A Modern History of Fiscal Prudence and Profligacy
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

We draw on a newly collected historical dataset of fiscal variables for a large panel of countries—to our knowledge, the most comprehensive database currently available—to gauge the degree of fiscal prudence or profligacy for each country over the past several decades. Specifically, our dataset consists of fiscal revenues, primary expenditures, the interest bill (and thus both the primary and the overall fiscal deficit), the government debt, and gross domestic product, for 55 countries for up to two hundred years. For the first time, a large cross country historical data set covers both fiscal stocks and flows. Using Bohn’s (1998) approach and other tests for fiscal sustainability, we document how the degree of prudence or profligacy varies significantly over time within individual countries. We find that such variation is driven in part by unexpected changes in potential economic growth and sovereign borrowing costs.

Abstract

We draw on a newly collected historical dataset of fiscal variables for a large panel of countries—to our knowledge, the most comprehensive database currently available—to gauge the degree of fiscal prudence or profligacy for each country over the past several decades. Specifically, our dataset consists of fiscal revenues, primary expenditures, the interest bill (and thus both the primary and the overall fiscal deficit), the government debt, and gross domestic product, for 55 countries for up to two hundred years. For the first time, a large cross country historical data set covers both fiscal stocks and flows. Using Bohn’s (1998) approach and other tests for fiscal sustainability, we document how the degree of prudence or profligacy varies significantly over time within individual countries. We find that such variation is driven in part by unexpected changes in potential economic growth and sovereign borrowing costs.

I. Introduction

The terms “fiscal prudence” and “fiscal profligacy” are often used, somewhat loosely, to denote whether fiscal policies tend to lead to a sustainable or unsustainable fiscal position. The latter would correspond, in more academic terms, to whether the government’s intertemporal budget constraint is met—that is, whether the expected present discounted value of all future fiscal surpluses matches the existing stock of public debt.

Although a precise definition of prudence or profligacy has not been established, policymakers, investors, and voters need to take a view all the time, in real time, on whether a country’s fiscal policies are appropriate to support economic growth and achieve other social objectives without causing a fiscal crisis. The focus is on the fiscal stance within the control of the government—usually proxied by the primary fiscal balance (i.e., the fiscal balance net of interest payments).

In practice, prudence and profligacy are medium-term concepts. Neither prudence nor profligacy is built up overnight: one or even a few years of expansionary fiscal policies do not necessarily cause a fiscal crisis, if a government’s initial position is sufficiently strong. Conversely, one cannot expect that, in real life, people will wait until infinity to check whether the intertemporal budget constraint is met. A few years of sustained deficits could well suggest that the intertemporal budget constraint is at risk. Thus, judgments regarding whether prudence or profligacy prevails are necessarily made over the course of a few years. Moreover, we believe (and show empirically below) that a country’s degree of prudence or profligacy is not constant forever; rather, it will change over time, as governments, citizens’ attitudes, and economic circumstances change.

This paper draws on a newly collected historical dataset of fiscal variables for a large panel of countries—to our knowledge, the most comprehensive database currently available—to gauge the degree of fiscal prudence or profligacy for each country over the past several decades. Specifically, our dataset consists of fiscal revenues, primary expenditures, the interest bill (and thus both the primary and the overall fiscal deficit), the government debt, and gross domestic product, for 55 countries for up to two hundred years. For the first time, a large cross country historical data set covers both fiscal stocks and flows.

Unfortunately, the economics profession has not yet developed a universally accepted indicator of fiscal sustainability. We rely heavily on the work of Bohn (1998, 2008), which we consider to be the “state of the art” in this area. Bohn’s sustainability criterion is based upon a time series regression of the primary fiscal surplus on the public debt and other controls.2 Thus far, empirical application of the Bohn test has been constrained by data limitations. Bohn’s own work analyzed long run historical time series data for the United States. A more recent study by Mendoza and Ostry (2008) analyzed panel data for 34 emerging markets and 22 advanced economies over the period 1990–2005, but the relatively short time period for which data were available required constraining the estimated fiscal policy response coefficient to be the same across the advanced economies and across the emerging markets. Our data collection effort makes it possible for the first time to run this test for individual countries, for a large number of countries.

A possible drawback of Bohn’s test is that it estimates the policy response over a long time frame of many years. Because our purpose is to explore variation in the fiscal policy response function across countries but also over time within a given country, we relax Bohn’s assumption of a constant long term fiscal policy response. Specifically, for each country, we employ three variations of the standard Bohn regression, including structural break tests, recursive searches for particularly influential observations, and iterations of the standard regression over rolling subsamples. We also complement these exercises with a simpler “policymakers’ criterion” widely used among practitioners, which consists of comparing the actual primary surplus to the primary surplus that would be required to stabilize the debt-to-GDP ratio.3 By algorithmically combining these criteria, we believe we provide a reasonable gauge of the degree of fiscal prudence or profligacy for each country at various points in time.

More generally, our paper follows a well established tradition in drawing lessons relevant to modern themes from long-run historical panel data sets (recent examples include Reinhart-Rogoff, 2009 and 2011 for public debt; and Schularick and Taylor, forthcoming, for credit aggregates).

While we use the terms “prudence” and “profligacy” for presentational simplicity, we attribute to them specific, positive, technical meaning as described in the body of the paper, rather than necessarily normative meaning. In other words, “profligate” fiscal policy responses may sometimes be justified from a normative standpoint—for example, to avoid plunging the economy into a deep and prolonged recession. The analysis in this paper is primarily positive.

Our main findings are the following:

  • For most advanced countries, particularly prior to the global economic and financial crisis that began in 2008, we find evidence that the response of the primary fiscal surplus to variation in government debt is consistent with meeting the intertemporal budget constraint, as well as stationarity of the debt.

  • Nevertheless, the evidence suggests that a given country’s fiscal policy response to changes in debt is by no means constant throughout its history (a working assumption that previous studies had made owing to data limitations). On the contrary, we document significant variation in such response, not only across countries, but also over time within a given country. Periods of a few or more years are distinguishable as clearly “prudent” or “profligate,” often with all techniques giving consistent messages. Indeed, one of the paper’s contributions is to document how individual countries fare with respect to fiscal prudence and profligacy, using each of the methods outlined above. Individual country results are reported in detail in this working paper’s tables and charts, with further information reported in the country pages in the accompanying Chartbook.

  • For example, the results suggest widespread fiscal prudence in most advanced economies during the mid-1990s until at least the mid-2000s; for the emerging economies, prudence becomes more widespread after the year 2000. Strong prudence is evident in the United States in the late 1990s (recalling contemporary discussion of a possible disappearance of the public debt); in Canada since the mid-1990s (beginning with an ambitious and successful fiscal adjustment plan); in several Euro area countries during the mid-1990s (coinciding with the Maastricht Euro entry process); in Ireland in the late 1980s and early 1990s (a well known fiscal adjustment episode); in Japan in the mid-1980s to early 1990s (as it sought to stabilize the debt); and in Turkey in the mid-1990s and at several points in the 2000s (as it improved its primary balance significantly). Conversely, notable episodes of fiscal stimulus are also evident, including the United States in 2009–11 and Spain in 2010. And Japan is found not to sufficiently improve its primary balance despite rising debts for several years starting in the late 1990s.

  • Finally, we show that a stronger response of the primary fiscal balance to changes in government debt is significantly associated with changes in long-run real GDP growth rates and long-term sovereign borrowing costs (measured by secondary market interest rates on long-term government debt). Declines in “potential” (or long-run) economic growth may not be fully apparent in real time to contemporary policymakers, who therefore often fail to respond to such declines through sufficient improvements in the primary balance. Conversely, increases in the cost of sovereign borrowing prompt policymakers to tighten fiscal policy in response.

The remainder of the paper is organized as follows. Section II outlines the data collection process and reports the summary statistics. Section III presents our empirical approach and its underpinnings. Section IV reports the empirical results. Section V concludes.

II. Data Sources and Basic Statistics

A. Data Sources for Fiscal and Other Macroeconomic Variables

The database covers an unbalanced panel of 55 countries (24 advanced economies—by present day definition from the IMF’s World Economic Outlook classification—and 31 non-advanced) over 1800–2011. The data consist of government revenue, non-interest government expenditure, and the interest bill (and thus also the overall fiscal balance and the primary balance), as well as gross public debt, all expressed as a share of GDP. Table 1 reports the list of countries and the corresponding period for which we have data for all the variables mentioned above.

Table 1.

Country Coverage of Main Variables in this Study

article image
Notes: This table provides the number of observations for which all key variables are available (Debt, Primary Balance and macro variables). It also contains the share of observations that were drawn from country-specific and other hand-collected sources.

This database covers not only public debt stocks, but crucially, the corresponding fiscal balance flows and their subcomponents. The availability of data on the primary balance to accompany the corresponding public debt observations makes it possible to apply well established tests or criteria that seek to gauge the degree of sustainability of a country’s fiscal policies and public debt position.

About half of the observations for the fiscal variables in our dataset are drawn from various cross-country sources, including the IMF’s World Economic Outlook (WEO) and International Financial Statistics (IFS) and the OECD Analytical Database for the past 20–50 years (subject to availability)4; the Statistical Yearbooks of the League of Nations and the United Nations (as well as their Public Debt Supplements) for the period between World War I and the 1970s (we collected these data by hand from various yearly reports); and Flandreau and Zumer (2004) for the pre-World War I era; in addition, long-run historical series are drawn from Mitchell’s International Historical Statistics and the Montevideo-Oxford Latin American Database (MOXLAD).

We hand-collected the other half of the data from country-specific sources, such as official government publications or economic histories that included public finance statistics. Examples of such data sources include Fregert and Gustafsson (2005) for Sweden over 1800–2004; Fernandez and Acha (1976) for Spain over 1850–1975; and Junguito and Rincon (2004) for Colombia over 1923–2003. The list of all sources, with complete bibliographical and coverage information, is provided in Appendix Table 1 (see electronic chartbook).

In collecting nominal GDP data for the distant past, we relied heavily on Mitchell and MOXLAD. For most countries, GDP data do not exist before World War I (indeed, the concept was not used by contemporaries), and in these cases we used proxy variables such as Gross National Product or Net National Product from Mitchell’s International Historical Statistics. In a few cases we used UN statistical yearbooks to fill in gaps in coverage between 1940 and 1975. GDP data were drawn from the OECD database for a few member countries beginning as early as 1960. For some countries, such as the United States, the United Kingdom, Italy, the Netherlands, Japan, Canada, and India, we used government publications or other country-specific sources. Starting in the mid 1990s, GDP data for almost all countries are taken from the WEO. Many sources, both cross-country and country-specific, provided fiscal data already expressed in terms of GDP as well.

Given the availability of multiple sources with significant overlaps for each country, we report in detail the “decision tree” (see Appendix Figure 1 in electronic chartbook) we used to splice together different sources to create continuous historical series. Within each country, we sought to preserve source continuity across time, to minimize jumps in the series that would have stemmed solely as a result of changes in sources. Whenever possible, we also sought to draw all variables (particularly the fiscal variables) for each given year from a single data source, to preserve consistency across concepts. The splicing process was straightforward for countries without much source overlap or source disagreement, and for countries for which we found a source offering long, nearly uninterrupted coverage for all concepts of interest. When these conditions did not hold, preserving source continuity across time sometimes became at odds with preserving source consistency across concepts. In such cases, because the primary balance usually had to be computed as the difference between the overall fiscal balance and interest payments, and because the response of the primary balance to changes in public debt is a key object of interest in this paper, we generally preferred source consistency across concepts over continuity across time. For instance, even though Mitchell provides fairly comprehensive data on revenue and expenditure, we often took revenue and expenditure data from UN statistical yearbooks where they were available, because these yearbooks also provide data on interest payments and debt. This said, we sought to use a given data source for continuous stretches of at least ten years, unless shorter stretches were the only way to fill a gap. Appendix Table 2 (see electronic chartbook) reports, by country, the data source used for each concept at each point in time.

Figure 1.
Figure 1.

Government Debt and Primary Fiscal Balance, 1850–2011, in percent of GDP

Citation: IMF Working Papers 2013, 005; 10.5089/9781616357825.001.A001

Sources: See Data Source Description in main text and accompanying electronic chartbook.Note: The top (middle) panel presents the unweighted and GDP-weighted mean, as well as the median, of the government debt (primary fiscal balance) as a share of GDP for the countries in the sample, whose number at each point in time is indicated by the bars in the bottom panel. The shaded areas show the 15th and 85th percentiles (not shown in the middle panel for the years corresponding to the major wars to preserve readability of the chart).
Table 2.

Summary Statistics for Full Sample (1800–2011)

article image
Source: Authors’ estimates; Full data sources reported in text and appendices.Notes: Summary statistics are presented for the unbalanced sample for the countries and years with available data, from 1800 to 2011, for up to 55 countries. The interest-growth differential is the difference between the implied nominal interest rate (this year’s interest payments divided by the average of this year and last year’s debt stock) and this year’s nominal GDP growth rate. All primary concepts are net of the public sector debt interest expenditure.

An important issue in the construction of long term public finance data series relates to the choice of government sector coverage. We sought to use data referring to the most comprehensive sector of government for which they were available. Accordingly, we report data at the general government level where these are available. In most cases, general government data are impossible to come by before 1960—not surprisingly, given that for most countries the share of spending by sub-national governments has risen significantly only since then. As a result, the sector reported switches (in most cases, simultaneously for all variables for a given country) from central government to general government in nearly all final spliced series, and this switch generally happens in the 1960s or 70s. For countries with large and active subnational governments, such as most advanced countries, this change in sector coverage resulted in significant breaks in the revenue and primary expenditure series; the breaks in the debt and fiscal balance series were smaller. Breaks in series are recorded in the database through dummy variables.

B. Summary Statistics for Main Fiscal Aggregates

Figure 1 reports the simple and GDP-weighted averages and the median of the public debt stock (top panel) and the primary balance (middle panel), in percent of GDP, from 1850 to the present. The shaded areas represent the range between the 15th and 85th percentiles. The number of countries in each year for which both debt and primary balance data are available ia also reported (bottom panel). We observe sharp decreases (increases) in the average primary balances (debts) during the World Wars. The range of both variables widens substantially during times of war, e.g. the U.S. Civil War. The primary deficits reverse quickly once the war episodes end, whereas postwar debts decline more gradually and over longer periods.

Focusing on the post-WWII era, debt declines continuously until the 1970s, through a combination of negative interest-growth differentials together with stable and slightly positive average primary balances. From the 1970s, debts generally increased and the primary balances improved somewhat further. Strong improvements in the primary balance occurred in the second half of the 1990s, largely reflecting Maastricht-related fiscal consolidations. Noticeable worsenings in the primary balance occurred in the late 1970s, early 1980s, early 1990s, and early 2000s. The global economic and financial crisis of the late 2000s resulted in the most pronounced and pervasive peacetime worsening of the primary fiscal balance experienced during our long term historical investigation: the average and GDP-weighted average primary deficits in 2008–09 were larger than at any other point in history aside from the World Wars. By way of comparison, the Great Depression is hardly noticeable in the chart.

Table 2 provides summary statistics for the full sample period (1800–2011, subject to data availability) for the major budgetary line items, including revenues, expenditures, the public debt interest bill, the overall fiscal balance, the primary balance, gross public debt, and the interest-growth differential.5 The dataset consists of about 5,700 observations for each of revenue, expenditure, overall fiscal balance, and debt. Interest expenditure data, however, are limited to approximately 4,800 observations (and consequently primary expenditure and primary balance as well). Both debt and primary balance data are available for about 4,500 country-years.

The revenue and expenditure ratios to GDP averaged 19 percent of GDP and 21 percent of GDP, respectively, while the public sector interest bill averaged 2½ percent of GDP. The primary surplus over the sample averaged ½ percent of GDP, debt averaged 50 percent of GDP, and countries generally faced a negative interest-growth differential. While both advanced and non-advanced countries maintain fiscal balances of similar magnitude, non-advanced economies report both lower revenues and expenditures. Advanced economies also report larger debt ratios, primary balances, and interest-growth differentials compared with the non-advanced. Primary surpluses in the top percentile are in excess of 9½ percent of GDP, though these largely correspond to commodity producers or countries with large government assets.6 Those in the top five percent of the distribution are above 5½ percent of GDP, and include several advanced economies with a well diversified production structure and relatively small government assets.

Table 3 reports the summary statistics for the post-World War II period, 1950-2011. Significant differences are again observed between advanced and non-advanced economies, notably: (i) size of the public sector nearly twice as large in the advanced economies compared with the non-advanced; (ii) slightly larger debt ratios in the advanced economies (based on the median across countries); (iii) larger primary balances in the advanced economies; and (iv) more negative interest rate-growth differentials in non-advanced economies.

Table 3.

Summary Statistics for Post-WWII (1950-2011)

article image
Source: Authors’ estimates; Full data sources reported in text and appendices.Notes: Summary statistics are presented for the unbalanced sample for the countries and years with available data, from 1950 to 2011, for up to 55 countries. The interest-growth differential is the difference between the implied nominal interest rate (this year’s interest payments divided by the average of this year and last year’s debt stock) and this year’s nominal GDP growth rate. All primary concepts are net of the public sector debt interest expenditure.

Table 4 compares the post-war period with the period up to 1950 but excluding wars, for the advanced economies. The table shows the tripling of public sector revenues and expenditures relative to the pre-1950 period. Despite worsening primary balances and primary expenditure increasing fourfold after 1950, debt declined, consistent with the negative interest rate-growth differential observed after World War II. The broader increase in primary expenditure could also be interpreted as consistent with government size increasing with economic development and over time (“Wagner’s law”).

Table 4.

Summary Statistics for Advanced Economies

article image
Source: Authors’ estimates; Full data sources reported in text and appendices.Notes: Summary statistics are presented for the unbalanced sample for the countries and years with available data, from 1800 to 2011, for up to 24 advanced countries. The interest-growth differential is the difference between the implied nominal interest rate (this year’s interest payments divided by the average of this year and last year’s debt stock) and this year’s nominal GDP growth rate. All primary concepts are net of the public sector debt interest expenditure. Wars with increases in the expenditure-to-GDP of at least 6 percentage points are excluded: Danish-Swedish, 1808-1809; United States Civil War, Greco-Turkish War, World War I, and World War II.

To make the case that our main object of interest, the primary fiscal balance, is a key driver of variation in the debt-to-GDP ratio, we recall identity (2.1), which decomposes the change in the debt-to-GDP ratio (Δdt) into the contributions from the primary fiscal balance (st), the interest rate-growth differential (rtgt1+gt)dt1, and the stock-flow residual (SFRt).

Δdt=st+(rtgt1+gt)dt1+SFRt(2.1)

As the variance of a sum can be expressed as the sum of the variances plus twice the sum of the covariances, we decompose the variance of the changes in the debt ratio into the sum of the variances and pairwise covariances of the primary deficit, the contribution from the interest-growth component, and the stock-flow residual.

Table 5 and Table 6 show the variance decomposition results for advanced and non-advanced economies with more than 30 observations for 1950-2011, and excluding changes corresponding to shifts in government sector coverage. For advanced economies, high stock-flow residual variances broadly correspond to countries with sizable asset accumulation (e.g. Norway, Sweden, Finland, and Japan), and to Greece, which experienced high inflation and defaults at various points in the 1950s and 1960s, and again debt restructuring in 2011. The average and median values for advanced economies suggest that fluctuations in the primary balance, interest rate-growth differentials, and stock-flow changes each explain roughly one-third of the variance of changes in the debt ratio. For non-advanced economies, more volatile debt changes are in most cases explained largely by the volatility in the stock-flow residual and interest-growth differential, consistent with greater prominence of defaults, episodes of high inflation, and more frequent exchange rate crises. Nevertheless, there is great heterogeneity among the non-advanced economies. The factors underlying variation in debt for macroeconomically stable non-advanced economies, such as Colombia and India, are similar to those for advanced economies, whereas debt changes in several other non-advanced economies are driven almost entirely by the stock-flow residual.

Table 5.

Decomposition of Variance of Debt Changes, Advanced Economies (1950-2011)

article image
Source: Authors’ estimates; Full data sources reported in text and appendices.Notes: This table reports the results of the variance decomposition of debt changes for advanced countries over 1950-2011, excluding debt changes corresponding to shifts in government sector coverage. We used the following identity to perform this decomposition exercise, derived from equation (1.1): Var(Δdt)=Var(st)+Var[(rtgt1+gt)dt1]+Var(SFRt)+2[CoVar(st,(rtgt1+gt)dt1)+CoVar(st,SFRt)+CoVar((rtgt1+gt)dt1,SFRt)], where Δd is the one-year difference in the debt to GDP ratio, s is the primary surplus to GDP ratio, r-g is the implied nominal interest rate-GDP growth differential, and SFR is the stock-flow residual. In the table, we label the interest-growth times outstanding debt component as “(r-g)d” and the stock-flow residual as “Residual”. Furthermore, C() indicates twice the covariance of the variables in parenthesis. Hence each column to the right of ΔDebt expresses a component of the variance of one-year debt changes; note that the values in the component columns sum to the values in the ΔDebt column.
Table 6.

Decomposition of Variance of Debt Changes, Emerging Economies (1950-2011)

article image
Source: Authors’ estimates; Full data sources reported in text and appendices.Notes: This table reports the results of the variance decomposition of debt changes for emerging countries over 1950-2011, excluding debt changes corresponding to shifts in government sector coverage. We used the following identity to perform this decomposition exercise, derived from equation (1.1):Var(Δdt)=Var(st)+Var[(rrgt1+gt)dt1]+Var(SFRt)+2[CoVar(st,(rtgt1+gt)dt1)+CoVar(st,SFRt)+CoVar((rtgt1+gt)dt1,SFRt)], where Δd is the one-year difference in the debt to GDP ratio, s is the primary surplus to GDP ratio, r-g is the implied nominal interest rate-GDP growth differential, and SFR is the stock-flow residual. In the table, we label the interest-growth times outstanding debt component as “(r-g)d” and the stock-flow residual as “Residual”. Furthermore, C() indicates twice the covariance of the variables in parenthesis. Hence each column to the right of ΔDebt expresses a component of the variance of one-year debt changes; note that the values in the component columns sum to the values in the ΔDebt column.

C. Data Sources for Other Variables

Wars

Years in which major wars led to extraordinary, temporarily high expenditures are excluded in Table 4 and the tests below. This is consistent with Barro’s (1979) “tax-smoothing” reasoning and with Bohn’s (1998) empirical analysis of United States data, which reported specifications excluding World War II and its immediate aftermath. In the multinational context, we identified a list of country-years involving participation in major wars (from various encyclopedias) and then excluded from the sample those country-years where war participation led to an increase in the expenditure-to-GDP ratio by at least six percentage points in one year. We also excluded up to two years after the war period, to allow for expenditures linked to demilitarization and reconstruction. Based on this criterion, we excluded country-years involving the Danish-Swedish War of 1808-1809, the United States Civil War, the Greco-Turkish War, World War I, World War II and the Indo-Pakistani War of 1971.7

Defaults

Years of default are drawn from Reinhart and Rogoff (2010) and are excluded from most empirical exercises. 8 Many Latin American economies in our sample experienced defaults, as did several European economies prior to and during World War II.

Output and expenditure gaps

We compute output and expenditure gaps as percent deviations of real output and expenditure series from their Hodrick-Prescott-filtered trends.9 For nearly all countries in the sample, real output series are computed from Maddison data for the early period, and the World Development Indicators (WDI) database for real GDP data for later years.10 For most countries, WDI real GDP data begin in the 1960s or 1970s. From 2009 through 2016, real GDP data and projections are drawn from the WEO.11 Real expenditure series result from the product of expenditure-to-nominal GDP and real GDP. For expenditure data gaps of three years or fewer, interpolated real expenditure values are used. For countries with blocks of at least 25 years of continuous expenditure data separated by a gap of at least 4 years, the HP-filter was applied to the separate sample periods. In cases where the expenditure data unnaturally changed as a result of a shift in coverage of the levels of government, all general government real expenditure figures were multiplied by the ratio of central government expenditure to general government expenditure in the year of the sector switch.

Commodity price indices

To control for the effects of commodity price swings on the primary balance of commodity producing countries, we include two world commodity price indices as additional regressors—one index includes petroleum prices and the other does not. The source is MOXLAD for 1900–1957 and the IMF Research Department thereafter. The list of countries dependent on commodity exports and those dependent on energy exports more specifically are drawn from the WEO (which in turn derived its country groupings from export composition data for 1962–2010).

Real long term interest rates

An objective of this paper is to explore potential determinants of variation—across countries and over time—in the fiscal policy response to increasing debt. To that end, in section V, we test the hypothesis that a higher marginal cost of sovereign borrowing (interest rates on government bonds observed on secondary markets) is associated with greater responsiveness of primary fiscal surpluses to government debt. Our real long term interest data consist of 2,270 observations and are mainly drawn from Bordo and others (2001), Dincecco (2011), and the IMF’s International Financial Statistics (IFS). Generally, Bordo and Dincecco cover the time period 1860–1947, whereas the IFS cover 1948–2011. In order to fill the gaps in our series we use several other cross-country databases as well as national sources (OECD; WEO; Mauro and others, 2006; IMF databases). Altogether, we have data for 27 countries with the series going back to the 1880s for the majority of the advanced economies.

III. Measuring Fiscal Prudence/Profligacy

We now outline how the degree of fiscal prudence or profligacy can be measured, both across countries and at different points in time, and how “fiscal reaction” regression analysis needs to be extended to explore how fundamental economic variables shape the degree of responsiveness of fiscal policy.

A. Bohn’s Fiscal Reaction Function and Policymakers’ Criterion

The literature on debt sustainability has identified a limited set of somewhat crude indicators of what may be labeled as fiscal prudence. In what follows, we rely heavily on an approach developed by Bohn (1998), which is based on estimating the following “fiscal reaction” regression on time series data for a given country:

st=ρdt+αZt+t,(3.1)

where st and dt are the primary surplus and the beginning-of-period public debt, respectively, both in percent of GDP; Zt captures other determinants of the primary balance, such as the business cycle or war expenditure shocks; εt is an error term.

Bohn (1998) shows that if ρ is estimated to be positive and significant, then fiscal policy is consistent with the intertemporal budget constraint under uncertainty, and that the test is robust to changes in interest rates, debt structure, and growth rates. Moreover, he shows that if ρ > (r–g/1+r) then the debt ratio is stationary: in the event of a shock to the debt ratio, the fiscal policy response would be sufficiently strong to bring the debt ratio gradually back to its initial level. In our empirical applications, we will use medium run averages of r and g, as detailed in the next sections.

Despite its many strengths, Bohn’s (1998) approach also has limitations. First, configurations of the debt ratio and ρ may emerge in which the primary fiscal surplus implied by the estimated fiscal policy reaction function is too high to be politically feasible or realistic. In the empirical applications, this limitation becomes relevant, as we show below. Second, the test was conceived against the background of a generally rising debt ratio. However, many countries experienced declining debt ratios for several decades: we would argue that, in that context, failure to obtain a positive and significant Bohn coefficient (which would require a worsening primary deficit) would indeed indicate an inconsistency with the intertemporal budget constraint but should be labeled as over-accumulation of assets rather than lack of fiscal prudence.12 We note such instances below when they occur.

Third, many years of data are necessary for a regression to be estimated, but policymakers and others often need to come to a judgment on whether fiscal policy is appropriate over a shorter horizon. Thus, policymakers and analysts in international financial institutions and the private sector often rely on comparisons between the actual primary surplus and the primary surplus that would be needed to stabilize the debt ratio—we label this the policymakers’ criterion.13 From the well known debt motion equation, the debt-stabilizing primary surplus is st = dt–1 (rtgt/1+gt). Note the close correspondence between the policymakers’ criterion and Bohn’s criterion for stationarity of the debt ratio, as a stable debt ratio is a special case of a stationary debt ratio.14 15

B. Methods to Gauge Variation in Prudence/Profligacy Over Time

Although Bohn’s original application of his test considered the longest available time series, in principle the response of the primary fiscal balance to changes in the public debt level (i.e., the slope coefficient in a Bohn regression) is unlikely to have remained the same throughout a country’s history.16 Rather, variation in the slope coefficient is a testable hypothesis and we deliberately chose to explore it here through various methods, given that each method carries both advantages and disadvantages, and answers slightly different questions.

  • 1) Structural break tests. We perform Bai and Perron (1998) structural break tests, which partition each country’s history into discrete sub-periods, varying in their degrees of fiscal policy response to debt changes. These tests allow the historical fiscal record to determine the dates and number of potentially unknown structural changes in the degree of policy response, within the constraints of the available data and minimum subsample size. This well established method will allow us to show that in many cases the fiscal policy response to an increase in debt changes significantly over time within a given country. However, it is not sufficiently flexible to capture outliers or important changes in behavior that may occur for a few specific years and in the proximity of the beginning or the end of the sample.

  • 2) Search for influential observations. This approach assumes a country’s response is largely the same throughout its history, but searches for “influential observation” years in which the response of the primary surplus to changes in debt is especially weak or negative (years which cause an otherwise prudent country to become imprudent) or especially strong and positive (the years that matter the most in rendering a country’s behavior prudent).

    For a country in which the estimated debt coefficient in the full sample regression is not positive and significant, the regression is recursively run excluding one observation at a time, searching for the observation whose omission leads to the largest decline in the p-value. Having dropped that observation, the procedure is run again on the remaining observations, iterating until the coefficient becomes positive and significant. Hence, the “most profligate” years, those whose omission is sufficient to restore the country to a finding of prudence, are identified. For example, in the case of the United States from 1950-2011, this procedure finds that dropping the years 2008-2011 is sufficient to return to a positive and significant slope coefficient.

    For a country in which the estimated debt coefficient is initially positive and significant, the regression is recursively run excluding one observation at a time, searching for the observation whose omission leads to the largest increase in the p-value. The process can alternate between designating “influentially profligate” and “influentially prudent” years, based on one’s choices of confidence levels to establish “threshold” p-values.

    While this procedure is somewhat less standard, it has greater flexibility to capture sudden changes in behavior and influential observations in opposite directions in close proximity to each other.

  • 3) Iterative estimation of Bohn regressions to rolling windows or an expanding sample. Bohn (1998) regressions are estimated over rolling windows of predetermined length (e.g., 25 years) or over an expanding sample period (beginning from, say, 1950). The rationale in this case is to gauge fiscal prudence or profligacy based only on information for specific periods (say, comparing 1955–80 with 1980–2005, in the case of windows of predetermined length) or on all information available to contemporaries as time progresses (in the case of expanding sample periods). Although this procedure imposes the constraint that the fiscal response coefficient is the same throughout a window of predetermined length, it is well established and makes for easy comparison across such windows.

C. The Drivers of Changes in Fiscal Prudence/Profligacy

Having established—as we do in Section IV—that the fiscal response coefficient to changes in debt varies significantly across countries and over time, we will turn to exploring the economic factors underlying such variation.17 We will thus relax the assumption of a simple linear relationship in which the single parameter ρ fully captures a country’s fiscal policy response and does not change over time. In particular, we consider the case where the fiscal policy response depends on unexpected changes in the real long-term growth rate, and on the marginal public sector borrowing rate. The economic rationale is that policymakers may fail to improve the primary balance in response to increases in the debt-to-GDP ratio if they fail to perceive that economic growth has slowed down in a permanent manner. Additionally, an increase in the marginal cost of borrowing will cause policymakers to improve the primary balance as a means of lowering debt. For this exercise, we work with the marginal cost of borrowing (yields on newly issued debt), because we expect this to have a much quicker impact on policymakers’ behavior than would be the case for the average cost of borrowing on all outstanding debt, which responds slowly to changes in market conditions. Specifically, we estimate:

A03eq02

Equation (3.2) includes two interaction terms that capture potential changes in the fiscal policy reaction function to growth surprises and marginal borrowing costs. The first term interacts debt with gt, which is the unexpected change in long-term real GDP growth. For a given country, if both estimates of ρ^>0andρ^β^1>0 are positive and significant, a country lowers its primary surplus as long-term growth unexpectedly slows. (In this case, imposing a linear functional form on countries where long-term growth slows over time would lead to lower estimates of ρ^>0 and, potentially, structural breaks.) The second term interacts debt with rt, the real long-term borrowing costs on new debt (proxied by secondary market yields). If estimates of both ρ^>0andρ^β^2>0 are positive and significant, a country increases its primary surplus as real long-term borrowing costs increase for a given level of debt. In this case, imposing a linear functional form on countries where long-term borrowing costs increase over time would lead to higher estimates of ρ^ and, again, possible structural breaks.)

IV. Estimates of Fiscal Prudence/Profligacy

A. Bohn (1998) Tests for Whole Sample Periods and Long Sub-Samples, Individual Countries

We begin by estimating Bohn (1998) equations for each country, excluding periods of default or major wars. Table 7 reports the estimated coefficient for the response of the primary fiscal balance to variation in debt (the Bohn coefficient) and its p-value. Separate tests are reported for large sub-sample periods: post-WWI (1920–2011), post-WWII (1948–2011); and post-WWII excluding the global financial crisis (1948–2007).

Table 7.

Bohn (1998) Test for Fiscal Policy Sustainability

article image
Source: Authors’ estimates; Full data sources reported in text and appendices.Notes: OLS estimation, robust standard errors. Debt coefficients reported; output gap and temporary spending controls included for all countries. Commodity price controls included for relevant countries based on latest WEO list of historical commodity exporters (index which includes oil prices applied to countries dependent on oil revenues). Default years and significant war years excluded; we report results for all countries with at least 25 non-default, non-war observations The number of observations reported refers to the full sample estimation.
Table 8.

Bohn Regression Results Iterated Over Twenty-Five Year Rolling Windows

article image
Source: Authors’ estimates; Full data sources reported in text and appendices.Notes: OLS estimations of equation (1.2), with robust standard errors: st = ρdt + αZt + εt. We include in Zt output and expenditure gap controls, and, where applicable, a world commodity price index that includes or excludes oil as appropriate. We consider only 25 year periods that are complete with data, and do not include significant war or default episodes. Dates listed refer to the final year of the corresponding 25 year window: hence 1990-1 refers to the 1966-1990 and 1967-1991 windows. The first column thus reports all 25-year windows for which ρ is positive and significant (at the 5 percent level), and the second column reports all windows in which ρ is negative and significant. The “change to positive” column reports 25-year windows in which ρ becomes significantly positive after not being so in the window immediately prior; the “change from positive” column reports windows in which ρ is no longer significantly positive after being so in the window immediately prior.
Table 9.

Bohn (1998) Fiscal Reaction Function with Expanding Sample

article image
Source: Authors’ estimates; Full data sources reported in text and appendices.Notes: Iterative Bohn (1998) OLS regressions of st = ρdt + αZt + εt, with dt the beginning of period debt ratio, st the primary balance, and Zt controls for cyclical and other factors, beginning 1950 with twenty-five observations, expanding by one observation, and excluding countries with postwar defaults. The dates listed correspond to the final year of the subsample. The India window starts in 1978, the year after its last default.