0
FRANCESCO LAGONA & FABIO PADOVANO*
Center for Economics of Institutions
and
Dipartimento di Istituzioni Politiche e Scienze Sociali,
Università Roma Tre
A NONLINEAR OPTIMAL SCORING ESTIMATE
OF THE RELATIONSHIP BETWEEN
BUDGET RULES AND FISCAL PERFORMANCE IN THE EUROPEAN UNION
ABSTRACT
Tests of the budget rules/fiscal performance relationship are metric-sensitive and arbitrary
in the evaluation of the stringency of the rules, aggregation of these evaluations in an index
and imposition of a linearly specified model. In this paper we propose a nonlinear principal
component analysis to solve these problems and evaluate the relative disciplinary power of
each rule. A battery of regressions on 1980-1999 optimally transformed data relative to 12 EU
countries confirms that more stringent rules reduce fiscal imbalances, but not budget size,
while increases in the degree of stringency of the rules are negatively correlated with public
expenditures growth.
JEL CLASSIFICATION CODES: H61, H62, C49
KEYWORDS: budget rules, fiscal performance, nonlinear principal components.
1
1. Introduction and literature review
Recent contributions to the literature on the determinants of public deficits focus on
budget procedures to explain the considerable cross country differences in fiscal performances
within highly interconnected and similarly developed economies, such as the OECD
countries, the Latin American countries and the U.S. States (Alesina and Perotti, 1994, 1996).
These contributions rest on the idea that democratic institutions allow policymakers to
partially internalize the political costs of their spending decisions, with consequent deficit bias
in financial choices (Buchanan and Wagner, 1977; Alesina and Perotti, 1994). Different
budget procedures, however, put similarly deficit-biased policymakers under different sets of
constraints. Budget outcomes should thus vary with the degree of stringency of these
constraints (von Hagen, 1991; von Hagen and Harden, 1995, 1996).
This literature has initially emphasized the effects of imposing numerical targets on fiscal
variables, such as budget deficit, public expenditures and debt. The provisions of the
Maastricht Treaty and the balanced budget rules adopted by almost all the U.S. states are
expression of this line of reasoning. Empirical tests of these theories have yielded mixed
results. Some studies find no significant correlation between measures of fiscal discipline and
single budget rules, such as veto power and balanced budget laws (Holtz-Eakin, 1988; Bunch,
1991). Others find that these rules “work”, provided that they are coordinated with other
constraints on fiscal discretion at other stages of the budgetary process (Poterba, 1994 and
1996; Alt and Lowry, 1994; Bayoumi and Eichengreen, 1995; Bohn and Inman, 1996;
Padovano, 1998). Otherwise, these rules prove not only ineffective, but generate incentives
for “creative accounting” and for a reduction of the transparency of the budget process
(Milesi-Ferretti, 1997).
These results led scholars to shift their attention from numerical to procedural budget
rules. These are the regulations that govern each stage of the budget process, from the
2
formulation of the budget proposal by the executive, to its discussion and approbation by the
legislature and to its final implementation (von Hagen, 1992; von Hagen and Harden, 1996;
Alesina and Perotti, 1996). These models generally predict that budget procedures lead to
greater fiscal discipline inasmuch as they strengthen the prerogative to the prime minister or
the finance minister over the spending ministers, limit universalism, reciprocity and
amendments in parliamentary budget sessions and constrain bureaucratic discretion in the
execution of the budget law (Baron, 1989, 1991; Baron and Ferejohn, 1989; von Hagen,
1992). Empirical analyses seem to lend support to these predictions: von Hagen (1992), de
Haan and Sturm (1994) and de Haan, Moessen and Volkerkink (1999) find indexes of
centralization of the budget process negatively correlated with budget deficits, government
debt and government expenditures in the EU countries. Alesina, Hausmann, Hommes and
Stein (1996) find similar results using a comprehensive sample of Latin American countries.
Though encouraging, these empirical studies suffer of several methodological
shortcomings that cast doubts on whatever finding they get. A first shortcoming arises from
the approximation of the degree of stringency of each rule. All studies capture this qualitative
dimension by assigning numerical evaluations to each rule (von Hagen, 1992; de Haan and
Sturm 1994; Alesina, Hausmann, Hommes and Stein, 1996; Padovano, 1998; de Haan,
Moessen and Volkerkink, 1999; and, among the official publications, ACIR, 1987). Although
these assignments are generally reasonable assessments of the rigorousness of each rule, this
procedure relies upon an arbitrary numerical coding of ordinal variables. Subsequent analyses
based on these variables are sensitive to the coding method chosen and thus produce spurious
results.
A second shortcoming derives from the aggregation of the rules. As the number of rules
that compose a budget process is usually very large, regressing fiscal variables on the entire
set of rules leads to highly parametric and unsatisfactory statistical models. To save degrees of
3
freedom, empirical studies estimate the rigorousness of the budget process as a whole; more
specifically, they construct some index that sums the numerical values assigned to each rule
of the process (von Hagen, 1992; de Haan and Sturm 1994; Alesina, Hausmann, Hommes and
Stein, 1996; Padovano, 1998). Implicitly, this means that the way a budget process constrains
fiscal choices can be represented by a linear additive function where all rules have equal
weight, i.e., they are perfect substitutes for each other in achieving the same degree of fiscal
discipline. It may be the case, however, that certain provisions restrict the choice set of fiscal
decision makers more than others. Yet, the arbitrariness and the high level of aggregation of
these indexes make it very difficult to learn the relative constraining power of single rules
within the budget process.
A third shortcoming is inherent to the form of the correspondence between budget rules to
fiscal outcomes. Theory (such as von Hagen, 1991) provides little guidance for the
specification of regression models in empirical tests. Most studies suppose that the degree of
stringency of a rule categorized in n variants increases linearly with the number of variants.
There are reasons to believe, however, that a nonlinear relationship is more appropriate. For
instance, Crain and Miller (1990) show that the general veto and the line item reduction veto
(the most and the least general form of veto power on a budget law) are less restrictive than
the intermediate form, the line item veto. This is a prima facie evidence of a U-shaped
relationship. These doubts lead Von Hagen (1992) and Padovano (1998) to use nonparametric
significance tests. While this approach is correct in principle, few studies adopt it
because it does not clearly expose the form of the relationship under test.
To improve on each of these shortcomings, our paper proposes a Nonlinear Principal
Component Analysis (hereafter, NLPCA) of the data set. NLPCA encloses a number of data
transformation procedures (see Kruskal and Shepard, 1974, Young, Takana and de Leeuw,
1978, Winsberg and Ramsay, 1983 and the extensive discussion in Gifi, 1990) which
4
generalize the standard principal component analysis to a method capable of both analyzing
qualitative data and reducing the number of qualitative exogenous variables for subsequent
use in the standard analysis of linear models. More precisely, NLPCA assigns to each country
a (small) number of scores that summarizes the degree of stringency of the overall budget
structure. This summary is the solution of an optimization process that: a) minimizes the loss
of explanatory power in the reduction process; b) keeps the ordinal properties of the
underlying data; c) yields evaluations of the degrees of stringency of the rules that are
invariant to monotone transformations, which implies that it is not sensitive to the interval
differences between the numerical evaluation of the data; d) highlights which rule has the
greatest disciplinary power; e) provides a non linear transformation of the independent
variables based on precise mathematical properties that yields the most appropriate
specification of the relationship between budget procedures and fiscal outcomes; f) permits a
quantitative summary of the budgetary changes occurred in each country during the study
period.
With these improvements in the specification of the estimates we reach three main results:
1) more stringent budget procedures limit deficit spending; 2) there is evidence of a nonlinear
relationship between budget rules and fiscal performance; 3) not all (sets of) budget rules
have the same disciplinary power with respect to a given indicator of fiscal performance. For
example, the negotiation of the budget proposal within the government seems to affect the
level of the deficit the most, while the regulation of the amendments to the budget proposals
and of the implementation of the budget bill appear the most important limit to the expansion
of the public outlays.
Lagona and Padovano (2000) is a first application of NLPCA to verify the relationship
between budget rules and fiscal performance in (some of) the EU countries during the 1980s.
Two are the crucial improvements of this paper with respect to Lagona and Padovano (2000).
5
First, we now dispose of a larger data set, which consider a greater variety of budget rules,
more countries and, most importantly, observations for two decades, the 1980s and the 1990s.
Second, the availability of two time periods allows us to check whether changes in the
stringency of the budget rules, which certain countries implemented between the 1980s and
the 1990s, have produced the predicted changes in the fiscal performance.
The rest of the paper is organized as follows. Section 2 illustrates the data set of the
budget rules underlying our analysis. Section 3 explains the motivations for implementing
NLPCA in this matter of inquiry. Specifically, Section 3.1 explains the shortcomings of the
approaches followed in the literature; Section 3.2 illustrates the main analytical properties of
NLPCA. The results of NLPCA are discussed in Section 4 and applied to the investigation of
the relationship between budget rules and fiscal performance in Section 5. Section 6
reassumes the main findings of our analysis.
2. Data description
2.1. Independent variables. Von Hagen (1992) still provides the most comprehensive and
coherent description and codification of the budget rules of a group of independent countries
characterized by homogenous and advanced economies, namely, 12 EU member countries in
the 1980s. De Haan, Moessen and Volkerkink (1999) extend and improve the data set to the
1990s, other rules and EU countries. Combining these sources we are able to base our
analysis on information about the budget procedures of Belgium, Denmark, France, Germany,
Greece, Ireland, Italy, the Netherlands, Portugal, Spain, Sweden and the United Kingdom for
the period 1980-1999.
The characteristics of each country’s budget procedures are reassumed according to: 1)
the internal organization of government; 2) the formulation of the budget proposal within the
government; 3) the discussion and approbation of the budget law in the parliament; 4) the
6
informativeness of the budget law; 5) the flexibility in the implementation of the budget law;
6) the stringency of long-term budget documents.
Each of these six “stages” is further disaggregated into several rules, up a total of 29.
Specifically, about the internal organization of government (stage O), the data set considers
how many government levels have fiscal power (variable O1), whether regional authorities
must balance the budget (O2), if they need the central government’s authorization to borrow
(O3), whether they are autonomous in planning their budget (O4), and how many ministries
participate in drafting the central government’s budget (O5). Information about the
formulation of the budget proposal (stage N) evidences whether it foresees a constraint on the
budget totals (variable N1), who has the power of setting the agenda (N2), if this power is
explicitly codified in the budget rules (N3) and what type of negotiations lead to the
formulation of the budget proposal (N4). Five characteristics of the discussion and
approbation of the budget by the legislative (stage P) are recorded: the parliament power to
amend the government proposal (variable P1), whether these amendments are required to be
offsetting (P2), if their approbation can cause the fall of the government (P3), whether the
parliament votes on the entire budget law or on its chapters (P4) and if the total budget size
must be voted on before or after the approbation of the single provisions (P5). The
informativeness of the budget law (stage I) is evaluated according to the inclusion of special
funds in the budget (variable I1), the existence of one or more budget documents (I2), the
transparency of the overall document (I3), the links made to national accounts (I4) and the
inclusion of government loans to non-government authorities (I5). The stage of the
implementation of the budget (called F) is disaggregated into six dimensions: the possibility
of the Minister of Finances to block expenditures (variable F1), the existence of cash limits
for spending ministers (F2), the requirement of the approbation of a controlling authority for
the disbursement of funds (F3), the possibility of transferring resources from one chapter to
7
another (F4), of changing the budget during its execution (F5) and of carrying unused funds
over to next year’s budget (F6). Finally, the information regarding long-term budget
documents (stage L) regards the type of fiscal variables targeted (L1), the length of the
planning horizon (L2), the forecasting method (L3) and the degree of commitment of the
long-term budget (L4). For a more detailed discussion of these variables, see von Hagen
(1992) and de Haan, Moessen and Volkerkink (1999).
These variables are given numerical values that increase in the degree of stringency of the
rule but vary for range and intervals. Table A.1 in Appendix A reassumes this information.
Table A.2 illustrates the rules adopted by each country.
Because of the severeness of the degrees of freedom problem, scholars fill the gaps in
information about the budget rule of each country by taking the average of the numerical
evaluation of the other rules of that same country (see von Hagen, 1992; de Haan and Sturm,
1994; Alesina, Hausmann, Hommes and Stein, 1996). These linear combinations, however,
reduce the true unobserved total variance of the data set. As in our data set the variance is of
considerable magnitude, this procedure appears inappropriate and likely to yield incorrect
parameter estimates. We instead choose to be as respectful as possible of the data.
Accordingly, we have supplemented the missing information with those published by OECD
(1987, 1995), at the cost of dropping Luxembourg from the data set, for which the OECD
sources proved unhelpful.
2.2 Dependent variables. Again to maximize the comparability of the results of our
approach with those of the literature, we choose as same dependent variables: 1) the ratio of
total budget deficit to GDP (measured as the difference between lending minus repayments),
called TDEF; 2) the ratio of primary deficit to GDP, PDEF; 3) the stock of the public debt
outstanding to GDP, DEB; 4) the ratio of the total public expenditures of the consolidated
government to GDP, EXP. We have also considered the ratio of total revenues to GDP, but it
8
proved multicollinear with total public expenditures, so we dropped this variable from the
analysis. All data on the dependent variables are form OECD Economic Outlook (1999).
Given the high persistence in the regressors, we take the average for the 1980s and the
1990s of the dependent variables as our regressands.
3. Regression on nonlinear optimal scores
3.1. Problems in summarizing budget rules. The general purpose of this analysis is to
test the alternative hypotheses of dependence between differences (cross countries and
between periods) in budget procedures and differences in fiscal performances, against the null
hypothesis of independence. To illustrate this matter in formal terms, let zist be the value of the
sth
variable in country i at time t, t=1,2 (e.g., the debt-to-GDP ratio), and let iktxˆ score the
degree of stringency of the kth
budget rule in country i (e.g., the veto power of the finance
minister). Furthermore, let zis= zis2 - zis1 and ikxˆ be (signed) measures of the change, from
period 1 to period 2, occurred in the ith
country with respect to the sth
fiscal variable and the
degree of stringency of the kth
budget procedure (the way we evaluate xik is the topic of
Section 3.2.) We consider two models:
M1: zist = f( iKtti xx ˆ...ˆ 1 ) + εist (1)
M2: zis = g( iKtti xx ˆ...ˆ 1 ) + ξis (2)
where f and g are monotone decreasing function of the changes in the budgetary process and
the errors ε and ξ are drawn from independent random variables with zero expectations. M1
tests the hypothesis that differences (cross country and between periods) in budget procedures
result in differences in fiscal performances; M2 tests the hypothesis that changes in the
budgetary processes determined changes of fiscal performance.
The efficiency of any statistical estimation of both M1 and M2 is an increasing
function of the degrees of freedom. In our analysis, the large number of the budget rules
9
relative to the total observations requires to restrict the available independent variables to a
smaller set of regressors. However, the standard methods for reducing the number of
independent variables present several problems. First, the high heterogeneity of budgetary
procedures in the EU countries in the two periods of our sample, shown in Table A2, does not
legitimate the deletion of regressors with low variability or/and with a high association with
other variables. NLPCA, instead, captures such heterogeneity. Secondly, clustering the
independent variables into homogeneous groups and conducting separate ANOVAs is also
unsatisfactory, in the absence of a theory that indicates how each rule influences fiscal
outcomes and, consequently, how to group the rules. Most analysts have resorted to indices
that summarize the information in the K variables, by assigning to each country i J<K linear
combinations ai of the x scores of the form
∑=
=
)1(
1
1
K
k
ikki xwa , …. , ∑=
=
)(
1
JK
k
ikkiJ xwa
where xik scores the stringency of the kth
budget rule in the ith
country, while K(j) is the
number of the budget variables clustered in the jth
groups, ΣjK(j)=K, and, finally, the wk are
weights that assess the relative importance of each rule in the budget process. In subsequent
regression analyses, the estimated coefficients on the ai regressors vary with the values of the
xs and the ws. The problem is that both the scores x and the weights w are a priori elicited,
i.e., they are not grounded on a theory or on an optimization problem that defines their
properties. In other words, the xs and the ws are essentially arbitrary transformations of the
data made by the analyst. As such, they are unsatisfactory. Furthermore, any coding of nonmetric
data, like those using rank numbers (von Hagen, 1992) or order statistics (Padovano,
1998), makes the estimates sensitive to the cardinal properties of the data, which in fact do
not exist. The data feature characteristics, or, more precisely, categories of which theory at
best suggests the ordinal ranking. Finally, in the empirical literature the prevailing weighting
schemes of the budget scores are binary, i.e., the weight is chosen to keep or exclude a
10
variable from the analysis (von Hagen, 1992; de Haan and Sturm, 1994; Alesina, Hausmann,
Hommes and Stein, 1996). Again the theoretical literature does not provide any explicit
indication to call these variables out.
It is to be noted, however, that if the xs were “truly” metric scores, a standard Principal
Components Analysis (hereafter, PCA) could be implemented in order to find optimal
weights and, therefore, an optimal summarization of the independent variables. This would
also yield insights on which variable carries the greatest explanatory potential, i.e., which rule
holds the greatest disciplinary power on the fiscal indicator under scrutiny. In this case the
problem would be to respect the ordinal nature of the data. We thus need a generalization of
PCA suitable for observations expressed in an ordinal scale.
3.2. Nonlinear optimal scoring of budget stringency. In formal terms, Table A2 is a
rectangular 2n×K matrix X, which contains the categories of K ordinal variables collected for
n countries during the 2 periods under consideration. The first n rows enter with the
stringencies relating to period 1981-90, while rows n+1 through 2n enter with the values
relating to the second period under consideration (1991-99). Hence, the (i,k)th
and (i+n,k)th
entries of matrix X are equal to the stringency of the kth
budget rule in country i during the
two periods.
11
Figure 1
To make an example, Figure 1a is a reduction of Table 2A to 2 countries (Belgium and
Sweden), 2 rules (T1 and T2) and 2 periods (the 1980s and the 1990s). We can denote Jk as
the number of the categories taken by the kth
rule (for instance, Jk = 4 if four degrees of
stringency are admissible) and let Ck be a 2n×Jk indicator matrix where the (i, j)th
entry is
equal to 1 if xik = j, and 0 otherwise. Merging these indicator matrices, a 2n×ΣkJk matrix C =
(C1…CK) is obtained (reported in Figure 1b), which is the connectivity matrix of a bipartite
graph where vertices on one side are countries, vertices on the other side are categories and
edges connect countries to budget categories (Figure 1c). Of course, it is always possible to
map this bipartite graph in a K-dimensional Euclidean space; categories would be drawn on
the K axes, countries become points with coordinates xik and edges are the perpendicular
projection segments of countries on these axes (Figure 1d).
12
At this point, however, several problems appear. First and least, the number of scores
for each country must be kept as small as possible, in order to save degrees of freedom for
both the models M1 and M2. At the same time, the resulting scatter plot would be
meaningless, since categories are ordinal, arbitrary scores and, accordingly, an Euclidean
distance between two points relating to the same country in the two different periods cannot
be used as a measure of the change of the stringency of the budgetary process occurred in that
country from period 1 to period 2.
Notice that constructing a dissimilarity index that takes in account the ordinal property
of the budget data could solve this distance problem. This would be helpful for M2 but not for
M1, for which a score xikt is needed for each combination of country, rule and time period.
Instead, the approach pursued here is based on finding a p-dimensional Euclidean space,
where p<<k, where countries and budget stringencies are positioned so to retain the maximum
amount of information from the original data, under the constraint of drawing this map in a
way that countries are close to their budget categories, and categories are close to the
countries that possess them. The selection of a dimension p<<k derives from the need to save
degrees of freedom for models M1 and M2. The choice of an Euclidean space stems from its
properties (projections, triangle inequality) that allow to measure the change in the budgetary
process of each country from one time period to another as the Euclidean distance between
two points. Accordingly, our dissimilarity index to be used in model M2 is simply a distance
between two optimal scores.
In formal terms, let X
~
denote any 2n×p matrix containing the p coordinates of the n
countries and let KYY
~
...
~
1 be any Jkt×p matrices containing the p coordinates of the budget
variables. We must find the Xˆ and KYY ˆ...ˆ
1 that minimize the average squared length of the
edge in the p-dimensional graph - our loss function
13
L ( X
~
, KYY
~
...
~
1 ) = 1/K ( )( )∑=




−
′
−
K
k
kkkktrace
1
~~~~
YCXYCX (3)
under the rank-one constraints
kkk λλλλ′= qY , k = 1 … K (4)
where qk is a Jk -column vector containing the category quantifications for the kth
budget rule
and λλλλk is a p-column vector of weights – the “component loadings”. In other terms, each
quantification matrix Yk is restricted to be one-dimensional, which implies that the category
quantifications become proportional to each other, after having been mapped onto the pdimensional
space. Since minimization of (3) is required over two matrices of variables, the
procedure we used is based on a reiterated application of an Alternating Least Squares
algorithm until convergence is reached. Briefly, each nth
step includes the following substeps:
S1) holding the X computed in the previous (n-1) step, computation of Y1
*
… YK
*
through the
minimization of (3); S2) holding the qk computed in the (n-1) step, computation of the
component loadings λλλλk
(n)
, imposing the rank-one restrictions (4) on the Yk ; S3) estimation of
the category quantifications q k
(n)
, using both the Y*
and the λλλλk
(n)
computed at substeps S1 and
S2, respectively; S4) account of the restrictions imposed by the ordinal measurement level of
the budget stringencies performing a weighted monotone regression in the metric Cj; S5)
computation of )()(
1
ˆ...ˆ n
K
n
YY , updating the Y1
*
… YK
*
by )(ˆ n
kY = qk
(n)
λλλλk
(n)
.
In particular, we emphasize that substep S4 takes into consideration the measurement
level of the variables under study (Gifi, 1990; de Leeuw and van Rijckevorsel, 1980), by that
making the score optimization invariant with respect to all the possible monotone
transformations of the stringencies. As a result, when convergence is reached, matrix Xˆ
contains optimal score transformations that: a) meet the proportionality requirements stated in
(2); b) minimize the loss of information due to the dimension reduction of data measured by
(1); c) are invariant for any preliminary, arbitrary numerical coding of the rigidity of budget
14
rules in matrix X. They are, however, sensitive to the choice of the dimension p of the space
over which the original country points are projected.
4. NLPCA estimates of budget rules rigidity
We have applied NLPCA to the data regarding all the stages of the budget process
reported in Table A2. The original K=29 dimension of the space of the budget rules has been
reduced to a p=2 space of their principal components (hereafter, PRIN1 and PRIN2). This
saves 27 degrees of freedom while keeping approximately 60% of the overall variance present
in the original data set, as shown by the first three Eigenvalues of the principal axes (Table 1).
The consideration of additional principal components would capture only 9% more of
variance in the dataset, at the cost of loosing the intuitive interpretation of a two-dimensional
projection of countries and of reducing the number of degrees of freedom available for the
subsequent analysis. Moreover, convergence of the ALS algorithm does not seem stable for
p≥3, an expected result, as high values of p increase the chance of several local stationary
points in the domain of (3). Conversely, p=2 drives (3) to a stable convergence after only 8
iterations. Hence, we choose to limit our analysis to p=2.
Table 1
Component Eigenvalue Cumulative
PRIN1 9.80 0.34
PRIN2 6.89 0.58
PRIN3 2.75 0.67
Graph B1 in Appendix B plots the original and the NLPCA monotone-transformed
evaluations of the categories. The non linear shape of many transformations shows that the
dependence structure between budget rules is nonlinear, i.e., an increase/decrease in the
degree of stringency of budget rule X in country A does not entail a proportional variation of
15
the degree of stringency of rule Y in the same country. Hence, running a linear regression of
(or applying any kind of linear estimator to) budget rules on indicators of fiscal performance
fails to capture the nonlinear structure of the explanatory variables. The likely and spurious
outcome of such estimates is low levels of statistical significance of the coefficients, like
those found by de Haan, Moessen and Volkerkink (1999).
The first result of the application of NLPCA to the original data is the biplot (Gabriel,
1981), reported in Graph 1. To interpret the biplot, one must keep in mind that it displays the
transformed data matrix Xˆ , which results after minimizing (3) and conducting an ordinary
PCA that leads to the identification of the first two principal components. More precisely, any
X
~
is approximated by the product PΛΛΛΛ−1/2
Q’, where the rows of the 24×2 matrix P are the
standardized PCA component scores (with respect to the first two principal components) and
ΛΛΛΛ−1/2
Q’ is the 2×29 structure matrix. Hence the (i,k)th
element of X
~
(i.e., the degree of
stringency of the kth
budget rule in the ith
country, after transformation) is approximated by the
inner product between the ith
row of P and the kth
row of QΛΛΛΛ−1/2
. As a result, in the biplot each
budget rule can be represented by a vector in the two dimensional plane spanned by PRIN1
and PRIN2, while each country can be identified by a point in that space.
Three types of information can be extracted from the biplot: 1) Just like any principal
component analysis, the coefficient of correlation between two variables, or between one
variable and one axis, is the cosine of the angle between the two arrows. Angles strictly wider
than 90° (evaluated counterclockwise) imply a negative correlation between the two variables,
while strictly narrower angles indicate a positive correlation. Angles at 180° indicate a
correlation coefficient of –1. Accordingly, the angle of the vector of each rule with respect to
the axes indicates which of the two principal component captures that rule. Since the two
principal components are orthogonal, a high correlation of a given rule with one principal
component implies a low correlation with the other. Thus, while one cannot say that one
16
principal component exclusively captures certain rules and the other exclusively the
remaining ones, it is legitimate to interpret one principal component as relatively more
correlated with the rules that the other component explains the less. 2) Since transformations
of categories are forced to be of rank one and ties are preserved, the distance from the origin
along the direction of the vector denotes increasing degrees of stringency of the rule
represented by that vector. 3) The orthogonal projection of a country point onto a rule vector
indicates how stringent is that rule in that country; e.g., the country that projects farthest along
the vector of a budget rule is the country with the highest degree of stringency with respect to
that rule. Similarly, the projection of the country point onto the axis of the two principal
components indicates how the country fares in terms of the groups of rules that each principal
component captures.
Graph 1 shows that PRIN1 captures the group of procedures N (structure of negotiations
within government), P (structure of parliamentary process), I (informativeness of the budget
draft) and most of the O (organization of general government). It thus can be taken as a
measure of the rigidity of the rules for elaborating and approving the budget document, from
its birth within the cabinet to its legification in the various government levels. PRIN2,
instead, best explains most of the F (flexibility in budget execution), that characterizes the
implementation of the budget law by the bureaucracy. Furthermore, most budget rules are
positively correlated to each other, while F6 (no carry-over of unused funds) and to a lesser
extent O4 (planning autonomy of local governments) present a negative correlation to the
majority of rules. This is evidence that countries with rigid budget procedures for what it
concerns, say, the negotiation of the budget proposal within the government and the
approbation of the budget law by the parliament tend to allow the possibility of crediting to
the next budget any resource not spent during the previous one and vice versa.
Coming to countries, France, Denmark and Sweden in the 1990s are those with the most
17
stringent budget rules, especially in terms of I3, P2, N1 and N2. The UK and the Netherlands
derive the constraining power from rules in set A and N3, i.e., mostly from tight control of
negotiations within the government, of the Parliament’s power to make amendments, as well
as from a good transparency of the budget document. Belgium and Greece seem to posses the
most idiosyncratic budget processes, as they are characterized by a high degree of stringency
of a few rules (especially O4 and F6) and a low degree of stringency in the other rules though
the situation has become relatively more equilibrated in Belgium during the 1990s.
Budget reforms in the 1990s produced a remarkable increase of disciplinary power in
Sweden, a moderate one in Italy, Ireland and Belgium, and little changes in the other
countries.
Graph 1
Biplot of budget stringencies after NLPCA transformation
Note: Label of variables indicate the direction of the corresponding vector; A stands for variables I1,I2,
I4, I5, P3-P5, N4, L3. Countries that reformed their budget approbation processes between the 1980s
and the 1990s have their acronym followed by the decade, while those which have not undergone a
process of reform have the same coordinates in the two decades and are thus indicated only by their
acronym.
18
5. Budget rules and fiscal performance
We are now in a position to test the relationship between fiscal performance of the 12 EU
countries of our sample and the estimated principal components of their budget procedures.
The analysis is carried out in two steps: first we insert the four indicators of fiscal
performance (public debt outstanding, total and primary public deficit and total public
outlays) in the space of the principal components to illustrate the relationship between the
dependent variables and each budget rule. By that we learn which rule carries the greatest
disciplinary power with respect to each fiscal variable and how the rules interplay with each
other in constraining the dynamics of the fiscal variable. To corroborate these findings, in the
second step of the analysis we estimate a regression model for each dependent variable in
their levels (model M1) and we check whether changes in budget rules have resulted in such
changes of fiscal performance as theory predicts (model M2).
5.1. Diagrammatic analysis. To detect nonlinear relations between variables of fiscal
performance and budget stringency, we have repeated NLPCA merging matrix X with the
column of the scores of the dependent variables, taking one at a time. The biplots obtained are
Graph 2 through 5. These plots cannot be overlapped, as the vectors of the independent
variables rotate to match the best two-dimensional projection after entering each fiscal
variable.
The results are strikingly different between financial variables (namely, total debt
outstanding, total and primary deficit) on the one hand, and the public expenditure variable,
on the other. As far as the financial variables are concerned, Graph 2 to 4 show that 15 out of
29 budget rules of our data set lie in the opposite half plane to that where the DEB, TDEF and
PDEF are evidence of a negative correlation between these rules and each financial variable
under scrutiny. This shows that a) budget rules do have a constraining power over financial
decisions, i.e., decisions on how to finance a given level of expenditures; b) they exert such
19
power conjointly, i.e., it is the stringency of a set of rules that produces fiscal discipline, rather
than a single rule in isolation. With respect to DEB, TDEF and PDEF this set of rules is
remarkably stable, as it is invariantly composed by O2; N1, N2, N4; P2, P3, P4, P5; I1, I2, I3,
I5; F2, F3, F4; L3. The rules with the greatest disciplinary power with respect to public debt
set a constraint on fiscal totals already in the budget proposal (N1), provide that amendments
be offsetting in order to be approved (P2) and regard the overall transparency of the budget
document (I3). This is a very plausible result, since it indicates that fiscal disequilibria tend to
increase when a) their variations are not made explicit from the beginning of the budget
process, i.e., from the proposal of the new budget; b) there are no institutional structures on
the parliamentary approbation of the budget that prevent the possibility of cycling majorities;
c) obfuscated budget documents reduce the controlling power of the taxpayer-voter. As for
total and primary deficits, the most stringent rules are, again, those that determine the
difference between fiscal totals in the budget proposal (N1) and in the budgets of the local
governments (O2), as well as those that increase the agenda setting power of the Finance
Minister (N2). These results are consistent with the predictions of the theoretical literature
(von Hagen, 1992; Baron, 1989, 1991; Baron and Ferejohn, 1989); however, our analysis
provides a much more precise corroboration of theory than those available so far, as the
biplots reveal the influence of single rules, rather than aggregated indexes.
Budget rules, instead, show a very limited constraining potential of the level of public
expenditures. Graph 5 shows that most of the rules lie in the same half-plane of the EXP
vector. The implication is that there is either no or a positive correlation between the
stringency of the budget rules and the level of public expenditures. This comes as no surprise,
since entitlements, such as health care, social security and unemployment insurance programs,
represent a high and a growing share of total budget outlays in our sample (Alesina and
Perotti, 1994). Entitlement programs are not subject to periodic revision in the budget
20
Graph 2
Biplot of budget rules, countries and DEB after NLPCA transformation
Note: Label of variables indicate the direction of the corresponding vector; A stands for variables I1, I5, P3P5,
F3, L3.
Graph 3
Biplot of budget rules, countries and TDEF after NLPCA transformation
Note: vector of the variable of fiscal performance is in bold; labels of variable indicate the direction
of the corresponding vector; A stands for variables I1, I5, P3-P5, F3.
21
Graph 4
Biplot of budget rules, countries and PDEF after NLPCA transformation
Note: vector of the variable of fiscal performance is in bold; labels of variable indicate the direction of the
corresponding vector; A stands for variables N4, I1, I2, I5, P3-P5, L3.
Graph 5
Biplot of budget rules, countries and EXP after NLPCA transformation
Note: vector of the variable of fiscal performance is in bold; labels of variable indicate the direction
of the corresponding vector; A stands for variables O4, I3, P2; B for variables O2, N1, N2, P4, I2, I5.
22
sessions, where the issue to be decided upon is only how to finance the disbursements that
they predetermine. The administration of these programs, however, is characterized by a
certain degree of discretion (Emerson, 1988; Alesina, Danninger and Rostagno, 1999);
interestingly, the two budget rules that approximate a –1 correlation coefficient with EXP are
those that limit bureaucratic discretion in the administration of the budget, namely F5
(changes in budget law during execution) and F6 (carry-over of unused funds to next year).
5.2. Regression results. We are now in a position to estimate models M1 and M2. As
we have argued in Section 4, in order to properly estimate the relation between indicators of
fiscal performance and NLPCA transformed values of the stringency of budget rules, we need
to use the NLPCA transformed values of the dependent variables as well. Graph b2
(Appendix B) illustrates how NLPCA transformed DEB, PDEF, TDEF and EXP.
These transformations take the form of a step function, since NLPCA clusters in a few
groups the original continuous observations of the indicators of fiscal performance of the
countries in our sample.
Given the small number of available observations (n=24), the set of independent
variables is limited to at most two regressors and the intercept. It is important to bear in mind
that, while our regression model is a standard OLS, the variables included in the estimating
equation are the optimally transformed scores of the original data on budget rules and fiscal
indicators. The fact that these transformations are non linear (as shown in Appendix B)
confirms that the actual correspondence between budget outcomes and budget rules is
nonlinear.
Table 2 reports the regression results for M1, testing the hypothesis that countries with
lax budget approbation procedures tend to have high fiscal imbalances. The results of the
estimates confirm what already found in the diagrammatic analysis. DEB, TDEF and PDEF
are negatively and significantly correlated with PRIN1, the principal component that best
23
captures the great majority of budget rules, except most of the F, on the flexibility allowed in
the implementation of the budget. In line with theory, the rules captured in PRIN1 are
strongly and negatively correlated with financial disequilibria. Conversely, the absence of
statistically significant coefficients on PRIN2 with respect to DEB, TDEF and PDEF imply
that a low stringency in the F rules does not affect the choice of how to finance a given level
of outlays. Rather, as the positive and significant sign of PRIN2 on EXP shows, it tends to
increase the volume of these outlays. However, the positive and significant coefficients of
PRIN1 and PRIN2 on EXP indicate that countries with more stringent rules are not
characterized by a lower share of public expenditures on GDP.
Table 2
ESTIMATES OF MODEL 1
Dependent variable
DEB TDEF PDEF EXP
C 0.7183***
(0.0432)
0.0463***
(0.004)
0.0941**
(0.0419)
0.4953***
(0.0109)
PRIN1 -0.1328***
(0.0442)
-0.0233***
(0.004)
-0.189***
(0.0428)
0.4953***
(0.0109)
PRIN2 -0.0678
(0.0442)
0.0065
(0.004)
-0.0249
(0.0428)
0.0358***
(0.0112)
F stat. 5.7
(0.0105)
17.89
(0.0000)
9.92
(0.0009)
11.66
(0.0004)
Note: Coefficient (Standard Error). F-statistics (p-value). ***
, **
denote a 1% and a 5% significance level,
respectively.
Model 2, instead, tests the hypothesis that an increase of the degree of stringency of
budget rules from one decade to another in the same country produces a reduction of fiscal
imbalances, and vice versa. It is the dynamic version of the hypothesis expressed in model 1.
To test it, we introduce the variable DIST in the model, that measures the distance between
the positions of the country points in the 1980s and in the 1990s in the space spanned by the
two principal components (Graph 1). The dependent variables are measured in the same way,
and are indicated as !DEB, !TDEF, !PDEF, !EXP. Given the small number of
24
observations, we exclude the outliers from the sample. Table 3 reports the outcomes of the
regression.
Table 3
ESTIMATES OF MODEL 2
Dependent variable
!DEB†
!TDEF††
!PDEF††
!EXP†
C 14.2399
(8.0377)
-0.3388
(0.7594)
-0.9035**
(1.0434)
1.6008
(1.5288)
DIST -14.6624
(20.2079)
-4.4241***
(1.2385)
-4.5887**
(1.8465)
-9.7891**
(3.9)
F stat. 0.53
(0.4888)
17.89
(0.006)
6.18
(0.0347)
6.3
(0.0364)
Note: Coefficient (Standard Error). F-statistics (p-value). ***
, **
denote a 1% and a 5% significance level,
respectively. (†)
Sweden and Italy excluded from the sample. (††)
Sweden excluded from the sample.
The most interesting result is the negative and statistically significant coefficient of DIST
on EXP. The combination of this result with those of Model 1 suggests that, while countries
with more stringent budget rules are not those with a lower expenditure to GDP ratio, there is
evidence that an increase of the stringency of these rules during the two decades yields a
slower growth of the public sector with respect to the whole economy. Therefore, while the
data do not seem to support the static version of the theory, they do for its dynamic
formulation. As for the other variables, the signs of the coefficients are negative, as expected,
although the coefficient on DEB is not statistically significant. As DEB is a stock variable, it
might take more than time elapsed after the budget reforms to obtain a change in the debt to
GDP ratio large enough to make the estimated coefficient statistically significant. However,
the negative and strongly significant coefficients on PDEF and TDEF are a reassuring
evidence that changes in the stringency of the budget rules reduce policymakers’ tendency to
finance public expenditures via debt.
25
6. Concluding remarks
The NLPCA approach followed in this paper provides results that enhance those of the
current literature along several dimensions. First, NLPCA bases the synthesis of the
evaluations of the single rules into aggregate variable(s) on precise mathematical properties.
Specifically, the number and the specification of these variables is optimized so as to
minimize the loss of explanatory power in the reduction process. Conversely, the approach
followed so far in the literature – the simple sum of the evaluations of the rules into one or
more indexes – is not based on an optimization process. It thus carries a greater loss of
information, ceteris paribus, and does not allow to know whether one single index, or more,
is the most reasonable synthesis. Second, NLPCA keeps the ordinal properties of the indexes
of the previous studies – the acceptable contribution of the numerical evaluations of budget
rules - but yields results that are invariant to monotone transformations of the numerical
coding of the rules. The other indexes, instead, are metric sensitive and are thus unsuitable for
regression analysis. Finally, with NLPCA we know the table of correspondences between the
single rules and the principal components. Knowing these correspondences and the
coefficients linking each principal components to the dependent variable, we can infer which
budget procedure carries the greatest disciplinary power on each indicator of budget
performance. In line with theory, with respect to the financial variables (the ratios of debt,
primary and total deficit to GDP) these are the ones that increase the agenda power of the
finance minister in the negotiations within the government that lead to budget proposal, that
increase the overall transparency of the budget bill and constrain the possibility for local
governments to finance their expenditures in deficit. Theory receives only a partial support
from the data with respect to the possibility that stringent budget rules place a binding
constraint on public expenditures. Our analysis shows that countries with more stringent rules
are not characterized by a smaller public sector compared to the size of the economy;
26
however, increases in the degree of stringency of the budget rules do produce a slower growth
of the expenditures to GDP ratio.
Finally, our analysis suggests two lines of refinement of the theory on the relationship
between budget rules and fiscal performance. First, the shape of the NLPCA transformations
of the original data on the degree of stringency of the budget rules suggests that their
disciplinary power often increases in a non linear way, and that the relationship between these
rules and the indicators of budget outcomes is also non linear. Second, we have pointed out
that different rules have different constraining powers with respect to different indicators of
budget performance; we thus need more disaggregated theoretical formulations of the
relationship between budget rules and fiscal performance.
References
- Advisory Commission on Intergovernmental Relations (ACIR) (1987), Fiscal Discipline
in the Federal System: Experience from the States, Washington D. C., ACIR.
- Alesina, A., Danninger, S. and Rostagno, M. (1999), “Redistribution Through Public
Employment: The Case of Italy”, Working Paper of the International Monetary Fund.
- Alesina A., Hausmann, R., Hommes, R. and Stein, E. (1996), “Budget Institutions and
Fiscal Performance in Latin America”, NBER Working Paper, n. 5586.
- Alesina, A. and Perotti, R. (1994), “The Political Economy of Budget Deficits”, IMF
Working Paper n. 94/85.
- Alesina, A. and Perotti, R. (1996), “Budget Deficits and Budget Institutions”, NBER
Working Paper n. 5556.
- Alt, J. and Lowry, R. (1994), “Divided Government and Budget Deficits: Evidence from
the States”, American Political Science Review 88, 811-828.
- Baron, D. (1989), “A Non-Cooperative theory of Legislative Coalitions”, American
Journal of Political Science 35, 1048-1084.
- Baron, D. (1991), “Majoritarian Incentives, Pork Barrel Programs and Procedural
Control”, American Journal of Political Science 35, 57-90.
- Baron D. and Ferejohn, J. (1989), “Bargaining in Legislatures”, American Political
Science Review 83, 1181-1206.
27
- Bayoumi, T. and Eichengreen, B. (1995), “Restraining Yourself: The Implications of
Fiscal Rules for Economic Stabilization”, IMF Staff Papers 42, pp. 32-48.
- Bohn, H. and Inman, R. P. (1996), “Balanced Budget Rules and Public Deficits: Evidence
from the U.S. States”, NBER Working Paper n. 5533.
- Buchanan, J. M. and Wagner, R. E. (1977), Democracy in Deficit: the Political Legacy of
Lord Keynes. New York, Academic Press.
- Bunch, B. S. (1991), “The Effect of Constitutional Debt Limits on State Governments’
Use of Public Authorities”, Public Choice 68, 57-69.
- Crain W. M. and Miller, J. C. III. (1990), “Budget Process and Spending Growth”,
William and Mary Law Review, 1021-1046.
- de Leeuw, J. (1977), “Correctness of Kruskal’s Algorithms for Monotone Regression with
Ties”, Psychometrika 42, 141-144.
- de Leeuw, J. and Van Rijckevorsel, J.L.A. (1980), “HOMALS and PRINCALS: some
generalizations of Principal Component analysis”, in: Diday et al. (eds.) Data Analysis
and Informatics. Amsterdam: North-Holland.
- - Emerson, M. (1988), What Model for Europe?, Cambridge, MIT Press.
- Gabriel, K. R. (1981), “Biplot Display of Multivariate Matrices for Inspection of Data
and Diagnosis”, in Barnett V. (ed.), Interpreting Multivariate Data, London, Wiley.
- Gifi, A. (1990). Nonlinear Multivariate Analysis. Wiley, New York.
- De Haan, J. Moessen, W. and Volkerkink, B. (1999), “Budgetary Institutions-Aspects and
Changes: New Evidence for Some European Countries”, in: Poterba, J. M. and de Haan, J.
Fiscal Institutions and Fiscal Performance, Chicago, University of Chicago Press, 265-
299.
- de Haan, J. and Sturm, J. E. (1994), “Political and Institutional Determinants of Fiscal
Policy in the European Community”, Public Choice 80, 157-172.
- von Hagen, J. (1991), “A Note on the Empirical Effectiveness of Formal Fiscal
Restraints”, Journal of Public Economics, 99-110.
- von Hagen, J. (1992), “Budgeting Procedures and Fiscal Performance in the European
Communities”, Economic Papers n. 96.
- von Hagen, J. and Harden, I. J. (1996), “Budget Processes and Commitment to Fiscal
Discipline”, IMF Working Paper n. 96/78.
- Holtz-Eakin, D. (1988), “The Line Item Veto and Public Sector Budgets: Evidence from
the States”, Journal of Public Economics 36, 269-292.
28
- Kruskal, J.B. and R.N. Shepard (1974), “A Nonmetric Variety of Linear Factor Analysis”
Psychometrika 39, 123-157.
- Lagona F., and Padovano, F. (2000), “Budget Rules and Fiscal Performance in the
European Union: A Nonmetric Principal Component Analysis”, in G. Sobbrio e V.
Dardanoni (eds.), Istituzioni Politiche e Finanza Pubblica , Milano, Franco Angeli: 126-
155.
- Milesi-Ferretti, G. M. (1997), “Fiscal Rules and the Budget Process”, CEPR Discussion
Paper Series n. 1664.
- Organization for Economic Cooperation and Development (1987), The Control and
Management of Government Expenditure. Paris, OECD.
- Organization for Economic Cooperation and Development (1995), Budgeting for Results:
Perspectives on Public Expenditure Management, Paris, OECD.
- Organization for Economic Cooperation and Development (1999), Economic Outlook,
Paris, OECD.
- Padovano, Fabio (1998), “Flessibilità e Programmazione Pluriennale nelle Procedure di
Approvazione del Bilancio e Performance di Finanza Pubblica dei Paesi OCSE”, in F.
Forte and S. Lo Faso (Eds.), “Le Regole della Costituzione Fiscale”, Notizie di Politeia,
49-50, 106-129.
- Poterba, J. M. (1994), “State Responses to Fiscal Crises: the Effects of Budgetary
Institutions and Politics”, Journal of Political Economy 102, 799-821.
- Poterba, J. M. (1996), “Do Budget Rules Work?”, NBER Working Paper n. 5550.
- Vinod, H.D. and Hullah, A., (1981) Recent Advances in Regression Methods, New York,
Marcel Dekker.
- Winsberg, S. and Ramsay, J. O. (1983), “Monotone Spline Transformations for
Dimension Reduction” Psichometrika 48, 575-595.
- Young, F.W., Takana Y. and de Leeuw, J. (1978), “The Principal Components of Mixed
Measurement Level Multivariate Data: An Alternating Least Squares Method with
Optimal Scaling Features”, Psychometrika 43, 279-281.
APPENDIX A
Table A.1
Numerical evaluations of budget rules
Stage Symbol Rule Range 0 1 2 3
O1 N. of government levels 0-3 4 + social security 3 + social security 2 + social security 2
O2 Local government balances
the budget
0-3 No requirement Requirement exists
but is not
considered binding
Requirement is
binding
Defici
exceed
expend
O3 Local government needs
authorization to borrow
0-1 No authorization is
required
Authorization from
higher government
level is required
O4 Local government has
planning autonomy
0-3 Lower level
governments are
autonomous
Lower level
governments may
be placed under the
surveillance of
higher level
governments
Lower levels
governments have
limited autonomy
Lower
govern
no aut
Organization of
general
government
O5 N. of ministries involved in
draft of overall balance
0-2 1 2 3
N1 Application of a constraint
on fiscal totals
0-4 None Debt to GDP Debt to GDP and
Deficit to GDP
Expen
GDP o
N2 Agenda setting for budget
negotiations
0-4 Minister of
Finances or cabinet
collects bids from
spending ministers
Minister of
Finances or cabinet
collects bids
subject to
guidelines
Cabinet decides on
budget norms first
Minist
propos
norms
on by
N3 Scope of budget norms in
the setting of the agenda
0-3 Expenditure or
deficit
Specific Broad and specific Broad
Structure of
negotiations
within government
N4 Structure of negotiations 0-2 All cabinet
members involved
together
Multilateral Bilateral Betwe
minist
Table A.1 (continued)
Stage Symbol Rule Range 0 1 2 3
P1 Amendments 0-1 Unlimited Limited
P2 Amendments required to be
offsetting
0-1 No Yes
P3 Amendments can cause the
fall of government
0-1 No Yes
P4 All expenditures passed in
one vote
0-2 Yes Mixed Votes chapter by
chapter
Structure of
parliamentary
process
P5 Global vote on total budget
size
0-1 Final only Initial
I1 Special funds included 0-4 No Some Most Yes, b
budge
I2 Budget submitted in one
document
0-2 No Recently yes Yes
I3 Assessment of budget
transparency by respondents
0-2 Hardly transparent Not fully
transparent
Fully transparent
I4 Link to national accounts 0-3 Not provided Possible Provided in
separate
Direct
Informativeness of
the budget draft
I5 Government loans to nongovernment
entities
included in budget draft
0-2 No Reported in
separate document
Yes
Table A.1 (continued)
Stage Symbol Rule Range 0 1 2 3
F1 Minister of Finances can
block expenditures
0-1 No Yes
F2 Spending ministers subject
to cash limits
0-1 No Yes
F3 Disbursement approval
required from Minister of
Finances or controller
0-1 No Yes
F4 Transfers of expenditures
between chapters
0-5 Unrestricted Limited Require consent of
Minister of
Finances
Requir
parliam
Flexibility of
budget execution
F5 Change in budget law
during execution
0-4 At discretion of
government
By new law which
is regularly
submitted during
fiscal year
At discretion of
Minister of
Finances
Requir
Minist
and pa
F6 Carry-over of unused funds
to next year
0-3 Unrestricted Limited Limited and
requires
authorization of
Minister of
Finances and
parliament
Not po
L1 Target variable 0-2 None Expenditures or
revenues
Total budget size
L2 Planning horizon (years) 0-4 One Two Three Four
L3 Forecasting method 0-3 Ad hoc forecast Fixed forecast Updated forecast,
but not based on
consistent
macromodel
Updat
consis
macro
Long-term
planning constraint
L4 Degree of commitment 0-4 None Internal orientation Indicative Weak
Source: von Hagen (1992); De Haan Moessen and Volkerkink (1999).
Table A.2
Summary of budget rules in the EU countries
Stage Symbol Rule Range Decade 0 1 2
1980s FRA, SPA BEL, DEN,
GER, ITA, NET,
SWE
GRE, PORO1 N. of
government
levels
0-3
1990s FRA, SPA BEL, DEN,
GER, ITA, NET
GRE, POR,
SWE
1980s IRL, ITA, POR, SPA,
SWE
BEL, GRE DEN, FRAO2 Local
government
balances the
budget
0-3
1990s IRL, ITA, POR,SPA GRE DEN, FRA,
1980s GER, IRL, NET,
POR, SPA, SWE, UK
BEL, DEN, FRA,
GRA, ITA
O3 Local
government
needs
authorization
to borrow
0-1
1990s GER, IRL, NET,
POR, SPA, UK
BEL, DEN, FRA,
GRE, ITA, SWE
1980s IRL, POR, SPA, SWE BEL, DEN,
GER, NET
FRA, ITA,
UK
O4 Local
government
has planning
autonomy
0-3
1990s POR, SPA, SWE BEL, DEN,
GER, NET
FRA, IRL,
ITA, UK
1980s BEL, GRE, FRA, GER, IRL,
NET, SPA, SWE,
UK
DEN, ITA,
POR
Organization of
general government
O5 N. of ministries
involved in
draft of overall
balance
0-2
1990s BEL, GRE FRA, GER, IRL,
NET, SPA, SWE,
UK
DEN, ITA,
POR,
Table A.2 (continued)
Stage Symbol Rule Range Decade 0 1 2
1980s BEL, GRE, SPA,
SWE
NET, POR IRL, ITAN1 Application of
a constraint on
fiscal totals
0-4
1990s GRE, SPA NET, POR BEL, IRL,
ITA
1980s BEL, GER, GRE,
IRL, ITA, SWE
POR, SPA DEN, NET,
UK
N2 Agenda setting
for budget
negotiations
0-4
1990s BEL, GER, GRE,
IRL, ITA
POR, SPA DEN, UK
1980s GRE, IRL, SWE BEL, DEN, ITA, NET,
POR
N3 Scope of
budget norms
in the setting of
the agenda
0-3
1990s GRE, IRL BEL, DEN ITA, NET,
POR
1980s BEL, GRE, IRL, SPA ITA DEN, FRA,
GER, NET,
POR, SWE,
UK
Structure of
negotiations within
government
N4 Structure of
negotiations
0-2
1990s GRE, SPA BEL, DEN,
FRA, GER,
IRL, ITA,
NET, POR,
SWE, UK
Table A.2 (continued)
Stage Symbol Rule Range Decade 0 1 2
1980s BEL, DEN, GER,
GRE, POR, SWE
FRA, IRL, ITA,
NET, SPA, UK
P1 Amendments 0-1
1990s DEN, GER, GRE,
POR, SWE
BEL, FRA, IRL,
ITA, NET, SPA,
UK
1980s BEL, GER, GRE,
IRL, ITA, NET, POR,
SPA, SWE, UK
DEN, FRAP2 Amendments
required to be
offsetting
0-1
1990s BEL, GER, GRE,
NET, POR, SPA, UK
DEN, FRA, IRL,
ITA, SWE
1980s BEL, ITA, SPA DEN, FRA,
GER, IRL, NET,
POR, SWE, UK
P3 Amendments
can cause the
fall of
government
0-1
1990s BEL, GRE, POR DEN, FRA,
GER, IRL, ITA,
NET, POR,
SWE, UK
1980s BEL, GER, GRE,
IRL, POR, SPA
FRA, ITA DEN, NET,
SPA, UK
P4 All
expenditures
passed in one
vote
0-2
1990s GER, GRE, POR,
SPA
FRA, ITA BEL, DEN,
IRL, NET,
SWE, UK
1980s BEL, DEN, GER,
GRE, IRL, ITA, POR,
SPA, SWE
NET FRA, UK
Structure of
parliamentary process
P5 Global vote on
total budget
size
0-1
1990s BEL, DEN, GER,
GRE, IRL, POR, SPA
NET FRA, ITA,
SWE, UK
Table A.2 (continued)
Stage Symbol Rule Range Decade 0 1 2
1980s POR IRL, ITA, SWE BEL, DEN,I1
Special funds
included
0-4
1990s POR IRL DEN
1980s BEL, GRE, IRL, ITA,
SWE, UK
DEN, FRA,
GER, NET, POR,
SPA
I2 Budget
submitted in
one document
0-2
1990s GRE, IRL, ITA, UK BEL, DEN, IRL,
NET, POR, SPA
1980s ITA BEL, DEN, IRL,
NET, POR, SPA,
SWE
FRA, GER,
GRE, UK
I3 Assessment of
budget
transparency
by respondents
0-2
1990s ITA BEL, DEN, IRL,
NET, POR, SPA
FRA, GER,
GRE, SWE
UK
1980s BEL, IRL, ITA, SWE DEN, GRE,
POR, SPA
FRAI4 Link to
national
accounts
0-3
1990s ITA DEN, GRE, POR BEL, FRA,
SWE
1980s ITA, POR GER, GRE, IRL,
SWE
BEL, DEN,
FRA, NET,
SPA, UK
Informativeness of
the budget draft
I5 Government
loans to non-
government
entities
included in
budget draft
0-2
1990s POR GER, GRE, IRL BEL, DEN,
FRA, ITA,
NET, SPA,
SWE, UK
Table A.2 (continued)
Stage Symbol Rule Range Decade 0 1 2
1980s BEL, DEN, IRL,
ITA, NET, POR,
SPA, SWE, UK
FRA, GER, GRE,F1 Minister of
Finances can
block
expenditures 1990s DEN, ITA, NET,
POR, SPA, SWE,
UK
BEL, FRA, GER,
GRE, IRL
1980S BEL, IRL, ITA,
SPA, SWE, UK
DEN, FRA, GER,
GRE, POR, UK
F2 Spending
ministers
subject to cash
limits
1990S IRL, ITA, SPA,
SWE, UK
BEL, DEN, FRA,
GER, GRE, POR,
SWE, UK
1980s DEN, GER, GRE,
IRL, ITA, SPA,
SWE, UK
BEL, FRA, GER,
NET, POR
F3 Disbursement
approval
required from
Minister of
Finances or
controller
1990S DEN, GRE, ITA,
SPA, SWE, UK
BEL, FRA, GER,
IRL, NET, POR
1980s ITA, NET, POR SPA GER, GREF4 Transfers of
expenditures
between
chapters
1990S NET, POR SPA GER, GRE
1980s NET ITA GRE
Flexibility of budget
execution
F5 Change in
budget law
during
execution 1990S NET ITA GRE
1980s BEL, DEN, ITA FRA, NET, SPA,
SWE, UK
GER, PORF6 Carry-over of
unused funds to
next year 1990S DEN FRA, NET, SPA,
UK
BEL,
GER, ITA,
POR, SWE
Table A.2 (continued)
Stage Symbol Rule Range Decade 0 1 2
1980s BEL, FRA, GRE,
ITA, POR, SPA,
SWE
DEN, UK GER, IRL,
NET
L1 Target variable
1990S BEL, FRA, GRE,
POR, SPA,
DEN, UK GER, IRL,
ITA, SWE
1980s BEL FRA, SWE DEN, GREL2 Planning
horizon (years)
1990S BEL FRA DEN, GRE
1980s BEL FRA, GRE, IRL,
ITA, POR, SPA,
SWE
DEN, NETL3 Forecasting
method
1990S BEL FRA, GRE, IRL,
ITA, POR, SPA
DEN, NET
1980s BEL FRA, SPA, SWE DEN,
GRE, POR
Long-term planning
constraint
L4 Degree of
commitment
1990S FRA, SPA DEN,
GRE, POR
Source: von Hagen (1992); De Haan Moessen and Volkerkink (1999).
39
APPENDIX B
Graph b1: NLPCA transformations of budget rules scores
0.1 1.2 2.3
O1
1.5
3.0
TO1
0 1 2 3
O2
1
3
TO2
0 1 2 3
O4
2
4
TO4
1.1 1.7 2.3 2.9
O5
1
2
TO5
0.1 1.2 2.3 3.4
N1
1
3
TN1
0 1 2 3
N2
1
3
TN2
0.1 1.2 2.3
N3
0.5
2.0
TN3
-0.2 0.6 1.4
N4
0.1
1.2
TN4
40
-0.2 0.6 1.4
P4
0.1
1.2
TP4
-0.2 0.6 1.4
P5
0.1
1.2
TP5
0.1 1.2 2.3 3.4
I1
2
4
TI1
-0.2 0.3 0.8 1.3 1.8
I3
1
2
TI3
-0.2 0.6 1.4
P4
0.1
1.2
TP4
-0.2 0.6 1.4
P5
0.1
1.2
TP5
0.1 1.2 2.3 3.4
I1
2
4
TI1
-0.2 0.3 0.8 1.3 1.8
I3
1
2
TI3
41
0.1 1.2 2.3
I4
1.2
2.4
TI4
-0.2 0.3 0.8 1.3 1.8
I5
1
2
TI5
1 3 5
F4
1
3
TF4
0.1 1.2 2.3 3.4
F5
1
4
TF5
0.1 1.2 2.3
F6
1.5
3.0
TF6
-0.2 0.3 0.8 1.3 1.8
L1
0
1
TL1
0.1 1.2 2.3 3.4
L2
2
4
TL2
0 1 2 3
L3
0
1
2
TL3
42
Graph b2: NLPCA transformations of fiscal variables
0.4 0.8 1.2
deb
0.3
0.8
Tdeb
0.4 0.5 0.6
exp
0.45
0.60
Texp
0.02 0.06 0.10
tdef
0.05
0.10
Ttdef
0.0 0.5 1.0
pdef
0.0
0.5
Tpdef
43
*
None of the authors is an individual subscriber to the Review of Economics and Statistics.
Corresponding author: Fabio Padovano, Dipartimento di Istituzioni Politiche e Scienze
Sociali, Università Roma Tre, Via C. Segre 4, 00146 Roma, Italy. Tel. +390655176402; Fax
+390655176248; E-mail: padovano@uniroma3.it. Paper presented at the seminars of the
CSEI, SIEP and EPCS2001. Partial funding by the Università Roma Tre through the Research
Grant 2000 “Budget Procedures and Public Debt: A Multivariate Nonlinear Analysis” is
gratefully acknowledged. We thank Jakob de Haan, Emma Galli and Bruno Bises for
comments on previous versions of this paper. The usual caveat applies.