Fiscal deficits, public debt, and government solvency: Evidence from OECD countries

JOURNAL

OF

THE

JAPANESE

AND

INTERNATlONAL

ECONOMIES

5,

354-380 (191)

Fiscal Deficits, Public Debt, and Government Evidence from OECD Countries

Solvency:

GIANCARLOCORSETTIAND NOURIEL ROUBINI*~ * Department

of Economics,

Yale University, New Haven, Connecticut and WEPR and NBER

06520;

Received January 9, 1991; revised April 1, 1991 Corsetti, Giancarlo, and Roubini, Nouriel- Fiscal Deficits, Public Debt, and Government Solvency: Evidence from OECD Countries This paper discusses different empirical tests of public sector solvency and applies them to a sample of 18 OECD countries. Under the maintained hypothesis that the government solvency constraint needs to be imposed, these tests develop from the idea of verifying whether the inter-temporal budget constraint of the public sector would be satisfied (a) if the fiscal and financial policy in the sample had been pursued indefinitely and (b) if the relevant macro and structural features of the economy were stable over time. If solvency is not supported by the empirical evidence, a change either in the policy or in the relevant macro and structura1 variables (growth, inflation, interest rates, demographic factors) must occur at some point in the future. Among the G-7 countries, public sector solvency seems a serious issue in Italy, whereas it does not appear to be a problem in Germany and Japan. The evidence for the United States is mixed. Problems of sustainability of the current path of fiscal policies are also present in Belgium, Ireland, The Netherlands, and Greece. .I. Japan. Znt. Econ., December 1991,5(4), pp. 354-380. Department of Economics, Yale University, New Haven, Connecticut 06520; and CEPR and NBER. o 1991 Academic Press. ~nc.

When constructing economic models, macroeconomic theorists generally require that fiscal and financial policies by the government satisfy an intertemporal budget constraint. An interesting question is then whether Paper prepared for the National Bureau of Economic Research and Tokyo Center of Economic Research Conference on Fiscal Policies in Open Macro Economies; Tokyo, January 7 and 8, 1991. We thank an anonymous referee, Willem Buiter, Kyoji Fukao, Lisa SchinelIer, and the participants in the Conference for helpful comments. Giancarlo Corsetti gratefully acknowledges financial support from the Consiglio Nazionale delle Ricerche.

0889- 158319 1 $3 .OO Copyright All rights

0 1991 by Academic Press, Inc. of reproduction in any form reserved.

354

SOLVENCY

AND

OECD

COUNTRIES

355

the empirical evidence indicates that this constraint is actually met in the experience of different countries. However, testing for government solvency is difficult because, per se, the solvency constraint imposes only mild restrictions on the behavior of the public sector. In practice, almost any short-run path of revenue and expenditure can satisfy it: large and persistent deficits today can always be offset with large surpluses at some time in the future. Nonetheless, while the solvency constraint does not rule out policies generating large (primary) deficits for prolonged periods of time, it does rule out the possibility that these policies be run forever. Cointegration tests then provide a useful tool with which to shed some light on the long-run implications of the fiscal and financial policy pursued by the government in a given time span. This is the fundamental insight of the literature on empirical tests for government solvency, starting with the contribution by Hamilton and Flavin (1986) and subsequently developed by Trehan and Walsh (1988)) Wilcox (1989)) Buiter and Pate1(1990)) Corsetti (1991), and Smith and Zin (1991). This paper discusses these different empirical tests for government solvency in a unified framework and applies them to a large sample of OECD countries, improving on the econometric methodology adopted in the previous literature. Under the maintained hypothesis that a government solvency constraint must be imposed on the economy, the class of tests under consideration stems from the following idea: verify whether the present value budget constraint of the public sector would be satisfied: (a) if the fiscal and financial policy in a given time period had been pursued indefinitely and (b) if the relevant features of the macroeconomic environment characterizing the sample period were stable over time. According to this approach, if solvency is not supported by empirical evidence, a change either in the policy or in the relevant macroeconomic variables (growth, inflation, interest rate, demographic factors) must occur at some point in the future. Note that these tests refer to thefeasibility rather than to the optima& of the fiscal and financial policy. Also, they give no information about the time in which the necessary changes should occur, l t Some contributions in the literature have approached solvency-related issues from a different vantage point. This literature argues that high levels of the public debt may be associated with a risk premium on government bonds, reflecting fears of repudiation. Sustainability of current policies can therefore be tested by focusing on the existence of such risk premia (see for instance Alesina ef al., 1989). The differences between such an approach and the one followed in this paper are evident. First, risk premia on government bonds may reflect much more than the violation of a long-run budget constraint: distributional and political considerations, for instance. In this case, we may observe risk premia on bonds regardless of government solvency. Second, unexpected temporary shocks within the sample may lead to rates of growth of the public debt which are not sustainable in the long run. If the market correctly perceives these rates of growth as temporary, insolvency of the public

CORSETTI AND ROUBINI

356

The plan of the paper is as follows. The first three sections discuss the logical and econometric structure of the solvency tests. Section 1develops the arithmetic of the public sector inter-temporal budget constraint and discusses its meaning and relevance in dynamically efficient economies. lt then relates solvency to traditional indicators of sustainability of public debt and states solvency-related implications for the pattern of fiscal deficit. Sections 2 and 3 discuss alternative testing frameworks and the statistical tests. The empirical evidence is presented in Section 4.

1.

THE PUBLIC SECTORBUDGETCONSTRAINT

This section has two main objectives. First, it will develop the arithmetic of the public sector’s budget constraint, with the goal of highlighting important accounting-related aspects of solvency. Second, it will discuss the conditions under which the intertemporal budget constraint for the public sector should or should not be imposed. The definition of the public sector includes central and local government, the central bank, public agencies, social security funds, and public enterprises. The consolidated public sector budget identity will then be B, - B,- 1 + M, - A4-1 + V,(& 4

pf

- B:- ,> + _ V,(F:: - F:-,) pt Pf

where B and B* are the stocks of domestic and foreign currency denominated public debt (nominal); M the stock of monetary base (nominal); F* the stock of foreign reserves (nominal); X the stock of public capital, evaluated at current reproduction costs; YX, public assets sold in the market at the price P,k, Cethe real consumption flow of the public sector within the period; 5 the real net current revenue; i and i* the domestic and foreign interest rates; V the foreign exchange rate, P the domestic price level; p the cash rate of return on public assets; and 6 the public capital depreciation rate. The left-hand side of (1.1) refers to stock adjustments, and the right-hand side to the current account. Note that the total net rate of return to capital supposedly consists of its cash rate net of sector stemming from the definition given above may not be reflected in the pricing of the public debt.

SOLVENCY

AND

OECD

357

COUNTRIES

depreciation (p - 6) plus a component in the form of either net tax revenue or reduction in expenditure. From an accounting point of view, the holding period rates i and i* are implicitly defined by the interest bill effectively paid during the period divided by the stock of net debt outstanding at the beginning of the period, in domestic and foreign currency, respectively. The available information refers almost exclusively to the netfinancial debt of the general government, defined as total interest-bearing liabilities minus financial assets owned by the sector. This measure is generally referred to as net debt. As far as solvency is concerned, however, the appropriate measure of the public sector net liabilities should also be net of the value of publicly owned real assets. The absence of data on the latter variable has important implications for the practical setup of the empirical tests. For instance, borrowing while investing changes net debt, but it may not change the debt corrected for the value of real assets, so that the evolution of the two series may be quite different over time. With a limited sample size, an investment project undertaken at some point in the sample may not recover its costs within the sample period. If public investment is not constant, this may lead either to an underestimation or to an overestimation of the government net liabilities. Let D, denote total financial liabilities less foreign reserves evaluated in domestic currency; that is, D, = B, + V,(@ - E;:). Let 9, = (M, - M,-,)/ P, (seigniorage) and .& = X, - (1 - S,- ,)X,- 1 - (P$YYC,)(net investment including capital gains). Expression (1. I) can be rewritten as

+ (1 t i,-,) +

I

+ (1 + i~-J~(B:-, t

- FFpl) (1.2)

or, more compactly, ii’1’9t-, %‘A*- yr+(I(I ++ nr-I)

= A, - Yp,+ (1 + &-,)S&,

(1.3)

where SD,= D,/P, and (1 + r,) = P,/P,- I ; At is the primary deficit; and if?!, is the nominal implicit interest rate paid on the net liabilities outstanding at the beginning of the period, with St- 1denoting the corresponding rate on real terms. Expression (1.3) is the usual dynamic equation for debt. The stock of government liabilities in real terms will grow in the period if the total deficit net of seigniorage is positive. As long as the nominal implicit interest rate is nonnegative, (1.3) can always be solved forward recursively. This yields

358

CORSETTI

AND

ROUBINI

aa,= j=oj=o 5 fi (1 + &+j)-’ (-A,+,+; + %+~+il + lim n (1 + i+m

[t+j)-*

23f+l+i. (1.4)

j=O

Consider now the definition of solvency. The public sector is solvent when the present discounted value of future primary surpluses minus seigniorage revenue is at least equal to the value of the outstanding stock of net financial debt, Therefore, solvency implies that the last term of (1.4) be nonpositive: lim fi (1 + i-rm

4,+j)-’

~d,+,+i ~ 0.

(1.5)

j=O

In other words, the public sector cannot be a net debtor in present value terms, so that, ultimately, the stock of debt cannot grow at a rate higher than the interest rate on the debt. Note that the measurement unit is irrelevant with respect to this criterion. Jn (1.4) and (1.5) both the stock of debt and the interest rate are in real terms, but we could redo the accounting in nominal terms (in either domestic or foreign currency) or in ratios to GDP. The crucial point is that Ponzi schemes in the form of systematic financing of the interest bill with additional borrowing are ruled out. Negative values of the limit (1.5) will result in a sort of supersoZuency, since in the limit the public sector will be a net creditor. This circumstance cannot be ruled out a priori, but, if it occurs, some other sector in the economy must be violating its solvency constraint. On the basis of this argument, (1.5) may well be written with an equality sign. Notice the analogy between (1 S) and a no bubble condition in asset pricing, which inspired the pioneering work of Hamilton and Flavin (1986). The transversality condition in asset pricing requires that in the limit the present discounted value of the asset’s terminal price go to zero. However, this analogy may obscure an important difference between the two conditions. In the asset pricing case, we price the income stream accruing from an asset by using a term structure which is independent of this stream. By the arithmetic of government solvency, the discount factor and the debt outstanding are not independent of each other. For instance, a permanent increase in the interest rate i will obviously decrease the discount factor. However, for a given path of primary deficits, it will also increase the interest bill and therefore the debt series. When the assumption of perfect foresight is dropped, an expectation operator conditional on the information available at time t is generally

SOLVENCY

AND

OECD COUNTRIES

359

added to expressions (1.4) and (1.5). In this case, the condition for solvency becomes limE, lj (1 + tl+j)-’ ?81+l+j = 0. i-P= j=O

Il.6)

Condition (1.6) is certainly met if the public sector can be expected to be solvent in every state of nature at any possible date. However, this interpretation of (1.6) is rather strong. It is relatively easy to provide examples in which the expression is true only if the public sector is a net financial creditor, a situation which bypasses the solvency issue altogether. Nonetheless, abandoning this excessively restrictive view of (1.6) leads us to accept (realistically) that solvency in expectation is compatible with actual insolvency in some particular evolution of the world. Consider now the conditions under which the public sector faces a present value budget constraint. It is well known that if an economy had too much capital and a decumulation could improve welfare-which is the definition of dynamic inefficiency -a benevolent government would actually play a Ponzi game, so that (1.6) would not hold. So the question becomes, do OECD economies suaer from an overaccumulation of capital? In a model a la Diamond, the stationary state for a dynamically inefficient economy is characterized by a net marginal productivity of capital which is lower than the growth rate of output. If there were neither uncertainty nor taxes, the net marginal product of capital and the real interest rate would coincide; a simple test of dynamic efficiency in a Diamond model-like economy would consist of checking whether the real interest rate is above the growth rate. This simple test is certainly not appropriate in practice. Indeed, while all OECD countries are characterized by prolonged periods in which the expost real interest rate falls below the growth rate, empirical evidence has indicated that the mean return to capital has constantly outpaced growth (see Abel et al., 1989). In this paper, the dynamic efficiency of the economies in the sample will not be tested, but will be taken as a maintained hypothesis. A second point, raised by Blanchard and Weil (1990), is the possibility for the government to provide intergenerational insurance by issuing debt in the presence of incomplete markets. In this case, permanent rollover of the debt can be feasible and welfare-improving even when capital is not overaccumulated. Again, we rule out this circumstance as a maintained hypothesis. It is important to stress that while a real interest rate below the growth rate does not provide a test of dynamic efficiency, it does have important implications for the dynamics of the debt to GDP ratio, which is the

360

CORSETTI

AND

ROUBINI

common indicator of the sustainability of fiscal and monetary policies. It is useful to discuss the relationship between this indicator and the solvency constraint. Solvency is a weak criterion by which to evaluate the sustainability of the public debt. It is well known that (1.6) can be satisfied also when the debt to GDP ratio increases without bound. All we need for solvency is that the growth rate of the debt be lower than the interest rate. Whether the debt to GDP ratio increases or not depends on whether the real interest rate is higher than the GDP growth rate and the latter is outpaced by the rate of growth of the real debt. A strict solvency criterion, therefore, does not bound the debt to GDP ratio. In some circumstances, this ratio may continue to increase without violating the public sector intertemporal budget constraint, In these cases, public sector solvency can be logically justified by stressing three important elements implied by the analysis. First, when the debt to GDP ratio grows unboundedly, the interest bill will at some point be larger than the whole GDP. Solvency thus requires that the tax base consist of both GDP and the interest income. Second, taxes must be nondistortionary and there should be no relevant costs of collection and enforcement. In this case, the increasing tax burden has no effect on the availability and allocation of resources. Third, the analysis ignores both the distributive effects which may accompany increasing debt and the political acceptability of those effects. These and similar considerations support the common view that only finite values of the debt to GDP ratio are sustainable. Therefore, it could be useful to use a solvency condition different from (1.6) and to make a distinction between a strict and a practical criterion (Buiter and Patel, 1990). The first is based on the discounted debt, and the second on the debt to GDP ratio. Infinite values of this ratio must be ruled out in the limit. Note that the two criteria impose quite different sets of restrictions on the path of deficits. Intuitively, a solvency constraint simply says that if the debt outstanding is positive today, primary surpluses must be run at some point in the future; this is a necessary condition for the strict solvency criterion to be satisfied. However, when the growth rate is higher than the interest rate, primary surpluses are not per se sufficient to rule out an increasing debt to GDP ratio. If this is the case, the practical criterion is not satisfied. 2.

TESTING

FOR SOLVENCY

As shown in Section 1, the present value budget constraint implies that the series of discounted debt be zero in expectation. In a rational expectations framework, the representative agent would evaluate govern-

SOLVENCY

AND

OECD COUNTRIES

361

ment solvency by solving the model of the economy making efficient use of all the available information. In principle, if we had a Lucas critiqueproof structural model of the economy, describing both economical and political processes, we could closely approximate the agent’s view about the issue. Nevertheless, a time-series analysis setting leads to a more direct (but rather mechanical) way of building the solvency test by focusing on the series of the discounted debt. Using sample information, we will estimate its distribution under the maintained hypothesis that this is stable. We will then test whether the unconditional mean of the series is zero (or nonpositive). As a preliminary step, the series of net debt should be discounted back to some base period. The next step will be to test whether the data generating process (DGP) which describes the behavior of the series over the sample period is covariance stationary. If it is, one can test whether the unconditional mean in the process is zero (or nonpositive). Either a positive drift or a time trend will eventually imply insolvency. If nonstationarity of the process cannot be rejected, we must consider two cases, depending on whether or not deterministic components also belong in the DGP of the series. A process with a unit root but no drift or time trend is in principle compatible with both insolvency and supersolvency, depending on the value of the debt in the sample. However, when the unit root coexists with positive deterministic components in the DGP of the series, this provides some evidence against solvency. A conservative albeit arbitrary approach to the interpretation of the test results consists of considering the case of a nondeterministic but nonstationary process as inconclusive. We will reject solvency only when a positive drift or time trend exists in addition to a unit root. Since nonstationarity per se rules out (1.6) for a nonzero debt, other contributions to the literature consider it sufficient to rule out the hypothesis of solvency regardless of the presence of a deterministic components in the DGP (Wilcox, 1989; Buiter and Patel, 1990). However, as we see later in the paper, this approach magnifies the power-related problems in the actual implementation of the test.2 In fact, given a small sample size, available tests for nonstationarity will tend to accept the null hypothesis of a unit root too often. * Nonstationarity of the series can be related to roots strictly greater than one. In this case, nonstationarity will still lead to insolvency for a positive value of the variable. This is the case discussed in Buiter and Pate1 (1990) with respect to India. On the other hand, the tests we perform are carried out under the maintained assumption that nonstationarity will be related only to the presence of a root equal to one, as opposed to roots strictly larger than one. Empirically, the estimate of the coefficient for the autoregressive term in the test regression is very close to one for all countries.

362

CORSETTI

AND

ROUBINI

As far as the interpretation of the test is concerned, therefore, insolvency will follow from positive deterministic components in the DGP of the series of the discounted debt-but only in the absence of structural changes in the process at some point in the future. In other words, either the fiscal and financial policy or some relevant features of the economy must change at some point in time for the intertemporal budget constraint of the public sector to be satisfied.? It should be clear by now that the test is not aimed at forecasting Staatsbankrott. Ex pust solvency can always be enforced by resorting to a variety of policy options: fiscal and monetary reforms (which can be more or less politically acceptable); policy measures which violate outstanding contracts between the public sector and the other sectors of the economy, such as a forced conversion of the debt; and policy measures which do not formally repudicate government financial liabilities but may conflict with previous government commitments, such as high inflationary debt monetization or a capital levy. Moreover, the test sheds no light about when the necessary fiscal and financial policy corrections should happen. These changes may have already occurred within the sample period. Unfortunately, traditional tests of structural breaks in the data generating process can be carried out only for covariance-stationary series. If nonstationarity cannot be rejected, the stability of the DGP in the whole sample will be implicitly assumed as a maintained hypothesis. Mutatis mutandis, the above methodology can be applied to the debt to DGP ratio, in the spirit of the practical solvency criterion. This will be our second set of tests for solvency. A third set can be obtained by using data on the fiscal balances rather than on stock of liabilities. Hamihon and Flavin’s work on empirical analysis of solvency focuses on the discounted primary deficit. Let X, be the series of the discounted cum seigniorage primary surplus. Expressions (1S) and (1.6) can be combined as

3 As an important example of structural variables affecting the public sector solvency constraint, consider the effects of the population age structure on social security. Given the forecasted increase in the number of retired people per worker over the next decades in the industrialized world, we may expect an important reversal in the pattern of budget surpluses that social security funds currently exhibit in many OECD countries. This of course challenges the out-of-sample stability of the observed features of our economies. However, note that including the social security funds in the public sector accounting framework without correcting for the present value of their future liabilities should bias the test toward accepting the null hypothesis of solvency.

SOLVENCY

AND

OECD

COUNTRIES

363

It is important to stress that in this case a purely nondeterministic stationary process for the series will not be sufficient to ensure solvency. The present value of the infinite series of the cum seigniorage primary surpluses must also be equal to C!$. As in the case of discounted debt, positive deterministic components in the series rule out solvency. A nondeterministic, nonstationary process is in principle compatible with both outcomes, depending on the initial conditions. We treat the latter result as inconclusive. Note that the Hamilton and Flavin (1986) test could be equivalently carried out in terms of the undiscounted cum seigniorage total surplus. By definition, we can write

(2.2)

and focus on the series of the cum interest surplus without discounting. This is the approach taken by Trehan and Walsh (1988). We perform this third class of tests for the following reason. As we discussed in Section 1, an important issue in the empirical implementation of the Wilcox-type test is that the public sector liabilities are not net of the value of real assets owned by this sector. Suppose that the net return on the public sector capital is equal to the interest rate on public debt. Then, borrowing in order to finance investment will not affect the government net worth. In this case, the increase in the discounted net financial liabilities does not imply the need for an active change in the fiscal and financial strategy at some point in the future. The trend will be modified automatically as soon as the acquired capital becomes productive. In other words, if the investment projects guarantee a rate of return (in cash or more generally in terms of net reduction in the primary deficit) at least equal to the borrowing rate, budget deficits to support capital expenditures are irrelevant in terms of the present value budget constraint. In general, government net worth will change with capital gains on real assets and the value of natural resource property rights, but, in principle, subtracting the value of public sector’s physical assets from the net financial debt rules out the risk of misinterpreting trends in the discounted debt. Indeed, the entire analysis in Section 2 can be recast in terms of this measure of net government liabilities. Setting aside the issue of the lack of information, note that in this case we would face a new problem, namely, how to value public assets. These could be valued at reproduction cost or at market value, under the assumption that each asset will remain publicly owned or will be sold to the private sector. When possible, valuation under alternative assumptions could give upper and lower

364

CORSETTI

AND

ROUBINI

boundaries to the debt, boundaries which per se would convey valuable information about the fiscal stance of the government. In the absence of data on the value of the public sector’s real assets, the Hamilton and Flavin-Trehan and Walsh (HF-TW) approach can provide a simple framework for clarifying the issue. Instead of estimating the stock of the debt net of real assets, it is possible to obtain some measure of the increase in financial debt controlling for the change in the stock of publicly owned real assets, as approximated by the net public capital formation. Consider the series of the first differences in the real undiscounted debt and subtract seigniorage revenue from it. This will produce a measure of the seigniorage-adjusted, inflation-adjusted real deficit. Then, if we disregard capital gains, the change in the real value of publicly owned assetsis given by the expenditure for capital formation net of depreciation. Of course, the approximation error will depend crucially on the nature of the public real assets. With this caveat in mind, we will perform a HF-TW type test on our measure of the real deficit minus net investment. 3. ECONOMETRIC METHODOLOGY

The test concerns stationarity of the data generating process of the series as well as the presence of deterministic components in it. Stationarity with or without deterministic components is tested by using the Phillips-Pert-on approach (Pert-on and Phillips, 1987; Phillips, 1987; Perron, 1988). The regression models to be estimated are Yt = QfIYr-I + 4

(3.1)

Yt = I4 + IYZYI-I + 4

(3.2) (3.3)

where T is the sample size. The conditions imposed on the sequence {~f}y=~ for i = 1,2,3 are very general. A wide variety of data generating mechanisms for yI are permitted, including virtually any ARMA with a unit root. The advantage of the Phillips-Perron test is that, without loss of generality, the test statistics require only the estimation of a firstorder autoregressive model by OLS and a correction factor based on the structure of the residuals from the regression. Two classes of statistics are possible. The null hypothesis for the first class is H,, : (~l~= 1. This is tested by using either Z(CX,),which is based on the standardized and centered least-squares estimates of ai, or Z&), based on the t statistics of LYE.

SOLVENCY

AND

OECD COUNTRIES

365

The second class of statistics builds on the regression I; tests studied by Dickey and Fuller (1981) for the general class of error process u’. The null hypothesis for Z(#,) in model (3.2) is H,-,: (IY~= 1, p2 = 0). In model (3.3), two statistics are possible: Z(&) for the joint null H, : (CQ = 1, p = 0) within a maintained hypothesis which permits a nonzero drift, and Z(w,) for the joint null H, : (~11~ = 1, p = 0, p3 = 0). The sources of critical values are Fuller (1976:371-373) for Z(q) and Z(&), and Dickey and Fuller (1981:1062-1063) for the other statistics.4 If nonstationarity can be rejected, a deterministic trend and drift can be tested by fitting an appropriate linear ARMA model. When the variable shows a monotonic trend, a possible testing strategy is that proposed by Perron (1988). Starting with the statistics related to the most general model (3.3), if rejection of the null is possible, the series is stationary and deterministic components can be tested by using OLS estimates. If the null is not rejected, it is still possible that either a deterministic drift or a trend belongs in the DGP. These can be tested by using the Z(+,) and Z(4,) statistics. If, again, the null H,, : (a3 = 1, p = 0, p3 = 0) cannot be rejected, we can make use of the statistics for both model (3.2) and, for a mean zero process, model (3.1). Further testing for a nonzero drift is possible by using Z(&). When the variable does not show a monotonic trend, model (3.3) may not be the most general one. Perron’s strategy may not be appropriate and we will make use of the full range of available statistics. Note three important caveats. First, the Phillips-Perron tests are asymptotic, while sample information is limited to yearly observations starting from 1960. Second, the power of these tests is very low against an alternative hypothesis of stationarity with a root close to one. Third, these tests requires that the data generating process be stable over the sample period.

4.

EMPIRICAL

EVIDENCE

In the analysis to follow, we perform three alternative solvency tests: (1) tests on the discounted debt (Wilcox, 1989; Buiter and Patel, 1990; Corsetti, 1990; (2) tests on the debt to DGP ratio along the lines of the “practical soIvency” criterion discussed in Buiter and Pate1(1990); and (3) tests on the real inflation-adjusted and seigniorage-adjusted current 4 These statistics are influenced by the choice of a lag truncation number (A). In some cases, results will depend on the number of lags allowed. These cases are highlighted in the tables.

366

CORSETTI

AND

ROUBINI

balance of the general government (the fiscal balance inclusive of interest payments but net of capital formation minus depreciation). The sample consists of 18 OECD countries. Results for the alternative tests are shown in Tables I through III. These report the probability value at which each null hypothesis can be rejected. A brief scheme containing the corresponding null hypotheses is also reported at the end of each table. Tables IV and V show the debt to GDP ratio and the seigniorage-adjusted, inflation-adjusted deficit (overall and net of public investmet# in selected years. The measure of the deficit in Table V is obtained by first differencing the end of period stocks of net (gross) debt in real terms and by subtracting seigniorage revenue from the differenced series. Note that this definition of the real deficit has the advantage of purging the government interest bill of its inflation-related component. Thefirst set of tests is based on a measure of discounted debt. Available OECD data refer to the general government financial liabilities, including and excluding government financial assets. These series are defined as gross and net debt, respectively. Since measurement errors look more severe in the case of the net debt, the test will be carried out by using both series when, from a logical point of view, only the second series would be appropriate. Also, because published data do not consolidate the central bank, we have subtracted the stock of base money from the series. The available OECD debt series is evaluated at par (rather than market) value, so that our analysis will reflect at least an error of approximation. We obtain implicit interest rates by dividing the interest bill in a given year by the stock of outstanding debt at the beginning of that year. Two series of the implicit interest rates will refer to the gross and the net government debt. Of course, while these average effective interest rates are formally more appropriate than marginal rates associated with new issues of government bonds, their correct estimate requires data on interest payments that exactly match the measure of liabilities used in the calculation. Due to the unavoidable measurement errors in estimating the implicit rates, we also construct a discount factor based on the gross yield on long-term government bonds. All interest rates are before-tax, leading to an underestimation of the discount factor. The magnitude of this underestimation varies across countries. Table I presents results of the Wilcox-type test on the discounted debt. According to the discussion in Section 2, we consider as evidence for insolvency the presence of a deterministic positive trend or drift in the DGP of the series. Solvency will be associated either with a nondeterministic stationary process (strict solvency) or with negative deterministic components in the DGP of the series (supersolvency). In the case of nondetermin5 Net investment refers to genera1 government gross capital formation minus depreciation.

Ho:a = Ho: a! = I-I,,:cr = Ho : a = Ha:& = H,:cr =

cm

cm cm GO cm cm GO -30 -00 00

40 -30
cm ‘30
-cm

cm -cm cm -cm -=m cm cm -cm (90 -cm

cm

490

00 -cm 40 X90 -30 cm -30 -30 -30

-cm

-==m

cm


40

-+O

cm

<90 >99 -30 cm 39 XW -30 <90 -GO

<90

-30


2?9

-30

00

cm 40

cm

40

C90

(90


00

39

WJl)

cm

z(q

Regression model (3.2)

1 1 for Z(a)and Z(t,) 1 andp = Ofor+, 1 for Z(a)and Z(r,) 1 andp = 001 = ?)forZ(&) 1 anda = Oandw = OforZ(&)

cm 40 <90 cm <90 cm cm -30

Zb3

’ Results depend on the lag truncation number. * Results are not consistent across definitions of debt. ’ Only the general government gross debt is available.

Model 3.3

Australia’ AnstiiaC Denmark Finland Greece Ireland Netherlands’ Norway Spain Sweden Null hypotheses Model 3.1 Model 3.2

United Kingdom United States

Japan

Belgium Canada France Germany Italy

Regression model (3.1)

40 -cm -30 cm C90 -30 -cm -cm <90 -cm

490 cm -30 -cm -30 -cm -cm <90

a4

cm

-=%I cm cm cm
-30 -=?Xl 40 -a0 cm GO -00 e?o

at,. 1


-30 >95 GO 40 739 233 -30 40 <90

>99 -30 40 00 39 7396 -00 -cm

zw

Regression model (3.3)

TABLE I TESTRESULTSFOR THE SERIES OF DISCOUNTED GENERAL GOVERNMENT NET INTEREST-BEARING LIABILITIES

cm

<90 -30 -30 40 229 >99 -30 00 <90

GO” -930 -30 -a0 399 HO* -==90 cm

Z(43)

= = = = =

1 1 for Z(o) and Z(t,) 1 andp = Ofor& 1 for Z(o) and Z(t,) landp = O(cl = ?)forZ(&) land/3 = Oande = OforZ(&)

a Results depend on the lag truncation number. b Results arc not consistent across definitions of debt. c -6unent gross debt is available.

Model 3.3

&:a= H, : a &:a Ho : o Ho:o H,:a

cm 40 cm

40 -30 40 40 40 cm

Australiac AustriaC Denmark Finland Greece< Ireland NetherlandsC Norway Spain Sweden Null hypotheses Model 3.1 Model 3.2

<90

cm -30 cm ‘30 40 <90 -30

<90 40 1906 -30 40 -SO 39 -30

em 40 GO <90 <90 40 40 <90

Belgium Canada France Germany Italy Japan United Kingdom United States

‘30

cm cm cm -cm cm em cm cm

C97.5”

>97.5” 40


Z(L,)

model (3.2)

II GOVERNMENT

TABLE

model (3.1)

GENERAL Regression

OF (UNDISCOUNTED)

Regression

SERIES

z&l

FOR THE

Z(t,,)

RESULTS

aa,)

TEST

cm cm cm (90

>95 >99

>99 FXJ
cm

cm 00 cm cm cm
NET INTEREST-BEARING AS SHARE

.a,, 1

cm -30 cm (90

39 >99

cm

>95” XJ9 -cm

39 >97.56

cm

cm

<90 190

cm <90 -30” >99b

190 -cm

2496 <97.5b

-=cm cm cm

39 >95” SW

xw

w#J3)

OF GDP

>97.5 ,996 >95b 39

.w#J2~

Regression model (3.3)

LIABILITIES

2 E 5

8

2

$ :

k# CQ

HO:o = Ho : LY= &:a = H, : (Y = H,,:o = HO:ti =

NA cm cm mb cm


cm cm cm


>rna

cm

cm


cm cm >95”sb

MINUS

-cm mb cm

cm6

cm >rna cm cm NA

Regression model (3.2)

DEFICIT

1 1 for Z(cu)and Z(r,) 1 and& = Oford, 1 for Z(a) and Z&J 1 and/3 = O@ = ?)forZ(&) 1 andp = Oandp = OforZ(&)

30 >97.50 >90b MO” NA 40 cm <90b >97.56 cm

40 >90b <90b >90b <90b >W 40 30

’ Results depend on the lag truncation number. b Results are not consistent across definitions of debt. c Data are derived from general government gross debt.

Model 3.3

GO" SO' -30 SO” NA 40 <90 <90b >mb

AustraIiac Anati-iaC Denmark Finland Greece Ireland Netherlandsc Norway Spain Sweden Null hypotheses Model 3.1 Model 3.2

cm

40 -30 40 40 <90 <90b <90 MO’

Belgium Canada France Germany Italy Japan United Kingdom United States

Regression model (3.1)

(OVERALL

NET

<90 tm cm cm NA cm <90 cm6 XW cm

cm

cm cm >95b cm cm <90 cm

INVESTMENT)

cm (90 cm -cm NA crnb cm cm6 cm cm

TESTRESULTSFORTHE SERIESOF REAL SEIGNIORAGE-ADJWTED INFLATION-ADJUSTED DEFICIT

cm

NA 39 <90
cm

cm

cm

cm

cm cm cm cm NA c90a.b cm cm6
cm
cm

Regression model (3.3)

CURRENT

490


-cm cm6

-cm cm NA
cm (90

8 : 3 i!

B

5 0

370

CORSETTI

AND

TABLE DEBT

United States Germany Japan France UK Italy Canada Belgium Australia Austria Denmark Finland Greece Ireland Netherlan Norway Spain Sweden

ROUBINJ IV

TO GDP RATIO

(*lOO)

1965

1970

1975

1980

1985

1988

1989

26.53 - 23.54 - 14.11 1.68 85.38 8.17 10.09 45 .?8 NA NA NA NA NA NA NA

19.73 - 18.17 - 14.99 0.435 64.83 17.35 0.13 40.30 33.99 6.24 -9.24 -8.23 5.37 NA 42.67 -7.73 9.10 -31.24

17.14 -9.51 - 12.09 2.80 49.65 31.% -2.06 37.06 21.34 11.22 - 15.82 - 12.27 4.44 55.95 34.29 -8.51 -11.64 - 36.28

13.45 4.42 7.72 2.052 42.63 37.71 5.99 58.53 19.75 25.67 3.39 - 10.37 9.95 73.56 39.09 -I,14 -6,72 - 20.95

21.10 12.77 17.07 9.50 43.17 45.88 26.92 103.44 19.81 36.97 27.54 -4.93 39.41 107.50 61.92 -25.35 6.80 9.43

23.89 13.17 7.25 12.02 32.97 76.87 32.94 116.00 18.31 42.19 17.68 -6.50 50.00 122.85 68.17 -31.17 8.60 -0.88

24.06 11.50 3.24 12.47 28.63 NA 33.85 115.10 15.19 NA NA NA 57.88 115.80 NA NA NA NA

iii NA

Note. Debt figures are net of base money. Gross debt (rather than net debt1 figures have been used in the following cases: Australia, Austria, Greece, Ireland, and The Netherlands.

istic but nonstationary processes, the test results will be considered inconelusive. Results in Table I refer to the net interest-bearing debt (net liabilities minus base money) discounted with the interest rate obtained by dividing the passive interest bill by the gross debt. This discount factor is preferred over the alternative one based on the published series of net interest payments and net debt because these series seem to exacerbate problems of mismatched accounting criteria and definitions. Nevertheless, we have performed the tests on a variety of measures of discounted debt, which we occasionally refer to in the analysis that follows. We first discuss the empirical evidence for the G-7 countries and Belgium. For these countries, available OECD data go back to the early 1960s. We begin with the two cases in which the results of the test are against solvency: Italy and Belgium. The test provides strong evidence against the sustainability of current policies in Italy.6 While Z(t,,) does not reject the null hypothesis of a unit root, both Z($,) and Z(&) reject the joint null of a unit root and a zero time trend (and a zero drift for the second statistic). Since in the case of Italy the estimated time trend is positive (the discounted debt is clearly trended upward), these results reject solvency. In principle, it is possible that the necessary steps toward a correction 6 See also Corsetti (1991).

SOLVENCY

INFLATION-ADJUSTED,

United States Germany Japan France UK Italy Canada Belgium

Australia Austria Denmark Finland Greece Ireland Netherlands Norway Spain Sweden

371

OECD COUNTRIES

TABLE V SEIGNIORAGE-ADJUSTED DEFICIT (% OF GDP)

1965

1970

-1.60

-0.87 - 1.48 -2.70 - 1.71 -6.69 2.23 -3.09 -0.83

0.42 NA 1.68 -2.01 -4.43 2.18 - 1.74 -0.72

AND

1975 1.98 5.25 2.62 4.52 4.16 3.64 2.22 1.74

1980 -0.31 3.37 2.43 - 1.38 0.45 -0.92 0.28 8.80

1985 2.03 0.55 0.37 0.71 0.50 6.80 6.41 8.75

1987

1988

1.96 0.88 -4.48 -0.13 -2.36 5.01 2.76 7.23

0.63 0.39 -3.76 0.66 -3.93 3.95

1.06 NA

1965

1971

1975

1980

1985

1987

1988

NA NA NA NA NA NA NA NA NA NA

-3.10 - 1.78 -2.77 -2.68 - 1.19 NA -0.63 -0.86 - 2.46 -4.42

- 1.01 4.33 1.64 0.39 -0.39 6.67 -0.82 0.96 - 1.94 0.28

-2.20 1.25 4.92 -0.69 -3.56 2.00 2.98 - 9.24 0.31 6.31

0.58 0.43 -5.45 -0.71 8.17 5.75 5.13 - 10.70 3.69 2.65

0.88 3.92 -5.16 0.16 1.37 5.84 0.38 -0.32 - 2.87 -3.80

-2.23 2.84 - 1.38 -2.64 4.10 3.58 3.87 -1.79 -0.05 NA

Current deficit (deficit net of government investment)

United States Germany Japan France UK Italy Canada Belgium

Australia Austria

Denmark Finland Greece Ireland Netherlands Norway

Spain Sweden

1965

1970

1975

1980

1985

1987

1988

-3.23 -3.47 NA -5.23 - 7.53 -0.21 -4.66 -2.43

-2.00 - 5.40 -6.72 -4.41 - 10.35 - 0.45 -5.31 -3.69

1.40 2.16 - ~2.21 2.13 0.77 0.70 0.01 - -0.86

-0.58 0.61 -3.09 -2.88 -0.59 -3.86 -0.91 6.22

1.72 - 0.95 -3.76 - 0.58 -0.19 3.31 5.19 7.48

1.54 - 0.68 -8.93 -1.38 -2.91 1.78 1.81 6.18

0.21 -1.17 -8.30 -0.67 NA 0.80 0.09 NA

1965

1971

1975

NA NA NA NA NA NA NA NA NA NA

-4.88 -6.06 -6.99 -5.48 NA NA -4.85 - 4.84 - 5.02 - 8.96

-3.13 -0.12 -1.43 -2.53 NA 3.14 - 4.02 - 3.05 -4.14 -2.79

1985 - 2.62 -2.17 2.26 -3.01 NA -1.46 0.44 - 12.44 - 1.00 3.46

Note. The “overall” inflation-adjusted, seigniorage-adjusted period stock of net (gross) debt in real terms minus seigniorage. minus government net investment.

-0.34 -2.33 -6.93 - 2.85 NA 2.42 3.17 - 12.60 0.73 0.81

1988 0.29 1.28 -6.56 -2.06 NA 3.69 1.29 -3.04 NA -5.26

-2.62 0.39 -3.03 -4.64 NA NA 2.22 -4.75 NA NA

deficit is the first difference of the end of The “current” deficit is the overall deficit

372

CORSETTI

AND

ROUBINI

of the fiscal stance were already undertaken within the sample period, but the test failed to detect their effects on the series of the discounted debt. However, Italy never achieved primary surpluses in the 198Os,so that the series of discounted debt is monotonically increasing. The real overall deficit was around 4% of the GDP at the end of the sample and even the current fiscal balance (i.e., net of public investment) showed a deficit throughout the 1980s(see Table V). .,Belgian discounted debt follows a nonstationary, nontrended process; both the statistics Z(Q and Z(&) accept the corresponding null hypotheses. However, the statistic Z(+,) suggests the presence of a deterministic drift, which is positive in the estimation. This is evidence against solvency. This country actually showed some degree of fiscal adjustment in the second half of the 198Os,running primary surpluses. Nonetheless, the size of the overall and current real deficit was always above 5% of the GDP in the past decade. In the case of Japan, results vary across definitions of discounted debt. While there is evidence of nonstationarity for all alternative series, the presence of a positive deterministic trend is supported in the series of net debt but not in that of gross debt. A nonzero deterministic component is also rejected by the statistics corresponding to the regression model (3.2). Since the failure to reject a unit root is accompanied by ambiguous results concerning deterministic components in the series, we consider this case inconclusive. Indeed, the Japaneseprimmy deficit shows three patterns within the sample period: positive between 1974 and 1983, virtually in balance for a few years, and, eventually, sharply negative beginning in 1986. For the other G-7 countries (United States, Germany, France, UK, and Canada),7 the discounted debt appears to follow a driftless and trendless nonstationary process. In this case supersolvency and insolvency are equally likely. Evidence for these countries is also interpreted as inconclusive. Next, consider the remaining 10 countries. Test results for Austria, Greece, and Ireland provide evidence for the presence of deterministic components in the DGP of discounted debt series. In the case of Austria, the statistic Z(C#B,) supports the existence of a nonzero drift in addition to a unit root. A positive drift is quite apparent in the corresponding data. Evidence against solvency is even stronger in the case of Greece and Ireland, where also the statistic Z(&) rejects the null H, : (cy3= 1, p = 0, ,X = ?). This suggests the presence of a nonzero time trend in the DGP of the series, which is positive in the estimation. The case of The Netherlands is interesting from the point of view of ’ For a detailed analysis of the Canadian case, see also Smith and Zin (191).

SOLVENCY

AND

OECD COUNTRIES

373

assessing the behavior of the test in the presence of within-sample structural changes. The statistics in Table I suggest the presence of a nonzero deterministic drift if we exclude the last 2 years in the sample (1987 and 1988), when the government started a program of fiscal retrenchment. However, the test statistics support a nondeterministic, nonstationary process when the full sample span is considered. For five countries (Norway, Finland, Denmark, Spain, Sweden, and Australia), discounted debt appears to follow a driftless and trendless nonstationary process, providing no clear evidence against or in favor of solvency. However, as Table V shows, four of these countries (Norway, Finland, Sweden, and Spain) are net financial creditors throughout most of the sample, while a fifth one (Australia) shows a declining pattern in the discounted debt during the second half of the sample. The sustainability of public debt does not appear to be an issue in these cases. We now move to the practical criterion for solvency and analyze the stochastic properties of the debt to GDP ratio. The results for the series of net interest-bearing liabilities (net debt minus base money) are presented in Table II. We consider first the G-7 countries and Belgium. Consistent with the results referring to discounted debt, there is clear evidence against solvency for Italy and Belgium. While both series appear to be nonstationary, the statistic Z(&) rejects the joint null hypothesis of a zero drift, a zero trend, and a unit root. In the case of Belgium, the presence of deterministic components in the series is detected also by Z(&), which suggests the presence of a time trend. The sharp increase in the Canadian debt to GDP ratio in the 1980smay help explain why the test statistics suggest the presence of a deterministic drift in the series (both Z(&) and Z(#,) reject the corresponding null). This violates the practical solvency criterion, but we have seen that results based on the discounted debt are inconclusive in the case of this country. This is not a contradiction; even if Canada were strictly solvent, in principle this could be compatible with an ever increasing debt to GDP ratio. Note, though, that for Canada this variable tends to stabilize slightly above 30% in the second half of the 1980s(see Table IV), possibly suggesting the presence of some corrective fiscal measures that the statistical test fails to detect. Test results are ambiguous for Germany and the United States, Using the regression model (3.3), there is some evidence supporting the presence of deterministic components in the DGP of the series, but only for net debt. The gross debt to GDP ratio appears to follow a nondeterministic, nonstationary process. Moreover, for the United States, a deterministic drift seems to be ruled out when we use statistics corresponding to the regression model (3.2). A conservative interpretation of these results suggests that this evidence is inconclusive.

374

CORSETTI

AND

ROUBINI

In the case of the UK, the statistic Z(+,) supports the presence of a deterministic component only in the nef debt to GDP ratio. However, note that, in this country, the debt to GDP ratio monotonically decreased from above 100% in 1960 to below 30% in 1989. The hypothesis of solvency based on the practical criterion should not be rejected. Finally, both France and Japan are characterized by a pure nonstationary process for the debt to GDP ratio: test results are therefore inconclusive. As far as the other countries in the sample are concerned, strong evidence against solvency is found for Austria, The Netherlands, Ireland, and Greece. A positive drift appears in the DGP of the series for these countries. In the case of Ireland, there is also evidence of a positive time trend. The case of Australia is more complex. The set of statistics referring to the regression model (3.3) suggests that the debt to GDP ratio is a nonstationary process with adeterministic drift. However, nonstationarity tends to be rejected by the statistics referring to regression models (3.1) and (3.2). Unambiguous conclusions cannot be drawn. For the remaining five countries (Sweden, Finland, Denmark, Norway, and Spain), a pure nonstationary process for the series makes the evidence inconclusive. However, as pointed out in the discussion of the results for the discounted debt, it is important to note that four of these countries were net financial creditors throughout most of the period. The third type of tests focus on the government budget deficit rather than on the stock of debt. Wilcox (1989) and Buiter and Pate1 (1990) convincingly argue that the only testable implication of the government intertemporal budget constraint is (1.6), so that tests should be carried out on the series of discounted debt. In fact, if we consider budget deficits, a pure nondeterministic stationary DGP of this series will not be sufficient for solvency. The infinite sum of the series should also be at most equal to the negative of the outstanding debt. The reason that we follow the Hamilton and Flavin-Trehan and Walsh approach is because, to some extent, it allows us to control for the public sector capital formation. The crucial point is that in principle the return on government investment projects may repay for their costs, without requiring increases in tax revenue. The benchmark case is one in which the rate of return on capital is equal to the interest rate on government debt. In this case, solvency should be assessedon a measure of debt net of government real assets. Taking the first differences of this series, we now focus on the real deficit net of public investment. Note that this approach implies the maintained assumption that capital gains on real assets are irrelevant. In addition, we again stress that the economic content of the government investment expenditure varies across countries. These considerations should be kept in mind when assessing the results that follow.

SOLVENCY

AND

OECD COUNTRIES

375

We performed tests on the real inflation-adjusted seigniorage-adjusted government deficit excluding net investment (the current fiscal balance). As in the case of the stock of debt, insolvency will follow from positive deterministic components in the series. Nevertheless, tests results are now inconclusive in all other cases. Table III presents the test results for the real fiscal balance net of public investment investmint computed by using the series of net debt. In no case were we able to provide clear evidence against solvency. In the group including the G-7 countries and Belgium, all countries but France and Canada are characterized by nondeterministic nonstationary processes for the adjusted current balance. Although this result does not allow us to draw any conclusions, it could be informative for analyzing the data in greater detail. Consider first the case of the United States. The U.S. data in Table V show a string of current fiscal surpluses in the period from 1960to 1981, followed by a string of current fiscal deficits up to the present. Yet, the period after 1985 is characterized by a phase of fiscal retrenchment with falling current deficit to GDP ratio. The inflationadjusted, seigniorage-adjusted budget is almost balanced by 1989.8While the continuation of the process of fiscal adjustment in the United States might be important for other macroeconomic reasons, the stabilization of the net debt to GDP ratio to a value around 24% since 1987could be an indication that the public sector solvency might not be an issue for the United States. However, there are a number of important caveats. First, our data refer to the consolidated general government; this includes state and local fiscal balances. It should be noted that for the U.S. general government a large fiscal deficit by the federal government is dampened by the fiscal surpluses of state and local authorities. Second, the transfer to off-budget accounts of many federal liabilities (such as the liabilities associated with the savings and loans bail-out) might understate the fiscal problems faced by the United States government. Third, the results do not account fully for future liabilities of the social security system. In conclusion, a conservative assessment of the test results and the other evidence suggests that the no unambiguous conclusions can be drawn for the solvency of the public sector in the United States. In the case of Japan, the current real fiscal balance has been in surplus in all the years in the sample. Moreover, even without considering the value of the public capital stock, the Japanesegeneral government is a net creditor by 1989. These elements support the view that public sector solvency is not an issue in the Japanesecase. On the other hand, as in the 8 It should be kept in mind that the data in Table V differ from the published conventional financial budget balance because the latter measure is not corrected for se&n&age revenue and includes interest payments which are not purged of their inflation component.

376

CORSETTI

AND

ROUBlNI

U.S. case, we do not consider fully in our test the future liabilities of the social security system deriving from demographic changes in the production age structure.9 For Germany, the data in Table IV for the current fisca1balance show surpluses in most years in the sample, with the exceptions of 1975, 1980, and 1981. Moreover, although the German government became a net financial debtor in 1978, the debt to GDP ratio has stabilized around lo-13% since 1982(see Table IV). While the test results and this additional evidence would suggest that public sector solvency is not an issue in the German case, it should be observed that our analysis has been based on past trends and is not able to capture the potential implications of the recent German unification for the future fiscal balances of the United German state. In the United Kingdom, the net debt to GDP ratio systematically fell from 116% in 1960 to 28.6% in 1989. The current fiscal balance was in surplus in all years between 1960 and 1989, with the exclusion of 1975, 1983, and 1984. The overall evidence is therefore quite consistent with public sector solvency in the UK. Both Italy and Belgium show large and often increasing current real fiscal deficits. Belgium has shown continuous current deficits since 1977. In these two countries, despite the high public investment to GDP ratio, even the current balance of the general government was negative the 1980s (see Table V). Given these large and persistent current fiscal deficits and a debt to GDP ratio around 100% in both countries, fiscal insolvency is very likely in the absence of a major shift in fiscal policies. The case of Canada is more complex: the current fiscal balances of Canada show surpluses in the years from 1960to 1981and current deficits since then. However, the large current deficits of the early 1980s have been significantly reduced and Canada reached a virtual current balance in 1989. The stabilization of the net debt to GDP ratio to a value around 33% since 1986is another signal of stable fiscal conditions. In the case of France, some statistics provide evidence in favor of the stationarity of the processes followed by the adjusted current balance. Moreover, Tables IV and V show that France had current fiscal surpluses in all but 4 years in the last three decades, plus stable and low levels of the debt to GDP ratio in the 1980s(around lo-12%). The overall evidence would therefore suggest that the French public sector is solvent. Among the smaller OECD countries, Norway, Finland, and Australia do not appear to need fiscal retrenchments. While Finland alone shows statistical evidence of stationarity, these countries are characterized by g See Iwata (1990) for a fine analysis of the fiscal policy implications structure in Japan.

of the changing age

SOLVENCY

AND

OECD COUNTRIES

377

current surpluses in all the years for which data are available (Australia shows minor current deficits in 1983, 1984, and 1987). Moreover, both Norway and Finland have been net creditors throughout the sample period, In the case of Australia only data on gross debt are available: its ratio to GDP has moved in a low and narrow range of 15-20% in the past decade. This, and the systematic evidence of current surpluses, suggests that solvency might not be an issue in the Australian case. The tests for Sweden, Austria, and Denmark fail to reject the unit root hypothesis in all cases. Sweden shows current surpluses or virtual balances in all years but 1980through 1983.Moreover, this country was a net creditor throughout the 1970s and became a net creditor again in 1988. In the case of Austria, we observe current fiscal surpluses in most of the years in the sample; the observed increase in the debt to GDP ratio (from 6.2% in 1970 to 42.1% in 1988) can be partially explained by the divergence beween current and overall fiscal balances. While the current fiscal balance is mostly in surplus, the overall (capital expenditures inclusive) fiscal balance is in deficit throughout most of the period. The assessment of general government solvency in Austria, therefore, could depend on the nature and characteristics of the stock of public capital. In the case of Denmark, years of severe fiscal imbalances in the 1979 to 1984 period have been followed by a serious fiscal retrenchment and surpluses in current (overall) fiscal balances since 1984 (1985). After peaking to 34% of GDP, the net debt to GDP ratio rapidly fell to 17.6% of GDP in 1988. The results for Spain do provide some evidence of stationarity. The data for Spain also show current fiscal surpluses for all years but 1984and 1985, a net creditor position for the public sector until 1983and a small net debt (8.6% of GDP as of 1988) since then. The overall evidence for these smaller OECD countries is therefore consistent with the conclusion that public sector solvency might not be a problem for these countries.

6.

CONCLUSIONS

This paper addressed both methodological and empirical aspects of tests for public sector solvency and provided an application to a large sample of OECD countries in a unified framework. Among the theoretical issues that have been mentioned throughout the paper, we want to stress the following. Meeting the intertemporal budget constraint is a long-run concern. The tests used in the paper are aimed at assessing whether sample policies will eventually lead to a rejection of solvency, if pursued over the distant future. Therefore, this class of tests relies on a steady-state assumption which allows the researcher to estimate

378

CORSETTI

AND

ROWBINI

the distribution of the debt series by using sample-related information. This is a crucial point in the logical structure of these tests. For instance, the presence of a forecastable future increase in public sector liabilities, such as an increase in the amount of transfers through the social security system or a future increase in health-related public expenditure due to demographic changes, can be expected to bias the test toward accepting the null of solvency. If social security funds currently build up reserves to meet future liabilities, their surpluses will reduce the size of the withinsample public sector deficits. When we reject solvency (as in the case of a number of countries in the sample), the explicit consideration of future liabilities only strengthens our conclusion. On the other hand, if results are inconclusive, it is certainly possible that correcting the debt and deficit figures for these liabilities could lead to a definite rejection of solvency. Note that this argument calls for a development of a new class of tests which focus on structural breaks rather than assuming stability of the data generating process. The statistical tests of the long-term solvency of the general government for a sample of 18 OECD countries in the period 1960-1989 show a wide variety of patterns of fiscal policy. We detected wide differences across countries in the long-term sustainability of present trends of spending and taxation. Among the G-7 countries, a serious problem of solvency was found only for Italy. With a debt to GDP ratio near 100%and the presence of still large primary deficits, a major improvement in the fiscal balances of Italy is required to ensure solvency. In the case of the United States, formal statistical tests of solvency do not allow us to draw unambiguous conclusions. Although a close examination of the data suggested that solvency of the general government might not be an issue for the trnited States, it is important to stress that this country faces future public sector liabilities that might seriously affect its fiscal and financial policy. In particular, the expected increase in social security liabilities and health expenditures deriving from the aging of the population and the liabilities associated with the savings and loan bail-out should be better evaluated in assessing the fiscal stance of the federal government. Solvency does not appear to be an issue for the other two major OECD countries, Japan and Germany. The large Japanese fiscal deficits of the 1970swere followed by a strong fiscal retrenchment in the 1980s;by 1989, the general government in Japan appeared as a virtual net creditor. Once again, our analysis does not consider fully the potential future social security liabilities deriving from the aging of the Japan population. With a debt to GDP ratio stabilized around 10% since 1982, Germany does not appear to face serious solvency problems either. The main caveat in this case refers to the implications of the recent German unification for the

SOLVENCY AND OBCD COUNTRIES

379

future fiscal balances of the united German state. The overall analysis of the data also suggests that solvency is not an issue in France, the United Kingdom, and Canada. Among the smaller OECD countries, problems of sustainability of the present paths of fiscal policy appear to exist in Belgium, Ireland, The Netherlands, and Greece. These countries have in common a large debt to GDP ratio (above 100% in Belgium and Ireland); they differ, however, in that two of them, Ireland and Belgium, started in the mid- 1980sa process of fiscal adjustment that has led to primary surpluses and, in the case of Ireland, to a minor reduction in the debt to GDP ratio. Conversely, primary deficits still persist in the other two countries (their size is larger in Greece than in The Netherlands); this is clearly inconsistent with long-term solvency . Finally, solvency does not appear to be a problem for the rest of the OECD countries in the sample. They are all characterized by primary fiscal surpluses plus low and falling debt to GDP ratios. Finland, Norway, and Sweden, in particular, appear to be net creditors.

DATA APPENDIX

The data used in the tests refer to the general government. The general government data consolidate the accounts of the central government, the social security agencies, and the state and local authorities. The sources of the data are as follows: OECD National Income Accounts for the General Government and GDP statistics; IMF IFS for base money and long-term government bond yields; OECD Economic Outlook and OECD (unpublished) for General Government debt data. All data are yearly. The largest sample span is 1960-1989 for the G-7 countries and Belgium; for the other OECD countries data on public debt are available only from 1970 on.

REFERENCES ABEL, A.B.,

MANKIW,N.G. SUMMERS, L. H., AND ZECKHAUSER, R.J.(1989). Assessing dynamic efficiency: Theory and evidence, Rev. Econ. Stud. 54, l-20. ALESINA, A., F'RATI, A., AND TABELLINI, G. (1989). “Public Confidence and Debt Management: A Model and a Case Study of Italy,” NBER W.P. 3135. BLANCHARD, 0. J., AND WEIL, P. (1990). “Dynamic Efficiency and Debt Ponzi Games under Uncertainty,” Unpublished, November. BUITER, W. H. (199O).“Principles of Budgetary and Fiscal Policy,” MIT Press, Cambridge, MA.

380

CORSETTI

AND

ROUBINI

BUITER, W. H., AND PATEL, U. R, (1990). “Debt, Deficits and Inflation: An Application to the Public Finances of India,” NBER W.P. 3287. CORSETTL G. (1991). Testing for solvency of the public sector: An application to Italy,” Econ. Notes, in press. DICKEY, D. A., FULLER, W. A. 11981). Likelihood ratio statistics for autoregressive time series with a unit root, Econometrica 49, 1057-1072. HAMILTON, J. D., AND FLAVIN, M. A. (1986). On the limitations of government borrowing: A framework for empirical testing, Empirical Econ. Reu. 76, 808-819. IWATA, K. (1990). “Budget, Aging, and External Imbalance,” presented at the Conference on Fiscal Policies in Open Macro Economies, Tokyo, Japan, January 7-8, 1991. PERRON, P. (1988). Trend and random walks in macroeconomic time series, J. Econ. Dynam. Control 12, 297-332. PERRON, P. AND PHILLIPS, P. C. B. (1987). Does GNP has a Unit Root?, Econ. Lett. 23, 139-145. PHILLIPS, P. C. B. (1987). Time series regression with a unit root, Econometrica 55,277-301. ROUBINI, N. AND SACHS, J. (1989). Government spending and budget deficits in the industrial countries, Econ. Pal. 8, 99-132. SMITH, G. W., AND ZIN, S. E. (1991). Persistent deficits and the market value of government debt, J. Appl. Economefrics 6, 31-44. TREHAN, B., AND WALSH, C, E. (1988). Common trends, the government budget constraint, and revenue smoothing, J. Econ. Dynam. Conrrol l2, 425-444. WERNER, M. (1990). “Is Mexico Solvent? Testing the Sustainability of the Government’s Fiscal Policy,” Yale University. WILCOX, D. W. (1989). The sustainabitity of government deficits: Implications of the present value borrowing constraint, JMoney, Credit, Banking 21, 291-306.