Journal Issue
Share
Article

Fiscal Deficits and Inflation

Author(s):
Luis Catão, and Marco Terrones
Published Date:
April 2003
Share
  • ShareShare
Show Summary Details

“A common criticism of this stress on the budget deficit is that the data rarely shows a strong positive association between the size of the budget deficit and the inflation rate.”

(Blanchard and Fischer, 1989, p. 513)

I. Introduction

A well-established theory in macroeconomics is that fiscally dominant governments running persistent deficits have sooner or later to finance those deficits with money creation (“seigniorage”), thus producing inflation (Sargent and Wallace, 1981). While this theory does not rule out the importance of other mechanisms through which inflation can be fueled and become persistent, fiscal imbalances have remained central to most models.2 The “fiscal view” of inflation has been especially prominent in the developing country literature, which has long recognized that less efficient tax collection, political instability, and more limited access to external borrowing tend to lower the relative cost of seigniorage and increase dependence on the inflation tax (Alesina and Drazen, 1991; Cukierman, Edwards, and Tabellini, 1992; Calvo and Végh, 1999).

Yet, as noted in the quote above, empirical work has had little success in uncovering a strong and statistically significant connection between the fiscal deficits and inflation across a broad range of countries and inflation rates. For instance, King and Plosser’s (1985) comprehensive analysis of the determinants of seigniorage in the United States and 12 other countries, using both single equation OLS regressions and VARs, indicates no generally significant causality running from fiscal deficits to changes in base money and inflation. In a more restricted sample of high inflation developing countries and using Granger-causality tests and variance decompositions in VARs, Montiel (1989) and Dornbusch, Sturzenegger, and Wolf (1990) find that fiscal deficits tend to accommodate rather than drive inflations—which instead they relate mainly to a combination of exchange rate shocks and inflationary inertia. Employing nonparametric correlation measures for 17 developing countries and dividing them into low- and high-inflation groups, de Haan and Zelhort (1990) find that seigniorage is weakly related to budget deficits except during very high inflation episodes. Click (1998) provides OLS estimates of the determinants of seigniorage in a cross section of 78 (mostly developing) countries and finds that fiscal variables play no significant role. More recently, Fisher, Sahay, and Végh (2002), using fixed effects in a panel of 94 developing and developed economies, conclude that fiscal deficits are main drivers of high inflations (defined in excess of 100 percent a year), and estimate that a 1 percentage point improvement (deterioration) in the ratio of the fiscal balance-to-GDP typically leads to a 4¼ percent decline (rise) in inflation, all else constant. However, they also find that changes in budget balances have no significant inflationary effects in low-inflation countries, or during low-inflation episodes in historically high-inflation countries. Finally, several cross-country studies on the determinants of inflation do not even include fiscal balances in their regressions, implicitly or explicitly assuming that fiscal balances play no role or that their effects are indirectly captured by other variables (Romer, 1993; Lane, 1995; Campillo and Miron, 1997; and Loungani and Swagel, 2001).

This paper takes a new look at this issue. Relative to previous studies, it uses a broad cross-country dataset and proposes a new approach to testing the theory which contains two main novelties. First, a simple intertemporal optimization model is used to show that equilibrium inflation is directly related to the fiscal deficit scaled by narrow money, where the latter stands for the size of the inflation tax base. We show that this distinction between the proposed specification and the standard practice of scaling deficit by GDP is not only theoretically appealing but also empirically relevant, since it introduces a key nonlinearity in the model—namely, it allows a given change in the deficit-to-GDP ratio to have a stronger impact in higher-inflation economies, where inflation tax bases are typically narrower. While previous work has acknowledged the existence of such a nonlinearity and tried to accommodate it through a semi-logarithm specification, the approach we propose is arguably less ad hoc and also empirically superior as shown later on.

Second, and also unlike previous studies, we model the deficit-inflation relationship as intrinsically dynamic, explicitly distinguishing between the short run and long run. Such a distinction is crucial, because fiscal deficits need not lead to higher money creation and inflation in the short run, as governments can temporarily finance their deficits with borrowing. Accordingly, econometric testing of the theory should ideally be capable of uncovering the relevant “equilibrium” or long-run parameters amidst a complex (and possibly noncausal) relationship between the two variables in the short run. As discussed in some detail below, this can be accomplished by specifying an autoregressive distributed lag (ARDL) model for each country, pooling them together in a panel, and then testing the cross-equation restriction of a common long-run relationship between the two variables using the “pooled mean group estimator” of Pesaran, Shin, and Smith (1999). This method, which explicitly models dynamics, is more germane to the spirit of the theory than the static fixed-effects estimator widely used in the literature, and its country-specific ARDL structure is capable of accommodating cross-country heterogeneity in inflation inertia.

The third main contribution of this paper lies in the use of very broad and up-to-date dataset, spanning 107 countries over 1960–2001 for a total of 3,607 observations. Such a panel is far more comprehensive than those found in previous studies, including Fisher, Sahay, and Végh (2002), which spans 94 countries for a maximum of 2,318 observations. Having a large panel allows us to slice the data into the various groups of interest without issues of sample representativeness or degrees of freedom becoming critical. Also, unlike the existing literature reviewed above, we consider both central and general government balance measures, and test the robustness of the deficit-inflation relationship to the inclusion of several conditioning variables.

The paper is organized as follows. Section II presents the theoretical model. Section III lays out the econometric methodology, while Section IV discusses the data and measurement issues. Estimation results are reported in Section V. Section VI concludes.

II. The Model

A central point of Sargent and Wallace (1981) is that the relationship between fiscal deficit and inflation is dynamic. Under fiscal dominance, deficits determine the present value of seigniorage but not necessarily current seigniorage. This is because borrowing allows governments to allocate seigniorage intertemporally, implying that fiscal deficits, seigniorage, and inflation need not be contemporaneously correlated. Moreover, because the short-run dynamics of the deficit-inflation relationship can be very complex (see, e.g., Dornbusch, Sturzenegger, and Wolf, 1990; Calvo and Végh, 1999), its direction and proximate magnitude are not amenable to theoretical predictions.

In contrast, the long-run relationship between the two variables is clearly spelled out by theory. This section shows how a parsimonious and testable specification can be simply derived from a small open economy version of the class of general equilibrium models surveyed by Ljungqvist and Sargent (2000). In this framework, money is assumed to play a role in determining macroeconomic equilibrium through a reduction in transactions costs (“shopping time”), enabling a fiscally dominant government to affect nominal money demand and inflation. The main features of this model economy and its steady-state equilibrium are as follows.

A. Households

The representative household maximizes the following lifetime utility function:

where β is the subjective discount factor (0 < β < 1) and where ct is period-t consumption, and ιt is period t-leisure. The current-period utility function, u(.,.), is assumed to be strictly increasing and strictly concave in its two arguments.

In each period, the household is endowed with a positive quantity of a good yt. Out of this endowment, the household pays taxes and can either consume or transfer the after-tax endowment over time through risk-free bond and money holdings. As result, the household is subject to a sequence of budget constraints given by:

where btp is the real value of the household holdings of one-period risk-free bonds that mature at the beginning of period t, these assets are denominated in period t consumption units; mt+1 denotes the household’s holdings of money balances between t and t+1; τt is a lump-sum tax at period t; pt is the price level; and Rt* is the international real gross rate of return on one-period bonds. The initial stocks of b0p and m0 are given and yt << ∞.

In each period t, the household has one unit of time which can be allocated to leisure, ιt, or shopping activities, st, so that ιt + st = 1. The amount of time spent on shopping is assumed to be directly related to the level of consumption, ct, and inversely related to the amount of real balances the household holds between t and t+1(mt+1/pt):

where S, Sc, Scc, Sm/p,m/p> 0 and Sm/p and Sc,m/p < 0. Because transaction costs are negatively related to money holdings, the return on money can be lower than the return in the risk free bond, as in the standard Baumol-Tobin money-demand function.

First order conditions with respect to ct, lt, bt+1, and mt+1 yield the following money demand function:

where Md is increasing on consumption (ct), and decreasing on the international real interest rate Rt* as well as on the domestic inflation rate πt=pt+1pt1.

B. Government

In each period t, the government spending gt is financed with tax collection, the issuance of one-period bonds, or by printing money. So, the respective budget constraint is given by:

where btg is the real value of the government’s net bond holdings denominated in consumption units of period t, and Mt is currency issued by the government at the beginning of the period t. Both b0g and M0 are given. Whenever btg<0, the government is a net borrower in period t.

C. Economy-Wide Budget Constraint and Stationary Equilibrium

With money supply equal to money demand (mt = Mt) and bt+1=bt+1p+bt+1g for all t, the economy wide budget constraint is thus:

where bt+1 is the net holdings of foreign bonds of the economy as a whole and b0 is given, so that the current account is defined as bt+1bt.

In the absence of trade restrictions and taxes, both purchasing power parity condition and the uncovered interest rate parity conditions hold, resulting in the equalization of onshore (Rt) and offshore real interest rates (Rt*). Stationary equilibrium in this small open economy then implies:

Substituting (7) into (5) yields:

which is the long-run relationship we shall examine in the remainder of the paper. It states that the rate of inflation is proportional to the ratio of gross-of-interest government deficit to the average stock of transaction or “narrow” money during the period; or equivalently, that inflation is proportional to the product of the ratio of gross-of-interest fiscal deficit to GDP by the inverse of the ratio of narrow money to GDP. With the demand for transaction money being negatively related to inflation, the size of the inflation tax base (M/GDP) will be lower (higher) as inflation is higher (lower). This implies that fiscal consolidation will be a more powerful instrument of price stabilization the higher the inflation rate.

III. Estimation Methodology

Allowing for generality and making use of the approximation π ≈ π/(1 + π), we consider the following empirical counterpart of equation (8):

where GTp(gτ+bg(R1)R) is the nominal equivalent of the real budget deficit concept underlying the theoretical model, and Ψ is the semi-elasticity parameter to be estimated. The empirical rationale for the nominal deficit approximation and other measurement issues are discussed in detail in Section 4.

To allow for rich dynamics in the way inflation adjusts to changes in the fiscal deficit or to any other variable, we nest equation (9) in an auto-regressive distributed lag (ARDL) structure where dependent and independent variables enter the right-hand side with lags of order p and q, respectively:

where πi,t stands for the observed inflation rate in group i at time t; μi represents fixed effects; and xi,t is a (k×1) vector of explanatory variables which includes the expression on the right-hand side of (9), i.e., xi,t=[(Gi,tTi,t)Mi,txi,t*], and x*i,tis a (k–1,1) vector which includes all other explanatory variables; λi,j are scalars and δi,l are (k×1) coefficient vectors. One well-known advantage of working with this ARDL specification, where all right-hand side variables enter the equation with a lag, is to mitigate any contemporaneous causation from the dependent to the independent variable(s) which might bias the estimates.3 This is an important consideration in the present context due to the presence of money on the right-hand side of (10) and the tight connection between money demand and inflation underlying the theoretical model.

Equation (10) can be re-parameterized and written in terms of a linear combination of variables in levels and first-differences:

where ϕi=(1j=1pλi,j),φi=j=0pδi,j,λi,j*=m=j+1pλi,m,δi,l*=m=l+1qδi,m, with j = 1,2,…, p − 1, and l = 1,2,…, q − 1. By grouping the variables in levels, this can be rewritten as:

where θi=ϕi1φi defines the long-run equilibrium relationship between the variables involved (i.e. Ψi, the coefficient on (Gi,tTi,t)Mi,t, is the first element of this vector) and ϕi the speed with which inflation adjusts toward its long-run equilibrium following a given change in xi,t.

The econometric literature suggests two approaches to consistent estimation of those parameters in dynamic panels with considerable heterogeneity across the distinct i’s and where T is large enough so that (11) can be estimated separately for each country. One is the so-called mean group (MG) estimator. It consists of estimating separate ARDL models for each country and derive θ and ϕ as simples averages of individual country coefficients θi and ϕi. This produces consistent estimates of the average of the parameters in heterogeneous panels provided that group specific parameters are independently distributed and the regressors are exogenous – Monte Carlo experiments confirming that this result holds generally, provided that N or T are not too small (Hsiao et al., 1999). However, it has also been shown that MG estimates will be inefficient if θi and ϕi are the same across groups, i.e., if the long-run slope homogeneity restriction holds (Pesaran, Shin, and Smith, 1999). In this case, Pesaran, Shin, and Smith (1999) propose a maximum likelihood-based “pooled mean group” (PMG) estimator which combines pooling and averaging of the individual regression coefficients in (11). This is shown to yield not only consistent but also considerably more efficient estimates that the MGE when the slope homogeneity restriction holds. By allowing the researcher to impose cross-sectionally long-run homogeneity restrictions of the form of θi = θ, ∀ i =1,2,..N, the PMG estimator also has the attractive feature of enabling one to test this restriction via standard Hausman-type tests.

Both the MG and PMG estimators have two key advantages over other estimators commonly used in the literature. Unlike the static fixed estimator, they allow for dynamics which is a well-known feature of inflationary processes. Relative to dynamic fixed effects (DFE) estimator, the MGE and the PMGE also have the advantage of allowing the short-run dynamic specification and error variances to differ across countries—a clear benefit since those variances may be quite different reflecting wide international disparities in historical inflation rates. Finally, the underlying ARDL structure dispenses with unit root pre-testing of the variables—a procedure which is marred by the low power of unit root tests and the controversy about their small sample properties in panels (O’Connell, 1998). Provided that there is a unique vector defining the long-run relationship among the variables involved, MG and PMG estimates of an ARDL specification such as in equation (11) yield consistent estimates of that vector—no matter whether the variables involved are I(1) or I(0)—once p and q are suitably chosen.4

IV. Data

The theoretical model of Section 2 indicates that the effects of budget deficits on inflation should vary across countries with significantly different inflation rates and levels of financial development, since both have a direct bearing on the size of the inflation tax base. So, sufficient heterogeneity in the country composition of the dataset is an important requirement for rigorous testing of the theory. Moreover, since the theory is mainly concerned with long-run equilibrium relationships and the proposed econometric methodology requires sufficiently long and uninterrupted time series, this is another important data requirement.

The dataset we have put together takes both requirements into account. It comprises all countries reported in the IMF’s International Financial Statistics (IFS) for which there exist no less than 20 years of continuous annual observations for the four variables featuring the theoretical model, i.e., inflation, the budget balance, GDP, and narrow money. Spanning 107 countries over the period 1960–2001, this dataset is the broadest and the most up-to-date we are aware of in the inflation literature.5Table 1 describes some of the main features of the data, reporting averages of the relevant ratios by country groups and decade-long sub periods.

Table 1.Selected Variable Averages, by Country Groups(Percent)
Inflation

(CPI)1
M1/GDPCentral

Gov. Balance/

GDP
General

Gov. Balance/

GDP
Openness2Oil Prices3
All countries
1961-19708.3818.98-2.1225.79-1.43
1971-198016.4017.93-3.68-4.1433.5346.31
1981-199076.3516.77-4.50-4.1535.09-1.73
1991-200148.8516.66-2.65-2.7038.993.18
Advanced countries
1961-19704.1727.03-1.2124.30-1.43
1971-198011.2824.35-3.42-2.0228.6846.31
1981-19908.3621.48-4.36-3.4732.00-1.73
1991-20012.8525.76-2.81-2.3134.153.18
Developing countries
1961-197010.1815.76-2.5826.39-1.43
1971-198017.8616.06-3.76-4.9934.9846.31
1981-199094.9715.48-4.54-4.3935.94-1.73
1991-200161.4514.17-2.60-2.8440.313.18
o/w: Emerging markets
1961-197020.4016.27-3.0623.33-1.43
1971-198028.5416.88-4.19-4.4729.0946.31
1981-1990124.0515.39-3.92-4.0831.42-1.73
1991-200136.8214.94-2.03-2.7336.313.18
Top 25 inflaters
1961-197021.5412.52-3.0822.58-1.43
1971-198033.6013.31-4.71-4.2323.7946.31
1981-1990276.3711.75-6.43-5.3122.98-1.73
1991-2001173.848.52-3.27-3.3529.123.18
Bottom 25 inflaters
1961-19702.8924.02-1.4133.96-1.43
1971-19807.9820.64-2.84-2.8243.5246.31
1981-19904.3619.52-3.28-2.6848.19-1.73
1991-20012.5522.27-2.02-1.6649.983.18
Sources: International Finance Statistics, IMF’s WEO and country desk databases, and Mitchell (1998a,b,c).

Average annual percent change.

½*(Exports plus Imports)/GDP.

Average annual percent change of the US dollar spot price.

Sources: International Finance Statistics, IMF’s WEO and country desk databases, and Mitchell (1998a,b,c).

Average annual percent change.

½*(Exports plus Imports)/GDP.

Average annual percent change of the US dollar spot price.

As in other studies, inflation is measured by the annual percent change in the consumer price index. The average money stock balance featuring on the right-hand side of (8) is the mean between the current year’s end-December stock and the preceding year’s end December stock of domestic M1. Since the latter is arguably the closest empirical equivalent for the transactions money concept in the theoretical model and is also a previously used measure of the inflation tax base (e.g., de Haan and Zelhorst, 1990; Rodrick, 1990; Metin, 1998), it seems preferable to other monetary aggregates.6

The main fiscal balance measure is the nominal deficit of the central government as reported in the IFS, i.e., including transfers and net interest payments and measured on a cash basis. One issue with this measure concerns its mapping to equation (8), where the term [gτ+bg(R1)R] strictly speaking measures changes in the real value of government debt. A well-known problem with the nominal deficit measure, as typically reported in the IFS and other statistical sources, is that it can be a misleading indicator of changes in real government debt during high and hyper inflations (regardless of whether nominal deficit is scaled by GDP or current money stock). This has led some practitioners to work with the “operational deficit” concept, defined as the primary balance plus real interest payments on the current debt stock (see, e.g., Tanzi, Blejer, and Teijero, 1993). However, this measure has not only the drawback of requiring certain assumptions and high-frequency data to be calculated with reasonable precision, but also is unavailable for most countries. Another potential criticism of the central government balance measure is that it fails to incorporate local governments, public enterprises, and central bank losses—entities deemed to play a considerable role in inflationary episodes in some countries, especially those where fiscal federalism prevails. In other words, broader deficit measures would be desirable. One problem is that sufficiently long series on public sector aggregates comprising local governments, public enterprises, and central bank losses are unavailable for all 107 countries. However, general government balances for a subset of 87 countries and spanning a shorter-time horizon (usually starting sometime in the mid to late 1970s) are available from the IMF’s World Economic Outlook and country desk databases. We use those series to test the robustness of the results, as described below.

V. Results

The second column of Table 2 reports MG and PMG estimates for the 107 country panel with the ARDL lag structure for each country being optimally chosen by the Schwartz Bayesian criterion (SBC).7 For the vast majority of countries (86 out of 107), specifications with no lagged dependent variables are rejected at conventional levels of statistical significance, indicating that dynamics is important and so that the static fixed-effects method is clearly inadequate to for the task at hand.8 As the null hypothesis throughout is that of no long-run relation between budget deficit and inflation, t-statistics yielded by both estimators reject the null at 5 or even 1 percent. The estimated error correction coefficient of just under 0.5 indicates that the adjustment of inflation to a given change in the fiscal balance has an average half-life of just over two years. Yet, the dramatic difference between the MG and the PMG estimates of the long-run elasticity parameter Ψ points to considerable sample heterogeneity. This is clear from the Hausman h-statistic of 5.6, which rejects the slope homogeneity restriction at 2 percent. Rejection of the homogeneity restriction implies that consistency of the PMG estimate is not warranted and so the MG estimate should be preferred.

Table 2.Dynamic Panel Estimates of Inflation on Central Government Deficit over M11/
A. By level of financial development
All CountriesAdvanced CountriesDeveloping Countries
MGPMGMGPMGAllEmerging Markets
MGPMGMGPMG
LR Elasticity (Ψ)1.43

(5.60)
0.02

(2.74)
1.69

(1.08)
-0.29

(-4.02)
1.40

(2.16)
0.02

(3.76)
2.26

(1.63)
0.38

(31.5)
EC Coefficient (Φ)-0.46

(-14.55)
-0.49

(-15.26)
-0.18

(-6.76)
-0.14

(-8.44)
-0.53

(-16.9)
-0.52

(-15.4)
-0.52

(-7.75)
-0.40

(-5.55)
h-statistic5.60

[0.02]
1.62

[0.20]
4.51

[0.03]
1.83

[0.18]
No. of Observations3607360788288227252725905905
B. By level of inflation
All CountriesTop 25Mid 50Bottom 25
MGPMGMGPMGMGPMGMGPMG
LR Elasticity (Ψ)1.43

(5.60)
0.02

(2.74)
4.46

(2.35)
0.40

(32.78)
0.18

(0.44)
0.03

(3.52)
0.89

(0.86)
0.00

(-0.42)
EC Coefficient (Φ)-0.46

(-14.55)
-0.49

(-15.26)
-0.49

(-7.54)
-0.43

(-5.95)
-0.53

(-11.27)
-0.52

(-11.38)
-0.53

(-11.58)
-0.52

(-11.56)
h-statistic5.60

[0.02]
4.57

[0.03]
0.13

[0.72]
0.75

[0.39]
No. of Observations3607360792492417651765918918

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

To gain some insight into the nature of this heterogeneity, we divide the panel into groups by level of financial development and inflation performance. Clearly, these two broad categories are not unrelated—more financially developed countries do typically display historically lower inflation rates—and not surprisingly about half of the countries we classify as low inflation economies are also classified as developed.9 However, since the overlapping between the two groups is not perfect (a few countries have both underdeveloped financial systems and stellar inflation performance), and given that other studies have considered low inflation countries a relevant sub-group in its own right, it seems important to consider those two ways of sub-dividing the panel. The respective MG and PMG estimates are reported in Table 2. They indicate that budget deficits are significant drivers of inflation in most groups, with the exceptions of low inflation economies and advanced countries for which one of the estimates actually yields the “wrong” sign.10 The estimated effect of changes of budget balance on inflation is very strong for developing countries in general, as the h-test indicates that the MG estimate of 1.40 should be preferred to the PMG estimate of 0.02. For an average M1/GDP ratio of 15.5 percent for these countries (See Table 1), this implies that a 1 percent reduction (increase) in the ratio of budget deficit to GDP lowers (raises) inflation by 8 ¾ percentage points on average, all else constant. For the more homogeneous emerging market group, the impact is less dramatic but still far from negligible. Given that the h-statistic cannot reject the cross-country slope homogeneity restriction and the PMG estimate of 0.38 ought to be preferred, a percentage point change in the ratio of budget balance to GDP is estimated to change inflation by some 2¼ percentage points for historical values of the M1/GDP ratio for this group of countries (see Table 1). Moreover, the t-ratio of 31.5 underlying the estimated coefficient indicates that this elasticity is very precisely calculated implying that the inferences for this group of countries are especially robust.

Breaking down by high vs. lower inflation groups, changes in the budget balance have a very strong effect in high inflation economies which, in our dataset, comprise several countries with average inflation rates above 100 percent during the 1980s and 1990s (see Table 1 and the Appendix). Far less strong but still statistically significant is the effect on “moderate” inflation countries. As the latter category comprises nine advanced countries out of 54 countries comprising the whole group, this raises the question of whether a significant relationship between budget deficits and inflation is also observed for that sub-group. Rerunning the separate regressions for this sub-group, we also cannot reject the existence of a positive relationship between deficits and inflation at 5 or 1 percent level of statistical significance.11 So, even among advanced countries with moderate levels of inflation one can still conclude that budget deficits matter for long-term inflation performance. Only for countries at the very bottom of the inflation distribution is there no evidence that Ψ is statistically significant.

This raises the question of why previous studies did not uncover such a significant positive effect of fiscal deficits on inflation across most country groups. Is it because of differences in econometric techniques, model specification, sampling, or a combination of all of the above? To shed light on this question, we re-estimate equation (9) using the static fixed effects estimator widely used in the literature.12 As discussed in Section 3, application of the method to cross-country inflation data is bound to be problematic not only due to its neglecting of inflation dynamics, but also because of the assumption of constant error variance across groups. The latter, in particular, is grossly violated in the present panel.13 Bearing these reservations in mind, the fixed effects estimates reported in panel A of Table 3 basically reinstate the broad inferences obtained with the MG and PMG estimators. The only exception is the developed country group for which the coefficient Ψ is now statistically significant and positive, albeit very small. As with the MG and PMG estimators, fiscal deficits do not seem to matter for inflation performance in economies with historically low inflation.

Table 3.Fixed-Effects Panel Regressions 1/
A. Inflation on Central Government Deficit/M1
All

Countries
Advanced

Countries
Developing CountriesBy inflation rates
AllEmerging

Markets
Top 25Mid 50Bottom 25
Slope1.35

(2.93)
0.05

(4.65)
1.41

(2.55)
1.51

(3.13)
1.71

(2.46)
0.03

(5.05)
0.01

(1.14)
R20.110.340.100.140.100.130.07
No. of Obs.:362388927349119251773925
B. Inflation on Central Government Deficit/GDP
All

Countries
Advanced

Countries
Developing CountriesBy inflation rates
AllEmerging

Markets
Top 25Mid 50Bottom 25
Slope19.10

(2.46)
0.37

(5.49)
21.97

(2.46)
12.73

(3.11)
45.68

(2.51)
0.37

(7.61)
0.01

(0.78)
R20.110.190.110.100.130.140.08
No. of Obs.362388927349119251773925
C. Log (1 + Inflation) on Central Government Deficit/GDP
All

Countries
Advanced

Countries
Developing CountriesBy inflation rates
AllEmerging

Markets
Top 25Mid 50Bottom 25
Slope1.83

(6.36)
0.33

(5.59)
2.06

(6.27)
2.95

(6.15)
4.05

(6.45)
0.33

(7.77)
0.03

(0.91)
R20.350.340.350.330.310.140.14
No. of Obs.:362388927349119251773925

Heteroscedasticity corrected t-ratios in parenthesis.

Heteroscedasticity corrected t-ratios in parenthesis.

Table 3 also reports fixed effects estimates of the more standard specification in which the budget deficit is scaled by GDP rather than by narrow money. They indicate that the inference that fiscal deficits generally matter can also be obtained with the more standard specification albeit with important quantitative differences relative to the results of Table 2. When inflation is not calculated by approximation log(1 + π) but simply as πt = 100(pt/pt-1-1) where pt is current CPI, the model does not allow for the inflation-budget deficit elasticity to change across inflation levels; accordingly, panel B estimates are very disparate for the different sub-panels. By taking such non-linearities into account and flattening outlier observations, the use of the log approximation in panel C yields more sensible estimates.14 Yet, the estimated magnitude of the fiscal effect is generally much lower than those using the (G-T)/M1 specification. For instance, a 1 percentage reduction (increase) in the deficit/GDP ratio is estimated to lower inflation by 1.83 percent—a figure about five times as low as that obtained with the MG estimate and the specification of equation (9). Likewise, the estimated semi-elasticity of 4.05 for the high inflation group, although virtually identical to that reported in Fischer, Sahay and Végh (2002) using a different dataset, is much lower than that previously using the dynamic panel methods and the (G-T)/M1 specification. Only when dynamic panel estimators are used instead of static fixed effects, can the combination of log approximation and the deficit/GDP specification yield significantly higher elasticities, as shown in Table 4. Yet, magnitude of the overall effect is still smaller than that yielded by the combination of the (G-T)/M1 specification and the dynamic panel estimators. This is especially the case regarding emerging market countries, for which the specification of Table 4 yields a semi-elasticity of inflation to the deficit-GDP ratio of 0.38, as opposed to 2¼ (=0.38/16.75) in Table 2.

Table 4.Dynamic Panel Estimates of Log (1+Inflation) on Central Government Deficit over GDP 1/
A. By level of financial development
All CountriesAdvanced CountriesDeveloping Countries
MGPMGMGPMGAllEmerging Markets
MGPMGMGPMG
LR Semi-Elasticity4.43

(2.15)
0.14

(3.36)
-1.15

(-1.54)
-1.28

(-4.50)
5.87

(2.25)
0.21

(4.60)
1.92

(0.26)
0.38

(2.90)
EC Coefficient (Φ)-0.45

(-14.67)
-0.43

(-14.52)
-0.18

(-5.50)
-0.14

(-9.76)
-0.52

(-15.10)
0.50

(-14.42)
-0.39

(-6.56)
-0.36

(-7.08)
h-statistic4.32

[0.04]
0.04

[0.84]
4.72

[0.03]
0.04

[0.84]
No. of Observations3612361288488427282725911911
B. By level of inflation
All CountriesTop 25Mid 50Bottom 25
MGPMGMGPMGMGPMGMGPMG
LR Semi Elasticity1.43

(5.60)
0.02

(2.74)
14.83

(2.08)
10.76

(12.67)
1.81

(1.06)
0.30

(4.93)
-0.11

(1.66)
-0.91

(-1.92)
EC Coefficient (Φ)-0.46

(-14.55)
-0.49

(-15.26)
-0.37

(-7.08)
-0.22

(-5.02)
-0.53

(-11.07)
-0.53

(-11.40)
-0.35

(-7.84)
-0.36

(-7.00)
h-statistic5.60

[0.02]
0.33

[0.57]
0.77

[0.38]
2.93

[0.09]
No. of Observations3604360492292217651765917917

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

It follows from these results that difficulties in uncovering a statistically significant and strong relationship between budget deficits and inflation stem from two main factors. One is the use of data samples with a disproportionately high weight on advanced countries and/or low inflation economies, since those would bias the evidence towards rejecting the theory. Second, from inadequate statistical methods such as the fixed effects estimator which, combined with specifications that do not account for differences in the size of the inflation tax base, impart a downward bias on the relevant cross-country estimates.

Before fully embracing the dynamic panel estimates of Table 2, however, it is important to test their robustness to the inclusion of other explanatory variables considered in the literature. One such a variable is oil price—a well-known source of inflationary pressures in the world economy (IMF, 2001; Loungani and Swagel, 2001; Hamilton and Herrera, 2002). As shown in Table 5, the statistical significance of oil price inflation as an explanatory variable varies somewhat across MG and PMG estimates and across country groups, but on the whole holds across much of the panel.15 Consistent with the findings of the above mentioned studies, the impact of oil prices on domestic long-term inflation is generally modest in developing countries but is quite strong among advanced countries, with a 1 percentage point increase in oil price inflation estimated to raise domestic inflation by some 0.2 percentage points. This role for oil price inflation in the regressions does not detract, however, from the strength of previous estimates on the fiscal deficit variable. With or without including oil price inflation, fiscal deficits continue to display a powerful effect on inflation in developing countries, emerging markets, and high-inflation economies and a much smaller effect amongst moderate inflation countries. Also as before, the estimated coefficient for advanced countries continues to yield an opposite sign as that predicted by theory, but the coefficient for low inflation countries now turns out to be statistically significant, albeit remaining very small.

Table 5.Dynamic Panel Estimates of Inflation on Central Government Deficit over M1 and Oil Price Inflation1/
A. By level of financial development
All CountriesAdvanced CountriesAll Developing CountriesEmerging Markets
MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1
LR Elasticities:
(G-T)/M11.14

(2.06)
0.03

(4.97)
4.05

[0.04]
-0.20

(-2.29)
0.07

(7.20)
10.1

[0.00]
1.51

(2.15)
0.03

(4.78)
4.46

[0.03]
2.47

(1.57)
0.38

(30.67)
1.75

[0.19]
Oil Price Inflation-0.02

(-0.05)
0.07

(20.18)
0.08

[0.78]
0.21

(3.25)
0.17

(15.99)
0.4

[0.53]
-0.08

(-0.19)
0.05

(13.33)
0.1

[0.75]
-0.27

(-0.32)
0.08

(6.46)
0.18

[0.67]
EC Coefficient (Φ)-0.50

(-15.39)
-0.47

(-15.88)
-0.29

(-7.22)
-0.24

(-5.65)
-0.56

(-14.8)
-0.53

(-14.25)
-0.39

(-5.94)
-0.41

(-6.68)
No. of Observations3607360784884827252725905905
B. By level of inflation
All CountriesAdvanced CountriesAll Developing CountriesEmerging Markets
MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1
LR Elasticities:
(G-T)/M11.14

(2.06)
0.03

(4.97)
4.05

[0.04]
4.83

(2.34)
0.40

(33.30)
4.61

[0.03]
0.11

(1.05)
0.06

(7.79)
2.6

[0.11]
0.07

(2.02)
0.03

(0.38)
4.56

[0.03]
Oil Price Inflation-0.02

(-0.05)
0.07

(20.18)
0.08

[0.78]
-0.35

(-0.27)
0.05

(2.09)
0.09

[0.76]
0.11

(3.26)
0.08

(14.07)
1.28

[0.26]
0.062

(4.59)
0.078

(14.54)
1.58

[0.21]
EC Coefficient (Φ)-0.50

(-15.39)
-0.47

(-15.88)
-0.45

(-7.75)
-0.40

(-5.96)
-0.56

(-10.9)
-0.56

(-12.09)
-0.41

(-8.99)
-0.44

(-8.84)
No. of Observations3607360792492417651765918918

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

The significance of the deficit-inflation relationship in developing and high inflation countries also appears to be robust to the potential omission of two other explanatory variables. One is openness to foreign trade. As argued in Romer (1993) and Lane (1995), the benefits of an expansionary monetary policy tend to be smaller in an economy with a larger share of trade in GDP because: (i) the weight of the home goods sector will be smaller implying that the impact of monetary expansion on domestic employment will be reduced; and (ii) the currency depreciation resulting from the monetary expansion will raise domestic inflation by more than in a closed economy. Hence, the more open the economy the less time-inconsistent the monetary policy, implying a negative relationship between openness and inflation, once other factors are held constant.16

Yet, Table 6 results lend limited support to the view. When openness (measured as the ratio of exports plus imports to GDP) enters the regression, MG estimates for the full panel yield the expected negative coefficient but this is statistically insignificant at 5 or 10 percent levels, whereas the PMG estimator yields a coefficient with the opposite sign as that predicated by theory. Only for the developed country group does the openness variable yield a statistically significant coefficient with the predicted sign, but respective effect is relatively small – a one percentage point increase in openness leading to 0.09 percentage point drop in the inflation rate. On the one hand, this suggests that the Romer (1993) and Lane (1995) results are sample specific, as argued in Terra (1998) and Bleaney (1999). On the other hand, Table 6 results indicate that fiscal deficit-inflation relationship is robust to the inclusion of openness among the regressors.

Table 6.Dynamic Panel Estimates of Inflation on Central Government Deficit over M1, Oil Price Inflation, and Openness
A. By Level of Financial Development
All CountriesAdvanced CountriesDeveloping CountriesEmerging Markets
MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1
LR Elasticities:
(G-T)/M10.96

(2.55)
0.02

(4.90)
6.21

[0.01]
-0.08

(-1.12)
0.02

(1.30)
2.04

[0.15]
1.24

(2.62)
0.03

(6.64)
6.54

[0.01]
1.99

(1.96)
0.07

(4.77)
3.58

[0.06]
Oil Price Inflation-2.00

(-0.95)
0.07

(19.74)
0.97

[0.32]
0.13

(6.61)
0.17

(13.01)
10.74

[0.00]
-2.60

(-0.97)
0.05

(13.49)
0.97

[0.32]
0.40

(2.36)
0.05

(8.22)
4.30

[0.04]
Openness-10.90

(-1.07)
0.01

(1.25)
1.15

[0.28]
0.06

(0.19)
-0.09

(-2.12)
0.22

[0.64]
-13.89

(-1.07)
0.02

(2.31)
1.15

[0.28]
3.29

(0.39)
0.03

(1.51)
0.15

[0.70]
EC Coefficient (Φ)-0.56

(-18.39)
-0.51

(-16.67)
-0.29

(-7.17)
-0.20

(-8.12)
-0.64

(-19.24)
-0.58

(-16.41)
-0.60

(-9.45)
-0.54

(-8.63)
No. of Observations3581358188288226992699899899
B. By Level of Inflation
All CountriesTop 25Mid 50Bottom 25
MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1
LR Elasticities:
(G-T)/M10.96

(2.55)
0.02

(4.90)
6.21

[0.01]
3.68

(2.68)
0.36

(21.19)
5.84

[0.02]
0.09

(2.01)
0.03

(4.53)
1.65

[0.20]
-0.80

(-2.15)
0.01

(0.11)
4.85

[0.03]
Oil Price Inflation-2.00

(-0.95)
0.07

(19.74)
0.97

[0.32]
-8.11

(-0.98)
0.03

(1.53)
0.96

[0.33]
0.07

(4..9S)
0.06

(12.53)
0.89

[0.35]
0.49

(2.51)
0.08

(14.39)
1.88

[0.17]
Openness-10.90

(-1.07)
0.01

(1.25)
1.15

[0.28]
-43.07

(-1.07)
0.23

(1.64)
1.16

[0.28]
-0.02

(-0.20)
0.09

(3.88)
0.94

[0.33]
-0.06

(-0.48)
-0.01

(-0.94)
0.15

[0.70]
EC Coefficient (Φ)-0.56

(-18.39)
-0.51

(-16.67)
-0.56

(-10.31)
-0.49

(-7.72
-0.6

(-12.72)
-0.56

(-12.15)
-0.47

(-9.17)
-0.41

(-9.04)
No. of Observations3581358191591517541754912912

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

The other explanatory variable we consider is the exchange rate regime. By tying domestic inflation to that of a low inflation country and being more conducive to fiscal and monetary discipline, fixed exchange rate regimes can arguably contribute to lower inflation (Gosh, Gulde, Ostry, and Wolf, 1997). Yet, it has also been argued that in allowing policy markers to lower temporarily inflation without a concomitant fiscal adjustment, fixed exchange rates can actually detract from fiscal discipline and give rise to a peso problem (Tornell and Velasco, 2000; Fatás and Rose, 2001). If so, no positive relationship between flexible exchange rates and inflation should be expected, at least in the medium- to long-run. We consider these hypotheses by including in the regressions the Reinhart-Rogoff (2002)de facto index of exchange rate flexibility, which is defined as ranging from 0 (complete inflexibility) to 15 (extreme floating). Our estimates show no evidence of a statistically significant relationship between exchange rate flexibility and inflation.17 As shown at the top of Table 7, the coefficient on exchange rate flexibility for the whole sample yields the expected positive sign but the associated t-ratio is well below usual levels of statistical significance for the whole panel as well as for most country, with notable the exception of emerging markets. But even for the latter, the respective MG and PMG estimates fall short of being statistically significant at 5 percent. Moreover, the estimated magnitude of the effect varies widely across different country groups and, in some of them, the estimates yield the opposite sign as expected by proponents of fixed exchange rates as instruments of price stabilization.

Table 7.Dynamic Panel Estimates of Inflation on Central Government Deficit over M1, Oil Price Inflation, and Exchange Rate Regime 1/
1. By level of financial development
All CountriesAdvanced CountriesDeveloping CountriesEmerging Markets
MGPMGh-stat.MGPMGh-stat.MGPMGh-stat.MGPMGh-stat.
LR Elasticities:
(G-T)/M10.68

(2.21)
0.04

(5.80)
4.3

[0.04]
-0.11

(-1.87)
0.33

(6.29)
205.6

[0.00]
1.08

(2.26)
0.06

(5.99)
4.62

[0.03]
1.66

(1.65)
0.38

(27.82)
1.62

[0.20]
Oil Price Inflation-0.02

(-0.12)
0.07

(18.04)
0.52

[0.47]
-0.14

(-0.51)
-0.35

(-4.99)
0.65

[0.42]
0.07

(0.55)
0.05

(10.31)
0.02

[0.89]
-0.45

(-0.56)
0.08

(6.47)
0.44

[0.51]
Exchange Rate Regime11.80

(0.96)
2.93

(1.18)
-0.06

(-1.36)
-0.22

(-1.20)
3.02

(1.05)
-0.65

(-0.47)
3.61

(1.75)
3.30

(1.84)
EC Coefficient (Φ)-0.50

(-12.57)
-0.45

(-12.65)
-0.35

(-7.69)
0.00

(-0.01)
-0.59

(-12.55)
-0.52

(-11.64)
-0.61

(-7.16)
-0.44

(-5.70)
No. of Observations2820282085285219681968872872
2. By level of inflation
All CountriesTop 25Mid-50Bottom 25
MGPMGh-stat.MGPMGh-stat.MGPMGh-stat.MGPMGh-stat.
LR Elasticities:
(G-T)M10.68

(2.21)
0.04

(5.80)
4.3

[0.04]
1.97

(1.75)
0.39

(31.33)
1.98

[0.16]
0.00

(0.37)
0.08

(8.30)
2.76

[0.10]
-0.08

(1.63)
0.00

(0.06)
2.79

[0.09]
Oil Price Inflation-0.02

(-0.12)
0.07

(18.04)
0.52

[0.47]
-0.55

(-0.63)
0.02

(0.69)
0.42

[0.52]
0.12

(4.31)
0.11

(15.88)
0.19

[0.67]
0.072

(5.70)
0.08

(12.26)
0.44

[0.51]
Exchange Rate Regime11.80

(0.96)
2.93

(1.18)
-7.04

(-1.24)
-6.89

(-1.24)
0.11

(0.91)
0.15

(1.15)
3.61

(1.75)
3.30

(1.84)
EC Coefficient (Φ)-0.50

(-12.57)
-0.45

(-12.65)
-0.49

(-5.17)
-0.37

(-4.42)
-0.5

(-8.56)
-0.47

(-9.29)
-0.61

(-7.16)
-0.44

(-5.70)
No. of Observations2820282081281213111311697697

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman statistic to test the cross-country long-run homogeneity restriction. Exchange rate regime measured by the Reinhalt-Rogoff (2002) index and treated as a fixed regressor.

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman statistic to test the cross-country long-run homogeneity restriction. Exchange rate regime measured by the Reinhalt-Rogoff (2002) index and treated as a fixed regressor.

Finally, we test the robustness of our results to the use of a broader measure of the fiscal balance. As noted in Section 3, general government data is only available for a subset of countries (86 out of the 107 countries) and for many of them the respective series do not start until the mid to late 1970s, so the panel size is now drastically reduced. Notwithstanding these changes in panel coverage, Table 8 shows that budget deficits remain a statistically significant driver of inflation for the whole sample as well as for developing and high inflation country groups. Moreover, estimates are on the whole similar to those previously obtained using the central government measure. The only noteworthy differences lie in the advanced country group (for which both MG and PMG estimates of Ψ are statistically insignificant) and the low inflation group (for which the h-statistic cannot reject the cross-country homogeneity restriction and the corresponding PMG estimate of 0.10 is statistically significant at 1 percent). Thus, if anything the use of general government balances yields even more favorable results to the theory than the central government measure.

Table 8.Dynamic Panel Estimates of Inflation on General Government Deficit over M1 and Oil Price Inflation.
A. By level of financial development
All CountriesAdvanced CountriesDeveloping CountriesEmerging Markets
MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1
LR Elasticities:
(G-T)/M11.41

(2.03)
0.02

(2.96)
4.03

[0.04]
-0.42

(-1.31)
-0.04

(-1.49)
1.23

[0.27]
2.08

(2.23)
0.04

(5.66)
4.90

[0.03]
2.47

(1.55)
0.37

(18.36)
1.58

[0.21]
Oil Price Inflation0.143

(1.20)
0.08

(16.35)
1.32

[0.25]
0.55

(1.94)
0.20

(9.32)
1.57

[0.21]
-0.89

(-1.27)
0.05

(11.42)
1.82

[0.18]
0.57

(0.47)
0.07

(3.86)
0.28

[0.60]
EC Coefficient (Φ)-0.50

(-13.40)
-0.55

(-12.92)
-0.22

(-8.32)
-0.16

(-8.36)
-0.66

(-11.04)
-0.52

(-11.59)
-0.53

(-7.34)
-0.37

(-5.23)
No. of Observations2368236867067016981698684684
B. By level of inflation
All CountriesTop 25Mid-50Bottom 25
MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1MGPMGh-stat.1
LR Elasticities;
(G-T)/M11.41

(2.03)
0.02

(2.96)
4.03

[0.04]
7.03

(2.34)
0.40

(25.63)
4.89

[0.03]
0.12

(1.37)
0.04

(3.91)
0.74

[0.39]
0.04

(0.65)
0.10

(2.64)
1.02

[0.31]
Oil Price Inflation0.143

(1.20)
0.08

(16.35)
1.32

[0.25]
-3.12

(-1.29)
-0.03

(-0.74)
1.65

[0.20]
0.07

(0.97)
0.09

(10.36)
0.06

[0.80]
-0.11

(-0.60)
-0.35

(-4.89)
1.97

[0.16]
EC Coefficient (Φ)-0.50

(-13.40)
-0.55

(-12.92)
-0.68

(-4.17)
-0.38

(-3.87)
-0.54

(-8.81)
-0.47

(-9.28)
-0.45

(-0.41)
-0.45

No. of Observations2368236850450411401140724724

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

t-ratios in parenthesis and p-values in brackets. The h-statistic refers to the Hausman test on the long-run homogeneity restriction.

VI. Conclusions

Economic theory postulates a clear causal connection between fiscal deficits and inflation. However, the strength of this relationship is not easy to measure. One reason stressed by Sargent and Wallace (1981) is that deficits need not cause inflation in the short run, since under fiscal dominance deficits determine the present value of seigniorage but not necessarily current seigniorage. This implies that proper empirical assessment of the theory requires sufficiently long time series and econometric techniques that can capture the dynamic dimension of this relationship. Second, theory also suggests that the inflation-deficit semielasticity may vary across countries with disparate inflation levels and distinct inflation tax bases. So, a suitable model should be also capable of accommodating such non-linearities. Third, the strength of the effect is likely to depend on the country’s level of financial development and central bank commitment to low inflation—deeper financial markets and more credible central banks in advanced economies tend to facilitate continuous rolling over of sizable debt stocks, obviating the need for seigniorage. So, it is important that empirical assessments of the theory take these differences into account.

This paper has sought to address each of these issues. By using dynamic panel data techniques, we modeled the deficit-inflation relationship as intrinsically dynamic. Also unlike previous studies, this relationship was modeled as nonlinear in the inflation tax base, leading to a distinct specification in which the fiscal deficit is scaled by narrow money rather than by GDP—a distinction shown to be empirically important in a panel comprising low-and high-inflation countries. Finally, working with a panel spanning 107 countries and 42 years of data has enabled us to test the theory over a reasonably lengthy horizon and across distinct institutional settings and inflation thresholds, without sample representativeness issues or degree of freedom limitations becoming critical.

On the other hand, the results are much more favorable to fiscal-based theories of inflation than previous research had found. Fiscal deficits have been shown to matter not only during high and hyperinflations but also under moderate inflation ranges, even though effects are substantially weaker in the latter case. Disaggregating by country groups, the deficit-inflation relationship comes out as surprisingly strong over a broad range of developing countries, also in contrast with the earlier literature. Moreover, the theory is also supported by the finding that none of the alternative explanatory variables considered undermine the strength of the effect or prove to be statistically significant across much of the panel, with the sole exception of oil prices. In particular, trade openness was found to matter for the developed country group but not for all countries; and little evidence was found that fixed exchange rate regimes help lower inflation on a systematic basis.

On the other hand, and similarly to previous studies, we did not detect any positive and strong connection between deficits and inflation in advanced economies and low-inflation country groups. This begs the question of why the theory seems to be violated in those two cases. Regarding the low-inflation country group, since half of its constituents consist mostly of very small, open economies with longstanding hard pegs or those that have given up their national currencies altogether (see the Appendix), the assumption of fiscal dominance underlying the theory is either severely weakened or nonexistent. The other half (13 out of 26) comprises advanced countries, taking us straight back to the question as to why the theory does not seem to hold for this group. Since we find evidence of a statistically significant relationship between budget deficits and inflation among advanced countries in the middle inflation ranges, violation of the theory appears to be more narrowly confined to the subgroup of low-inflation advanced countries. As noted above, greater monetary policy credibility, institutional constraints to fiscal dominance, and deeper financial markets seem to be at least part of the answer.

Last but not least, this paper has also shown that the statistical significance of the fiscal deficit-inflation relationship in most countries is relatively robust to alternative specifications and the use of standard panel data techniques. At the same time, we have also shown that the measured strength of the effect is not: the dynamic panel estimators and the econometric specification proposed in this paper yield considerably higher elasticities and seem to fit the data better than the standard specification using static fixed effects. This suggest that previous failures in uncovering a strong relationship between budget deficits and inflation stem, at least in part, from using a model specification and/or econometric techniques which do not accommodate key features of the theory. In sum, this paper’s findings indicate that the continuing prominence of fiscal-based theories of inflation does not seem unjustified, after all.

Appendix

Appendix: List of Countries and Country Groups

A. By level of financial developmentBy Inflation level
AdvancedEmerging MarketsOther DevelopingTop 25Mid-50Bottom 25
AustraliaArgentinaBahamasArgentinaBarbadosAustralia
AustriaBrazilBahrainBoliviaBelizeAustria
BelgiumChileBarbadosBrazilBhutanBahamas
CanadaChinaBelizeChileBotswanaBahrain
CyprusColombiaBhutanColombiaBurundiBelgium
DenmarkCzech RepublicBoliviaCongo, Dem Rep.CameroonBurkina Faso
FinlandEgyptBotswanaEcuadorChadCanada
FranceHungaryBurkina FasoGhanaCosta RicaChina
GermanyIndiaBurundiGuyanaCzech RepublicCyprus
GreeceIndonesiaCameroonIcelandDenmarkSt. Vincent & Gren.
IcelandIsraelChadIndonesiaDominican RepublicFrance
IrelandJordanCongo, Dem. Rep. ofIsraelEgyptGermany
ItalyKoreaCosta RicaMalawiEl SalvadorJapan
JapanMalaysiaDominican RepublicMexicoEthiopiaMalaysia
NetherlandsMexicoEcuadorNicaraguaFijiMalta
New ZealandPakistanEl SalvadorNigeriaFinlandMorocco
NorwayPeruSt. Vincent & Gren.PeruGabonNetherlands
PortugalPhilippinesEthiopiaPolandGambiaNorway
SpainPolandFijiRomaniaGreeceOman
SwedenSingaporeGabonSierra LeoneGuatemalaPanama
SwitzerlandSouth AfricaGambia, TheTanzaniaHaitiSingapore
United KingdomThailandGhanaTurkeyHondurasSt. Kitts and Nevis
United StatesTurkeyGuatemalaUgandaHungarySweden
UruguayGuyanaUruguayIndiaSwitzerland
VenezuelaHaitiVenezuelaIranThailand
ZimbabweHondurasZambiaIrelandUnited States
IranZimbabweItaly
KenyaJordan
LesothoKenya
MadagascarKorea
MalawiLesotho
MaldivesMadagascar
MaltaMaldives
MauritiusMauritius
MoroccoMyanmar
MyanmarNepal
NepalNew Zealand
NicaraguaPakistan
NigeriaPapua New Guinea
OmanParaguay
PanamaPhilippines
Papua New GuineaPortugal
ParaguayRwanda
RomaniaSolomen Islands
RwandaSouth Africa
Sierra LeoneSpain
Solomon IslandsSri Lanka
Sri LankaSt. Lucia
St. Kitts and NevisSwaziland
St. LuciaSyrian Arab Republic
SwazilandTonga
Syrian Arab RepublicTrimdad and Tobago
TanzaniaTunisia
TongaUnited Kingdom
Trinidad and Tobago
References

    Alesina, Alberto and AllanDrazen, 1991, “Why Are Stabilizations Delayed?” American Economic Review, Vol. 81 (December), pp. 1170–88.

    • Search Google Scholar
    • Export Citation

    Banerjee, Anindya, JuanDolado, J.W.Galbraith, and David F.Hendry, 1993, Co-integration, Error Correction and the Econometric Analysis of Non-Stationary Data (Oxford, England: Oxford University Press).

    • Crossref
    • Search Google Scholar
    • Export Citation

    Blanchard, Olivier, and StanleyFischer, 1989, Lectures on Macroeconomics (Cambridge, Massachusetts: MIT Press).

    Bleaney, Michael, 1999, “The Disappearing Openness-Inflation Relationship: A Cross-Country Analysis of Inflation Rates,” IMF Working Paper 99/161 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    Campillo, Martha, and JeffreyMiron, 1997, “Why Does Inflation Differ Across Countries?” in Reducing Inflation: Motivation and Strategy, edited by ChristinaRomer and DavidRomer(Chicago: University of Chicago Press).

    • Search Google Scholar
    • Export Citation

    Calvo, Guillermo, and CarlosVégh, 1999, “Inflation Stabilization and BOP Crises in Developing Countries,” in Handbook of Macroeconomics, ed. by JohnTaylor and MichaelWoodford, Volume C (Amsterdam: North Holland).

    • Crossref
    • Search Google Scholar
    • Export Citation

    Click, Reid, 1998, “Seigniorage in a Cross-Section of Countries,” Journal of Money, Credit and Banking, Vol. 30 (May), pp. 154–63.

    Cukierman, Alex, SebastianEdwards, and GuidoTabellini, 1992, “Seigniorage and Political Instability,” American Economic Review, Vol. 82 (June), pp. 537–55.

    • Search Google Scholar
    • Export Citation

    De Haan, Jakob, and DickZelhorst, 1990, “The Impact of Government Deficits on Money Growth in Developing Countries,” Journal of International Money and Finance, Vol.9 (December), pp. 455–69.

    • Crossref
    • Search Google Scholar
    • Export Citation

    De Haan, Jakob, and WillemKooi, 2000, “Does Central Bank Independence Really Matter? New Evidence for Developing Countries Using a New Indicator,” Journal of Banking and Finance, Vol. 24 (April), pp. 643–64.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Dornbusch, Rudiger, FedericoSturzenegger, and HolgerWolf, 1990, “Extreme Inflation: Dynamics and Stabilization,” Brookings Papers on Economic Activity: 2, Brookings Institution, pp. 1–84.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Fatás, Antonio, and AndrewRose, 2001, “Do Monetary Handcuffs Restrain Leviathan? Fiscal Policy in Extreme Exchange Rate Regimes,” CEPR Discussion Paper No. 2692 (London: Center for Economic Policy Research).

    • Search Google Scholar
    • Export Citation

    Fischer, Stanley, RatnaSahay, and CarlosVégh, 2002, “Modern Hyper—and High Inflations,” Journal of Economic Literature, Vol. 40 (September), pp. 837–80.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Franco, Gustavo, 1990, “Fiscal Reforms and Stabilization: Four Hyperinflation Cases Examined,” The Economic Journal, Vol. 100 (March), pp. 176–87.

    Ghosh, Atish, Anne-MarieGulde, Jonathan D.Ostry, and HolgerWolf, 1997, “Does the Nominal Exchange Rate Regime Matter?” NBER Working Paper 5874 (Cambridge, Mass.: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    Goodfriend, M.1997, “Monetary Policy Comes of Age: A 20th Century Odyssey,” Federal Reserve Bank of Richmont Economic Quarterly, Vol. 83/1, pp. 1–22.

    • Search Google Scholar
    • Export Citation

    Hamilton, James, and AnaHerrera, 2002, “Oil Shocks and Aggregate Macroeconomic Behavior: The Role of Monetary Policy,” Journal of Money Credit and Banking, forthcoming.

    • Search Google Scholar
    • Export Citation

    Hsiao, C., M.H.Pesaran, and A.K.Tahmiscioglu, 1999, “Bayes Estimation of Short-Run Coefficients in Dynamic Panel Data Models,” in Analysis of Panels and Limited Dependent Variables: A Volume in Honor of G.S. Maddala, ed. by C.Hsiao, K.Lahiri, L.F.Lee, and M.H.Pesaran(Cambridge, England: Cambridge University Press).

    • Crossref
    • Search Google Scholar
    • Export Citation

    International Monetary Fund. 2001, World Economic Outlook, May, (Washington: IMF).

    Ireland, Peter N., 1999, “Does the time-consistency problem explain the behavior of inflation in the United States?” Journal of Monetary Economics, Vol. 44 (October), pp. 279–91.

    • Crossref
    • Search Google Scholar
    • Export Citation

    King, Robert, and CharlesPlosser, 1985, “Money, Deficits, and Inflation,” Carnegie Rochester Conference Series on Public Policy, Vol. 22 (Spring), pp. 147–96.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Lane, Philip, 1995, “Inflation in Open Economies,” Journal of International Economics42, pp. 327–47.

    Ljungqvist, Lars, and ThomasSargent, 2000, Recursive Macroeconomic Theory (Cambridge, Mass.: MIT Press).

    Loungani, Prakash, and PhillipSwagel, 2001, “Sources of Inflation in Developing Countries,” IMF Working Paper 01/198 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    Metin, Kivilcim, 1998, “The Relationship Between Inflation and the Budget Deficit in Turkey,” Journal of Business and Economic Statistics, Vol.16 (October), pp. 412–22.

    • Search Google Scholar
    • Export Citation

    Mitchell, B.R., 1998a. European Historical Statistics, 4th edition, London.

    Mitchell, B.R.1998b. International Historical Statistics: Asia and Africa, 3rd edition, London.

    Mitchell, B.R., 1998c. International Historical Statistics: The Americas and Australasia, 4th edition, London.

    Montiel, Peter, 1989. “An Empirical Analysis of High-Inflation Episodes in Argentina, Brazil, and Israel,” Staff Papers, International Monetary Fund, Vol. 36 (September), pp. 527–49.

    • Crossref
    • Search Google Scholar
    • Export Citation

    O’Connell, Paul G., 1998. “The Overvaluation of Purchasing Power Parity,” Journal of International Economics, Vol. 44 (February), pp. 1–19.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Pesaran, M. Hashem, and YongcheolShin, 1998, “An Autoregressive Distributed Lag Modelling Approach to Cointegration Analysis,” in Econometrics and Economic Theory in the 20thCentury: The Ragnar Frisch Centennial Symposium, edi. By SteinarStrom(Cambridge, Mass.: Cambridge University Press).

    • Crossref
    • Search Google Scholar
    • Export Citation

    Pesaran, M. Hashem, and RonSmith, 1999, “Pooled Estimation of Long-Run Relationships in Dynamic Heterogeneous Panels,” Journal of the American Statistical Association, Vol. 94 (June), pp. 621–34.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Reinhart, Carmen, and KennethRogoff, 2002, “The Modern History of Exchange Rate Arrangements: A Reinterpretation,” NBER Working Paper No. 8963 (Cambridge, Mass.: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    Rodrik, Dani, 1991, “Premature Liberalization, Incomplete Stabilization: The Ozal Decade in Turkey,” in Lessons of Economic Liberalization and its Aftermath, ed. by M.Bruno, S.Fischer, E.Helpman, and N.Liviatan(Cambridge, Mass.: MIT Press).

    • Search Google Scholar
    • Export Citation

    Romer, David, 1993, “Openness and Inflation: Theory and Evidence,” Quarterly Journal of Economics, Vol. 108 (November), pp. 869–903.

    Sargent, Thomas, 1982, “The Ends of Four Big Inflations,” in Inflation, Causes, and Effects, ed. by RobertHall(Chicago, Ill.: University of Chicago Press).

    • Search Google Scholar
    • Export Citation

    Sargent, Thomas, and NeilWallace, 1981, “Some Unpleasant Monetarist Arithmetic,” Federal Reserve Bank of Minneapolis Quarterly Review, Fall.

    • Search Google Scholar
    • Export Citation

    Stock, James, 1987, “Asymptotic Properties of Least Squares Estimators of Cointegrating Vectors,” Econometrica, Vol. 55 (September), pp. 1035–56.

    Tanzi, Vito, MarioBlejer, and MarioTejeiro, 1993, “Effects of Inflation on Measurement of Fiscal Deficits: Conventional Versus Operational Measures,” in How to Measure the Fiscal Deficit, ed. by MarioBlejer and AdrienneCheasty(Washington: International Monetary Fund.

    • Search Google Scholar
    • Export Citation

    Terra, Maria Cristina, 1998, “Openness and Inflation: A New Assessment,” The Quarterly Journal of Economics, Vol. 113 (May), pp. 641–48.

    Tornell, Aaron, and AndrésVelasco, 2000, “Fixed versus Flexible Exchange Rates: Which Provides More Fiscal Discipline?” Journal of Monetary Economics, Vol. 45 (April), pp. 399–436.

    • Crossref
    • Search Google Scholar
    • Export Citation
1

We thank Tam Bayoumi, Ken Kletzer, Ayhan Kose, Jonathan Ostry, Ken Rogoff, and participants in an IMF Research Department seminar for comments on an earlier draft. Emily Conover and Ben Sutton provided excellent research assistance.

3

An extensive survey of ARDL models is provided in Banerjee, Dolado, Galbraith, and Hendry (1993). The time-series properties of ARDL models in the estimation of long-run cointegrating relationships are discussed in Pesaran and Shin (1998).

4

If the variables are I(1), the superconsistent property of OLS estimates holds and reverse causality becomes a non-issue (Stock, 1988). If the variables are I(0), the fact that left-hand side variable enter the regression in lagged form helps mitigate endogeneity biases. Moveover, reverse causality in fiscal deficit-inflation relationship seems to be more of an issue only in very high inflation episodes or during hyperinflations (Sargent, 1982; Franco, 1990; Dornbusch, Sturzenegger, and Wolf, 1990).

5

While the IFS was the main data source, some gaps in the series were filled with data from IMF’s country desks and World Economic Outlook databases, and Mitchell (1998a,b,c).

6

While the change in high-powered money is also a widely used measure of seigniorage, it is less germane to theoretical concept of demand for transactions money in the model. Moreover, high powered money as a measure of the inflation tax base is not unproblematic: it overestimates the inflation tax base when reserve requirements held at the central bank are remunerated (as is the case in some countries in our panel, and underestimates it when the government finds a way of extracting from banks the gains yielded by negative real interest rates paid on sight deposits.

7

To allow for reasonably rich dynamics without wasting too many degrees of freedom, we impose the condition that p,q ≤ 3. A shorter lag structure (p, q ≤ 2) does not change qualitatively the results but lowers the t-ratios in several cases.

8

The same conclusion holds if other standard model selection criterion—the Akaike Information Criterion—is used.

9

See the Appendix for the list of countries comprising each group. The developed v. developing country breakdown is based on the IMF’s World Economic Outlook classification, whereas the definition of “emerging markets” follows IMF (2001). High inflation countries comprise those in the upper quartile of the inflation distribution (averaged over the 1960–2001 period) and low inflation countries comprise the bottom 25 percent. Using a breakdown by quintiles rather than by quartiles does not change the thrust of the results. Period averages of the inflation rate and of other relevant variables for those various groups are provided in Table 1.

10

Another distinctive feature of the advanced country estimates is the substantially lower coefficient on the error correction term, suggesting that the dynamics of the inflation-deficit relationship takes much longer to unravel.

11

Specifically, the PMG estimator yields a coefficient of 0.1 with a t-ratio of 3.05 and a h-statistic of 0.29, which clearly does not reject the cross-country slope homogeneity assumption. Full details of these estimates are not reported to conserve on space but are available from the authors upon request.

12

We have also estimated (9) using pooled OLS without country-specific fixed effects but standard hausman tests clearly favored the fixed-effect specification relative to pooled OLS at any conventional level of statistical significance.

13

Standard errors of individual country regressions vary from as low as 1 (Austria) to as high as over 300 as in high inflation economies such as Argentina, Brazil, and Turkey. Such a dispersion of error variances is reflected in the disparate R2,s between the distinct groups as shown in Table 3.

14

The way the log approximation accomodates non-linearities in the data can be readily seen by taking the derivative of inflation with respect to the deficit in In(l + π) = μ + Ψy(G − T)/GDP + ε. This yields: π[(GT)/GDP]=ψy(1+π) which states that, for a given estimate of Ψy, the effect of a percentage change in the ratio of fiscal deficit to GDP will be higher as π increases.

15

The average fit of individual country regressions also improves significantly with the inclusion of oil prices. The full panel country averages of the adjusted R2 for the model with oil is 0.38, as opposed to 0.27 without oil. The full panel average unadjusted R2 with oil is 0.45.

16

The other widely studied hypothesis derived from the time inconsistency theory of monetary policy is that inflation should be lower in countries with more independent central banks or with central banks which are credibly committed to a low inflation mandate (Cuckierman and others, 1992; de Haan and Kooi, 2000). For evidence that central bank behavior helps explain historical swings in inflation rates in the United States, see Goodfriend (1997) and Ireland (1999). Lack of long time series on central bank independence measures for most countries in our panel unfortunately prevents us from evaluating this hypothesis on a broad cross-country basis.

17

As the index is not available for all countries in the data set, the panel size drops to 78 countries.

Other Resources Citing This Publication