Fiscal forecasts and slippages: the role of the SGP and domestic fiscal frameworks.
Martins, Patricia ; Correia, Leonida
INTRODUCTION
The Stability and Growth Pact (SGP) was designed to enforce fiscal
discipline in the member states of the European Union (EU). The recent
crisis, mainly the sovereign debt crisis in some countries of the euro
area, led to a deep reform of the EU's economic governance
framework. The 2011 and 2013 SGP reforms aim to deliver longer-term
improvements in EU public finances by tightening enforcement and
enshrining EU rules at the national level.
The preventive arm of the SGP requires countries to achieve
medium-term budgetary objectives, ensuring the sustainability of public
debt and providing room for fiscal maneuver in the economic
stabilization required in periods of recession without violating the
budget deficit limit. In the annually established Stability and
Convergence Programs (SCPs), the governments of both euro area and
non-euro area member states determine, respectively, what adjustment
paths are required to achieve their medium-term budgetary objectives and
the economic assumptions that underpin them.
The persistence of budget deficits above 3% of gross domestic
product (GDP) in several European Monetary Union (EMU) countries between
1999 and 2011 suggests that some countries may systematically have
failed to implement the adjustment paths defined in their SCPs. Until
the 2011 SGP reform, the preventive arm functioned as an ex ante fiscal
rule: there was no ex post verification of compliance with the planned
adjustment paths, nor did sanctioning procedures exist to punish
unjustified slippages. Under these circumstances, governments may have
been tempted to present forecasts in their SCPs that overestimated what
could in fact be achieved in order to be sure of meeting ex ante
criteria.
Studies of budgetary slippages can be divided into two groups. In
the first, authors have investigated forecasting errors (eg, Strauch et
al., 2004; Annett, 2006; Briick and Stephan, 2006; Pina and Venes,
2011), while in the second, they have been more concerned with
implementation errors (eg, Moulin and Wierts, 2006; EC, 2007; Beetsma
and Giuliodori, 2010; Beetsma et al, 2009; Pina, 2009; von Hagen, 2010;
Frankel and Schreger, 2013). Both concepts refer to the difference
between budgetary outcomes and forecasts. However, when the budgetary
slippages are defined as implementation errors, it is stressed that
their main cause is a lack of implementation of planned measures. The
authors recognize that the forecasts that governments submit in their
SCPs may contain a strategic element. It should not be expected that
governments will consider all the available information, or that their
forecasts will be accurate and unbiased. This means that the
implementation of the SGP may aggravate rather than mitigate the problem
of time inconsistency identified by Kydland and Prescott (1977). In the
annual SCPs, while governments may design their adjustment paths with a
view to positively influencing the evaluation of the European Commission
(EC), the implementation of expansionary policy measures can exacerbate
the problems of large deficits and public debts. Therefore, the second
group of studies highlights the importance of investigating at the same
time both the determinants of forecasts and subsequent deviations from
plans. This paper pursues the type of research undertaken by these
studies.
Our main objects of study are the budget balance, public debt, and
economic forecasts incorporated into the SCPs submitted between
1998/1999 and 2008/2009 by the governments of the 15 EU member states
before the 2004 enlargement. We intend to examine if these forecasts and
the corresponding forecast errors are explained by the definition and
application of the SGP until the recent crisis.
So far, to our knowledge, no published paper has examined the
determinants of public debt forecasts and/or their corresponding errors.
The relation between public debt and deficits is to a large extent an
identity. However, we are interested in testing whether the governments
were more engaged in the preparation of budget balance forecasts than
public debt forecasts. This seems a possibility as the public debt rule
of the SGP does not seem to have been a matter of great concern until
2011.
Given the importance of real GDP growth forecasts in constituting
fiscal forecasts and in explaining any subsequent slippages, and the
fact that governments are more likely to make growth forecasts
deliberately optimistic if their budget deficits are large, it was
thought important also to analyze the determinants of real GDP growth
forecasts and their respective errors. Moreover, we drew on both SCP
forecasts and EC autumn forecasts because EC forecasts have a prominent
role in the SGP.
In addition, we are interested in testing how domestic fiscal
frameworks can explain the different performance of EU governments'
forecasts. After the onset of the crisis, legislative developments at
the EU level recognized the importance of fiscal governance in the
member states (EC, 2014). The 2011 reform of the SGP established a
collection of new laws, known as the Six Pack, which includes a
directive on the requirements for budgetary frameworks of the member
states (Directive 2011/85/EU). This directive reinforces the injunction
that realistic forecasts should be used for fiscal planning. In 2013,
the Two Pack went further and introduced the requirement for national
medium-term fiscal plans and draft budgets of the EMU countries to be
based on independent and public macroeconomic forecasts. Several authors
(Jonung and Larch, 2006; Frankel and Schreger, 2013; Debrun and Kinda,
2014) have highlighted that the independent fiscal institutions, usually
termed fiscal councils, involved in the provision or assessment of
forecasts, contribute to less biased macroeconomic and budgetary
forecasts.
In this study, we focus on forecasts published in SCPs for the
years 1999 2009. In the period between 1998 and 2010, dates for the
submission of SCPs varied from country to country and from year to year,
usually falling between the October and February following the year to
which they referred. Thus, for example, forecasts for the year 2009 were
published at the end of year 2008 or in the first quarter of 2009 and so
to compute the respective forecast errors, we use the estimates from
SCPs summited at the end of 2009 or at the beginning of 2010. After
2011, the introduction of the European Semester required member states
to publish their SCPs during the first semester of the year, usually in
March. As governments did not publish SCPs at the end of 2010, it is not
possible to extend our sample after 2009 with directly comparable data.
The remainder of this paper is structured as follows: The next
section reviews the main results in the empirical literature on the
determinants of fiscal forecasts and respective forecasting errors; the
subsequent section presents the dependent variables, the statistical
properties of forecasting errors, and describes the methodology used;
the penultimate section presents the models and discusses the results of
our econometric analysis; and the final section summarizes our
conclusions.
DETERMINANTS OF FISCAL FORECASTS AND FORECASTING ERRORS
The principal variables determining forecasts and forecasting
errors have an economic, political, or institutional nature: economic
determinants are related to cyclical circumstances and/or to the state
of public finances; political determinants are associated with electoral
cycles, partisan cycles, and/or whether one or several parties form the
government, often collectively referred to as political fragmentation in
the literature; and institutional determinants have their origins in the
structure and dynamics of domestic institutions.
Economic variables
Until 2011, the SGP had a preventive arm consisting of an ex ante
fiscal rule and a corrective arm that imposed a 3 % limit on the
deficit-to-GDP ratio; this may have encouraged biased forecasting on the
part of governments, particularly in member states where public finances
had attained excessive deficits. As enforcement of the SGP is related to
economic variables, in this study we opted to analyze simultaneously
both the economic determinants and the contribution of the SGP.
Some research indicate that observed budget deficits are higher
than forecasts when growth has been unexpectedly poor (Brack and
Stephan, 2006; Beetsma et al, 2009; Pina and Venes, 2011). This suggests
that governments often resort to window dressing, that is, they present
rosy rather than realistic output growth forecasts in order to boost
both their projected revenues and their predicted public expenditure
(EC, 2005). (1) This strategy is also convenient for governments because
by underestimating the output gap, a larger share of the budget deficit
is made to appear cyclical in nature, thereby facilitating adjustment
toward medium-term SCP objectives (EC, 2008). (2)
Large budget deficits at the time the forecasts are made correspond
to optimistic fiscal forecasts, and more ambitious budget balance
forecasts frequently result in large and negative forecasting errors.
With respect to the effect of SGP, Beetsma and Giuliodori (2010) and
Beetsma et al. (2009) state that governments running deficits above the
limit of 3% of GDP have tended to publish less optimistic budget balance
forecasts. However, the results of Frankel and Schreger (2013) go in the
opposite direction: the relationship between deficits and over-optimism
is stronger for euro area countries for which the limits of the SGP are
relevant than for countries outside the euro area.
Political variables
The influence of political variables, such as those related to
electoral cycles, partisan cycles, and the degree of fragmentation of
the political system, have been examined by various authors, several of
whom (Bruck and Stephan, 2006; Giuliodori and Beetsma, 2008; von Hagen,
2010) have concluded that deviations from the budget balance forecasts
are primarily because of the effects of electoral cycles. While
government forecasts are generally more optimistic in the run-up to
elections, the timing of elections appears to play no statistically
significant role in explaining errors in real GDP growth forecasts
according to Strauch et al. (2004) and von Hagen (2010). The literature
nonetheless suggests that the so-called degree of fragmentation of the
political system may have two effects: budget deficit forecasts are more
optimistic when made by recently elected governments, probably as a
means of signaling to European and other international institutions
their commitment to the consolidation plans; also, political instability
and ideological swings from governments of the right to those of the
left tend to produce larger negative errors.
Institutional variables
The influence of the domestic fiscal framework on the quality of
forecasts has attracted researchers' attention. According to the EC
(2010), the elements constituting that framework include the rules,
regulations, and procedures that underlie the planning and
implementation of budgetary policies, and its main components consist of
national numerical fiscal rules, medium-term budgetary frameworks for
multiannual fiscal planning, and the various independent institutions
involved in budgetary policy. First, to solve the problems posed by
common pool resources, the budgetary process can be centralized by
imposing one of two forms of fiscal governance: delegation or contract.
Strauch et al. (2004) and von Hagen (2010) conclude that governments
operating under contracts have stronger incentives to publish prudent
forecasts than governments operating under delegation as unexpectedly
negative outcomes regarding GDP growth and the budget deficit will
almost certainly impose the inconvenience of renegotiating government
coalitions. Second, more appropriate national fiscal rules and
medium-term budgetary frameworks usually cause smaller budget slippages:
while budget forecasts tend to be more optimistic, forecasting errors
are smaller, because the proposed policy measures are more effectively
implemented (Beetsma et al, 2009). Finally, the size of the budget
balance forecast errors can be reduced if governments use GDP growth
forecasts provided by independent institutions (Annett, 2006). Frankel
and Schreger (2013) state further that EMU governments in an excessive
deficit situation and with an independent fiscal institution that
prepares budget forecasts publish fiscal forecasts that are less
optimistic than governments in the same situation but without such an
independent institution.
DATA AND METHODOLOGY
Dependent variables and data
Our research focuses on the determinants of the forecasted changes
in the budget balance and public debt presented in SCPs by the
governments of the IS countries that were members of the EU in 1999, and
the determinants of any subsequent deviations from those predicted
changes. In addition, we study the real GDP growth forecasts prepared by
both governments and the EC, and the respective forecasting errors.
Moulin and Wierts (2006) emphasize the advantages of considering
forecasted changes and differences between observed and forecasted
changes rather than forecasted values and slippages between observed and
predicted values, respectively. When forecast errors are computed as the
difference between observed and forecasted values, the introduction of
methodological changes after the elaboration of the forecasts
contributes to higher errors that undermine correct interpretation. In
addition, with regard to SCP forecasts, the analysis of forecasted
changes in budget balance makes it easier to assess the effort planned
by governments to achieve their medium-term budgetary objectives.
Beetsma et al. (2009) use the expression implementation errors to
stress that deviations from forecasted changes in budget balance result
mainly from failures in the implementation of policy measures underlying
the projections. In this study, albeit we compute the forecasting errors
in budget balance and public debt in a way similar to that proposed by
those authors, we opt to use the expression errors in forecasted changes
in preference to implementation errors because we consider it more
appropriate when the fiscal variable under scrutiny is public debt.
Therefore, the six dependent variables used in this study are
defined as follows: (3)
* Forecasted changes in budget balance and in public debt for
country i in year t, [DELTA][B.sup.t-1.sub.i,t] and
[DELTA][Debt.sup.t-1.sub.i,t] are the difference between the government
forecast for year t, submitted in the SCP usually at the end of year
t-1, and the estimate for year t-1, published at the end of that year;
(4)
* Errors in forecasted changes in the budget balance and in the
public debt for country i in year t, E[B.sup.t,t-1.sub.i,t] and
[EDebt.sup.t,t-1.sub.i,t] the observed change in year t minus the
forecasted change in year t-1;
* Real GDP growth forecasts presented in year t-1 by the government
or by the EC, [y.sup.t-1.sub.i,t].
* Real GDP growth forecast errors, [Ey.sup.t,t-1.sub.i,t], are the
difference between the estimate published in year t and the forecast
presented in year t-1.
To compute the dependent variables, we use two sources of
statistical information, the SCPs and the EC. We focus on the budget
balance, public debt, and GDP growth forecasts published in SCPs for the
period 1999-2009. However, we use the GDP forecasts and estimates
published by both governments and by the EC to compare the performance
of their growth forecasts and to confirm the robustness of the results.
EC data have the advantage of being published regularly in October or
November each year and therefore are made known to all governments at
the same time. (5)
The time covered by our study is the period extending from the
launch of the EMU to the beginning of the European economic crisis and
consequent economic governance reforms. We use data from SCPs submitted
between 1998/1999 and 2009/2010, forecasts for year f = 1999 and
estimates for year t = 2009, respectively. We do not extend our sample
further because governments did not publish SCPs at the end of 2010 and
since 2011 the European Semester has required the publication of SCPs
during the first semester of the year, usually in March.
Statistical properties of forecasting errors
We analyze the accuracy of forecasts for the whole sample and by
country over the period 1999-2009 using three summary statistics: mean
error (ME), mean absolute error (MAE), and root mean square error
(RMSE). (6) The MAE considers negative forecast errors as positive so
the errors do not cancel each other out as happens in the calculation of
the ME. The RMSE penalizes forecasters that make few but large forecast
errors more heavily than forecasters that make many small forecast
errors. We also test for unbiasedness and for autocorrelation. For each
country, we estimate the autoregressive specification of order 1 [AR(1)]
presented below:
[E.sup.t,t-1.sub.t] = [alpha] + [mu][E.sup.t-1,t-2.sub.t-1] +
[v.sub.t] (1)
When the estimated coefficient [alpha], corresponding to the
average forecasting error, is not statistically different from 0, there
is no systematic bias in the forecasts. If the null hypothesis [mu] = 0
is not rejected, there is no persistence of forecasting errors. (7) The
results are presented in Tables 1-3 for the budget balance, public debt,
and real GDP growth forecast errors, respectively. (8) Because we use
two sources of statistical information for estimates and for GDP
forecasts, there are two different ways of calculating the fiscal
forecasting errors, with SCP estimates or EC estimates, and three
different ways of computing the GDP growth forecast errors, with SCP
estimates and growth forecasts, EC estimates and SCP growth forecasts,
or EC data.
In the period 1999-2009, the average errors in forecasted changes
in budget balance were negative in all countries. When the SCP estimates
are used, the null hypothesis of no bias is rejected in the case of
Italy, Luxembourg, and the United Kingdom, and there is a positive and
significant degree of persistence of forecasting errors in Germany,
Spain, Luxembourg, and the United Kingdom.
The errors in forecasted changes in public debt are, on average,
positive, meaning that the observed changes were greater than expected,
except in Austria, Denmark, and Finland when EC estimates are used. The
results suggest underestimation of public debt in the case of Greece,
Italy, and Portugal, when the SCP estimates are used, and in the case of
Belgium, Greece, France, and Italy when the EC estimates are used. There
are also problems of positive and high autocorrelation in Denmark and
Spain, regardless of the source of data used, and in Ireland and the
United Kingdom, when we use the estimated changes from SCPs.
The average errors in real GDP growth forecasts are negative in all
countries. Irrespective of the data source used, there is a systematic
bias in the output growth forecasts in France, Italy, and Portugal.
As expected, the analysis of the results for the EU15 suggests
optimistic forecasts: overestimation of the observed change in the
budget balance and of the output growth, and underestimation of the
observed change in the public debt. In the case of fiscal variables, the
estimated coefficients [mu] are positive and statistically significant,
and thus there are systematic patterns of demeaned forecast errors over
time, which account for the recurrent delays in the implementation of
fiscal consolidation plans. In addition, the estimated coefficients of
bias and autocorrelation are higher, mainly in the case of public debt,
when data from the SCPs are used than when using EC estimates. We
investigate if these results for the EU15 are explained by the
definition and application of the SGP until the recent crisis.
Methodology
Our study provides six models, one for each dependent variable: (1)
forecasted changes in budget balance, (2) forecasted changes in public
debt, (3) errors in forecasted changes in budget balance, (4) errors in
forecasted changes in public debt, (5) forecasts of real GDP growth, and
(6) forecast errors of real GDP growth. As we use two different data
sources, each model is estimated according to different specifications,
depending on the source of information used in the definition and the
construction of the variables. The model for real GDP growth forecasts
is estimated for two specifications, using government and EC forecasts,
respectively. In case of the other five models, we have three
specifications: Specification A uses SCP data, Specification B uses
government forecasts and EC estimates, and Specification C employs
estimates and GDP growth forecasts published by the EC.
The aim of the econometric analysis is to identify the main
economic and political determinants of each dependent variable, using
panel data models with country and time fixed effects. We use country
fixed effects to control for heterogeneity and time fixed effects to
control for any time-varying biases that are common across all
countries.
The results of the Hausman test, reported in Tables 4-6, reject the
alternative of a random effects estimator in the models used to explain
the forecasted changes in budget balance and public debt, and models of
GDP growth forecasts. With respect to models of errors in fiscal
forecasted changes, the results are mixed. Thus, we decided to apply
fixed effects estimators to all specifications because the random
effects estimator is inconsistent in the case of errors in forecasted
changes in budget balance when SCP estimates are used, and because the
same estimation technique should be used to compare the respective
estimated coefficients.
To explain the GDP growth forecast errors, the results of pooled
ordinary least squares (OLS) models are also presented. In this case,
the country fixed effects have no statistical significance, the
estimated coefficients from the fixed effects model suggest that all
explanatory variables are statistically insignificant in Specifications
B and C, and the null hypothesis of the Breusch-Pagan LM test is not
rejected in all specifications.
In the two models used to explain forecasted changes in fiscal
variables, we tested for endogeneity in the real GDP growth forecasts
variable. In the case of models of errors in forecasted changes in
budget balance and public debt, we repeated this test for the variable
forecast errors of real GDP growth. Under the null hypothesis of the
endogeneity test, the endogenous explanatory variable can be considered
exogenous. This hypothesis was rejected only for the budget balance
forecasted changes. Therefore, we opted for a model with instrumental
variables only in this case.
The results of the endogeneity tests seem to confirm that over the
period of study governments were more engaged in the preparation of
budget balance forecasts than public debt forecasts. When governments
elaborate their GDP growth forecasts, they have in mind the forecasted
changes of the budget balance, but not the forecasted changes of public
debt, probably because the public debt rule was not been a matter of
great concern until the sovereign debt crisis.
As variables related to domestic institutions evolve very slowly
over time and thus including them in our country effects models could
lead to unreliable estimated coefficients, the three-step procedure
proposed by Beetsma et al. (2009) was followed to examine their
influence. First, for each institutional variable we computed a mean
value for each country over time. Then, for each of the dependent
variables, we estimated the country fixed effects of the regressions
including only the economic and political variables that were
significant. Finally, we investigated whether the mean of each
institutional variable contributes significantly to the explanation of
those country fixed effects.
EMPIRICAL RESULTS
Economic and political determinants and the role of the SGP
On the basis of published studies on the determinants of budgetary
plans cited earlier in the paper, two models are constructed to explain
forecasted changes in budget balance and public debt, respectively:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (3)
The models include country [v.sub.i] and time [z.sub.t] fixed
effects. The first three explanatory variables are common to both models
and are real GDP growth forecasts for year t, [y.sup.t - 1.sub.i,t],
budget balance estimates for year t-1, [B.sup.t-1.sub.i, t-1], and
public debt estimates for year t-1, [Debt.sup.t-1.sub.i, t - 1]. In
these models, real GDP growth forecasts is an explanatory variable, but
we also investigate their determinants in this paper to verify whether
governments used the growth forecasts for window dressing. Concerning
the state of public finances, we use the estimates of the two EMU fiscal
indicators, budget balance and public debt, to compare their relative
importance in explaining the fiscal forecasted changes.
The further three explanatory variables related to the fiscal
variable used in each equation are designated excessive budget
deficit/public debt, [SGPb.sup.t-1.sub.i,t-1] and
[SGPdebt.sup.t-1.sub.i, t], difference from EC budget balance/public
debt forecasts, [DiffECb.sup.t-1.sub.i,t] and
[DiffECdebt.sup.t-1.sub.i,t], and forecasted change in budget
balance/public debt in the year t-2 program, [DELTA][B.sup.t-2.sub.i,t]
and [DELTA][Debt.sup.t-2.sub.i,t]. These three variables are related to
the role of the SGP and so are of special interest in this study. The
excessive budget deficit and excessive public debt variables are used
only for EMU countries. The variables have a positive value if the
figures published by the EC at the time at which the forecasts were made
indicate a breach of the Maastricht Treaty limits in year t-1. These
variables are included to assess whether non-compliant EMU countries
submitted forecasted changes consistent with fiscal consolidation plans.
The difference from EC budget balance/public debt forecasts is defined
as the difference between the government forecasts inscribed in SCPs and
those prepared by the EC. The variables have a positive sign when the
state of public finances predicted by national governments is more
favorable than that predicted by the EC. The SGP gives a prominent role
to EC forecasts and in the case of significant differences from EC
forecasts, governments are asked to provide detailed explanations.
Because SCPs are rolling and flexible multiannual budgetary frameworks,
the variables forecasted change in budget balance/public debt in the
year t-2 program aim to specify the relationship between the forecasts
in year t-2 and those made in year t-1, that is, thereby determining the
level of governments' commitment to the fiscal targets enshrined in
their SCPs.
Finally, Pol designates a set of political variables. For the
purposes of this study, these are divided into three groups. For
variables of the first group, we use the Golinelli and Momigliano (2009)
database, which we have updated and completed. The political variables
of the second and third groups were collected from the database
published by Armingeon et al. (2009). The first group consists of
variables underpinning what have been termed opportunistic motivations
for governments introducing bias into their forecasts. The first
variable signals whether legislative elections took place or not in year
t, [elect.sub.i,t]. The other two indicate the type of elections
involved: regular elections, [regular.sub.i,t], and early elections,
[snap.sub.i,t]. (9) The second group embraces variables linked to
potential ideological motivations for biasing forecasts: party
composition of government, [gov_party.sub.i,t], and ideological gap
between current and previous governments, [gov_gap.sub.i,t]. The final
group refers to variables reflecting the degree of fragmentation of the
political system: the variable type of government, [gov_type.sub.i, t],
is related to size fragmentation, while the variables number of
government changes per year, [gov_chan.sub.i,t], and new party
composition of government, [gov_new.sub.i,t], reflect the temporal
dimension of political fragmentation.
As we are interested in the explanatory power of political
variables that relate to the year in which forecasts are published, in
the study of the opportunistic motivations we consider only the variable
regular elections, referring to elections that policymakers know will
take place during the following year at the time forecasts are included
in the SCPs. Given the high correlation coefficients between some of
these political variables, we add each of them to the model one at a
time to avoid problems of multicollinearity that might lead to incorrect
estimates of statistical significance.
Columns A, B and C in Table 4 present the coefficients for the
statistically significant variables for each of the three specifications
used to estimate forecasted changes in budget balance and public debt.
Our objective is to determine the existence of significant differences
between the estimated coefficients of the two models.
Improvements in economic growth perspectives, [y.sup.t-1.sub.i,t],
are reflected both in more favorable budget balance and public debt
forecasted changes, with the exception of public debt forecasts when SCP
estimates are used. If, at the end of year t-1, the budget deficit rises
by 1% of GDP, a further fiscal adjustment of about 0.3% of GDP will be
planned in year f, with the remaining two-third provoking an increase in
the public debt forecasted for year t. The public debt estimates for
year t-1, [debt.sup.t-1.sub.i,t-1], are not statistically significant in
explaining budget balance forecasted changes. In the case of public
debt, and despite its low absolute value, the estimated coefficient of
public debt is statistically significant when EC estimates are used.
These results indicate that governments were little concerned with
exerting restraint over the growth of their public debt ratios.
In Specification A, the estimated coefficient of the excessive
budget deficit, [SGPb.sup.t-1.sub.i,t-1], is positive and statistically
significant, showing that the more excessive the budget deficit, the
greater the adjustment planned in the budget balance. However, when EC
data are used, this variable loses its statistical significance. The
difference between specifications may be explained by EMU governments
with excessive deficits intentionally signaling their commitment to the
deficit rule. The estimated coefficient of excessive public debt,
[SGPdebt.sup.t-1.sub.i,t-1], has no statistical significance in any of
the specifications. This seems to confirm that, in the period under
scrutiny, the national governments and the European institutions
attached much less importance to situations of excessive debt than to
excessive budget deficits.
The size and significance of the variables difference from EC
budget balance and public debt forecasts depends to a large extent on
the source of the statistical information. As these differences
increase, the budget balance forecasted changes are more favorable when
EC estimates are used. In the case of public debt, the estimated
coefficients are negative and statistically significant in the three
specifications, but their absolute values are higher when EC estimates
are used. This outcome indicates that there are differences between
forecasts and also between the estimates produced by governments and by
the EC, which affect the size of forecasted changes in the budget
balance and public debt. (10)
In the case of forecasted change in budget balance/public debt in
the year t-2 program, the interpretation of estimated coefficients
suggests that forecasted changes inscribed in SCPs submitted in the
previous year were revised downward. A lack of government commitment is
more striking in the case of the budget balance, probably because
governments tended to produce more ambitious deficit-reduction plans as
they were conscious of the close scrutiny to which these would be
subjected by European institutions.
Most political variables proved to have no significant influence,
with two exceptions. First, changes in the party composition of the
government in the year in which the budget balance forecasts are
published presage more ambitious fiscal consolidation plans. Second,
when we use SCP estimates for public debt, it is left-wing, rather than
right-of-center or center parties, that tend to present more prudent
forecasts.
Our analysis of the determinants of errors in the forecasted
changes in budget balance and public debt is based on the following
models:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)
The forecast error of real GDP growth, [Ey.sup.t, t-1.sub.i,t], is
common to both models. For the specific fiscal variable being studied in
each equation, three other explanatory variables are considered: the
lagged error in forecasted change, [EB.sup.t-1, t-2.sub.i, t-1] or
[EDebt.sup.t-1, t-2.sub.i, t-1], the forecasted change,
[DELTA][B.sup.t-1.sub.i,t] or [DELTA][Debt.sup.t-1.sub.i,t], and the
revision of the excessive budget deficit or public debt,
[DSGPb.sup.t,t-1.sub.i, t-1] or [DSGPdebt.sup.t, t-1.sub.i, t-1]. For
EMU countries, the variables revision of excessive budget deficit/public
debt, [DSGPb.sup.t, t-1.sub.i, t-1] and [DSGPdebt.sup.t, t-1.sub.i,
t-1], have positive signs when a revision of deficit or debt estimates
points to an even more unfavorable outcome, and negative signs when the
outcome--albeit still excessive--is less serious than the first
estimates previously indicated. The aim of this exercise is to assess
whether governments were more effective in the implementation of their
respective adjustment paths, leading to smaller forecasting errors, if
their revised estimates revealed a more serious situation than
forecasted.
To understand whether shifts in the political context contribute to
the explanation of forecasting errors, we examine political variables
that describe government behavior in the year following the publication
of forecasts. To reveal cases of opportunism, the variables elections
and early elections are now considered.
The estimated coefficients for Specifications A, B, and C are shown
in Table 5. Negative real GDP growth forecast errors, [Ey.sup.t,
t-1.sub.i, t], constrain the effective implementation of the planned
changes in budget balance and particularly in public debt.
The lagged forecasting error, [EB.sup.t-1, t--2.sub. i, t--1], only
has significance in explaining errors in forecasted changes in budget
balance if SCP estimates are used, although some persistence seems to
occur. Conversely, forecasted changes in budget balance and public debt,
[DELTA][B.sup.t-1.sub.i, t] and [DELTA][Debt.sup.t-1.sub.i,t], are only
statistically significant when EC estimates are used. Taken together,
these two results suggest that, in the case of budget balance, negative
lagged forecasting errors partly account for negative errors in
forecasted changes when SCP statistical information is used. However,
when the analysis is carried out with EC estimates, forecasted changes
that are overly optimistic appear to induce negative forecasting errors.
In both situations, the absolute value of the estimated coefficients is
between 0.2 and 0.3. This inverse relationship between the optimism of
forecasted changes and the capacity to implement them effectively
corroborates our argument that any serious attempt to identify the
determinants of forecasting errors must examine all the explanatory
variables that underpin the respective forecasts.
The estimated coefficient of the variable revision of the excessive
budget deficit is high, positive, and statistically significant in
explaining budget balance forecasting errors. This suggests that when a
government identifies a situation of excessive deficit that is worse
than initially estimated, it feels compelled to undertake additional
restrictive fiscal measures, resulting in a greater-than-planned
improvement in the budget balance. (11) Again, this interpretation does
not hold when there are unfavorable revisions of public debt estimates
in excess of 60% of GDP.
While the statistical significance of political variables depends
on the source of the data used, a combination of ideologically motivated
opportunistic behavior and the fragmentation of the political system
exacerbate the difficulties of concretizing the forecasted fiscal
changes. Variables related to the electoral cycle appear to function in
a number of ways. If SCP estimates are used, the proposed fiscal
consolidation is more difficult to achieve when forecasts refer to a
year in which legislative elections are to take place. Early elections,
[snap.sub.i,t], contribute to an increase in errors in forecasted
changes in public debt of at least 1.3% of GDP. As expected, the
coefficients related to political variables with an ideological
dimension, [gov_party.sub.i,t] and [gov_gap.sub.i,t], show that the
substitution of a more right-wing government with a left-wing one tends
to harms budget implementation when EC estimates are used. An increase
in the number of parties represented in the government, [gov_type.sub.i,
t], and an increase in changes in government per year,
[gov_chan.sub.i,t], contribute to larger and more negative errors in
forecasted changes in budget balance when SCP and EC estimates,
respectively, are used. Finally, a new party composition of government,
[gov_new.sub.i,t], contributes to an increase in the errors in
forecasted changes in public debt, except in Specification B. These
results confirm that coalition governments and political instability
undermine the implementation of planned measures.
In the preceding analysis, we confirm that real GDP growth
forecasts and economic forecast errors are important in explaining the
forecasted changes in budget balance and public debt and errors in those
forecasted changes, respectively. Therefore, we have constructed
identical models to those used for fiscal variables to investigate
whether the SGP and economic and political variables affect growth
forecasts and respective errors in a similar way.
To explain real GDP growth forecasts, the following model is
employed:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (6)
The explanatory variables are the real GDP growth estimate,
[y.sup.t-1.sub.i, t-1], the budget balance estimate, [B.sup.t-1.sub.i,
t-1], in both cases for year t-1, the excessive budget deficit,
[SGPb.sup.t-1.sub.i, t-1], and the difference from EC economic
forecasts, [OPFy.sup.t-1.sub.i,t]. As the last variable corresponds to
the difference between the real GDP growth forecasts provided by
governments and by the EC, its sign is positive when forecasts of
official governments are relatively more optimistic.
The model for real GDP growth forecast errors is:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (7)
Four economic variables are analyzed: real GDP growth forecasts,
[y.sup.t-1.sub.i,t], the difference in real GDP growth estimates for the
year t-1 published in t and in t-1, [Dy.sup.t, t-1.sub.i, t-1], the
difference in budget balance estimates for the year t-1 published in t
and in t-1, [DB.sup.t, t-1.sub.i, t-1], and revisions of excessive
budget deficit, [DSGPb.sup.t,t-1.sub.i, t-1]. With regard to the
political determinants of economic forecasts and respective errors, we
include the same aforementioned variables used to account for forecasted
changes and for forecasting errors in the fiscal variables,
respectively.
The results presented in Table 6 for Specifications A and B, which
draw on government and EC forecasts, respectively, lead us to conclude
that all the explanatory variables, except the budget balance estimate,
[B.sup.t-1.sub.i, t-1], have a statistically significant influence over
real GDP growth forecasts. This conclusion does not support the idea
that governments use window dressing to subvert the fiscal rules; put
another way, the evidence does not show that less favorable budget
deficit situations systematically encourage governments to publish more
optimistic economic growth forecasts as some authors (Milesi-Ferretti
and Moriyama, 2006; van den Noord, 2007) have argued.
Each 1% over-estimate of real GDP growth in year t-1 contributes to
an increase of 0.5% in the forecasted real GDP growth in year t. For EMU
countries, however, a further worsening of an already excessive deficit
situation reduces forecasted real GDP growth, probably because of
government recognition that fiscal consolidation policies will need
further strengthening. As expected, the estimated coefficients for the
variable difference from EC economic forecasts show that the government
forecasts of real GDP growth contained in SCPs tend to be more
optimistic than the forecasts made by the EC.
Of all the political variables analyzed, only the number of changes
in government, [gov_chan.sub.i, t-1], proves to be statistically
significant in explaining real GDP growth forecasts. In the year a given
forecast is published, changes in government appear to increase the
likelihood that forecasts will be slightly more unfavorable, which could
signal the intention of new governments to implement more restrictive
fiscal measures.
With respect to real GDP growth forecast errors, when we use models
with country and time fixed effects, only in Specification A is there at
least one significant estimated coefficient. Specifications B and C are
omitted because none of the coefficients are statistically significant.
The difference in real GDP growth estimates for the year f-1 published
in t and in t-1, [D.sup.t, t-1.sub.i, t--1], has a positive and
statistically significant coefficient in Specification A, leading us to
conclude that downward revisions of economic growth estimates will
induce greater negative errors in real GDP growth forecasts of SCPs.
This causality derives from the fact that, as has already been
mentioned, economic growth forecasts tend to be more optimistic when
growth estimates for year t-1 have proven to be more favorable.
Because none of the variables are statistically significant in
Specifications B and C, and because country fixed effects are not
statistically significant in any of the three specifications, a pooled
OLS model is also applied. The results are presented in Columns A',
B\ and C' of Table 6. According to the estimated coefficients, the
difference in real GDP growth estimates is also the only variable with
statistical significance, and the resulting forecasting errors are
greater when SCP estimates and forecasts are used.
Finally, the results indicate that none of the political variables
have any statistically significant influence over forecasting errors
related to economic growth.
Robustness check
The crisis in the years 2008 and 2009 had a great impact on the
real economy of most European countries and respective public finances
that was not expected at the time of the publication of the first
forecasts. Thus we divide the sample into two sub-samples, 1999-2007 and
2008-2009, and use Chow tests to verify the robustness of results. (12)
The estimated coefficients show that in the crisis years, the
forecasted fiscal changes are mainly explained by the differences
between the government forecasts and those produced by the EC, because
some of the SCPs submitted at the end of 2008 were later updated by the
respective governments. Regarding the errors in the forecasted changes
in budget balance, another important result is that although the
economic and financial crisis worsened the excessive deficits, the
governments were more effective in the implementation of their planned
consolidation policies.
Institutional determinants
Five indicators are used to attach relative magnitudes to the
variables associated with the domestic fiscal framework, characterizing
member countries in terms of governance, the presence of multiannual
budgetary frameworks, the type of fiscal rules applied at the national
level, the existence of independent fiscal institutions and whether
forecasts are produced by independent institutions.
Delegation countries, [delegation.sub.i,t], is a dummy variable
that takes the value 1 for countries operating under delegation and 0
for countries operating under contracts. The countries classified as
delegation countries are Austria, Germany, Greece, Spain, France, Italy,
Portugal, and the United Kingdom. For the 15 European countries studied,
with the exception of Portugal, von Hagen's (2010) classification
is used. In the case of Portugal, and in line with the EC (2006) and
Debrun et al. (2009), we judge it more appropriate to classify it as a
delegation country. The objective of this dummy variable is to verify
whether countries operating under contracts publish more prudent
forecasts than delegation countries because of the transaction costs of
renegotiating coalition agreements.
The variable multiannual budgetary frameworks, [MTBF.sub.i, t],
refers to the medium-term budgetary framework index attributed to each
member state by the EC. Similarly, the variable national fiscal rules,
[rule.sub.i, t], corresponds to the EC's fiscal rule index. The
variable independent fiscal institutions, [institution.sub.i, t], uses
an index computed by Debrun and Kumar (2008), consisting of an
unweighted average of four criteria: the legal influence of the fiscal
institution over the budget process, the extent of its independence from
government influence, the perceived utility of its forecasts, and its
perceived effectiveness. The third criterion, regarding the utility of
its forecasts, corresponds to our independent forecast variable,
[indfor.sub.i, t]. The higher the values for these four variables, the
greater the expected contribution to more accurate forecasts and to
reducing forecasting errors. (13)
To avoid problems of multicollinearity, we individually test the
mean of each institutional variable for its contribution to the
explanation of the country fixed effects u, of the models that include
the statistically significant economic and political variables.
On the basis of the estimated coefficients for the mean of each
institutional variable, presented in Table 7, we are able to draw the
following conclusions. While countries with better multiannual budgetary
frameworks and with independent forecasts make more optimistic forecasts
of budget balance, their effective implementation is higher. For public
debt, the quality of the multiannual budgetary frameworks is not
relevant in explaining the forecasted changes, but it reduces the size
of forecasting errors when EC estimates are used.
Stronger national fiscal rules lead to forecasts of additional
public debt reduction when the EC estimates are used and reduce the size
of errors in forecasted changes in public debt when government estimates
are used.
Governments operating under delegation publish more prudent budget
balance forecasts in Specification B. However, as expected, delegation
countries produce more optimistic public debt forecasts in Specification
A and the errors in the forecasted changes in public debt in
Specifications B and C are higher. Except for the variable delegation,
all institutional variables reduce the forecasted growth in real GDP.
Thus the better the quality of domestic institutions, the less
optimistic are the real GDP forecasts. However, these variables are not
statistically significant in explaining economic growth forecast errors.
Multiannual budgetary frameworks display the highest determination
coefficient ([R.sup.2]) in the regressions of forecasted changes in
budget balance and real GDP growth forecasts. The change in its
respective index explains about half of the estimated country fixed
effects.
Conclusions
This paper undertakes an analysis of the determinants of fiscal
forecasts and respective errors. We focus on the budget balance, public
debt, and GDP growth forecasts presented by the 15 European governments
in the Stability and Convergence Programs for the period 1999-2009. Our
results suggest that governments were more engaged in the preparation of
budget balance forecasts than public debt forecasts, probably because
the debt rule of the corrective arm of the SGP was not effective until
2011, when it was tightened.
As deficits are higher at the time of the publication of fiscal
forecasts, the budget balance forecasts are more optimistic and the
public debt forecasts are more prudent. The public debt estimates only
influence the respective forecasted changes that are more prudent. Thus,
governments seem less concerned with presenting plans for the reduction
of their public debt ratios than of their budget balance ratios. The
Maastricht limits do not influence the respective forecasted changes,
except in the case of budget balance when SCP estimates are used: EMU
governments with excessive deficits present adjustment plans with the
intention of signaling their commitment to the deficit rule. The
forecasted fiscal changes do not adequately reflect the planned changes
inscribed in the SCPs submitted in the preceding year. The lack of
commitment of governments to the fiscal objectives previously
established has accounted for the recurrent delays in the implementation
of fiscal consolidation plans.
As expected, the errors in forecasted fiscal changes are higher
when the respective forecasts are more optimistic, which confirms that
it is more appropriate to investigate simultaneously the role of the SGP
and of other variables in explaining the government forecasts submitted
in SCPs and respective errors, as is done in this paper.
When there is a situation of excessive deficit that exceeds initial
estimates, governments have a strong incentive to implement further
fiscal adjustments, which results in an improvement in the budget
balance greater than forecasted. Thus, the deficit limit established in
the corrective arm of SGP seems to have been effective. The same does
not seem to be true for the debt limit: errors in forecasted changes in
public debt are not explained by unfavorable revisions of public debt
estimates in excess of 60% of GDP.
The real GDP growth forecasts are not explained by the estimates of
budget balance at the time of their elaboration, except when there are
excessive deficits that incentive the governments to present more
prudent economic forecasts. Thus, governments do not seem to have used
the economic growth forecasts to subvert the rules of the SGP as some
authors have argued. The GDP growth forecast errors are greater when the
estimates and economic growth forecasts elaborated by governments are
used, but they are mainly the result of information problems concerning
the real economic situation at the time the forecasts are elaborated.
Finally, our findings point out that better domestic fiscal
frameworks, mainly multiannual budgetary frameworks, contribute to more
prudent GDP growth forecasts and smaller errors in forecasted fiscal
changes. Given the prominence that domestic institutions have received
in recent years, mainly in Europe, we highlight that in countries where
independent institutions prepare the forecasts, the economic growth
forecasts are more prudent and the forecasts of budget balance are more
optimistic, but their effective implementation is higher.
Acknowledgements
This work is supported by national funds provided by the
FCT--Portuguese Foundation for Science and Technology, through its
project UID/SOC/04011/ 2013. The authors are grateful to two anonymous
referees for comments and suggestions. They are also thankful for
comments of Joao Loureiro (University of Porto, Portugal), Antonio
Afonso (Lisboa School of Economics and Management, Portugal) and
Christopher Gerry (University of Tras-os-Montes and Alto Douro,
Portugal). The usual disclaimer applies.
REFERENCES
Annett, A. 2006: Enforcement and the stability and growth pact: How
fiscal policy did and did not change under Europe's fiscal
framework. IMF Working Paper 6/116, International Monetary Fund:
Washington.
Armingeon, K, Gerber, M, Leimgruber, P and Beyeler, M. 2009:
Comparative political data set 1960-2007. Institute of Political
Science, University of Berne.
Beetsma, R and Giuliodori, M. 2010: Fiscal adjustment to cyclical
developments in the OECD: An empirical analysis based on real-time data.
Oxford Economic Papers 62(3): 419-441.
Beetsma, R, Giuliodori, M and Wierts, P. 2009: Planning to cheat:
EU fiscal policy in real time. Economic Policy 24(60): 753-804.
Briick, T and Stephan, A. 2006: Do eurozone countries cheat with
their budget deficit forecasts? Kyklos 59(1): 3-15.
Castro, F, Perez, J and Vives, MR. 2013: Fiscal data revisions in
Europe. Journal of Money, Credit and Banking 45(6): 1187-1209.
Debrun, X and Kinda, T. 2014: Strengthening post-crisis fiscal
credibility-fiscal councils on the rise: A new dataset. IMF Working
Paper 14/58, International Monetary Fund: Washington.
Debrun, X and Kumar, MS. 2008: Fiscal rules, fiscal councils and
all that: Commitment devices, signaling tools or smokescreens? Prepared
for the Banca d'Italia Workshop on Public Finance, March Rome.
Debrun, X, Hauner, D and Kumar, MS. 2009: Independent fiscal
agencies. Journal of Economic Surveys 23(1): 44-81.
EC. 2005: Public finances in EMU--2005. European Economy:
Luxembourg.
EC. 2006: Public finances in EMU--2006. European Economy:
Luxembourg.
EC. 2007: Public finances in EMU--2007. European Economy:
Luxembourg.
EC. 2008: EMU @10--Successes and challenges after ten years of
economic and monetary union. Office for Official Publications of the
European Communities: Luxembourg.
EC. 2010: Public finances in EMU--2010. European Economy: Brussels.
EC. 2014: Report on public finances in EMU--2014. European Economy:
Brussels.
Frankel, JA and Schreger, J. 2013: Over-optimistic official
forecasts and fiscal rules in the eurozone. Review of World Economics
149(2): 247-272.
Gilbert, ND and de Jong, JFM. 2014: Does the Stability and Growth
Pact induce a bias in the EC's fiscal forecasts? DNB Working Paper
451, Nederlandsche Bank: Amsterdam.
Giuliodori, M and Beetsma, R. 2008: On the relationship between
fiscal plans in the European Union: An empirical analysis based on
real-time data. Journal of Comparative Economics 36(2): 221-242.
Golinelli, R and Momigliano, S. 2009: The cyclical reaction of
fiscal policies in the euro area: The role of modelling choices and data
vintages. Fiscal Studies 30(1): 39-72.
Jonung, L and Larch, M. 2006: Improving fiscal policy in the EU:
The case for independent forecasts. Economic Policy 21(47): 491-534.
Kydland, FE and Prescott, EC. 1977: Rules rather than discretion:
The inconsistency of optimal plans. Journal of Political Economy 85(3):
473-491.
Merola, R and Perez, JJ. 2013: Fiscal forecast errors: Governments
versus independent agencies? European Journal of Political Economy 32:
285-299.
Milesi-Ferretti, GM and Moriyama, K. 2006: Fiscal adjustment in EU
countries: A balance sheet approach. Journal of Banking and Finance
30(12): 3281-3298.
Morris, R, Ongena, H and Schuknecht, L. 2006: The reform and
implementation of the Stability and Growth Pact. European Central Bank
Occasional Paper 47, European Central Bank: Germany.
Moulin, L and Wierts, P. 2006: How credible are multiannual
budgetary plans in the EU? Paper presented at the Banca d'Italia
Workshop on Fiscal Indicators, March.
Pina, AM. 2009: Elusive counter-cyclicality and deliberate
opportunism? Fiscal policy from plans to final outcomes. Banco de
Portugal, Working Paper 6/2009, Economics and Research Department:
Lisboa.
Pina, AM and Venes, N. 2011: The political economy of EDP fiscal
forecasts: An empirical assessment. European Journal of Political
Economy 27(3): 534-546.
Strauch, R, Hallerberg, M and von Hagen, J. 2004: Budgetary
forecasts in Europe--The track record of stability and convergence
programmes. European Central Bank Working Paper 307, European Central
Bank: Germany.
van den Noord, P. 2007: Fiscal policies in EMU at the crossroads.
In: Breuss, F. (ed). The Stability and Growth Pact--Experiences and
Future Aspects. Springer: Wien, New York.
von Hagen, J. 2010: Sticking to fiscal plans: The role of fiscal
institutions. Public Choice 1443(4): 487-503.
APPENDIX
Description of variables
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
PATRICIA MARTINS & LEONIDA CORREIA
Department of Economics, Sociology and Management (DESG), Centre
for Transdisciplinary Development Studies (CETRAD), University of
Tras-os-Montes and Alto Douro (UTAD), Escola de Ciencias Humanas e
Sociais - Polo II, Vila Real 5000-801, Portugal.
(1) However, the causation may be in the opposite direction.
Countries running higher deficits are more likely to present optimistic
output growth forecasts to subvert fiscal rules (Milesi-Ferretti and
Moriyama, 2006; van den Noord, 2007).
(2) Morris et al (2006) claim that incentives to underestimate the
output gap in real time increased with the 2005 SGP reform, because the
budgetary effort is now dependent on the phase of the business cycle in
which the national economy finds itself.
(3) The analytical expressions of the variables are presented in
the Appendix.
(4) Throughout this paper, the lower index refers to the country
and the year to which the data correspond, whereas the upper index
indicates the date of publication. The forecasts are published usually
at the end of t-1, that is, just before the beginning of the year t to
which they refer, and estimates are published at the end of the year in
question and are subject to subsequent revision.
(5) When governments prepare their forecasts, they have private
information not available to international agencies. Merola and Perez
(2013) point out that the fiscal forecasts of the EC and OECD present
political biases in terms of governmental forecasts, and Gilbert and de
Jong (2014) show that for EMU countries the fiscal forecasts of the EC
are over-optimistic when the budget deficit threatens to exceed the
limit of 3% of GDP.
(6) The MAE and the RMSE are given by [MAE.sub.i] =
[[summation].sup.T.sub.t=1] [absolute value of [E.sub.i,t]]/T and
[RMSE.sub.i] = [square root of [[summation].sup.T.sub.t-1]
[E.sup.2.sub.i,t]/T], respectively, where E = [EB.sup.t,t-1.sub.i,t],
[EDebt.sup.t,t-1.sub.i,t], [Ey.sup.t,t-1.sub.i,t].
(7) Given the small number of observations per country, we use an
AR(1), as do Beetsma and Giuliodori (2010).
(8) Statistical results by country should be interpreted with
caution as the number of observations is limited and the higher
volatility of forecasting errors makes it less likely that the null
hypothesis of no bias is rejected.
(9) Regular elections and early elections are dummy variables,
taking the value 1 in the years in which elections occur regularly or
are called early, respectively. Elections are considered to be regular
when they occur at the end of the mandate or no more than 6 months
before its end. Otherwise, elections held outside of the normal
electoral calendar are considered early.
(10) After analyzing government revisions of previous estimates of
fiscal balance, Castro et at. (2013) concluded that the previously
published information is biased and non-efficient, and that budget
balance estimates have tended to be revised upwards.
(11) While Pina (2009) arrived at a similar conclusion, for Beetsma
et al. (2009) the variable proved to have no statistical significance.
(12) The results are available from the authors on request.
(13) Indices regarding multiannual budgetary frameworks, national
fiscal rules, independent fiscal institutions and independent forecasts
have been normalized to the [0,1] interval.
Table 1: Statistical properties of errors in
forecasted changes in budget balance (1999-2009)
[EB.sup.t, t-1, SCPs.sub.i, t]
ME MAE RMSE [alpha] [mu]
AT -0.06 0.40 0.48 -0.07 0.39
BE -0.48 0.59 0.90 -0.18 1.10
DE -0.20 0.74 0.87 -0.06 0.66 ***
DK -0.45 0.80 1.17 -0.41 0.95
EL -0.68 1.10 1.71 -0.58 0.96
ES -0.77 1.17 2.14 -0.64 0.77 ***
FI -0.31 1.09 1.60 -0.34 -0.47 *
FR -0.60 0.63 1.19 -0.53 0.42
IE -0.49 1.62 2.06 -0.53 0.28
IT -0.68 0.72 0.91 -0.52 * 0.40
LU -1.41 1.67 2.08 -1.19 ** 0.30 **
NL -0.35 0.66 1.07 -0.43 -0.24
PT -0.91 1.07 1.83 -1.15 -0.45
SE -0.08 0.81 1.11 -0.35 0.90
UK -0.89 1.09 1.43 -0.64 * 0.64 **
EU15 -0.56 0.94 1.46 -0.50 *** 0.44 ***
[EB.sup.t, t-1, EC.sub.i, t]
ME MAE RMSE [alpha] [mu]
AT -0.44 0.60 0.87 -0.39 0.25
BE -0.06 0.92 1.24 -0.31 -0.17
DE -0.24 1.05 1.21 -0.16 0.22
DK -0.17 0.81 0.98 -0.31 0.46
EL -0.46 1.01 1.41 -0.50 0.29
ES -0.59 0.85 1.36 -0.50 0.54
FI -0.21 1.21 1.66 -0.14 -0.63 *
FR -0.79 0.79 1.36 -0.56 0.65
IE -0.82 1.45 1.76 -0.88 0.19
IT -0.66 0.90 1.02 -0.51 0.36
LU -0.68 1.14 1.56 -0.75 0.14
NL -0.37 0.65 1.07 -0.43 0.11
PT -0.74 0.95 1.53 -0.96 -0.35 **
SE -0.43 0.97 1.28 -0.36 0.30
UK -0.04 1.09 1.58 -0.30 0.70
EU15 -0.45 0.96 1.35 -0.48 *** 0.27 **
Notes: *, **, and *** indicate statistical significance at the
10%, 5%, and 1% levels, respectively.
In columns [alpha] and [mu], the coefficients and the respective
level of statistical significance are presented (the robust
standard errors are omitted).
Table 2: Statistical properties of errors in
forecasted changes in public debt (1999-2009)
[EDebt.sup.t, t-1, SCPs.sub.i,t.]
ME MAE RMSE [alpha] [mu]
AT 0.57 1.39 1.92 0.87 -0.56 *
BE 1.67 2.25 3.37 1.43 0.18
DE 0.81 1.57 1.85 0.69 0.33
DK 0.44 2.47 3.70 0.68 0.81 ***
EL 2.45 2.51 4.16 1.70 * 0.63
ES 0.77 1.74 2.94 0.79 1.15 ***
FI 1.01 2.21 2.85 0.71 0.17
FR 1.47 1.93 2.76 0.78 0.86
IE 3.82 5.76 9.79 1.94 1.29 **
IT 2.02 2.24 2.70 2.15 ** 0.06
LU 0.69 1.44 2.66 0.74 0.02
NL 0.61 2.12 2.85 0.68 0.06
PT 2.20 2.44 3.28 2.31 * 0.06
SE 0.52 1.99 2.93 0.86 0.06
UK 1.79 2.28 4.05 1.15 0.92 ***
EU15 1.40 2.30 3.92 0.94 *** 0.70 ***
[EDebt.sup.t, t-1, EC.sub.i,t.]
ME MAE RMSE [alpha] [mu]
AT -0.46 1.57 2.19 -0.59 -0.20
BE 1.26 1.61 2.13 1.28 * 0.03
DE 0.93 1.44 1.66 0.86 0.09
DK -1.53 1.80 2.54 -0.77 0.86 *
EL 2.64 2.64 3.80 3.62 * -0.45
ES 0.43 1.26 1.92 0.46 0.99 ***
FI -0.28 2.14 2.58 -0.23 -0.14
FR 1.33 1.69 2.12 1.07 * 0.39
IE 0.46 2.50 3.44 0.67 -0.10
IT 2.08 2.48 2.83 1.57 * 0.21
LU 0.49 1.36 2.43 0.61 -0.03
NL 1.35 2.70 3.77 1.04 0.98
PT 0.67 1.87 2.48 0.85 0.00
SE 0.95 1.95 2.66 1.24 0.01
UK 0.66 2.17 2.96 0.46 0.45
EU15 0.74 1.96 2.71 0.66 *** 0.31 **
Notes: *, **, and *** indicate statistical significance at the
10%, 5%, and 1% levels, respectively.
In columns a and fi, the coefficients and the respective level
of statistical significance are presented (the robust standard
errors are omitted).
Table 3: Statistical properties of errors
in real GDP growth forecasts (1999-2009)
[Ey.sup.t, t-1, SCPs, SCPs.sub.i,t]
ME MAE RMSE [alpha] [mu]
AT -0.26 0.75 0.86 -0.25 -0.14
BE -0.37 0.90 0.96 -0.46 -0.42
DE -0.62 1.03 1.28 -0.61 0.07
DK -0.55 0.77 1.35 -0.40 0.84
EL -0.35 0.51 0.82 -0.27 0.63
ES -0.38 0.69 0.93 -0.29 0.55
FI -0.94 1.75 2.84 -1.03 -0.12
FR -0.70 0.93 1.20 -0.72 ** 0.14
IE -0.36 1.85 2.26 -0.56 0.23
IT -1.09 1.33 1.55 -1.03 * 0.06
LU -0.18 2.11 2.54 -0.37 0.22
NL -0.50 1.13 1.46 -0.59 0.01
PT -0.92 0.98 1.27 -1.15 ** -0.21
5E -0.67 1.53 2.25 -0.83 0.57
UK -0.45 0.88 1.28 -0.47 0.70
EU15 -0.56 1.14 1.64 -0.57 *** 0.19
[Ey.sup.t, t-1, EC, SPs.sub.i,t]
ME MAE RMSE [alpha] [mu]
AT -0.34 0.81 0.94 -0.33 -0.12
BE -0.29 0.85 0.92 -0.35 -0.43 *
DE -0.54 1.07 1.30 -0.55 0.00
DK -0.45 0.84 1.43 -0.46 0.27
EL -0.42 0.56 0.81 -0.34 0.36
ES -0.44 0.76 0.96 -0.32 0.53
FI -0.95 1.71 2.68 -1.03 0.01
FR -0.82 1.02 1.31 -0.87 ** 0.06
IE -0.67 1.87 2.30 -0.78 0.18
IT -1.05 1.33 1.51 -1.00 ** 0.03
LU -0.17 1.85 2.29 -0.37 0.26
NL -0.70 1.33 1.70 -0.85 0.00
PT -0.99 1.06 1.34 -1.33 ** -0.33
5E -0.67 1.62 2.2 -0.80 0.57
UK -0.4 0.88 1.24 -0.46 0.57
EU15 -0.60 1.17 1.63 -0.60 *** 0.18
[Ey.sup.t, t-1, EC, EC.sub.i,t]
ME MAE RMSE [alpha] [mu]
AT -0.65 1.12 1.56 -0.61 0.09
BE -0.58 1.11 1.33 -0.70 -0.37
DE -0.66 1.26 1.81 -0.67 -0.03
DK -0.56 0.95 1.53 -0.56 0.11
EL -0.34 0.61 1.14 -0.34 1.63
ES -0.40 0.82 1.26 -0.35 1.02
FI -1.03 1.85 2.87 -1.17 0.05
FR -0.59 0.86 1.07 -0.72 ** -0.19
IE -1.05 2.07 2.94 -0.94 0.36
IT -0.99 1.25 1.72 -0.90 ** 0.15
LU -0.45 1.95 2.29 -0.61 0.48
NL -0.82 1.40 1.90 -1.02 -0.21
PT -0.88 1.06 1.37 -1.05 ** -0.17
5E -0.56 1.29 1.77 -0.66 0.30
UK -0.53 0.87 1.27 -0.55 0.39
EU15 -0.67 1.23 1.81 -0.66 *** 0.23
Notes: *, **, and *** indicate statistical significance at the
10%, 5%, and 1% levels, respectively.
In columns [alpha] and [mu], the coefficients and the respective
level of statistical significance are presented (the robust
standard errors are omitted).
Table 4: Forecasted changes in budget balance and public debt
[DELTA][B.sup.t-1.sub.i,t]
A B C
[y.sup.t-1.sub.i,t] 0.392 * 0.478 *** 0.638 ***
(0.20) (0.15) (0.18)
[B.sup.t-1.sub.i,t-1] -0.255 ** -0.340 *** -0.355 ***
(0.10) (0.08) (0.06)
[Debt.sup.t-1.sub.i,t-1]
[SGPb.sup.t-1.sub.i,t-1] 0.199 **
(0.08)
[DiffECb.sup.t-1.sub.i,t-1] 0.446 ** 0.668 ***
(0.16) (0.10)
[DELTA][B.sup.t-2.sub.i,t] 0.355 *** 0.400 *** 0.320 ***
(0.12) (0.12) (0.13)
[gov_new.sup.i,t-1] 0.253 *** 0.265 ** 0.248 ***
(0.07) (0.10) (0.11)
[R.sup.2] 0.80 0.84 0.86
N 150 150 150
Country effects 4.44 *** 4.64 *** 5.48 ***
Time effects 10.24 *** 3.59 ** 8.69 ***
Hausman test 35.99 *** 35.97 *** 36.90 ***
[DELTA][Debt.sup.t-1.sub.i,t]
A B C
[y.sup.t-1.sub.i,t] -0.514 *** -0.670 **
(0.14) (0.27)
[B.sup.t-1.sub.i,t-1] -0.730 *** -0.759 *** -0.698 ***
(0.13) (0.11) (0.14)
[Debt.sup.t-1.sub.i,t-1] -0.079 *** -0.072 ***
(0.02) (0.02)
SGP[debt.sup.t-1.sub.i,t-1]
DiffEC[debt.sup.t-1.sub.i,t] -0.214 *** -0.692 *** -0.725 ***
(0.06) (0.13) (0.13)
[DELTA][Debt.sup.t-2.sub.i,t] 0.463 *** 0.468 * 0.547 **
(0.15) (0.22) (0.22)
[gov_party.sub.i,t-1] 0.170 **
(0.08)
[R.sup.2] 0.81 0.89 0.89
N 146 146 146
Country effects 2.41 * 2.63 ** 2.69 **
Time effects 45.73 *** 10.39 *** 4.23 ***
Hausman test 59.21 *** 37.69 *** 42.33 ***
Notes: *, **, and *** indicate statistical significance at the 10%,
5%, and 1% levels, respectively. As regards budget balance
forecasted changes, the models incorporate instrumental variables
with country and time fixed effects. The 'real GDP growth
forecasts' are instrumented by using the 'real GDP growth estimates
for the year t-1,' published at the end of year t-1 by national
governments (in Specification A) and by the EC (in Specifications B
and C). Panel data models with country and time fixed effects for
public debt forecasted changes do not include instrumental
variables. Standard errors corrected by the jackknife method are
given in parentheses.
Table 5: Errors in forecasted changes in budget balance and public debt
[EB.sub.t,t-1.sub.i,t]
A B C
[EB.sup.t-1, t-2i, t-1] 0.306 ***
(0.10)
[Ey.sup.t,t-1.sub.i,t] 0.490 *** 0.452 *** 0.349 **
(0.09) (0.10) (0.13)
[DELTA] [B.sup.t-1.sub.i,t] -0.247 ** -0.277 **
(0.10) (0.11)
[DSGPb.sup.t,t-1.sub.i,t-1] 1.079 * 0.959 *** 1.111 ***
(0.57) (0.24) (0.23)
[elect.sub.i,t] -0.418 *
(0.21)
[gov_party.sub.i,t] -0.135 **
(0.06)
[gov_gap.sub.i,t] -0.210 *
(0.10)
[gov_type.sub.i,t] -0.309 *
(0.16)
[gov_chan.sub.i,t] -0.383 ** -0.411 **
(0.15) (0.17)
[R.sup.2] 0.60 0.65 0.60
N 150 165 165
Country effects 1.9 1.74 1.13
Time effects 4.71 *** 25.56 *** 16.86 ***
Breush-Pagan LM test 0.00 0.93 0.06
Hausman test 55.92 *** 14.69 2.06
[EDebt.sub.t,t-1.sub.i,t]
A B C
[EDebt.sup.t-1, t-2.sub.i, t-1]
[Ey.sup.t, t-1.sub.i,t] -0.989 ** -0.710 ** -0.771 *
(0.41) (0.30) (0.37)
[DELTA] [Debtor.sup.t-1.sub.i,t] -0.319 *** -0.374 ***
(0.10) (0.10)
[DSGPdeb.sup.t,t-1.sub.i,t-1]
[snap.sub.i,t] 1.728 * 1.583 ** 1.264 *
(0.93) (0.73) (0.66)
[gov_new.sub.i,t] 0.679 * 0.837 *
(0.38) (0.46)
[R.sup.2] 0.58 0.49 0.50
N 162 162 162
Country effects 1.89 3.07 ** 3.12 **
Time effects 8.12 *** 4.30 *** 4.19 ***
Breush-Pagan LM test 4.61 ** 21.21 *** 21.03 ***
Hausman test 0.87 1.20 4.68
Notes: *, **, and * ** indicate statistical significance at the
10%, 5%, and 1% levels, respectively. The models are estimated
with country and time fixed effects. Standard errors corrected
by the jackknife method are given in parentheses.
Table 6: Forecasts and forecast errors of real GDP growth
[y.sup.t-1.sub.i,t]
A B
[y.sup.t-1.sub.i,t-1] 0.532 *** 0.522 ***
(0.06) (0.04)
[B.sup.t-1.sub.i,t-1]
[SGPb.sup.t-1.sub.i,t-1] -0.241 *** -0.196 ***
(0.06) (0.06)
[DffECty.sup.t-1 0.763 *** -0.131 **
.sub.i,t] (0.05) (0.05)
[gov_chan.sub.i,t-1] -0.123 * -0.101 *
(0.06) (0.05)
[R.sup.2] 0.94 0.91
N 165 165
Country effects 3.20 ** 4.86 ***
Time effects 42.26 *** 32.85 ***
Breush-Pagan LM 0.00 11.84 ***
test
Hausman test 80.16 *** 98.54 ***
[Ey.sup.t,t-1.sub.i,t]
A A' B' C'
[y.sup.t-1.sub.i,t]
[Dy.sup.t,t-1.sub.i,t-1] 0.833 * 0.818 *** 0.491 ** 0.358 *
(0.44) (0.24) (0.24) (0.21)
[DB.sup.t,t-1.sub.i,t-1]
DSGPb.sup.t,t-1
.sub.i,t-1]
[R.sup.2] 0.69 0.68 0.65 0.76
N 165 165 165 165
Country effects 0.78
Time effects 48.93 *** 14.14 *** 14.80 *** 21.41 ***
Breush-Pagan LM 0.00 0.00 0.00 0.00
test
Hausman test
Notes: *, **, and *** indicate statistical significance at the
10%, 5%, and 1% levels, respectively. The models are estimated
with country and time fixed effects in Columns A and B. Pooled
OLS models appear in columns A', B', and C'. Standard errors
corrected by the jackknife method are given in parentheses.
Table 7: Institutional variables
A B C
[v.sub.i] [DELTA][B.sup.t-1.sub.i,t]
[delegatiori.sub.i] -0.221 -0.545 * -0.436
(0.06) (0.23) (0.11)
[MTBF.sub.i] 0.747 *** 1.019 *** 1.226 ***
(0.44) (0.50) (0.56)
[rule.sub.i] 0.640 0.812 1.019
(0.18) (0.18) (0.22)
[institution.sub.i] 0.151 -0.025 0.079
(0.01) (0.00) (0.00)
[indfor.sub.i] 0.451 ** 0.551 ** 0.527 *
(0.07) (0.06) (0.05)
[v.sub.i] de [EB.sup.t,t-1.sub.i,t]
[delegation.sub.i] -0.297 0.240 0.116
(0.07) (0.11) (0.04)
[MTBF.sub.i] 0.912 *** 0.439 0.405 **
(0.43) (0.23) (0.33)
[rule.sub.i] 0.392 0.29 0.190
(0.04) (0.06) (0.04)
[institution.sub.i] -0.331 0.114 0.203
(0.03) (0.01) (0.05)
[indfor.sub.i] 0.150 0.270 * 0.419 ***
(0.01) (0.04) (0.16)
A B C
[v.sub.i] de [DELTA][Debt.sup.t-1.sub.i,t]
[delegatiori.sub.i] -1.418 *** 0.162 -0.027
(0.46) (0.00) (0.00)
[MTBF.sub.i] 0.580 0.415 0.144
(0.05) (0.01) (0.00)
[rule.sub.i] 1.350 -2.270 ** -2.277 **
(0.15) (0.22) (0.25)
[institution.sub.i] -0.788 -1.116 -1.174
(0.05) (0.06) (0.07)
[indfor.sub.i] -0.547 1.540 1.420
(0.02) (0.08) (0.08)
[v.sub.i] de [EDebt.sup.t,t-1.sub.i,t]
[delegation.sub.i] 0.112 0.969 * 1.088 **
(0.00) (0.24) (0.30)
[MTBF.sub.i] -1.331 -1.157 ** -1.203 **
(0.26) (0.22) (0.23)
[rule.sub.i] -2.061 * -1.420 -1.376
(0.35) (0.18) (0.17)
[institution.sub.i] -0.973 0.501 0.415
(0.08) (0.02) (0.02)
[indfor.sub.i] 0.183 0.334 0.060
(0.00) (0.01) (0.00)
A B C
[v.sub.i] de [y.sup.t-1.sub.i,t]
[delegatiori.sub.i] -0.102 -0.153
(0.03) (0.06)
[MTBF.sub.i] -0.493 ** -0.541 **
(0.48) (0.45)
[rule.sub.i] -0.533 ** -0.540 *
(0.31) (0.25)
[institution.sub.i] -0.421 * -0.443 *
(0.21) (0.18)
[indfor.sub.i] -0.313 * -0.324 *
(0.08) (0.07)
[v.sub.i] de [Ey.sup.t,t-1.sub.i,t]
[delegation.sub.i] -0.154 -0.142 0.158
(0.05) (0.04) (0.09)
[MTBF.sub.i] -0.393 -0.319 0.030
(0.19) (0.13) (0.00)
[rule.sub.i] -0.050 -0.017 0.191
(0.00) (0.00) (0.05)
[institution.sub.i] 0.030 0.003 0.260
(0.00) (0.00) (0.10)
[indfor.sub.i] -0.093 -0.017 0.028
(0.00) (0.00) (0.00)
Notes: *, **, and *** indicate statistical significance at the
10%, 5%, and 1% levels, respectively. Robust standard errors are
omitted. R2 values are given in parentheses. The number of
observations corresponds to the number of countries (15) in all
estimations.
