovereign credit rating and domestic public debt exhibit no or a very weak U-shaped savings offset. The results for the domestic currency sovereign credit ratings show that the estimation results in columns (1) and (2) of Table 3 are statistically insignificant for the interaction terms. Compared with the results of the foreign currency sovereign credit ratings used in the baseline regressions, noisier inferences are obtained on the left tail of the U-shaped function (which includes -1 in the confidence intervals). Columns (3)–(8) of Table 3 and the remaining graphs in Figure 3 show the results with different variants of debt-to-GDP measures. The results between domestic public debt and external public debt show a clearer distinction as we observed in the results between foreign currency sovereign ratings and domestic currency ratings. The results with external debt-to-GDP ratios in columns (7) and (8) of Table 3 and the right and bottom panel in Figure 3 show more significant U-shape than those with domestic debt measures do. In particular, the right tails of the estimated U-shaped offset coefficients with public debt measures, shown in columns (4) and (6), are statistically insignificant and noisy. Overall, these results support the ownership hypothesis in constituting the U-shaped pattern. For the measures that are not sensitive to measuring foreign investors’ exposure to sovereign default, i.e., the domestic currency sovereign credit ratings and domestic sovereign debt to GDP, no or very weak evidence for a U-shaped pattern of the private-public savings offset is obtained. For the fiscal sustainability measures representing foreign investors’ exposure to default risk, i.e., external public debt to GDP, we find robust evidence of a U-shaped pattern of 19

the savings offset in sovereign solvency. That is, the U-shaped pattern seems to be constituted by (partial) foreign ownership of sovereign debt. [Insert Table 3] [Insert Figure 3]

4. Alternative channels of the U-shaped pattern of the savings offset In this section, we suggest a number of alternative channels that may constitute the U-shaped pattern of the savings offset. First, we distinguish between developed and emerging and developing countries. Second, we consider the role of the country’s history of sovereign default. Third, we focus on financial openness and the size of the domestic banking sector.19

4.1 Developed vs. emerging and developing countries The major contribution in this study is that RE breaks down when a country’s foreign currency sovereign credit rating takes on extremely poor values, which confirms the existence of the left tail of the U-shaped pattern of the savings offset. In this sub-section, we implement a sub-sample analysis for developed and emerging and developing countries to examine whether such a Ushaped pattern prevails in both country sets.20 Table 4 reports the results of the sub-sample regression. In columns (1) of Table 4, for developed countries, the estimated coefficients of the public saving rate and its interaction terms are statistically insignificant. Thus, we exclude the interaction term of public saving rate and a

19

Note that the system GMM estimation is preferred in a short panel structure. However, the following sub-sample analysis leads to the proliferation of instruments by reducing the size of N more but remaining that of T, and casts doubt on the validity of instruments. Thus, we use restrict the number of instruments by restricting their lags randomly. Furthermore, we report system GMM results together with the baseline panel regression results. 20 Developed and emerging/developing countries are defined according to the IMF classification (see Appendix, Table A1).

20

quadratic rating term in columns (2) and (3). The results in columns (2) and (3) show that for developed country sample, the private-public saving offsets are greater as sovereign rating becomes lower, which echoes previous theoretical studies such as Sutherland (1997) and Perotti (1999) findings and are also consistent with Nickel and Tudyka (2014). However, the results for the emerging markets and the developing country sample in columns (4) and (5) show that the public saving rate and its interaction terms exhibit significant responses, which confirms the finding of a nonlinear private–public savings offset. Figure 4 plots the marginal coefficient of the savings offset as a function of sovereign credit ratings. The result suggests that a significant Ushaped pattern is detected only for emerging and developing economies, but not for developed countries. This finding is sensible, because sovereign defaults are, by and large, an emerging and developing countries phenomenon. Thus, consumers in developed countries do not make saving decisions based on sovereign credit ratings. In contrast to developed countries, consumers in emerging and developing economies are well aware of the fact that governments may default on sovereign debt, owing to a history of sovereign defaults in these countries (typically produced by macroeconomic and political instability, ineffective tax regimes, and investors’ reluctance to roll over sovereign bonds in periods of fiscal stress). Moreover, the foreign ownership issue in sovereign debt largely applies to emerging markets, whereas for developed countries a much larger share of domestic sovereign debt is held by domestic investors. Thus, our result of a breakdown of RE for poor sovereign credit ratings seems to be driven by emerging and developing economies. [Insert Table 4] [Insert Figure 4] 4.2 History of debt crises 21

In this section, we examine the effect of sovereign debt crisis “memory” on the savings offset. Here, we re-estimate the regressions for a sample split of countries that have (not) experienced a sovereign debt crisis in the past. In order to examine how different lengths of sovereign debt crisis memories influence the savings offset, we test the savings offset for the following sets of countries: those that have never experienced a debt crisis; those that have experienced a debt crisis since 1970;21 those that have experienced a debt crisis since 1989, the beginning of our sample; and those that have experienced a debt crisis in the last 5, 10, or 15 years. Data on sovereign debt crises are taken from Laeven and Valencia (2012). Please see the Appendix, Table A5, for the dates of historical sovereign debt crises in our sample. Table 5 and Figure 5 report the results. In particular, the marginal effects depicted in Figure 5 reveal that the right tail of the U-shaped pattern of the savings offset shift upward (further away from RE) when a country’s historical sovereign debt crises are further in the past. The savings offset for the range of poor sovereign credit ratings stays relatively unchanged. However, the offset for medium and good ratings indicate that as the number of years since the last debt default increases, RE starts to break down (i.e., the offset coefficient becomes greater than -1) at the same ratings level. This result suggests that the memory of consumers plays an important role in shaping the savings offset. When consumers remember a sovereign default, they may not see large amounts of free fiscal resources, even though rating agencies release optimistic ratings, moving the savings offset further away from RE. However, in our small sub-sample analysis in Table 5, the specification tests of system GMM such as weak instruments and overidentifying restrictions fail to provide satisfactory statistics. Thus, for comparison purposes, Appendix Table 6 and Appendix Figure 1 report the results with panel fixed effects regression.

21

More precisely, this definition records sovereign defaults since the start of the Systemic Crises Database of Laeven and Valencia (2012) in 1970.

22

[Insert Table 5] [Insert Figure 5] 4.3 Financial openness and the size of banking sector In this section, we test the relevance of financial openness and size of banking sector (financial development) to the U-shaped pattern of the savings offset. In our sample, most emerging and developing countries that rely on external public debt restrict capital accounts and show low de jure financial openness, while developed countries attain the highest level of financial openness. For example, given growing expectations of sovereign default, a country with no capital controls would lose foreign capital including public debt liabilities abruptly. Finally, sovereign default burdens are levied on mostly domestic agents, relaxing the U-shaped pattern. However, a country with more restrictive capital accounts is likely to prevent such capital flight and to dilute foreign sovereign bondholders. This has the important implication of a pronounced U-shaped pattern of the savings offset, particularly in the range of poor sovereign credit ratings. We use a sample split approach, dividing our full sample into financially open and financially closed countries. First, the mean of the financial openness measures for 80 individual countries is computed. Then if the mean of a country’s financial openness is above the median of the 80 countries’, it is considered a high financially open country. For de jure measure, the threshold that splits sub-sample is 0.69. We use the de jure capital account openness, taken from Chinn and Ito (2006). Table 6 and the upper panel of Figure 6 show the results for financial openness. Interestingly, countries with low financial openness exhibit a clear U-shaped savings offset, while those with high financial openness show a statistically insignificant U-shape. This result suggests that capital controls, often imposed in emerging market economies, lead to a lock in of foreign 23

investors in times of sovereign distress leading to a more pronounced U-shaped pattern in countries with low capital account openness.22 Similarly to financial openness, we consider the effect of the size/development of the domestic banking sector. If a country has a large banking sector (i.e., a high ratio of banking assets to GDP), the government may be better able to sell its sovereign bonds in times of sovereign distress, thus, reducing the U-shaped pattern. This reasoning is in line with that of Caballero and Krishnamurthy (2004), who posit a more pronounced crowding out of fiscal expansion on private investment for countries with low degrees of financial depth. The data on banking assets-to-GDP ratios are drawn from the World Bank’s Financial Structure Database. First, we use the deposit bank assets-to-GDP ratio as the size of the banking sector. Again, we divide our full sample into two sub-samples in terms of the degree of financial market size. Table 7 and the lower panel of Figure 6 show the results. Similarly to the results for financial openness, we find that countries with large banking sectors show a much less pronounced U-shaped savings offset pattern than in the case of countries with a small banking sector. This finding suggests that governments of countries with large financial sectors can better refinance themselves when relying on a large domestic banking sector. For governments in small financial sector regimes, the risk of sovereign default in stressed situations is higher, leading to a more pronounced U-shaped savings offset pattern. [Insert Table 6] [Insert Table 7] [Insert Figure 6] 22

Our finding is consistent with previous study by Cho and Pyun (2016). They argue that greater financial integration prevents the domestic interest rate from rising to a certain extent (given the world interest rate), and thus reduces the degree of private-public savings offset. They link the patterns of international financial integration with the savings offset coefficients for advanced and emerging market countries and particularly show that the savings offset of advanced countries has declined as their financial market openness has increased.

24

5. Conclusion This study considers the role of sovereign default risk in the breakdown of Ricardian equivalence (RE). RE posits that rational consumers will increase private savings when fiscal deficit is increased, leading to a private-public savings offset. Previous studies have shown that while RE generally breaks down, RE starts to hold (fiscal policy becomes ineffective) when a country is running huge fiscal deficit or accumulating large public debts. A novel feature of this study is that we reveal a new reason why RE can break down although fiscal space is very tight. When a country is near fiscal insolvency, as indicated by poor sovereign credit ratings, especially in foreign currency, consumers may anticipate that the government will default on sovereign debt rather than increasing future taxes. If some fraction of domestic sovereign debt is owned by foreign creditors, the wealth losses associated with sovereign default would not entirely be made by domestic consumers and thereby their net wealth effect after considering their savings on future taxes is positive. Thus, the presence of foreign creditors of sovereign debt, constitutes a deviation from RE in the sovereign default risk state. In this state of high sovereign default risk, higher fiscal deficits need not be accompanied by an increase in private savings (contrary to the prediction of RE). Using a panel of 80 countries for the period 1989–2010, we find robust evidence of a Ushaped pattern in the savings offset, with respect to foreign currency sovereign credit ratings. RE breaks down in states of very high solvency, but also in states of high sovereign default risk. Moreover, we examine various channels that constitute the U-shaped pattern of the savings offset. We find that the U-shaped pattern prevails in the following cases: i) the U-shape pattern is restored with respect to external pubic debt (not domestic public debt), ii) emerging and 25

developing economies (but not in developed economies); iii) countries that have experienced sovereign debt crises in the past; and iv) countries that have small and segmented financial sectors. Overall, consumers appear to monitor sovereign default risk when adjusting their saving behavior after fiscal shocks.

26

References Abbas, S.M., Belhocine, N., El-Ganainy, A., Horton, M., 2010. A historical public debt database. IMF Working Paper 10/245, 1–26. Acharya, V.V., Drechsler, I., Schnabl, P., 2014. A pyrrhic victory? Bank bailouts and sovereign credit risk. Journal of Finance 69, 2689–2739. Afonso, A., Furceri, D., Gomes, P., 2012. Sovereign credit ratings and financial markets linkages: Application to European data. Journal of International Money and Finance 31, 606–638. Agca, S., Celasun, O., 2012. Sovereign debt and corporate borrowing costs in emerging markets. Journal of International Economics 88, 198–208. Alonso-Borrego, C. and Arellano, M. 1996. Symmetrically Normalised Instrumental-Variable Estimation Using Panel Data. Journal of Business and Economic Statistics 17, 36–49. Arellano, M. and Bond, S. 1991. Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. Review of Economic Studies 58, 277– 297. Arellano, M. and Bover, A. 1995. Another Look at the Instrumental-Variable Estimation of Error-Components Models. Journal of Econometrics 68, 29–51. Arteta, C., Hale, G., 2008. Sovereign debt crises and credit to the private sector. Journal of International Economics 74, 53–69. Barro, R. J. 1989. The Ricardian Approach to Budget Deficits. Journal of Economic Perspectives 3(2), 37–54. Bazzi, Samuel, and Michael A. Clemens. 2013, Blunt instruments: avoiding common pitfalls in identifying the causes of economic growth. American Economic Journal: Macroeconomics 5(2), 152–186. Beetsma, Roel. 2008. A Survey of the Effects of Discretionary Fiscal Policy. Studier i Finanspolitik (2), 1–37. Bernheim, B.D., 1987. Ricardian equivalence: An evaluation of theory and evidence. NBER Macroeconomics Annual, 263–304. Bertola, G., Drazen, A., 1993. Trigger points and budget cuts: Explaining the effects of fiscal austerity. American Economic Review 83, 11–26. Bi, H., 2012. Sovereign default risk premia, fiscal limits, and fiscal policy. European Economic Review 56, 389–410. 27

Blanchard, O., Perotti, R., 2002. An empirical characterization of the dynamic effects of changes in government spending. Quarterly Journal of Economics 1329–1368. Blundell, R. and Bond, S. 1998. Initial Conditions and Moment Restrictions in Dynamic Panel Data Models. Journal of Econometrics 87, 115–143. Borensztein, E., Panizza, U., 2009. The costs of sovereign default. IMF Staff Papers 56, 683–741. Bowsher, C. G. 2002. On Testing Overidentifying Restrictions in Dynamic Panel Data Models. Economic Letters 77, 211–220. Brooks, R., Faff, R.W., Hillier, D., Hillier, J., 2004. The national market impact of sovereign rating changes. Journal of Banking & Finance 28, 233–250. Caballero, R.J., Krishnamurthy, A., 2004. Fiscal policy and financial depth. National Bureau of Economic Research Working Paper No. 10532. Chinn, M.D., Ito, H., 2006. What matters for financial development? Capital controls, institutions, and interactions. Journal of Development Economics 81, 163–192. Cho, D.Y., J.H. Pyun, 2016. Measuring the Time-varying Effect of Fiscal Policy on Private Saving with Financial Integration, SKKU working paper. Corbo, V., Schmidt-Hebbel, K., 1991. Public policies and saving in developing countries. Journal of Development Economics 36, 89–115. Correa, R., Lee, K.H., Sapriza, H., Suarez, G.A., 2014. Sovereign credit risk, banks’ government support, and bank stock returns around the world. Journal of Money, Credit and Banking 46, 93–121. Corsetti, G., Meier, A., Müller, G., 2012. What determines government spending multipliers? Economic Policy 27, 521–565. Corsetti, G., Kuester, K., Meier, A., Müller, G.J., 2013. Sovereign risk, fiscal policy, and macroeconomic stability. The Economic Journal 123, 99–132. Demirgüç-Kunt, A., Huizinga, H., 2013. Are banks too big to fail or too big to save? International evidence from equity prices and CDS spreads. Journal of Banking & Finance 37, 875–894. Furceri, D., Zdzienicka, A., 2012. How costly are debt crises? Journal of International Money and Finance 31, 726–742. Gande, A., Parsley, D.C., 2005. News spillovers in the sovereign debt market. Journal of Financial Economics 75, 691–734. 28

Gennaioli, N., Martin, A., Rossi, S., 2014. Sovereign default, domestic banks, and financial institutions. Journal of Finance 69, 819–866. Ghosh, A.R., Kim, J.I., Mendoza, E.G., Ostry, J.D., Qureshi, M.S., 2013. Fiscal fatigue, fiscal space and debt sustainability in advanced economies. The Economic Journal 123, F4–F30. Giavazzi, F., Jappelli, T., Pagano, M., 2000. Searching for non-linear effects of fiscal policy: Evidence from industrial and developing countries. European Economic Review 44, 1259– 1289. Ilzetzki, E., Mendoza, E.G., Végh, C.A., 2013. How big (small?) are fiscal multipliers? Journal of Monetary Economics 60, 239–254. Jappelli, T., Pagano, M., 1994. Saving, growth, and liquidity constraints. Quarterly Journal of Economics 109, 83–109. Kim, S., 2015. Country characteristics and the effects of government consumption shocks on the current account and real exchange rate. Journal of International Economics 97, 436–447. Klein, M.W., Shambaugh, J.C., 2006. Fixed exchange rates and trade. Journal of International Economics 70, 359–383. Laeven, L., Valencia, F., 2012. Systemic banking crises database: An update. IMF Working Paper 12/163. Levy-Yeyati, E., Panizza, U., 2011. The elusive costs of sovereign defaults. Journal of Development Economics 94, 95–105. Loayza, N., Schmidt-Hebbel, K., Servén, L., 2000. What drives private saving across the world? Review of Economics and Statistics 82, 165–181. Masson, P.R., Bayoumi, T., Samiei, H., 1998. International evidence on the determinants of private saving. The World Bank Economic Review 12, 483–501. Mendoza, E.G., Oviedo, P.M., 2009. Public debt, fiscal solvency and macroeconomic uncertainty in Latin America. Economía Mexicana NUEVA É POCA 18(2). Modigliani, F., Brumberg, R., 1955. Utility analysis and the consumption function: An Interpretation of cross-section data, in: Kurihara, K.K. (ed.), Post-Keynesian Economics, Rutgers University Press, New Brunswick, NJ. Mountford, A., Uhlig, H., 2009. What are the effects of fiscal policy shocks? Journal of Applied Econometrics 24, 960–992. Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica, 1417-1426. 29

Nickel, C. and Tudyka, A., 2014. Fiscal Stimulus in Times of High Debt: Reconsidering Multipliers and Twin Deficits. Journal of Money, Credit and Banking, 46(7), 1313-1344. Ostry, J.D, Ghosh, A.R., Kim, J.I., Qureshi, M.S., 2010. Fiscal space. International Monetary Fund, Research Department. Perotti, R., 1999. Fiscal policy when things are going badly. Quarterly Journal of Economics 114, 1399–1436. Pyun, J.H., Rhee, D.E., 2015. Fiscal multipliers during the global financial crisis: Fiscal and monetary interaction matters. Contemporary Economic Policy 33, 207–220. Ramey, V.A., Shapiro, M.D., 1998. Costly capital reallocation and the effects of government spending. Carnegie Rochester Conference on Public Policy 48, 154–94. Reinhart, C.M., Rogoff, K.S., 2011. From financial crash to debt crisis. American Economic Review 101, 1676–1706. Reinhart, C.M., Rogoff, K.S., 2013. Banking crises: An equal opportunity menace. Journal of Banking & Finance 37, 4557–4573. Ricciuti, R., 2003. Assessing Ricardian equivalence. Journal of Economic Surveys 17, 55–78. Roodman, D. 2009a. How to do xtabond2: An introduction to difference and system GMM in Stata. Stata Journal, 9(1), 86–136. Roodman, D. 2009b. A Note on the Theme of Too Many Instruments. Oxford Bulletin of Economic and Statistics, 71, 135–158. Röhn, O., 2010. New evidence on the private saving offset and Ricardian equivalence. OECD Economics Department Working Papers 762. Rose, A.K., 2005. One reason countries pay their debts: Renegotiation and international trade. Journal of Development Economics 77, 189–206. Seater, J.J., 1993. Ricardian equivalence. Journal of Economic Literature 31, 142–190. Sutherland, A., 1997. Fiscal crises and aggregate demand: Can high public debt reverse the effects of fiscal policy? Journal of Public Economics 65, 147–162. Staiger, D. and Stock, J.H., 1997. Instrumental variables regression with weak instruments. Econometrica 65(3), 557. Stock, J.H. and Yogo, M., 2005. Testing for Weak Instruments in Linear IV Regression. Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg.

30

Table 1. Baseline results Dependent variable

Lagged private saving rate Public saving rate Public saving rate × Rating Public saving rate × Rating^2 Rating log of GDP per capita GDP per capita growth

Private savings/GDP Sparse controls

add income variables

Full specification

(1)

(2)

(3)

--

--

--

-0.3328 (0.226) -0.0881*** (0.033) 0.0042*** (0.001) -0.0015 (0.001) 0.0402*** (0.015) 0.0841 (0.054)

-0.4602** (0.218) -0.0629** (0.032) 0.0032*** (0.001) -0.0013 (0.001) 0.0236 (0.016) 0.0651 (0.052) -0.1421*** (0.053) -0.5573*** (0.084) -0.0173*** (0.005)

Yes Yes

Yes Yes

-0.4548** (0.221) -0.0633* (0.032) 0.0031*** (0.001) -0.0013 (0.001) 0.0273* (0.016) 0.0517 (0.051) -0.1310** (0.053) -0.5550*** (0.084) -0.0114 (0.007) -0.0006 (0.001) 0.0616*** (0.020) -0.0026 (0.004) Yes Yes

Dependency ratio, young (of working-age population) Dependency ratio, old (of working-age population) Life expectancy at birth (years) M2/ GDP Change in terms of trade Crisis dummy Country fixed effects (FEs) Year fixed effects (FEs) AR(1) test AR(2) test Weak instruments (F-stat.) Hansen over id (p-val.) # of instruments/ # of country group Observations R-squared

1,308 0.799

1,308 0.809

1,307 0.812

Private savings/GNI Full Full specification specification System System GMM GMM (4) (5) 0.3661*** 0.3576*** (0.096) (0.089) -0.0981 0.1495 (0.491) (0.554) -0.1312* -0.1682** (0.077) (0.085) 0.0058** 0.0071** (0.003) (0.003) 0.0042** 0.0045*** (0.002) (0.001) -0.0112 -0.0091 (0.009) (0.007) 0.1490** 0.1405** (0.069) (0.058) -0.1982*** -0.1945*** (0.057) (0.058) -0.4906*** -0.5532*** (0.120) (0.125) 0.0010** 0.0011* (0.000) (0.001) -0.0009 -0.0006 (0.001) (0.001) 0.0677** 0.0767*** (0.029) (0.024) 0.0109* 0.0156*** (0.006) (0.005) Yes Yes Yes Yes 0.002 0.002 0.883 0.683 19.8 13.32 0.99 1.000 117/80

114/74

1,279

1,212

Note: Two-step system GMM estimators are reported. Collapse command is used to avoid the proliferation of instruments (Roodman, 2009a, 2009b). However, we do not limit the number of lags of the instruments arbitrarily. In column (5), we use savings measure divided by gross national income (GNI), instead of GDP, as the dependent variable. Constant is included but not reported. Robust standard errors are in parentheses. *,**,and *** are respectively significance level at 10%, 5% and 1%.

31

Table 2. Robustness: Alternative Identification Dependent variable

Lagged national savings to GDP Discretionary Gov’t spending/GDP (percent change) Discretionary Gov’t spending/GDP × Rating Discretionary Gov’t spending/GDP × Rating^2 Rating log of GDP per capita GDP per capita growth Age dependency ratio, young (of working-age population) Age dependency ratio, old (of working-age population) Life expectancy at birth (years) M2/ GDP Change in terms of trade Crisis Country fixed effects (FEs) Year fixed effects (FEs) AR(1) test (p-val.) AR(2) test (p-val.) Weak instruments (F-stat.) Hansen over id (p-val.) # of instruments/ # of country group Observations R-squared

National savings to GDP

Percent change in NS/GDP year-by-year change

(1) --

(2)

-0.2461** (0.115) 0.0523*** (0.020) -0.0026*** (0.001) -0.0001 (0.001) 0.0429*** (0.016) 0.0669 (0.054) -0.0163 (0.054) -0.7522*** (0.084) -0.0193** (0.008) -0.0001 (0.001) 0.0657*** (0.021) -0.0059 (0.004) Yes Yes

-1.3492** (0.659) 0.2476** (0.110) -0.0106** (0.004) -0.0095** (0.004) 0.0125 (0.037) 0.4230** (0.198) 0.0390 (0.356) -0.2589 (0.438) 2.2262* (1.225) -0.0285 (0.026) 0.4112*** (0.105) 0.0228 (0.015) Yes Yes

National savings to GDP w/ lagged dependent variable (3) 0.6888*** (0.029) -0.2099* (0.113) 0.0439** (0.019) -0.0022*** (0.001) -0.0016** (0.001) 0.0208** (0.009) 0.0611 (0.040) 0.0300 (0.032) -0.2138*** (0.047) -0.0110 (0.007) 0.0005 (0.001) 0.0975*** (0.019) 0.0009 (0.003) Yes Yes

1,278 0.850

1,275 0.166

1,278 0.927

National savings to GDP System GMM

(4) 0.7577*** (0.061) -0.3629* (0.217) 0.0679** (0.035) -0.0030*** (0.001) 0.0012 (0.001) -0.0014 (0.003) 0.0649 (0.052) -0.0558** (0.025) -0.1818*** (0.066) 0.0003 (0.000) -0.0003 (0.000) 0.1217*** (0.029) 0.0080*** (0.003) Yes Yes 0.000 0.924 17.37 0.99 115/80 1,278

Note: Two-step system GMM estimators are reported. Collapse command is used to avoid the proliferation of instruments (Roodman, 2009a, 2009b). However, we do not limit the number of lags of the instruments arbitrarily. Constant is included but not reported. Robust standard errors are in parentheses. *,**,and *** are respectively significance level at 10%, 5% and 1%.

32

Table 3. Alternative fiscal sustainability measures Dependent variable Fiscal solvency indicator

Lagged Private savings to GDP Public savings to GDP Public savings to GDP × Fiscal solvency Public savings to GDP × Fiscal solvency^2 Fiscal solvency log of GDP per capita GDP per capita growth Age dependency ratio, young Age dependency ratio, old Life expectancy at birth, total (years) M2/ GDP Change in terms of trade Crisis

Sovereign ratings, domestic currency System GMM (1) (2)

-0.5053* (0.282) -0.0306 (0.035) 0.0016 (0.001) -0.0015 (0.001) 0.0413** (0.019) -0.0245 (0.051) -0.194*** (0.064) -0.375*** (0.097) -0.0004 (0.006) -0.0019** (0.001) 0.0367* (0.022) 0.0054 (0.004)

0.522*** (0.093) -0.6830 (0.642) -0.0397 (0.095) 0.0025 (0.003) 0.0025** (0.001) -0.0042 (0.007) 0.0878 (0.057) -0.1126** (0.054) -0.319*** (0.110) 0.0011 (0.001) -0.0009 (0.001) 0.0541** (0.027) 0.0094* (0.006)

Private savings to GDP Gross Public debt to Domestic public debt GDP (IMF) to GDP (WDI) System System GMM GMM (3) (4) (5) (6)

-0.479*** (0.086) -0.4980** (0.203) 0.2938** (0.131) 0.0120 (0.008) 0.0233 (0.016) 0.0360 (0.053) -0.0624 (0.052) -0.742*** (0.098) -0.0054 (0.007) -0.0013* (0.001) 0.057*** (0.020) -0.0057 (0.004)

0.444*** (0.106) -0.538** (0.245) -0.4938 (0.707) 0.3980 (0.520) -0.0050 (0.018) 0.0013 (0.008) 0.1532* (0.086) -0.1732** (0.068) -0.3793** (0.183) 0.0000 (0.001) 0.0002 (0.002) 0.072*** (0.027) 0.0071 (0.005)

-0.556*** (0.081) -0.5842** (0.245) 0.363*** (0.139) 0.0243** (0.010) 0.0391** (0.017) 0.1312** (0.060) 0.0361 (0.065) -0.825*** (0.117) -0.0325 (0.127) -0.0016 (0.001) 0.0248 (0.026) -0.0081** (0.004)

0.552*** (0.138) -0.3391 (0.271) -0.6186 (1.091) 0.5176 (0.694) -0.0212 (0.027) -0.0018 (0.014) 0.1632 (0.133) -0.1498* (0.080) -0.2760 (0.256) 0.1265 (0.158) 0.0014 (0.005) -0.0018 (0.042) 0.0009 (0.005)

External public debt/GDP (WDI) System GMM (7) (8)

-0.1084 (0.151) -4.283*** (0.875) 5.268*** (1.307) 0.0299 (0.034) -0.0020 (0.025) 0.183*** (0.068) 0.2325** (0.108) -2.099*** (0.361) 0.1570 (0.171) 0.0021 (0.021) 0.0406* (0.024) 0.0017 (0.007)

0.646*** (0.126) 0.9846 (0.813) -10.739** (4.888) 11.445** (5.005) 0.0861 (0.060) -0.0168 (0.012) 0.1053 (0.075) -0.1034 (0.075) -0.4093** (0.188) 0.2241 (0.163) 0.0276 (0.032) 0.0515 (0.061) 0.0143* (0.007)

Country fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes AR(1) test (p-val.) 0.000 0.001 0.002 0.000 AR(2) test (p-val.) 0.292 0.754 0.318 0.807 Weak instruments (F6.8 9.8 --※ --※ stat.) Hansen over id (p-val.) 0.947 0.990 0.139 0.998 # of instruments/ # of 106/80 117/80 50/ 43 54/40 country group Observations 1,120 1,120 1,287 1,260 809 809 549 546 R-squared 0.825 0.813 0.842 0.852 Note: Two-step system GMM estimators are reported. Collaps command is used and we limit the number of lags of the instruments at 5 in column (6) to avoid the proliferation of instruments. In columns (1) and (2), we include domestic currency sovereign credit rating as an alternative fiscal solvency measure. From column (3) to (4), we introduce gross public debt-to-income ratio as a proxy for fiscal solvency from IMF database. In columns (5) and (6), we include domestic public debt to GDP and in columns (7) and (8), external public debt-to GDP ratio are included. Constant is included but not reported. Robust standard errors are in parentheses. *,**,and *** are respectively significance level at 10%, 5% and 1%. ※ fails to compute weak instrument statistics.

33

Table 4. Savings offset in Developed vs. Emerging and Developing countries Dependent variable

Lagged private savings to GDP Public savings to GDP Public savings to GDP × Rating Public saving to GDP × Rating^2 Rating log of GDP per capita GDP per capita growth Age dependency ratio, young (of working-age population) Age dependency ratio, old (of working-age population) Life expectancy at birth, total (years) M2/ GDP Change in terms of trade Crisis dummy Country fixed effects Year fixed effects AR(1) test (p-val.) AR(2) test (p-val.) Weak instruments (F-stat.) Hansen over id (p-val.) # of instruments/ # of country group Observations

R-squared

Private savings to GDP Emerging Markets and Developed countries Developing countries System System Panel FEs Panel FEs Panel FEs GMM GMM (1) (2) (3) (4) (5) 0.5525*** 0.4518*** (0.084) (0.159) -1.2900 -1.9793*** -1.6194*** -0.3888* -0.4221 (1.027) (0.272) (0.533) (0.232) (0.422) -0.0197 0.0726*** 0.0524** -0.0807** -0.1414* (0.133) (0.014) (0.025) (0.039) (0.075) 0.0029 0.0046*** 0.0068** (0.004) (0.002) (0.003) -0.0007 -0.0010 0.0003 0.0006 0.0071** (0.002) (0.002) (0.003) (0.001) (0.003) 0.0525** 0.0546** 0.0287* 0.0193 -0.0124 (0.026) (0.026) (0.016) (0.021) (0.012) 0.0175 0.0135 0.1374 0.0381 0.1030 (0.095) (0.092) (0.090) (0.062) (0.073) 0.3930*** 0.4043*** -0.3237*** 0.0635 -0.1685* (0.102) (0.101) (0.113) (0.108) (0.087) -0.3506*** -0.3669*** -0.5474*** -1.5688*** -0.6358*** (0.096) (0.096) (0.172) (0.363) (0.214) 0.2466 0.2239 -0.1288 -0.0137 0.0009 (0.293) (0.294) (0.380) (0.009) (0.001) -0.0007 -0.0005 -0.0020 -0.0015 -0.0004 (0.001) (0.001) (0.001) (0.001) (0.002) 0.0503* 0.0512* 0.0682** 0.0504** 0.0663 (0.029) (0.029) (0.029) (0.023) (0.053) -0.0087* -0.0086* -0.0003 0.0048 0.0217** (0.005) (0.005) (0.006) (0.006) (0.010) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 0.003 0.001 0.95 0.357 7.9 8.8/ 7.7※ 1.00 0.277 42/29 613 0.838

613 0.837

588

38/51 694 0.813

691

Note: Two-step system GMM estimators are reported in column (5). But in column (3), we report one-step system GMM results (two-step system GMM estimators fail to support the validity of dynamic specification, which cannot reject AR(1) test. We use collapse command and limit the number of lags of instruments at 3 in column (3) and at 1 in column (5) to avoid the proliferation of the instruments. Constant is included but not reported. For the sample split, we follow the IMF classification for advanced and emerging/developing countries. See Appendix Table I for the country list. Robust standard errors are in parentheses. *,**,and *** are respectively significance level at 10%, 5% and 1%. ※ indicates Stock-Yogo’s weak IV critical values at the 10 % level.

34

Table 5. Savings offset and a country’s debt default history Dependent variable

History of debt crisis

Countries that have never had sovereign default before

(1) Lagged private savings/GDP

Public savings to GDP Public savings to GDP × Rating Public savings to GDP × Rating^2 Rating log of GDP per capita GDP per capita growth Dependency ratio, young Dependency ratio, old Life expectancy at birth, total (years) M2/ GDP Change in terms of trade Crisis (except debt crisis)

0.5236*** (0.052) -0.2564 (1.071) -0.1070 (0.143) 0.0050 (0.005) 0.0059*** (0.002) -0.0131 (0.009) 0.0992 (0.071) -0.2078*** (0.042) -0.5765*** (0.118) -0.0640 (0.129) 0.0000 (0.003) 0.0315 (0.026) 0.0116* (0.006)

Private savings/GDP Countries that Countries that Countries that had debt have have default since experienced experienced 1970 more than 5 more than 10 year debt year debt crisis crisis (2) (3) (5)

0.4762*** (0.142) -0.8113** (0.379) -0.0410 (0.068) 0.0049 (0.004) -0.0005 (0.001) -0.0057 (0.009) 0.1439** (0.065) -0.0636 (0.083) -0.3351* (0.198) 0.0006 (0.001) -0.0003 (0.001) 0.0995** (0.039) 0.0142** (0.006)

0.4383*** (0.135) -0.6309** (0.315) -0.0653 (0.056) 0.0031 (0.002) 0.0014 (0.002) -0.0033 (0.009) 0.0852 (0.067) 0.0657 (0.082) -0.1141 (0.208) 0.1374 (0.096) -0.0122 (0.022) 0.1112** (0.046) 0.0125* (0.007)

0.1962* (0.114) -0.5343* (0.319) -0.1019 (0.069) 0.0070 (0.006) -0.0010 (0.002) 0.0000 (0.021) 0.0404 (0.105) -0.1399 (0.130) -0.1467 (0.266) -0.4879 (0.383) -0.0558 (0.035) 0.0340 (0.028) 0.0178** (0.008)

Countries that have experienced more than 15 year debt crisis (6)

0.3700*** (0.140) -0.3568** (0.156) -0.0959*** (0.034) 0.0092** (0.004) 0.0009 (0.001) -0.0655* (0.035) -0.0971 (0.166) -0.0831 (0.179) 1.1328* (0.629) 0.2072 (0.320) 0.1216 (0.128) 0.0227 (0.050) 0.0131* (0.007)

Country fixed effects Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes AR(1) test (p-val.) 0.000 0.000 0.000 0.004 0.05 AR(2) test (p-val.) 0.305 0.72 0.559 0.985 0.608 Weak instruments (F-stat.) 5.86 7.85 8.86 4.61 0.36 Hansen over id (p-val.) 1.00 1.00 1.00 1.00 1.00 # of instruments/# of countries 75/50 75/30 75/26 51/15 35/5 Observations 887 420 378 204 68 Note: In columns (1)-(5), we report one-step system GMM results because the two-step system GMM estimators fail to support the validity of dynamic specification, which cannot reject AR(1) test. We use collapse command and limit the number of lags of instruments at 10 in columns (1)-(3), at 5 in column (4) and at 1 in column (1) to avoid the proliferation of the instruments. Constant is included but not reported. Robust standard errors are in parentheses. *,**,and *** are respectively significance level at 10%, 5% and 1%.

35

Table 6. Savings offset and financial openness Dependent variable Financial openness

Private savings to GDP de jure FO measure (Chinn & Ito) High Low

Low Methods Lagged private savings/GDP Public savings to GDP Public savings to GDP × Rating Public savings to GDP × Rating^2 log of GDP per capita GDP per capita growth Age dependency ratio, young Age dependency ratio, old Life expectancy at birth, total (years) M2/ GDP Change in terms of trade Crisis Country fixed effects Year fixed effects AR(1) test (p-val.) AR(2) test (p-val.) Weak instruments (F-stat.) Hansen over id (p-val.) # of instruments/ # of country group

Observations R-squared

(1)

(2)

--

--

-0.3920* (0.203) -0.0923** (0.037) 0.0037** (0.002) -0.0008 (0.001) 0.0511** (0.024) 0.1800*** (0.067) -0.1577 (0.131) -1.0453*** (0.232) -0.0128* (0.007) -0.0010 (0.001) 0.0796*** (0.022)

-0.6172 (0.668) 0.0365 (0.090) -0.0013 (0.003) -0.0021 (0.002) 0.0363 (0.026) -0.0944 (0.071) -0.0555 (0.085) -0.5090*** (0.095) 0.5232 (0.437) 0.0387*** (0.007) 0.0345 (0.041)

Yes Yes

Yes Yes

40

40

566 0.855

741 0.768

High

System GMM (3) 0.3783*** (0.115) -0.5635* (0.340) -0.0814 (0.055) 0.0043* (0.002) 0.0038 (0.003) -0.0181** (0.009) 0.2397*** (0.078) -0.2326** (0.093) -0.7584*** (0.236) 0.0009 (0.001) -0.0008 (0.001) 0.0962** (0.039)

System GMM (4) 0.5788*** (0.062) 0.0405 (1.381) -0.0580 (0.189) 0.0018 (0.006) 0.0008 (0.001) 0.0098 (0.009) 0.0299 (0.067) -0.1793*** (0.047) -0.4048*** (0.112) -0.1961* (0.114) 0.0170*** (0.006) 0.0390 (0.037)

Yes Yes 0.000 0.974 7.9 0.944 62/40

Yes Yes 0.000 0.598 2.12 1.00 62/40

561

718

Note: In columns (3) and (4), we report one-step system GMM results because the two-step system GMM estimators fail to support the validity of dynamic specification, which cannot reject AR(1) test. We use collapse command and limit the number of lags of instruments at 7 to avoid the proliferation of the instruments. We divide our full sample into two sub-samples in terms of the degree of financial openness: First, the mean of the financial openness measures for 80 individual countries is computed. Then if the mean of a country’s financial openness is above the median of the 80 countries’, it is considered a high financially open country. For de jure measure, the threshold that splits subsample is 0.69. Robust standard errors are in parentheses. *,**,and *** are respectively significance level at 10%, 5% and 1%.

36

Table 7. Savings offset and size of financial sector Dependent variable Financial market size

Private savings to GDP Deposit money bank assets to GDP Small

Lagged private savings/GDP Public savings to GDP Public savings to GDP × Rating Public savings to GDP × Rating^2 log of GDP per capita GDP per capita growth Age dependency ratio, young Age dependency ratio, old Life expectancy at birth, total (years) M2/ GDP Change in terms of trade Crisis

Large

(1)

(2)

--

--

-0.2910 (0.227) -0.1260*** (0.046) 0.0070*** (0.002) 0.0014 (0.002) 0.0375 (0.031) -0.0243 (0.069) 0.3057** (0.123) -2.2336*** (0.675) -0.0249*** (0.009) -0.0017 (0.001) 0.0615** (0.026)

1.4301* (0.776) -0.3179*** (0.100) 0.0109*** (0.003) 0.0002 (0.001) -0.0051 (0.021) 0.0097 (0.057) 0.1003 (0.062) -0.4724*** (0.085) 0.1766* (0.104) -0.0014 (0.001) 0.0455* (0.024)

Small

Large

System GMM (3) 0.4468*** (0.113) -0.6194 (0.377) -0.0811 (0.065) 0.0049* (0.003) 0.0036** (0.002) -0.0045 (0.007) 0.0838 (0.062) -0.1626** (0.068) -0.5772*** (0.153) 0.0012*** (0.000) -0.0015** (0.001) 0.0861** (0.038)

System GMM (4) 0.4446*** (0.068) 4.2363* (2.192) -0.6482** (0.282) 0.0205** (0.009) 0.0069*** (0.002) -0.0180 (0.011) 0.1148** (0.051) -0.1284* (0.074) -0.5438*** (0.171) 0.2868*** (0.082) -0.0005 (0.002) 0.0374 (0.025)

Country fixed effects Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes AR(1) test (p-val.) 0.000 0.001 AR(2) test (p-val.) 0.941 0.851 Weak instruments (F-stat.) 12.7 4.6 Hansen over id (p-val.) 1.00 1.00 # of instruments/ # of country group 38/43 38/37 62/40 61/39 Observations 636 643 638 749 Note: In columns (3) and (4), we report one-step system GMM results because the two-step system GMM estimators fail to support the validity of dynamic specification, which cannot reject AR(1) test. We use collapse command and limit the number of lags of instruments at 7 to avoid the proliferation of the instruments. We divide our full sample into two sub-samples in terms of the degree of financial market size: First, the mean of the banking sector size measures for 79 individual countries is computed. Note that China is excluded owing to no data on banking sector. Then if the mean of a country’s banking sector size is above the median of the 79 countries’, it is considered a country with a large banking sector. For banking assets measure, the median threshold is 0.54. Robust standard errors are in parentheses. *,**,and *** are respectively significance level at 10%, 5% and 1%.

37

Figure 1. U-shaped Ricardian offset coefficient -0.1

SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

Marginal effect of public saving on private saving

Column (3) of Table 1

0

-0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1

Sovereign credit rating SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

Marginal effect of public saving on private saving

Column (4) of Table 1

1 0.5 0 -0.5 -1 -1.5

Sovereign credit rating SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

Marginal effect of public saving on private saving

Column (5) of Table 1

1 0.5 0 -0.5 -1 -1.5

Sovereign credit rating

Note: These figures are drawn using the estimation results of Table 1. Bold line indicates the marginal effect of public saving on private saving, which is Ricardian offset ranged from -1 (full offset) to 0 (no offset). In the lower panel, we use savings measure divided by gross national income (GNI), instead of GDP, as the dependent variable. Dotted lines indicate 90% confidence intervals.

38

Figure 2. Different identification (two-step approach)

0 -0.1

SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

Marginal effect of discretionary public spending on total national savings

Column (1) of Table 2

0.1

-0.2 -0.3

-0.4 -0.5

0.05

0 -0.05 -0.1

SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

Marginal effect of discretionary public spending on total national savings

Column (3) of Table 2

0.1

-0.15 -0.2 -0.25 -0.3 -0.35 -0.4 -0.45

0.1 0 -0.1

SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

Marginal effect of discretionary public spending on total national savings

Column (4) of Table 2

0.2

-0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8

Sovereign credit rating Note: These figures are drawn using the estimation results of Table 2. Bold lines indicate the marginal effect of public saving rate on “national” saving rate, which is the alternative Ricardian offset ranged from -1 (no offset) to 0 (full offset). Dotted lines indicate 90% confidence intervals.

39

Figure 3. Alternative measures for fiscal sustainability Sovereign rating in domestic currency

Gross public debt to GDP

Column (2) of Table 3

Column (4) of Table 3

0.5

2.5

2.0 SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

0.0

-0.5

1.5 1.0 0.5

-0.5

1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0 17.0 18.0 19.0 20.0 21.0 22.0 23.0

0.0

-1.0

-1.0

-1.5

-1.5 -2.0

-2.0

Domestic public debt to GDP

External public debt to GDP

Column (6) of Table 3

Column (8) of Table 3

2.0

3.0

1.5

2.0

1.0

1.0 0.5 0.0

1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.011.0

0.0 0

0.2 0.4 0.6 0.8

1

1.2 1.4 1.6 1.8

2

-1.0

-0.5 -1.0

-2.0

-1.5

-3.0

Note: These figures are drawn using the estimation results of Table 3. Bold lines indicate the marginal effect of public saving on private saving, which is Ricardian offset ranged from -1 (full offset) to 0 (no offset). Dotted lines indicate 90% confidence intervals.

40

Figure 4. Developed vs Emerging Markets/Developing countries (Table 4) Panel FEs

System GMM Developed economies Column (3) of Table 4

Column (2) of Table 4 B B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

Marginal effect of public saving on private saving

-0.2

-0.2

-0.4

-0.4

-0.6

-0.6

-0.8

-0.8

-1.0

-1.0

-1.2

-1.2

-1.4

-1.4

-1.6

-1.6

-1.8 -2.0

B B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

0.0

0.0

-1.8 -2.0

Sovereign credit rating

Sovereign credit rating

Emerging market and developing economies 0.2

-0.1

SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AA-

Marginal effect of public saving on private saving

0.0

0.0

-0.2

-0.2

-0.3

-0.4

-0.4

-0.6

-0.5

Column (5) of Table 4

SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AA-

Column (4) of Table 4

-0.8

-0.6

-1.0

-0.7 -1.2

-0.8

-1.4

-0.9 -1.0

-1.6 Sovereign credit rating

Sovereign credit rating

Note: These figures are drawn using the estimation results of Table 4. Bold line indicates the marginal effect of public saving on private saving, which is Ricardian offset ranged from -1 (full offset) to 0 (no offset). Dotted lines indicate 90% confidence intervals. For the sample split, we follow the IMF classification for advanced and emerging/developing countries. See Appendix Table I for the country list.

41

Figure 5. History of debt crisis 0.5

AA-

A+

A

A-

BBB+

BBB

BBB-

BB+

BB

BB-

B+

B

B-

CCC+

CCC

CCC-

CC

C

SD,D

0

-0.5

-1

-1.5

Sovereign credit rating in foreign currency Countries have had more than 5 year debt crisis Countries have had more than 10 year debt crisis Countries have had more than 15 year debt crisis

Note: These figures are drawn using the estimation results of Table 5. Bold lines with markers denote the marginal effect of public saving on private saving, which is Ricardian offset ranged from -1 (full offset) to 0 (no offset).Dotted lines indicate 90% confidence intervals.

42

Figure 6. Variations in REs: Financial openness and Size of financial sector

0.5

0

SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

Marginal effect of public saving on private saving

w/ de jure financial openness

-0.5

-1

-1.5

Sovereign credit rating Low financial openness

High financial openness

1

0.5

0

-0.5

SD,D C CC CCCCCC CCC+ BB B+ BBBB BB+ BBBBBB BBB+ AA A+ AAAA AA+ AAA

Marginal effect of public saving on private saving

w/ deposit money banks assets to GDP

-1

-1.5

Sovereign credit rating Small banking sector

Large banking sector

Note: These figures are drawn using the estimation results of Tables 6 and 7. Bold line indicates the marginal effect of public saving on private saving, which is Ricardian offset ranged from -1 (full offset) to 0 (no offset). Dotted lines indicate 90% confidence intervals. We divide our full sample into two sub-samples in terms of the degree of financial openness and market size: First, the means of these two financial variables for 80 individual countries are computed. Then if the mean of a country’s measure is above the median of the 80 countries’, it is considered a country with high financial openness or a large financial sector.

43

Appendix Table 1. Country list (IMF classification) Country

starting year (data available)

starting year (data available)

Country

Country

starting year (data available)

Developed (IMF advanced economies) Australia

1989

Argentina

1993

Pakistan

1994

Austria

1989

Bahamas, The

2003

Panama

1997

Belgium

1989

Bahrain

2002

Paraguay

1995

Canada

1989

Barbados

1999

Peru

1997

Cyprus

1994

Bolivia

1998

Philippines

1993

Czech Republic

1993

Botswana

2001

Poland

1995

Denmark

1989

Brazil

1994

Romania

1996

Finland

1989

Bulgaria

1998

Russia

1996

France

1989

Burkina Faso

2004

Saudi Arabia

2003

Germany

1989

Cameroon

2003

Senegal

2000

Greece

1989

Chile

1992

South Africa

1994

Iceland

1989

China

1992

Sri Lanka

2005

Ireland

1989

Colombia

1993

Suriname

1999

Israel

1989

Costa Rica

1997

Thailand

1989

Italy

1989

Dominican Republic

1997

Trinidad and Tobago

1996

Japan

1989

Ecuador

2000

Tunisia

1997

Korea, Rep.

1989

Egypt, Arab Rep.

1997

Turkey

1992

Latvia

1997

El Salvador

1996

Uruguay

1994

Netherlands

1989

Estonia

1997

Venezuela

1989

New Zealand

1989

Ghana

2003

Norway

1989

Guatemala

2002

Portugal

1989

India

1990

Singapore

1989

Indonesia

1992

Slovak Republic

1994

Jordan

1995

Spain

1989

Kenya

2006

Sweden

1989

Madagascar

2004

Switzerland

1989

Malaysia

1989

United Kingdom

1989

Mali

2004

United States

1989

Mexico

1992

Morocco

1998

Mozambique

2004

Nigeria

2006

Emerging/Developing

44

Appendix Table 2. Variable description Variable Private saving rate

Description gross private savings over GDP

Source World Economic Outlook (WEO)

Public saving rate

gross public savings over GDP.

World Economic Outlook (WEO)

Sovereign credit rating Standard & Poor’s sovereign credit ratings (ranging from AAA to SD,D) are converted into numerical values using the rating migration table of Correa et al. (2014). Baseline regressions use the Foreign Currency Rating. Robustness checks in Table 6 (specification 1) use Domestic Currency Ratings. Public debt/GDP

Standard & Poor’s: Sovereign Rating And Country T&C Assessment Histories, own calculations

Abbas et al. (2010) https://www.imf.org/external/pubs/ cat/longres.cfm?sk=24332.0 WDI

Domestic public debt to GDP External public debt to GDP (log) GDPPC

Outstanding domestic public debt securities to GDP

GDP per capita is converted into constant US dollar

WDI

GDPPC Growth

GDP per capita growth is calculated in real terms.

WDI

Dependency ratio, young

A ratio of the population aged 0-14 to the population aged 15-64

Dependency ratio, Old

A ratio of the population aged 65+ to the population aged 15-64

WDI and the United Nations Population Division http://www.un.org/popin/data.html WDI and the United Nations Population Division

Life expectancy

Total years of life expectancy at birth

WDI

M2/GDP % change in the terms of trade de jure financial openness Size of banking sectors

WDI and the United Nations Population Division WDI and the OECD statistics.

A percent change in the terms of trade in goods and service

World Economic Outlook (WEO) Chinn and Ito (2006)

Deposit money bank assets divided by GDP

45

World Bank’s Financial Structure Database

Appendix Table 3. Summary statistics Full sample

Developed countries

Emerging and Developing countries

Variable

Obs.

Mean

S.D.

Min

Max

Obs.

Mean

S.D.

Min

Max

Obs.

Mean

S.D.

Min

Max

Private saving/GDP

1308

0.193

0.073

-0.317

0.473

614

0.210

0.059

-0.009

0.438

694

0.178

0.080

-0.317

0.473

Public saving/GDP

1308

0.029

0.054

-0.155

0.479

614

0.017

0.050

-0.155

0.222

694

0.040

0.054

-0.095

0.479

1387

14.427

5.160

0.000

21.000

645

18.851

2.822

1.800

21.000

742

10.581

3.320

0.000

18.000

1197

15.303

5.063

0.000

21.000

527

19.588

2.363

1.800

21.000

670

11.932

3.950

0.000

20.200

1290

0.530

0.296

0.037

2.204

611

0.621

0.326

0.068

2.204

679

0.449

0.238

0.037

1.650

853

0.346

0.243

0.007

2.189

568

0.402

0.267

0.424

2.189

285

0.234

0.126

0.007

0.548

586

0.246

0.159

0.010

0.984

--

--

--

--

--

586

0.246

0.159

0.010

0.984

(log)GDPPC

1384

8.733

1.316

5.476

10.643

645

9.859

0.524

7.918

10.643

739

7.751

0.966

5.476

9.986

GDPPC Growth

1384

0.025

0.036

-0.176

0.162

645

0.020

0.030

-0.176

0.129

739

0.030

0.039

-0.143

0.162

1387

0.399

0.168

0.195

0.932

645

0.282

0.057

0.197

0.532

742

0.501

0.167

0.195

0.932

1387

0.151

0.074

0.026

0.369

645

0.210

0.048

0.070

0.369

742

0.099

0.050

0.026

0.259

Life expectancy

1345

1.465

6.908

0.478

68.800

617

0.779

0.024

0.690

0.829

728

2.047

9.354

0.478

68.800

M2/GDP

1367

1.063

3.745

0.146

52.680

629

0.838

1.109

0.243

19.500

738

1.255

4.987

0.146

52.680

% change in the terms of trade

1307

0.004

0.066

-0.365

0.456

613

0.001

0.036

-0.138

0.272

694

0.007

0.083

-0.365

0.456

Sovereign credit rating, foreign currency Sovereign credit rating, domestic currency Gross public debt/GDP Domestic public debt/GDP External public debt/GDP

Dependency ratio, young Dependency ratio, Old

46

Appendix Table 4: Coding of Standard and Poor’s Sovereign Credit Ratings Rating AAA AA+ AA AAA+ A ABBB+ BBB BBBBB+ BB BBB+ B BCCC+ CCC CCCCC C SD,D

Numerical Value 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0

Outlook Numerical Value Positive 0.2 Watch developing 0.1 Stable 0 Watch negative -0.1 Negative -0.2 Note: Appendix Table, Correa et al. (2014)

47

Appendix Table 5. Dates of sovereign debt crises Country

Sovereign debt crises (Laeven and Valencia 2012)

Argentina

1982-1993, 2001-2005

Bolivia

1980-1992

Brazil

1983-1994

Bulgaria

1990-1994

Cameroon

1989-1992

Chile

1983-1990

Costa Rica

1981-1990

Dominican Republic

1982-1994, 2003-2005

Ecuador

1982-1995, 1999-2000, 2008-2009

Egypt,

1984-1992

Indonesia

1999-2002

Jordan

1989-1993

Madagascar

1981-1992

Mexico

1982-1990

Morocco

1983-1990

Mozambique

1984-1991

Nigeria

1983-1992

Panama

1983-1996

Paraguay

1982-1992

Peru

1978-1996

Philippines

1983-1992

Poland

1981-1994

Romania

1982-1987

Russia

1998-2000

Senegal

1981-1996

South Africa

1985-1993

Trinidad and Tobago

1989

Turkey

1978-1982

Uruguay

1983-1991, 2002-2003

Venezuela

1982-1990

48

Appendix Table 6. Savings offset and a country’s debt default history Dependent variable

Private savings to GDP Countries that have experienced more than 10 year debt crisis (4)

Countries that have experienced more than 15 year debt crisis

(2)

Countries that have experienced more than 5 year debt crisis (3)

0.6422

-0.2983

-0.4396**

-0.3985*

-0.3780

(0.449)

(0.234)

(0.213)

(0.205)

(0.230)

Public savings to GDP × Rating

-0.1969***

-0.1471***

-0.0655*

-0.0936

0.0029

(0.061)

(0.047)

(0.039)

(0.059)

(0.090)

Public savings to GDP × Rating^2

0.0070***

0.0102***

0.0023

0.0067

0.0022

(0.002)

(0.003)

(0.002)

(0.005)

(0.009)

-0.0009

-0.0003

-0.0006

-0.0009

-0.0033

(0.001)

(0.002)

(0.002)

(0.002)

(0.003)

0.0629***

-0.0357

-0.1177***

-0.0249

-0.0302

(0.018)

(0.037)

(0.036)

(0.040)

(0.091)

-0.0245

0.0809

0.1423**

-0.0133

-0.0241

(0.073)

(0.068)

(0.068)

(0.073)

(0.123)

0.0199

0.0187

0.1905

0.6883***

4.3904***

(0.060)

(0.148)

(0.146)

(0.156)

(1.075)

-0.5029***

-0.8871

0.8007

-1.0573

6.3591**

(0.089)

(0.731)

(0.685)

(0.795)

(2.350)

-0.0054

-0.0934

-0.0118

1.0189

2.2421*

(0.007)

(0.191)

(0.163)

(0.786)

(1.278)

-0.0013*

-0.0646**

-0.0315

0.0704

0.2659***

(0.001)

(0.026)

(0.023)

(0.043)

(0.081)

0.0504*

0.0555**

0.0585**

0.0194

0.0446

(0.026)

(0.028)

(0.027)

(0.031)

(0.038)

-0.0067

0.0019

0.0003

0.0107*

-0.0040

History of debt crisis

Public savings to GDP

Rating log of GDP per capita GDP per capita growth Age dependency young

ratio,

Age dependency ratio, old Life expectancy at birth, total (years) M2/ GDP Change in terms of trade Crisis

Countries that have never had sovereign default before

Countries that had debt default since 1970

(1)

(5)

(0.005)

(0.007)

(0.007)

(0.006)

(0.008)

Country FEs

Yes

Yes

Yes

Yes

Yes

Year FEs

Yes

Yes

Yes

Yes

Yes

Observations

872

420

378

204

68

R-squared 0.841 0.762 0.792 0.827 0.885 Note: Robust standard errors are in parentheses. Constant is included but not reported. *,**,and *** are respectively significance level at 10%, 5% and 1%.

49

Appendix Figure 1. Marginal effects from Appendix Table 6 1

0.5

AA-

A

A+

A-

BBB+

BBB

BBB-

BB+

BB

BB-

B

B+

B-

CCC+

CCC

CCC-

CC

C

SD,D

0

-0.5

-1

-1.5 Sovereign credit rating

Countries have had more than 5 year debt crisis Countries have had more than 10 year debt crisis Countries have had more than 15 year debt crisis

Note: Bold lines with markers denote the marginal effect of public saving on private saving, which is Ricardian offset ranged from -1 (full offset) to 0 (no offset). Dotted lines indicate 90% confidence intervals.

50