European Economic Review 66 (2014) 1 15

Contents lists available at ScienceDirect

European Economic Review

journal homepage: www.elsevier.com/locate/eer

Public debt and the limits of fiscal policy to increase economic growth
Vladimir K. Teles a,n, Caio Cesar Mussolini b
a b

Sao Paulo School of Economics - FGV, Brazil Insper, Brazil

a r t i c l e i n f o
Article history: Received 22 August 2011 Accepted 6 November 2013 Available online 25 November 2013 JEL classification: O23 O41 O54 Keywords: Public debt Endogenous growth Fiscal policy

abstract
This study proposes a theoretical model of endogenous growth that demonstrates that the level of the public debt-to-gross domestic product (GDP) ratio should negatively impact the effect of fiscal policy on growth. This effect occurs because government indebtedness extracts a portion of young people's savings to pay interest on the debts. Therefore, the payment of debt interest requires an allocation exchange system across generations that is similar to a pay-as-you-go pension system, which results in changes in the savings rate of the economy. The major conclusions of the theoretical model were verified using an econometric model that provides evidence of the validity of this conclusion. Our empirical analysis controls for time-invariant, country-specific heterogeneity in growth rates. We also address endogeneity issues and allow for heterogeneity across countries in the model parameters and for cross-sectional dependence. & 2013 Elsevier B.V. All rights reserved.

1. Introduction This paper examines how the size of the public debt-to-GDP ratio limits the effects of productive government expenditures on long-term growth. We conduct this analysis by proposing a model with overlapping generations and endogenous growth, wherein the government can go into debt to increase its productive expenditures. Various studies present models in which public expenditures affect growth within a framework of endogenous growth (Chen, 2006; Devarajan et al., 1996; Glomm and Ravikumar, 1997; Barro, 1990). The effect of these expenditures varies according to their nature, their composition, and the size of the tax burden. Other studies demonstrate that public debt can negatively affect growth (SaintPaul, 1992; Brauninger, 2005). In this paper, we establish a link between these two classes of models and demonstrate that the effect of productive expenditures on growth is limited not only by the size of the tax burden and the rate of indebtedness, as predicted by these models, but also by the debt-to-GDP ratio. Like the models of Barro (1990) and Saint-Paul (1992), the model that we propose demonstrates that an increase in public expenditures can have three direct effects on growth. An increase in public expenditures positively affects economic productivity but also has negative effects in that it increases the tax burden and public indebtedness as necessary to finance the expenditures, which results in a decrease in savings. Furthermore, as we will demonstrate here, there is an additional indirect effect. An increase in productive expenditures results in an increase in the equilibrium interest rate due to the increase in productivity. This increase, in turn, results in an increase in expenditures for interest on public debt and an additional reduction in savings because government indebtedness extracts a portion of young people's savings to pay interest on the debts of the older generation, which is no longer saving. Therefore, the payment of debt interest requires an

Corresponding author. E-mail address: vladimir.teles@fgv.br (V.K. Teles).

0014-2921/$ - see front matter & 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.euroecorev.2013.11.003

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

allocation exchange system across generations that is similar to a pay-as-you-go pension system, which results in changes in the savings rate of the economy. In the second part of the paper, we estimate a growth equation using the specifications proposed by the theoretical model for a panel of countries to provide valid evidence for the theoretical model's conclusions. Our empirical analysis controls for time-invariant, country-specific heterogeneity in growth rates. We also address endogeneity issues and allow for heterogeneity across countries in the model parameters and for cross-sectional dependence. Various studies have found empirical evidence that the allocation of public funds to education, health, and infrastructure expenses positively impacts economic growth (Aschauer, 1989; Easterly and Rebelo, 1993; Gupta et al., 2005). However, there is no consensus regarding this issue, as many studies have yielded insignificant results (Devarajan et al., 1996; Agell et al., 2006). We suggest that the disagreement among these studies can be reconciled if we more closely examine the theory. We use a specified theoretical model that clarifies the proper non-linear specification of our growth regression and allows us to disregard the role of public debt in this relationship. In addition, we find evidence that the magnitude of the effect of public debt in this context is significant. 2. Theoretical model In this section, a simple overlapping-generations model (OLG) of endogenous economic growth is developed, wherein it is established that public expenditures can affect economic productivity. The model is an extension of the work of Barro (1990) and Glomm and Ravikumar (1997) in that the government can incur debt to increase its productive expenditures. We closely follow Brauninger (2005) in the sense that we use an AK overlapping-generations model with public debt, but we allow for productive government expenditures. We also follow Brauninger (2005) in the sense that we assume that unproductive government spending is given exogenously, so the government has two options to meet its budgetary restrictions. The first one is the case where the government fix the deficit ratio, so the tax rate becomes endogenous. The second one is the case where the government fix the tax rate, so the adjusting variable is the deficit ratio. Our model is complementary to that of Adam and Bevan (2005), which is an overlapping-generation model in which public spending can be productive and may be financed by debt. Thus, it has a structure that is very similar to that of the model used here. However, whereas Adam and Bevan (2005) focus on a subject other than the balanced growth path and assess the immediate impacts (in one period) of fiscal deficits on growth, we focus on the long term, examining how the size of the debt-to-GDP ratio limits the effects of productive spending on long-term growth and how the predictions of endogenous growth models using fiscal policy may be affected. 2.1. Individuals Consider an overlapping-generations model in which each generation lives for two periods and consists of a continuum of identical individuals within the interval 0; 1. The utility function of an agent who is born in the period t 0; 1; is defined as follows:
t U ln ct t ln ct 1 ;

where is the consumption in period i of an individual who is born in t and 0 o o 1. The initial generation of older people is endowed with k0 units of capital. The following generations are each endowed with one unit of labour. The young people of generation t have the following restrictions:
t ct t st r 1 wt ; t ct t 1 r 1 r t 1 st ; t ct t ; c t 1 Z 0;

ct i

2 3 4

where kt 1 dt 1 is savings and dt are government bonds that are owned by private agents. Each individual takes the wage rate, wt, the real interest rate, r t 1 , and the tax rate, , as given. Clearly, one unit of labour is inelastically supplied by each young individual, and no older individual wishes to save. st t 2.2. Firms There is a representative firm that seeks to maximise its profits in an environment of perfect competition, and it has the following production function:
yt Az1 kt lt t 1

where yt is the output, zt are the productive government expenditures that is, the expenditures that affect the marginal product of capital lt is the labour force, and kt is the capital stock that is rented by the firm following the law of accumulation kt 1 1 kt it , where is the depreciation rate and it is the level of investment. For the sake of simplicity, let us assume 1. This production function is identical to Barro's (1990) if we assume that the labour supply is the same as the size of the

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

younger generation, i.e., lt 1. In turn, the government limits the quantity of productive expenditures as a part of the aggregate product. zt xyt ; where x is the fraction of production that is designated for productive expenditures. Using (6) in (5), we now have yt A1= x1 = kt : 7 6

Therefore, the economy's production function exhibits constant marginal returns for capital, despite the fact that the returns for firms are decreasing. This outcome is a result of the externality of productive government expenditures. The firm does not perceive that by increasing its capital stock (and, consequently, its output), it will generate an increase in productive spending by the government, and this oversight leads to an increase in the marginal products of labour and capital. Following Barro (1990) and Glomm and Ravikumar (1997), we consider productive public expenditure as spending flow rather than public capital accumulation. We use this strategy because the sum of the coefficients of kt and zt in the production function is 1, setting the AK model. Therefore, the productive public spending affects the level of aggregate productivity, as can be seen in (7). The implication is that an increase in the rate of spending will cause an increase in productivity, permanently raising the level of the marginal product of capital and hence its rate of growth. Because the rate of expenditure of the product is constant, it is ensured that the relationship between the product and both the production costs and private capital will remain constant, which greatly simplifies the design and reduces the necessary conditions for equilibrium with balanced growth. Futagami et al. (1993) show that when considering public expenditures as investments, we require a more complex system where public capital must also grow at the same rate of public debt and private capital. This modification may make steady growth less likely. If we consider public expenditures as investments that increase the growth rate of the stock of public capital, the rate of productivity growth can increase without limits, which implies that the marginal product of capital can grow indefinitely given the nature of the AK aggregate production function. Hence, the investment in physical capital and economic growth assumes an explosive trajectory. In this case, there will not be a balanced growth path in the model. 2.3. Government The government spends a fixed fraction 0 o g o 1 of output on consumption, which we assume does not affect the utility of private agents; the government also spends a fraction x on productive expenditures and levies a tax rate on the income of agents. In addition, it borrows resources from the private sector by issuing one-period domestic bonds dt that pay the risk-free gross interest rate 1 r t 1 , where d0 0. Changes in government debt are expressed as the primary deficit g xyt wt added to the debt service payments r t dt . Thus, the equation below represents the inter-temporal government budget constraint: dt 1 dt r t dt g xyt wt : 8

2.4. Competitive equilibrium

t t Given k0, a competitive equilibrium for this economy involves a sequence of allocations fkt 1 ; dt 1 ; yt ; ct t ; ct 1 ; st gt 0 and prices fwt ; r t g1 , such that t0 1

1. 2. 3. 4.

t t Given wt and r t 1 , the allocation ct t ; ct 1 ; st solves the maximisation problem for the younger generation t. Given wt and rt, the allocation yt ; kt solves the maximisation problem for the representative firm. st t kt 1 dt 1 . t 1 yt ct kt 1 gyt xyt . t ct

The solution to the optimisation problem for the younger generation in period t yields the following: st t 1 wt =1 : The profit maximisation of the firms leads to yt 1 Az1 kt rt : t kt Substituting (6) and (7) into this equation, we find r t A1= x1 = 10 9

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

and, given that technology exhibits constant returns to scale with the private factors, there is no economic profit; therefore, yt yt 1 kt 1 Az1 kt wt t kt

and analogously wt 1 A1= x1 = kt : 11

Therefore, because productive government expenditures affect the productivity of the economy, they naturally also affect interest rates and wages. Because the present model is represented in the AK form, savings affect long-term growth. Eq. (9) demonstrates that shocks to the equilibrium wage lead to greater savings, and Eq. (11) indicates that shocks to productive government expenditures increase the wage level. Therefore, increases in productive expenditures lead to permanent productivity shocks and higher wages. This change produces an increase in savings, investments, and subsequently, economic growth. However, it is important to note that in (9), savings depend negatively on ; as decreases, the net gain from productivity increases. This result is consistent with those obtained by Barro (1990) and Glomm and Ravikumar (1997). 2.5. Fixing the deficit ratio In this section we assume that the government adjusts its tax on income to meet the budget constraint. In such a case, the deficit ratio is exogenous, and is defined as dt 1 dt =yt , whereas taxation is endogenous, responding to changes in spending. Therefore, we can write Eq. (8) as yt wt g xyt r t dt : We can use Eqs. (7) and (10) in Eq. (12) to express the tax rate as
t g x d kt

12

13

and use Eqs. (7), (11) and (13) in Eq. (9) to express the capital growth rate as follows: kt 1 kt g x 1 A1= x1 = 1 kt   dt A1= x1 = A1= x1 = 1 1: 1 kt

14

Eq. (14) demonstrates the dynamic equilibrium of capital and, consequently, that of the output. There are three components on the right-hand side of the equation that show us the three paths by which an increase in productive government expenditures can affect growth. The first part of the equation demonstrates productive expenditure effect over savings, and consequently, economic growth. As mentioned earlier this effect is ambiguous once it raises the marginal product of labour, but to keep the deficit ratio constant, the government must increase taxes. The second component of the equation will be zero if the deficit-to-GDP ratio is zero. This component also shows that the marginal effect of productive government expenditures varies depending on the deficit, which demonstrates that when the government increases its productive expenditures there is a negative effect on growth because this directly increases government indebtedness and reduces the savings that are designated for private investment. The third component of the equation reproduces the indirect impact of productive government expenditures on indebtedness and, therefore, growth. Because increases in x raise interest rates and because the government pays the interest rate on public debt, an increase in productive expenditures influences the costs of the debt, which indirectly raises the rate of government indebtedness. As the public debt increases, this relationship becomes more perverse, and productive expenditures have a lower marginal effect on growth. Models with budgetary equilibrium, like those found in Barro (1990) and Glomm and Ravikumar (1997), only provide results for the first part of the equation. The second component of the equation follows the same logic as the models of debt and endogenous growth that are found in Saint-Paul (1992) and Brauninger (2005), wherein increases in the public debt lead to greater indebtedness, which decreases the amount of savings allocated to capital accumulation. In our approach, as in Adam and Bevan (2005), we find an additional effect (represented by the third part of the equation) that makes the size of the debt relevant. We can understand this effect as similar to the dynamics of a pay-as-you-go pension system in that a portion of young people's savings is reallocated to pay the older generation (through greater debt interest payments) instead of to the accumulation of capital, which diminishes economic growth. Thus, the magnitude of the impact of an increase in productive expenditures on economic growth varies based on the size of the deficit and the public debt. In other words, it is more likely that increases in infrastructure, education, and health expenditures, or increases in any other productive expenditures will stimulate economic growth when the unproductive portion of the government and the debt-to-GDP ratio are lower.

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

, d k k

d d k
Fig. 1. Equilibrium.

To analyse whether there is a steady-state equilibrium in the economy, we must examine the debt dynamics. Using (7) and the deficit-to-GDP definition, we attain dt 1 dt kt A1= x1 = : dt dt 15

Thus, for the growth rates of kt and dt to be constant or for there to be an equilibrium, the ratio dt =kt needs to be constant such that kt 1 =kt dt 1 =dt . By setting Eqs. (14) and (15) equal to one another, we obtain a quadratic equation for the equilibrium debt-to-capital ratio, the solution to which is defined as follows: p d WA1= x1 = 1 7 J H ; k 2J where we defined W, J and H as W g x 1 1 A1= x1 = J  1 1 " #2 1 WA1= x1 = A1= x1 = : H 4 J J

Thus, if the solution above has at least one real positive root, there is a steady-state equilibrium. Naturally, the existence of an equilibrium depends on the values of the model parameters, and as demonstrated by Brauninger (2005), an equilibrium may be obtained for plausible parameter values. ^ and d ^ are the capital and debt growth rates, Eqs. (13) and (14) are plotted in an example in which we In Fig. 1, where k assume that the two roots are real and positive and 4 0. We obtain a result with two equilibriums. The first, which is characterised by a low debt-to-capital ratio, a low debt-to-GDP ratio, and a high economic growth, is locally stable, whereas the second, which is characterised by low growth and a high debt-to-GDP ratio, is unstable. In Fig. 2, we evaluate a scenario in which an economy in equilibrium experiences an increase in unproductive government expenditures. In this case, the capital growth curve shifts downward and the debt growth curve remains the same. If the economy is initially at stable equilibrium with high growth, this leads to a decline in steady-state economic growth and an increase in the debt-to-GDP ratio.1 In an extreme situation with an excessive increase in unproductive expenditures, the two curves no longer intersect, meaning that the growth rate for debt will be larger than that of the economy, suggesting an explosive increase in debt and an absolute economic collapse. This would be the case if there were no steady-state equilibrium. This result is essentially the same as that obtained by Brauninger (2005). However, the case that most interests us, and the one that this paper attempts to clarify, involves increases productive government expenditures. We can see in Fig. 3 that the capital growth curve slope rises but the debt growth curve does not move. The result is a higher economic growth and a lower debt-to-GDP ratio, for the stable equilibrium. This increase in economic growth will occur because the fiscal cost of increasing this type of spending is still small in comparison to the
1 Because capital is proportional to GDP (see Eq. (7)), the debt-to-capital ratio and the debt-to-GDP ratio always move in the same direction due to changes in the economic environment.

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

k,d

d d k
Fig. 2. Increase in government consumption with endogenous taxes.

k, d

d k
Fig. 3. Increase in productive expenditures with endogenous taxes.

resulting productivity gains. Consequently, the economy still has the necessary fiscal strength, and this type of policy is very successful.2

2.6. Fixing the tax rate Now we assume that tax rate is exogenous. In this case, if the government wants to raise its spending he must do that by increasing his deficit. We can use Eqs. (7)(10) to express the capital growth rate as follows: kt 1 kt 1 1 A1= x1 = 1 kt  d t g x 1 A1= x1 = 1 A1= x1 = 1: kt 16

Thus, we arrive at an equation for the capital-growth rate that is analogous to that of the previous model. Therefore, the main conclusions are very similar: (i) A rise in productive spending increases economic growth through productivity and savings, however, once again, the size of this effect decreases as the size of the tax burden increases, as in Barro (1990); (ii) The primary deficit ratio has a negative impact on growth by increasing government indebtness; (iii) The level of debt reduces growth because interest expenditures are proportional its level.
2 However, if the economy is at the unstable steady-state equilibrium, the debt growth rate will exceed the capital growth rate, leading to an explosive increase in the debt-to-capital ratio.

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

, d k

d d k
Fig. 4. Increase in government consumption with exogenous taxes.

Substituting Eqs. (7), (10), and (11) into Eq. (8), and dividing by dt, we obtain the debt growth rate: dt 1 dt kt A1= x1 = g x 1 A1= x1 = : dt dt 17

Proceeding as before, by setting Eqs. (16) and (17) equal to one another, we obtain a quadratic equation for the equilibrium debt-to-capital ratio, the solution to which is defined as follows: p d W A1= x1 = 1 7 J H k 2J where we defined W , J and H as W  1 1 g x 1 1

J  1 A1= x1 = " #2 1 W A1= x1 = A1= x1 = H  4g x 1 J J

However, as mentioned by Brauninger (2005), a fixed tax rate usually does not lead to a stable steady state, for plausible parameters. Ignoring that for a moment, we can proceed as in Section 2.5, showing the main results graphically. Assuming that g x 1 4 0, we have an equilibrium that looks very similar to Fig. 1, so we do not repeat that figure again. In Fig. 4 we show the effects of an increase in unproductive government expenditures. As in Section 2.5, the capital growth curve shifts downward, but now the debt growth curve shifts upward. Again there is a decline in economic growth if we consider the stable equlibrium. The outcome of an increase in productive government expenditures can be seen in Fig. 5. The capital growth curve slope rises and the debt growth curve shifts upward. A possible result for the stable equilibrium is a higher economic growth and a higher debt-to-GDP ratio, as depicted on that figure, but we cannot assure that, once it depends on the parameter values. 2.7. Summary There are at least two important conclusions that we can extract from this model. First, unlike in the models that have been developed thus far, such as those of Saint-Paul (1992) and Brauninger (2005), increases in the debt-to-GDP ratio may be linked to increases in the growth rate under certain circumstances.3 Second, the marginal effect on growth that is caused by productive expenditures, such as expenditures on infrastructure, education, and health, depends on both the primary surplus of the government and the size of the debt. Saint-Paul (1992) model includes government debt in an overlapping-generations model of perpetual youth with a production function with externalities similar to those used in Romer (1986), i.e., with an aggregate production function AK. In Saint-Paul's (1992) model, the implications of social security and public debt are revisited, and the role of these instruments in avoiding the possible over-accumulation of capital in typical OLG models disappears. Consequently, any
3 Although Adam and Bevan (2005) model is similar to ours, they do not explore the issue of financing productive government expenditures by increasing debt.

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

k,d

d k d k
Fig. 5. Increase in productive expenditures with exogenous taxes.

change in the public debt will be Pareto-inefficient. Moreover, a higher level of debt implies lower growth once the debt decreases the proportion of savings allocated to productive investment, which is the engine of growth in the AK model. Our model can be seen as an extension of Saint-Paul (1992) that includes productive public spending. This extension changes some results and generates new insights. With this change, an increase in debt under certain fiscal conditions and parameter values may be linked to increased growth, and thus, changes in debt now may be Pareto efficient. Moreover, the inclusion of productive expenditures in the production function implies a new channel through which debt affects growth. Now the interest rate depends on productive spending, so there is a crowding-out effect that is proportional to the size of the debt-to-GDP ratio. This effect can also be seen as a transfer of resources from the young generation, the savers, to the old generation, which reduces the impact of productive government expenditures on growth. 3. Empirical evidence The theoretical model has some important empirical implications. Most importantly, the model indicates that the magnitude of the effect of productive expenditures on growth depends on the debt-to-GDP ratio. In countries with a high debt-to-GDP ratio, an increase in productive expenditures implies an increase in the marginal product of capital and hence in the interest rate, which, in turn, has a significant impact on spending on debt service, generating a crowding-out effect. As the debt grows larger, the crowding-out effect of the interest rate also increases. Similarly, the tax burden and the size of the fiscal deficit also affect the magnitude of the impact of productive expenditures on growth, as illustrated in the theoretical model. In this section, we conduct an empirical exercise to determine whether we can find the correlations predicted by the model. Of course, because of the difficulty of solving various problems, which certainly feature significant endogeneity, we cannot perform a test of the model or make any assertions about the direction of causality. Moreover, there are many other possible explanations for the relationship between fiscal variables and growth that are not described in the model; therefore, the empirical analysis is unable to prove that our model is true. Instead, our aim is to clean the data of econometric problems and to determine whether the correlations are consistent with the theoretical model, providing some evidence that its implications make sense. Our estimation strategy employs the principles outlined in Adam and Bevan (2005), in which the source of financing for expenditures implies the non-linearity of productive expenditures in the specification used. However, we include the effects of the debt-to-GDP ratio to observe the crowding-out effect suggested by the theoretical model. The various empirical problems that have been outlined in growth econometrics are considered to provide adequate empirical evidence. In particular, it is necessary to address the likely problems of parameter heterogeneity; cross-section dependence; endogeneity; and time-invariant, country-specific heterogeneity (Durlauf et al., 2005; Bond et al., 2010). To avoid the problem of endogeneity, we use a standard GMM estimator. However, problems such as heterogeneity of parameters and cross-section dependence can generate bias and inconsistency for the estimators, even if the endogeneity problem has been solved. The non-linear nature of a model that includes interactions between key variables may resolve a large portion of the heterogeneity of the parameters if the theoretical model is correct; however, it is likely that other institutional characteristics such as corruption and bureaucracy also affect the productive expenditure coefficient in the growth equation. Therefore, using the strategy that was adopted by Lee et al. (1997), Bond et al. (2010), Gemmell et al. (2011), who employ growth regressions using a panel that is similar to ours, we use the mean group approach of Pesaran and Smith (1995) to attempt to resolve the heterogeneity of the parameters. Because it is likely that countries suffer from cross-section dependence during the shocks in the growth process, we also use the multifactor error structure proposed by Pesaran (2006).

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

These methods require the use of a series of annual country growth data. This approach is possible in our case because the growth and fiscal data vary from year to year.4 A potential problem with utilising annual data for growth studies is the possibility that the results reflect fluctuations in the series rather than fluctuations in the long term. Following Bond et al. (2010), we rely on dynamic econometric specifications to implicitly filter out these higher-frequency influences while acknowledging the limitations of this approach. Using country-level panel data, we employ a specification in which growth depends on its lagged values in addition to the variables of interest in a linear version of the parameters based on Eq. (12). Because Eq. (12) is no longer an equilibrium equation, an alternative is to simultaneously perform the estimation using Eq. (13); however, because there is no variable in Eq. (13) that is not present in Eq. (12), this approach does not help us to identify the equation. Therefore, we opt to estimate Eq. (12) using a two-step GMM process. Nonetheless, we recognise that this is not an equilibrium relationship; rather, we only verify whether the relationship that is predicted by the model has some basis in the data. In the theoretical model, the impact of productive expenditures or of any other increase in productivity on growth exists because such policies increase the marginal return of capital and thus stimulate investment. However, the size of the gain crucially depends on fiscal variables. Therefore, we define the empirical model with a non-linear specification with interactions to better fit the implications of the theoretical model. For example, in Eq. (9), one can see that a shock to productive spending affects savings but that the size of this effect depends on the size of the tax burden because the higher the income taxation, the lesser the degree to which the productivity shock translates into net marginal returns. Thus, the effect of taxation it is not just the increase in taxation that was increased by spending, which would be corrected by a linear specification; rather, the channel through which productivity shocks affect growth depends on the size of the tax burden. The same can be said for the size of the fiscal deficit, which determines how much of the savings will be allocated to closing the public accounts. A productivity shock increases the savings, but a portion of this increase will not generate private investment because it will be used to cover the deficit. This crowding-out effect also influences the impact of debt size because an increase in productivity raises interest rates, which results in a higher payment on debt service that is proportional to the size of the debt. Thus, the channel through which productive government spending shocks affect growth depends on the components of the financing of public spending, which in turn determine its magnitude. A simple correlation cannot be used to close the budget; rather, a nonlinear specification is required, with components of interaction that are better able to capture the relationship between productive spending and growth following the theoretical model. Therefore, the regression model is defined as follows: gr it 0 s gr it s 0 xit 1 1 xit 1 n it 1 2 xit 1 n surplusit 1
s1 T

3 xit 1 n dit 1 =yit 1 4 dit 1 =yit 1 i t eit ;

18

where grit is the per capita GDP growth rate, which is expressed as a percentage of country i in year t, and where T is the number of lags of the dependent variable that are included among the explanatory variables. The variables of interest are xit and its interaction with the tax burden, debt, and primary surplus. The variable xit is the sum of central government spending on education, health, transportation, communications, and energy divided by the GDP. We chose these as productive expenditures following the criteria that are traditionally used by Aschauer (1989), Barro and Sala-i-Martin (1995), Devarajan et al. (1996), Kneller et al. (1999), and Adam and Bevan (2005), who classify these expenditures as productive because they are used to form physical and human capital. Conversely, spending in the categories of social insurance, recreation, and economic services, as well as unclassified spending, was not included in the specification because these categories of spending are considered growth neutral. We recognise that if these other expenses are not growth neutral, the interpretation of the coefficients will change. The effect of a unit change in a relevant variable could be offset by that of a unit change in an omitted category, which may be an implicit financing element (Kneller et al., 1999). The variable is the tax burden on income as a fraction of the GDP. The variable surplusit is the primary surplus of the central government relative to the GDP, and the variable dit =yit is the debt-to-GDP ratio. All of these values are expressed as percentages. Dummy variables for the year were included to show the common effects over time across countries, t. Timeinvariant, country-specific heterogeneity was captured using fixed-effect, cross-section dummy variables i with the goal of controlling for the effects of institutional, geographic, and other differences that did not change during the study period. In the steady-state equilibrium, with constant fiscal variables and without shocks, per capita production grows at a constant country-specific rate that is defined as follows: gr i 0 0 xi 1 xi n i 2 xi n surplusi 3 xi n di =yi i : 1 T s 1 s 19

4 If the series data do not contain annual frequency averages (e.g., data for human capital), do not vary annually (e.g., data on political systems), do not vary over time (e.g., data on colonial origins), or vary little over time (e.g., data on income inequality), this method may not be appropriate. For further discussion, see Bond et al. (2010) and Durlauf et al. (2005).

10

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

Therefore, the long-term effect of an increase in productive government expenditures on the GDP will be defined as follows: 0 1 n i 2 n surplusi 3 n di =yi : 1 T s 1 s 20

We wish to determine whether the interactions of xit with the tax burden, debt, and primary surplus are significant in the long term. Therefore, we test whether the relationships i =1 T s 1 s , i 1; 2; 3 are significant and have the correct signs. As expected, the dynamic structure of the model itself is such that using the least squares method produces biased and inconsistent estimators. Consistent estimators can be obtained if shocks eit are serially uncorrelated using the lagged values of the endogenous variables that are not correlated with eit and eit 1 as instrumental variables. In addition, the use of instrumental variables makes it possible to correct for the endogeneity of the current values of the fiscal variables. 3.1. Pooled results Eq. (15) was estimated using unbalanced panel data from 74 countries during the period 19722004. The data for economic growth, GDP, investment as a percentage of GDP, and trade are obtained from the Penn World Tables 6.3; the inflation data come from the World Development Indicators (WDIs); the information on productive expenditures, tax burden, and the primary surplus comes from the Government Financial Statistics (GFS) released by the IMF; and the debtto-GDP ratios are obtained from the Inter-American Development Bank.5 The results of the estimations for the baseline specifications that were obtained using pooled annual data are shown in Table 1, in which the same slope parameters are used for all of the included countries. In the GMM specifications, grit-1, xit 1 , xit 1 n it 1 , xit 1 n surplusit 1 , and xit 1 n dit 1 =yit 1 are treated as endogenous. The instrument set consists of grit-5, lags 5 and 6 of the GDP per capita, lags 24 of investment as a percentage of the GDP, lags 2 and 3 of trade (the sum of exports and imports) as a percentage of GDP, and lags 2 and 3 of inflation. These variables are among those that have been used as instrumental variables in similar estimations in the literature, for instance, in studies by Kneller et al. (1999), Agell et al. (2006) and Bond et al. (2010)).6 The validity of these instruments was not rejected when we conducted the SarganHansen test for over-identifying restrictions. The first two columns report that the results of the estimations obtained using OLS analysis with up to four lags and with and without the interactions of the expenditure variable with the other fiscal variables. Columns three and four depict the results for the same specifications using a two-step GMM analysis. Various specifications using different numbers of maximum lags of the dependent variable, T, were employed.7 We consistently found similar results, as shown in the last two columns, in which it is possible to compare the results of T 4 and T 1. The comparison of the results obtained using OLS and GMM suggests that the results obtained using OLS considerably underestimate the degree of growth rate persistence and therefore underestimate the long-term effects of the fiscal variables on growth. For this reason, we focus on the GMM results. The results obtained without the interactions demonstrate the existence of a negative and insignificant effect of productive expenditures on growth for both of the longterm coefficients. Conversely, by including the interactions of the fiscal variables, we find that public expenditures have a positive and significant effect; however, this effect depends on the values of the other fiscal variables in the predicted direction of the theoretical model. These results provide us with some evidence that the data demonstrate the same direction as the proposed results of the theoretical model. In part, the results show that the inclusion of additional elements from the government budget makes the model more complete and therefore allows us to better identify the effects of productive spending (Kneller et al., 1999). We believe that these results also provide evidence that the heterogeneity of the effects of spending on the productivity growth of countries depends on financing conditions and the initial fiscal health of each country. The size of the effect of productive expenditures on growth is calculated assuming a primary surplus of 3% and a tax burden of 23%. In this scenario, the marginal effect of an increase in productive expenditures of 1% of GDP on long-term growth varies from 0.36% to 0.26% for values of the debt-to-GDP ratio that range between 10% and 120%, as shown in Fig. 4. Changes in the values of the tax burden and the surplus shift this relationship upward or downward; however, given that the coefficients of these interactions are very small, these shifts are imperceptible. This result shows that the interactions with the tax burden and with the surplus, although they are significant and have the expected direction, are not large. The results suggest that the size of the debt-to-GDP ratio is the most relevant variable in the effect of productive expenditures on growth, as this variable is quantitatively and statistically significant.
A table in the Appendix presents the names of the included countries. Other estimations were derived using the system GMM process with an alternative strategy that has been used in the literature, as in Bond et al. (2001) and Gregoriou and Ghosh (2009). We found that the results were similar and were robust to changes in the selection of the instrumental variables and in the estimation method. However, once we used the pooled annual data, the number of time series observations in the panel increased vis-a-vis the number of countries, making the system GMM process an inadequate estimator. Therefore, we opted to conduct an analysis using the same strategy as Bond et al. (2010) and used a two-step GMM estimator. 7 The lags were included up until the point at which they ceased to be significant: 4 years. We tested specifications including lags 1-4 and obtained the same results.
6 5

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15 Table 1 Baseline specifications. Variables (i) OLS 4 lags 4.49e 06 ( 0.35) (ii) OLS 4 lags 0.0418 (0.70) 0.0081nn ( 1.86) 1.48e 08 (0.69) 1.04e 07 ( 0.73) 0.0001 ( 0.70) 0.0467 [0.70] 1.65e 08 [0.69] 1.16e 07 [ 0.73] 0.0001 [ 0.71] (iii) GMM 4 lags 0.0002n ( 3.14) (iv) GMM 4 lags 0.4361n (2.15) 0.0107 (0.26) 1.56e 07n (2.15) 1.03e 06n ( 2.15) 0.0011n ( 2.16) 0.3745n [3.32] 1.34e 07n [2.15] 8.87e 07n [ 2.12] 0.0009n [ 3.35] 3.322 [0.5055] 0.41 [0.68] 1.37 [0.17] (v) GMM 1 lag 0.4522n (2.62) 0.0031 (0.10) 1.62e 07n (2.62) 1.07e 06n ( 2.61) 0.0011n ( 2.63) 0.3754n [3.50] 1.34e 07n [2.62] 8.89e 07n [ 2.58] 0.0009n [ 3.52] 2.217 [0.8184] 0.64 [0.52] 1.09 [0.27]

11

Lags included xt 1 dt 1 =yt 1 xt 1 n surplust 1 xt 1 n t 1 xt 1 n dt 1 =yt 1 Long-run effects x x n surplus xn x n d=y Hansen J Stat p-value Test of first-order serial autocorrelation p-value Test of second-order serial autocorrelation p-value

6.05e 06 [ 0.35]

0.0016 [ 1.17]

11.198 [0.1907] 2.10 [0.03] 1.50 [0.13] 1.09 [0.27] 0.34 [0.73] 3.27 [0.001] 1.06 [0.28]

Notes: A full set of year and time dummies were included. t-ratios are in parentheses. z-ratios of tests of the long-run effects, the p-values of the serial correlation tests and p-values of the SarganHansen test of over-identifying restrictions are in square brackets. In the GMM specifications grit-1, xit 1 , xit 1 n it 1 , xit 1 n surplusit 1 , xit 1 n dit 1 =yit 1 are treated as endogenous. The instrument set consists of gr it 5 , lags 5 and 6 of GDP per capita, lags 24 of investment as percentage of GDP, lags 2 and 3 of trade (sum of exports and imports) as percentage of GDP and lags 2 and 3 of inflation. n Significance level at 0.05. nn Significance level at 0.10.

The level of debt can affect growth because it implies that there will be higher taxes in the future, which discourages investment. However, if the most important effect is the intertemporal substitution of taxation, the debt level should be the most relevant tax variable. Nevertheless, the debt level is insignificant in the estimates, and only the interaction of debt with productive expenditures is significant. This provides us with some evidence that the channel via the interest rate is relevant. Fig. 6 3.2. Subsample results Table 2 separately reports pooled IV estimates for our baseline model in the last column of Table 1 for subsamples of high income OECD countries and non-OECD countries based on the classification of the World Bank. The results show that the effects of fiscal policy on growth are heterogeneous across countries and that these effects are correlated with per capita income. The results for the non-OECD countries are very similar to the regression results for all countries. However, the results for the high-income countries show no significant relationship between the fiscal variables and economic growth. This result corroborates the results of other studies such as those of Jones (1995), Aghion et al. (2006) and Bond et al. (2010), who find no significant relationship between savings and long-term growth for developed countries but find a strong relationship between them for poor countries. When the relationship between fiscal variables and growth in our model is based on an AK model in which savings are the driving force of growth, the effect of the fiscal variables on growth is also restricted to poor countries. 3.3. Results for mean group estimates In this section, the baseline model is re-estimated while relaxing the hypothesis that the slope parameters are common across countries. If this restriction is invalid, then the inferences from the previous sections will also be invalid. The heterogeneity of the

12

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

0.4

0.35

0.3

0.25

0.2

0.15

10

20

30

40

50

60

70

80

90

100 110 120

debt / GDP (%) surplus = -3; tax = 23

Fig. 6. Marginal effect of productive expenditures (x) in economic growth. Table 2 OECD and non-OECD subsamples. Variables xt 1 dt 1 =yt 1 xt 1 n surplust 1 xt 1 n t 1 xt 1 n dt 1 =yt 1 Long-run effects x x n surplus xn x n d=y Hansen J Stat p-value Test of first-order serial autocorrelation p-value Test of second-order serial autocorrelation p-value OECD 0.3793 (1.12) 0.0958 (1.30) 0.0263 ( 0.68) 0.0131 (0.97) 0.1192 ( 1.40) 0.4716 [0.75] 0.0328 [ 1.10] 0.0163 [1.24] 0.0148 [ 0.98] 2.878 0.5784 3.50 [0.0005] 1.42 [0.15] Non-OECD 0.5320nn (1.86) 0.2001 ( 0.89) 1.90e 07nn (1.87) 1.26e 06nn ( 1.86) 0.0013nn ( 1.87) 0.3846n [2.13] 1.38e 07nn [1.87] 9.10e 07nn [ 1.84] 0.0009n [ 2.14] 3.147 0.5336 3.60 [0.0003] 0.32 [0.74]

Notes: A full set of year and time dummies were included. t-ratios are in parentheses. z-ratios of tests of the long-run effects, the p-values of the serial correlation tests and p-values of the SarganHansen test of overidentifying restrictions are in square brackets. In the GMM specifications grit 1 , xit 1 , xit 1 n it 1 , xit 1 n surplusit 1 , xit 1 n dit 1 =yit 1 are treated as endogenous. The instrument set consists of gr it 5 , lags 5 and 6 of GDP per capita, lags 24 of investment as percentage of GDP, lags 2 and 3 of trade (sum of exports and imports) as percentage of GDP and lags 2 and 3 of inflation. n Significance level at 0.05. nn Significance level at 0.10.

parameters across countries is plausible because the efficiency of public expenditures may be different under alternative institutional conditions. Therefore, given the likely importance of heterogeneous coefficients, we use the Mean Group Estimator of Pesaran and Smith (1995) and employ the same strategy that Lee et al. (1997), Bond et al. (2010), Gemmell et al. (2011) use for growth regressions with a panel that is similar to ours.

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15 Table 3 Mean group estimates. Long-run effects Allowing for cross-sectional dependence x x n surplus xn x n d=y Robust mean No 0.0250 [0.01] 7.98e 09 [0.09] 7.54e 08 [0.02] 0.0001 [0.02] Robust mean Yes 0.0225 [0.02] 9.10e 09 [0.04] 6.99e 08 [0.03] 0.001 [0.01]

13

Note: Robust Means are reported. p-values are in square brackets. All specifications include grit 1 , xit 1 , xit 1 n it 1 , xit 1 n surplusit 1 , xit 1 n dit 1 =yit 1 treated as endogenous. See Table 1 for the list of instruments used (here in deviations-from-means form).

In simple terms, the Mean Group Estimator individually estimates Eq. (15) using a two-step GMM, the same instrument that was previously used for each country, and transforms the variables by subtracting the sample mean values for the same year from the original series to correct for the possible effects of common shocks, at which point the average of the estimated coefficients is obtained for the different countries. Here, we individually obtain the estimated long-term effects for the various countries and estimate the robust means together with their standard errors.8 Another possible challenge of these estimations is the potential existence of cross-sectional dependence. Cross-sectional dependence will occur if the economic growth of a country affects the growth of other countries such that the residual eit is not independent for individual countries. This effect is plausible if we believe that there are spillovers that impact technology and the accumulation of physical or human capital (Conley and Ligon, 2002) or if we believe that the economic development of one country affects the terms of trade of other countries (Acemoglu and Ventura, 2002). To correct for this potential problem, we follow Pesaran (2006) in including the annual average growth of each country as an explanatory variable. This method also allows us to include more flexible tendencies in the individual estimations for the specific countries. Table 3 depicts the results obtained using the Mean Group Estimator for the long-term parameters that were estimated using the variables of interest in the model. The columns depict the specifications with and without correction for crosssectional dependence. The values are lower in magnitude for the effects of fiscal policy on growth, showing some evidence that the pooled estimations may be biased if parameter heterogeneity is not considered. Nevertheless, all of the long-term coefficients have the signs that were previously suggested by the theoretical model and are statistically significant. Furthermore the marginal effect of productive spending becomes negative for a debt-to-GDP ratio greater than 22.5%.

4. Conclusion This paper proposes a theoretical model of endogenous growth in which the level of the public debt-to-GDP ratio can negatively impact the effects of productive public expenditures on growth. The main conclusions obtained from the theoretical model are verified using an econometric model that provides evidence of the validity of the theoretical model. Our empirical analysis controls for time-invariant, country-specific heterogeneity in the growth rates. We also address endogeneity issues and allow for heterogeneity across countries in terms of the model parameters; we also allow for crosssectional dependence. Our approach has enabled us to verify the existence of effects that have already been predicted in the literature, such as the non-linear effects of productive expenditures on growth given the size of the tax burden, as shown by Barro (1990), or given the indebtedness rate. Such effects are negative for direct capital accumulation because they lead to diminishing marginal net returns of capital or savings extracted from the economy to finance public expenditures. In addition to isolating the above effects, we were able to observe an additional effect: the impact of productive expenditures on growth depends on the size of the debt-to-GDP ratio because an increase in the magnitude of productive expenditures leads to an increase in the productivity of the economy and, thus, to an equilibrium of interest rates. The latter occurs because there is no decreasing marginal return for aggregate capital in the endogenous growth models. This increase in interest rates leads to higher government spending for debt servicing; as the size of the debt increases, so does the impact of this increase on interest rates. For this reason, a higher debt-to-GDP ratio corresponds to a smaller impact of productive expenditures on economic growth. We can also understand this effect as a process of income transfer between generations: the income flows from the younger generation, which has a portion of its savings invested in government securities and thus decreases its own capital
8

This is a robust variant of the Mean Group Estimator that is suggested by Bond et al. (2010).

14

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15 Table A1 List of Countries. South Africa Germany Argentina Australia Austria Finland Barbados Belarus Belgium Bolivia Brazil Bulgaria Burundi Bhutan Cameroon Canada Kazakhstan Chile Kuwait Lesotho Costa Rica Luxembourg Egypt El Salvador Slovakia Mauricio Spain United States Estonia Ethiopia Norway Panama Gambia Greece Guyana Netherlands Czech Republic Rwanda Yemen Indonesia Ireland Iceland Israel Cyprus Tajikistan Latvia Denmark Maldives Zambia Morocco Slovenia Mexico Mongolia Nepal Nicaragua Zimbabwe France Pakistan Paraguay Poland UK Honduras Hungary Senegal Sri Lanka Sweden Switzerland Suriname Thailand Colombia Tunisia Turkey Uruguay Malta

accumulation, to the older generation, which does not save and has its interest paid in this way. In this sense, the observed effect is similar to that of the pay-as-you-go pension system in overlapping-generation models, in which income is transferred between generations and decreases the accumulation of capital. In addition to incorporating the effect of public debt on the relationship between productive expenditures and economic growth, the model also demonstrates that increases in the size of the debt can lead to greater economic growth; the status quo is a healthy fiscal situation, and indebtedness is associated with an increase in productive expenditures. This finding runs contrary to those of previous models, in which debt increases always lead to decreased growth. This result shows that changes in the public debt can be Pareto optimal, leading to benefits for all generations. This conclusion is quite different from the one encouraged by the results obtained using endogenous models of debt, in which expenditures are always unproductive. We used the econometric specifications that were derived from the implications of the theoretical model to validate the main conclusions of the model. The results suggest that the size of the debt-to-GDP ratio is a relevant variable in the effects of productive expenditures on growth.

Appendix A See Table A1. References

Acemoglu, D., Ventura, J., 2002. The world income distribution. Q. J. Econ. 117, 659694. Adam, C.S., Bevan, D.L., 2005. Fiscal deficits and growth in developing countries. J. Public Econ. 89, 571597. URL http://ideas.repec.org/a/eee/pubeco/ v89y2005i4p571-597.html. Agell, J., Ohlsson, H., Thoursie, P.S., 2006. Growth effects of government expenditure and taxation in rich countries: a comment. Eur. Econ. Rev. 50, 211218. Aghion, P., Comin, D., Howitt, P., 2006. When does domestic saving matter for economic growth? NBER Working Papers 12275, National Bureau of Economic Research, Inc. URL http://ideas.repec.org/p/nbr/nberwo/12275.html. Aschauer, D.A., 1989. Is public expenditure productive? J. Monet. Econ. 23, 177200. Barro, R.J., 1990. Government spending in a simple model of endogenous growth. J. Polit. Econ. 98, S103S126. Barro, R., Sala-i-Martin, X., 1995. Economic Growth. MIT Press. Bond, S., Leblebicioglu, A., Schiantarelli, F., 2010. Capital accumulation and growth: a new look at the empirical evidence. J. Appl. Econom. 25, 10731099. Bond, S.R., Hoeffler, A., Temple, J., 2001. GMM estimation of empirical growth models, CEPR Discussion Papers 3048, CEPR Discussion Papers. Brauninger, M., 2005. The budget deficit, public debt, and endogenous growth. J. Public Econ. Theory 7, 827840. Chen, B.-L., 2006. Economic growth with an optimal public spending composition. Oxf. Econ. Pap. 58, 123136. Conley, T.G., Ligon, E., 2002. Economic distance and cross-country spillovers. J. Econ. Growth 7, 157187. Devarajan, S., Swaroop, V., Heng-fu, Z., 1996. The composition of public expenditure and economic growth. J. Monet. Econ. 37, 313344. Durlauf, S.N., Johnson, P.A., Temple, J.R., 2005. Growth econometrics. In: Aghion, P., Durlauf, S. (Eds.), Handbook of Economic Growth, vol. 1. , Elsevier, pp. 555677. chapter 8. Easterly, W., Rebelo, S., 1993. Fiscal policy and economic growth: an empirical investigation. J. Monet. Econ. 32, 417458. Futagami, K., Morita, Y., Shibata, A., 1993. Dynamic analysis of an endogenous growth model with public capital. Scand. J. Econ. 95, 607625. URL http:// ideas.repec.org/a/bla/scandj/v95y1993i4p607-25.html.

V.K. Teles, C. Cesar Mussolini / European Economic Review 66 (2014) 1 15

15

Gemmell, N., Kneller, R., Sanz, I., 2011. The timing and persistence of fiscal policy impacts on growth: evidence from OECD countries. Econ. J. 121, F33F58. Glomm, G., Ravikumar, B., 1997. Productive government expenditures and long-run growth. J. Econ. Dyn. Control 21, 183204. Gregoriou, A., Ghosh, S., 2009. On the heterogeneous impact of public capital and current spending on growth across nations. Econ. Lett. 105, 3235. Gupta, S., Clements, B., Baldacci, E., Mulas-Granados, C., 2005. Fiscal policy, expenditure composition, and growth in low-income countries. J. Int. Money Finance 24, 441463. Jones, C.I., 1995. Time series tests of endogenous growth models. Q. J. Econ. 110, 495525. URL http://ideas.repec.org/a/tpr/qjecon/v110y1995i2p495-525.html. Kneller, R., Bleaney, M.F., Gemmell, N., 1999. Fiscal policy and growth: evidence from OECD countries. J. Public Econ. 74, 171190. URL http://ideas.repec. org/a/eee/pubeco/v74y1999i2p171-190.html. Lee, K., Pesaran, M.H., Smith, R., 1997. Growth and convergence in multi-country empirical stochastic Solow model. J. Appl. Econom. 12, 357392. Pesaran, M.H., 2006. Estimation and inference in large heterogeneous panels with a multifactor error structure. Econometrica 74, 9671012. Pesaran, M.H., Smith, R., 1995. Estimating long-run relationships from dynamic heterogeneous panels. J. Econom. 68, 79113. Romer, Paul, 1986. Increasing returns and long-run growth. J. Politic. Econ. 94 (5), 10021037. Saint-Paul, G., 1992. Fiscal policy in an endogenous growth model. Q. J. Econ. 107, 12431259.