THE RELATIONTHIP BETWEEN TRADE OPENNESS

AND PUBLIC EXPENDITURE. THE SPANISH CASE,

1960-2000

Mª Dolores Gadea Rivas*

Marcela Sabaté Sort*

Estela Sáenz Rodríguez*

Department of Applied Economics

School of Economics

University of Zaragoza

Gran Vía, 2

Zaragoza, 50010

Phone: +34 976761841

Fax: +34 976761840

lgadea@unizar.es

msabate@unizar.es

essaenz@unizar.es

* Research Group External Sector and Monetary Integration (SEIM)

seim.unizar.es
ABSTRACT:

Recently, greater attention has been paid to the possible relation between trade
openness and public expenditure. Most of the articles of this kind of literature are based
on the central argument of the work of Rodrik (1998): more open economies are
exposed to a greater risk as a result of the possible turbulences in the international
markets, which can affect their domestic economy. The public sector can exert an
isolation function over this external risk, increasing its participation in the whole
economy. Today, this is a controversial and open subject, in spite of the many empirical
studies carried out.

The aim of this paper is to empirically determine whether there is a relation

between trade openness and public expenditure in Spain in the period 1960-2000, a
relatively short period that has witnessed an accelerated process both of openness to the
international markets of goods and services and of public sector growth. The time series
analysis for the Spanish economy in the period 1960-2000, based on the cointegration
test of Johansen, reveals a long-term relationship between public expenditure and both
trade openness and several protection indicators. We complete our analysis with several
econometric techniques which reveal that the aforementioned relation is both robust and
stable.

CLASSIFICATION JEL: F41, H11

KEY WORDS: trade openness, public expenditure, cointegration

2
1. INTRODUCTION

The aim of this paper is to empirically determine whether there is a relation

between trade openness and public expenditure in Spain in the period 1960-2000, a
relatively time period that has witnessed an accelerated process both of openness to the
international markets of goods and services and of public sector growth. We begin in
1960, after the application of the Plan of Stabilization and Liberalization of 1959, which
initiates the change of the rigidities of the previous autarkic system. The period contains
key dates for the Spanish economy like the restoration of democracy and Spain’s
integration into the European Union. We want to assess whether the parallel evolution
of our two variables is a coincidence or not. That is to say, we want to find out if there
is really a long-term relation between the openness rates or, alternatively, trade policy
decisions and the Spanish public sector.

Our starting hypothesis is based on the seminal and much cited work of Rodrik
(1996, 1998). Most experts who support a positive relation between international
economic integration and the public economy base their arguments on the hypothesis of
compensation proposed by Rodrik. More open economies are exposed to a greater risk,
as a result of the possible turbulences in the international markets, which can affect their
domestic economy. As the public sector is "the safe" sector of the economy –both in
terms of employment and income-, it can exert an isolation function over the external
risk that affects the other sectors, increasing its participation in the whole economy.
Today, this is a controversial and open subject, in spite of the many empirical studies
about it.

In our opinion, the increase of international integration is a long-term dynamic

process. In addition, the relation between trade openness, external risk and public
expenditure may be idiosyncratic, due to the specific characteristics of each country.
Thus, we believe that a time series analysis is more suitable than the usual cross-country
analyses. The Spanish case in 1960-2000 might illustrate the fulfilment of the argument
of Rodrik because, in this period, Spain experienced an important process of openness
to the international economy. This integration involved a risk for many sectors,
especially for those that had been more protected. The definitive integration came with
the entry of Spain into the European Union, opening her economy to others that, for
several reasons, were more developed and competitive. Given this situation, the public

3
budget was probably used to tackle this new situation. Furthermore, trade liberalization
is also based on a "social contract". So, the political authorities had to reach agreements
with the economic agents and satisfy their protection requirements.

There are several applied papers about the Spanish economy that deal with the
relation between trade openness and government size. One of them was carried out by
Gadea (1993), who found, through the cointegration test of Engle and Granger, a
positive relation between the two variables in 1964-1991. Another is that of Molana et
al. (2004), who used the cointegration test of Johansen. Their result was that there was
no cointegration between trade openness and public expenditure on good and services in
1948-1998 and that there was no long-term causality. One of the possible reasons for
this result, as Molana et al. explain, might be the unsuitable measurement of the
variables used in their analysis, especially in the case of trade openness.

In this same line, our work is based on a bivariant analysis, to avoid the results
being affected by the explanatory variables chosen. We make several contributions.
First, we believe that the sample period is more adequate because it begins in 1960,
when Spain really initiates the liberalizing process 1 . Second, apart from considering
several measures of trade openness, we take into account an issued that is novel in the
literature, namely, the link between the protection level and public expenditure 2 .
Finally, we apply two tests never before used in the case of Spain. The first is a
robustness analysis based on the application of Levine and Renelt (1992) of the extreme
bounds test of Leamer (1983, 1985), adding other potentially explanatory variables of
expenditure to the bivariant VAR. The second is a Hansen and Johansen (1993, 1999)
stability analysis to reveal whether the relation is stable or affected by some structural
change. We find a positive and significant long-term relation between several indicators
of trade openness and public expenditure, and this relation is both robust and stable.

1
Molana et al. (2004) carried out a study of 22 OECD countries apart from Spain, in which the
integration process was earlier. Besides, the considered range of countries forced them to use the public
expenditure on goods and services.
2
In general, the most used measure is total trade as a percentage of GDP in current terms.

4
 This paper is divided as follows. In the second section, we carry out a survey of
the literature about the relation between trade openness and public expenditure,
emphasizing arguments that have provoked an intense debate. In the third section we
describe the trade liberalization and public sector growth in the Spanish economy for
the period under study. The empirical results are shown in the fourth section. Finally,
we sum up the main conclusions of this work.

2. EMPIRICAL LITERATURE

Cameron (1978) was the pioneer empirical analysis, where the author directly
linked the level of openness and public sector size. The motivation for his analysis was
the spectacular increase of the public sector in the developed countries, especially after
the Second World War, when the welfare State began to be consolidated. One of the
main results of his analysis, through a sample of 18 developed capitalist countries in
1960-1975, is that trade openness is one the most important variables to explain the
public economy expansion. The reason, as Cameron (1978) said, is that open economies
are very competitive and have a high industrial concentration. This higher level of
specialization favours union organization, which increases social public incomes. Based
on the same period -after the WWII-, Swank (1983) considers several explanations of
the welfare State increase in 17 developed democracies. The aim of this author is to test
the argument of Cameron (1978). He includes variables of interest groups in the
reference model of Cameron, and finds that the openness variable remains positive and
significant. So, the author argues, the link through which trade openness affects the
welfare State may be the intervention of the State to tackle the pressures of the global
market.

Schmidt (1983), through a cross-country and time series analysis in 22 developed

economies, finds a positive relation between trade openness and taxes and social
security contributions -as a percentage of GDP- in three different periods. The first is
that of the post-war reconstruction, 1950-1960; the second is that used by Cameron,
1960-1975; and the third is that of the world economic crisis, 1974-1978. Using the
panel data technique, Hicks and Swank (1992) find a positive relation between trade

5
openness and welfare spending in 18 developed countries in 1960-1982 3 . The same
result is found in the paper of Huber et al. (1993), which is also applied to 18 developed
economies, although they study 1956-1988.

As the papers mentioned so far that analyze the link between openness and public
expenditure share a political approach and were usually carried out in countries with a
high level of economic and social development (like those of the OECD).

In the seminal and much cited work of Rodrik (1996, 1998), a new view of the
link between openness and expenditure through external risk was developed. The idea
behind it is that more open economies are exposed to a higher external risk derived from
the possible shocks in the international markets and that, to mitigate this external risk,
governments increase public expenditure. Through a cross-country analysis, he finds a
positive relation between the openness of the eighties (1980-89) and the public
expenditure on goods and services of the early nineties (1990-92), for a sample of 103
countries 4 . As there is a wide range of explicative variables, this is not a spurious
relation due to the omission of variables. The sample is very large, so this relation is not
due to either to the choice or source of data. The relation holds for different periods and
different measures of public sector. The analysis also includes a wide range of control
variables to test the robustness of the link between openness and expenditure. Later, in
larges samples but with longer periods of study, other works have supported the positive
relation between trade openness and public expenditure, like those of Garrett (2001),
Martínez-Mongay (2002), Shelton (2007) and Gemmell et al. (2008).

Because of the intense debate about this subject, it is necessary to explain the
contrary or alternative arguments to the paper of Rodrik (1998). Authors like Ferris and
West (1996) find a negative relation between trade openness and public expenditure in
the United States between 1959 and 1986 5 . They propose that international integration
involves more tax competition and, because of this, governments have less capacity to
increase taxes -especially capital tax- so the size of the public sector is also restricted. In
this same line, Ferris (2003) and Borcherding et al. (2004) support this negative link in

3
Again in line with the postulates of Cameron (1978), that is to say, the idea of that the link openness-
expenditure is derived from specialization and interest groups.
4
The aforementioned positive relation between trade openness and public expenditure on goods and
services is found in a panel data analysis as well.
5
On the contrary, Abizadeh (2005) finds a positive relation between trade openness and public
expenditure in the United States in period 1960-2000.

6
20 OECD countries in 1970-1997 6 . However, this idea had already been considered by
Rodrik (1997) himself, who found empirical evidence -in 18 OECD countries in 1965-
1991- that more economic integration moves the tax burden from mobile factors
(capital) to relatively immobile factors (labour). Other authors, like Iversen and Cusack
(2000), Kittel and Winner (2005) and Dreher (2006, 2008) show that there is no any
relation between globalization and public sector size in OECD countries.

Another alternative argument about the theorical link between trade openness and
expenditure, is derived from the work of Saunders and Klau (1985), and has to do with
economies of scale in the provision of public goods and services. In Alesina and
Wacziarg (1998) the idea that trade openness and public expenditure on goods and
services are linked through external risk, as Rodrik (1996) suggests, is questioned. In
their opinion, the link between openness and expenditure is country size. Firstly, smaller
economies can not obtain benefits from access to big markets unless they adopt open
trade policies. Furthermore, smaller countries that can not take advantage of scale
economies in the provision of public goods have bigger public sector. They add the
population variable to the reference specification of Rodrik (1996) and obtain some
evidence of this idea. Nevertheless, the authors point out that there is a high level of
colinearity between openness and country size, so it is difficult to draw definitive
conclusions. To study the sensitivity of the relation between openness, country size and
public sector size proposed by Alesina y Wacziarg (1998), Ram (2008) carries out a
panel data study of 154 countries and the period 1960-2000 7 . His main result is that
trade openness exerts a positive and significant effect on public expenditure on goods
and services, once country size is controlled for.

In any case, as Alesina and Wacziarg (1998) state, these points of view are not
necessarily contradictory. Country size may be negatively related to public expenditure
on goods and services at the same time as trade openness exerts a direct effect on public
transferences. The important thing, because it reconciles both points of view, is that the
stabilizing role of government in open economies should be specifically analyzed
through public expenditure on transfers.

6
Álvarez, Pascual and Romero (2003) find a negative relation between trade openness and total public
expenditure in the EU-15 between 1998 and 2000. Nonetheless, this empirical evidence is based on a
simple analysis of coefficients of correlation.
7
Simplifying the reference equation of Alesina and Wacziarg (1998) and Rodrik (1998), because the only
control variable is GDP per capita.

7
3. PROCESS OF OPENNESS AND PUBLIC EXPENDITURE GROWTH IN
SPAIN

1959 is a key date in the Spanish economy because the Plan of Stabilization and
Liberalization (PEL) ended the autarkic period of Franco and initiated a liberalizing
course. Within the field of trade policy, a progressive reduction of quantitative
restrictions was developed, which were replaced by border taxes. From then on and
until the beginning of the third stage of the Economic and Monetary Union, external
openness has been growing, although with a slightly irregular path.

Between 1960 and 1975, the main trade barriers which were used to protect
national production were import taxes, regimens of globalised, bilateral and State trade
and the Compensation Tax of Internal Burden. The process of trade liberalization
started from this very regulated trade system.

After 1975, and in a context of international economic crisis, the openness of

Spanish economy underwent a reversal, with levels of protection -import taxes divided
by non-energy imports- similar to those of 1966. This backward step was due to the
elimination of prior measures -tariff exemptions and subsidies- in order to avoid
substantial trade deficit. In 1977, adhesion to the European Economic Community
(EEC) was applied for again and the Moncloa Pacts were signed, which allowed and
advance in the liberalizing process. However, between 1978 and 1985, protection only
decreased a little, being trade policy influenced by the preagreement to EEC access.

By 1985, trade openness had increased importantly, as nominal protection had

decreased from 16.5% in 1960 to 9.3%. Nevertheless, the integration of Spain into the
EEC was the definitive impulse for trade liberalization, with a reduction of protection
which Spain has never gone back. Agreements in the Treaty of Adhesion involved a
great transformation of the protection system. First, with the extinction of the regimen
of administrated trade. Second, with the extinction of border tax adjustments, that were
replaced by VAT, and a more active policy of export promotion. And finally, with a
dismantling of tariffs for all EEC members, who are our principal suppliers, and the
adoption of a Common External Tariff applied to other countries outside the EEC,
which had a lower average level than the Spanish one. From then on, trade openness

8
increased as never before and, in the early nineties, the process of trade liberalization
was practically completed.

In a parallel process, a stage of public sector growth started in 1960, reaching

similar levels to those of other developed countries at present. In 1960-1975, a slight
modernization of Spanish public sector is undertaken, decreasing classic functions of
the State and increasing the participation of preferential and economic goods. However,
as with the expansion of openness, an institutional event, in this case the restoration of
democracy caused the greatest growth in expenditure. The maximum point was
achieved in 1993, when a policy of budgetary control was initiated to advance towards
the euro. In this sense, we can talk about a parallel evolution of public expenditure and
trade openness, as both of them started from reduced levels and then underwent a
dramatic increase in a relatively short time (Figure 1). Or alternatively, we can talk
about an asymmetric evolution of public expenditure over GDP and nominal protection
rate (Figure 2).

9
 0
10
20
30
40
50
60

0
10
20
30
40
50
60
1960 1960

1962 1962

1964 1964

1966 1966

1968 1968

1970 1970

1972 1972

1974 1974

1976 1976

1978 1978

10
TPEGDP

TPEGDP
1980 1980

1982 1982

1984 1984
XMGDP

TMMNE
1986 1986

1988 1988

1990 1990
Figure 1. Trade openness and public expenditure

Figure 2. Nominal protection and public expenditure

1992 1992

1994 1994

1996 1996

1998 1998

2000 2000
4. RESULTS OF THE ECONOMETRIC ESTIMATION

A very important part of time series study is analyzing the long-term relation
between variables which economic theory predicts. This is what we will tackle in this
section. Having revised the literature about the relation between trade openness and
public expenditure and having carried out a historic and graphic analysis of both series,
we will check here if there was a stable relation in the evolution of the two variables in
Spain in 1960-2000. First, we describe the variables of external openness, trade policy
and public expenditure that we have used. Second, we present the methodology of
cointegration used in the econometric analysis and the results derived from it. Third, we
apply a robustness test to these results. Finally, we analyze the stability of the link
between the two variables.

4.1 Variables

In order to capture trade openness we use six variables: total trade and imports of
goods over GDP in current terms (XMGDP and MGDP), imports of non-energy goods
relative to GDP in current terms (MNEGDP), openness of trade sector (XMGDPCOM)
(X+M relative to GDP minus services, in current terms) and total trade and imports of
goods relative to GDP, in real terms (XMGDPREAL and MGDPREAL).

We also propose other measures that are supposed to better reflect the protection
system of a country. First, import taxes relative to non-energy imports, as an objective
measure of Spanish trade policy (TMMNE). Second, we add the component of
protection of the Compensation Tax of Internal Burden to the previous measure, in
order to reflect the real degree of protection of the economy (TMCPMNE). Third, non
tariff barriers, that include both the percentage of imports that were subjected to some
sort of control (REGTRADE) (regulated trade) and exports tax deduction (XTD) as a
percentage of total exports.

The variable representing government size is the traditional ratio of total public
expenditure as a percentage of GDP, in current terms (TPEGDP) and social public
expenditure as a percentage of GDP, in current terms (SPEGDP).

11
 4.2 Analysis of cointegration

The method of cointegration used in this work is the multivariant technique of

Johansen, based on the VAR model. The main advantage compared to uniequational
methods is that it does not suppose that there is just one direction in the relation studied,
as it is a system of equations in which all variables are endogenously fixed. We will first
specify a model of two endogenous variables (openness/protection and public
expenditure) 8 . The optimum length of the VAR in accordance with the AIC and SC
criteria, which allows the residuals fulfil the requirements of normality,
homoscedasticity and absence of correlation is one lag. The next step involves choosing
one out of the five cases proposed by Johansen (1995), in order to make some
suppositions about the underlying trend in the data. The analysis of integration order
says that series are I(1). Nevertheless, taking into account the arguments of Juselius
(2006) 9 , the graphic analysis of the variables in levels leads us to consider two
possibilities. The first is that they have no trend (model 2) and the second is that the
have a stochastic trend (model 3). The SC and AIC criteria select model 2 for
XMGDPREAL, MGDPREAL and MNEGDP and model 3 for XMGDP, MGDP and
XMGDPCOM.

With the purpose of establishing the number of cointegration vectors, the method
of Johansen considers trace and eigenvalue tests. Results of these are shown in Table 1.
In the case of XMGDPREAL, MGDPREAL and MNEGDP both trace and eigenvalue
tests reject the null hypothesis of no cointegration in favour of a relation of
cointegration, since the result is higher than the critical value. However, for XMGDP,
MPIBGDP and XMGDPCOM, there is no cointegration.

8
Previously, we carry out a stationarity analysis through unit root tests, which reveal that all of the
indicators used are I(1).
9
We should think that out time series are stationary because, being ratios, they are bounded and, thus,
have no trend. However, it is also very important to take into account temporal dimension when carrying
out a macroeconomic study. If the perspective is long-term -for example a century-, the series are usually
stationary. But in medium and short-term, most macroeconomic variables have a trend, because there has
not been time for the deviations to adjust to adjust to the average. Such non stationary behaviour should
be taken into account in statistical analysis so as not to reach mistaken conclusions. According to Juselius
(2006), considering these variables as non stationary also offers a good empirical approximation so as to
be able to use the properties of cointegration technique.

12
 Table 1. Cointegration test of Johansen:
Total public expenditure and trade openness, 1960-2000
Cointegration based on max eigenvalues:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 3.45 14.26 0.91
MGDP r=0 r≥1 9.76 14.26 0.23
XMGDPCOM r=0 r≥1 1.48 14.26 0.99
r=0 r≥1 58.03 15.89 0.00
XMGDPREAL
r≤1 r=2 0.86 9.16 0.97
r=0 r≥1 46.72 15.89 0.00
MGDPREAL
r≤1 r=2 1.88 9.16 0.80
r=0 r≥1 36.03 15.67 0.00
MNEGDP
r≤1 r=2 1.32 9.24 0.90
Cointegration based on trace of stochastic matrix
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 4.72 15.49 0.84
MGDP r=0 r≥1 12.69 15.49 0.13
XMGDPCOM r=0 r≥1 1.93 15.49 0.99
r=0 r≥1 58.88 20.26 0.00
XMGDPREAL
r≤1 r=2 0.86 9.16 0.97
r=0 r≥1 48.59 20.26 0.00
MGDPREAL
r≤1 r=2 1.88 9.16 0.80
r=0 r≥1 37.35 20.26 0.00
MNEGDP
r≤1 r=2 1.32 9.16 0.90

The relation between the cointegrated variables adjusts, according to the first
vector of the cointegration test, to the following terms:

LTPEGDP = -2,71 + 1,10LXMGDPREAL

(1,98) (2,57)
LTPEGDP = -2,35 + 1,56LMGDPREAL
(2,31) (4,18)
LTPEGDP = -5,19 + 2,82LMNEGDP
(2,56) (3,55)

with t-ratios in brackets.

Of the variables in current terms, only the non-energy imports (MNEGDP) present
cointegration. On the contrary, total trade and imports in real terms (XMGDPREAL and
MGDPREAL) have a long-term relation with public expenditure and, in addition, their
coefficients are significant and have the expected sign. We have determined, therefore,
that there is a positive and long-term relation between trade openness and public

13
expenditure. Nevertheless, as can be seen in Table 2, the positive sign of the coefficient
of the error correction term makes this cointegration relation doubtful.

Table 2. Coefficients of adjustment in equation ∆GPTPIB (α2)

XMGDPREAL MGDPREAL MNEGDP

0.005 0.008 0.008
(2.36) (2.21) (2.24)

All these results are supported when the measures of openness are replaced by
measures of trade policy. Table 3 shows the results of the test of Johansen, where we
can see that with both statistics there is a relation of cointegration between public
expenditure and all the measures of trade policy.

Table 3. Cointegration test of Johansen:

Total public expenditure and measures of trade policy, 1960-2000
Cointegration based on max eigenvalues:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
r=0 r≥1 19.05 15.89 0.02
TMMNE
r≤1 r=2 6.37 9.16 0.16
r=0 r≥1 21.91 15.89 0.01
TMCPMNE
r≤1 r=2 8.58 9.16 0.06
r=0 r≥1 18.37 15.89 0.02
REGTRADE
r≤1 r=2 2.32 9.16 0.71
r=0 r≥1 23.13 15.89 0.00
XTD
r≤1 r=2 5.62 9.16 0.22
Cointegration based on trace of stochastic matrix
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
r=0 r≥1 25.42 20.26 0.01
TMMNE
r≤1 r=2 6,37 9.16 0.16
r=0 r≥1 30.49 20.26 0.00
TMCPMNE
r≤1 r=2 8.58 9.16 0.06
r=0 r≥1 20.69 20.26 0.04
REGTRADE
r≤1 r=2 2.32 9.16 0.71
r=0 r≥1 28.75 20.26 0.00
XTD
r≤1 r=2 5.62 9.16 0.22

The relation between the cointegrated variables, according to the first vector of the
cointegration test, is the following:

14
 LTPEGDP = 3.45 - 0.26LTMMNE
(14.92) (2.21)
LTPEGDP = 3.50 - 0.21LTMCPMNE
(20.42) (2.55)
LTPEGDP = 3.77 - 0.22LREGTRADE
(48.59) (7.34)
LTPEGDP = 3.82 - 0.36LXTD
(32.81) (5.91)

Their coefficients are significant and have the expected negative sign. According
to Table 4, as with openness, the correction term is positive.

Table 4. Coefficients of adjustment in equation ∆GPTPIB (α2)

TMMNE TMCPMNE REGTRADE XTD

0.04 0.04 0.06 0.05
(2.60) (2.52) (2.35) (3.81)

Because of these results, we decided to assess the possibility that by not having
considered the important institutional change of the restoration of democracy in 1977,
the estimation was distorted. For this reason, we decided to repeat the earlier analysis,
adding a dummy variable (DEMO), which takes value 0 for any year between 1960 and
1977 and value 1 from 1978 on. With this we try to capture the effect of the restoration
of democracy on the growth of expenditure. When introducing this dummy, the results
of the test of Johansen remain the same, that is to say, XMGDPREAL, MGDPREAL
and MNEGDP are the variables that have a relation of cointegration with expenditure,
as can be seen in Table 5.

15
 Table 5. Cointegration test of Johansen:
Total public expenditure, trade openness and democracy, 1960-2000
Cointegration based on max eigenvalues:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 6.38 14.26 0.57
MGDP r=0 r≥1 11.45 14.26 0.13
XMGDPCOM r=0 r≥1 5.11 14.26 0.73
r=0 r≥1 39.00 15.89 0.00
XMGDPREAL
r≤1 r=2 3.67 9.16 0.46
r=0 r≥1 34.40 15.89 0.00
MGDPREAL
r≤1 r=2 5.08 9.16 0.27
r=0 r≥1 23.68 15.89 0.00
MNEGDP
r≤1 r=2 3.96 9.16 042
Cointegration based on trace of stochastic matrix
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 7.69 15.49 0.49
r=0 r≥1 16.57 15.49 0.03
MGDP
r≤1 r=2 5.12 3.84 0.02
XMGDPCOM r=0 r≥1 5.81 15.49 0.72
r=0 r≥1 42.68 20.26 0.00
XMGDPREAL
r≤1 r=2 3.67 9.16 0.46
r=0 r≥1 39.49 20.26 0.00
MGDPREAL
r≤1 r=2 5.08 9.16 0.27
r=0 r≥1 27.64 20.26 0.00
MNEGDP
r≤1 r=2 3.96 9.16 0.42

LTPEGDP = 11.41 - 1.42LXMGDPREAL

(3.86) (1.29)
LTPEGDP = 13.79 - 2,90LMGDPREAL
(5.29) (2.62)
LTPEGDP = 10.56 - 2.43LMNEGDP
(4.66) (2.58)

However, the relation between expenditure and openness is negative, in contrast to

the positive nexus found previously. In the case of the protection measures, only tax
deduction maintains its relation of equilibrium. And, although the sign of this variable is
the expected one and causality runs from protection to expenditure, the adjustment
coefficient is positive.

16
 The following step was to investigate whether this lack of influence was due to the
relation having undergone significant changes during those four decades. The procedure
consists of elaborating graphs of the coefficients of the significant variables derived
from a recursive estimation, that is to say, a repeated estimation that uses ever wider
subsamples from the sample period. If these coefficients fluctuate a lot, it will be an
indication of a possible instability. Figures 3, 4 and 5 show the evolution of the
coefficient that relates different measures of openness to expenditure, without
considering the institutional change of 1977. The results point to a clear lack of
stability, which could be related to the unexpected sign of the error correction term.
With the inclusion of the variable DEMO in Figures 6, 7 and 8, a stable relation
between the two variables is found until the mid nineties. From then on, there is an
instability that seems to be the cause of the aforementioned negative sign.

17
 Figure 3. Recursive estimation of XMGDPREAL
.7

.6

.5

.4

.3

.2

.1

.0
-.1

-.2
1970 1975 1980 1985 1990 1995 2000

Recursive C(2) Estimates ± 2 S.E.

Figure 4. Recursive estimation of MGDPREAL

.8

.6

.4

.2

.0

-.2
1970 1975 1980 1985 1990 1995 2000

Recursive C(2) Estimates ± 2 S.E.

Figure 5. Recursive estimation of MNEGDP

1.5

1.0

0.5

0.0

-0.5
1970 1975 1980 1985 1990 1995 2000

Recursive C(2) Estimates ± 2 S.E.

18
 Figure 6. Recursive estimation of XMGDPREAL (including DEMO)
.45

.40

.35

.30

.25

.20

.15

.10
1984 1986 1988 1990 1992 1994 1996 1998 2000

Recursive C(2) Estimates ± 2 S.E.

Figure 7. Recursive estimation of MGDPREAL (including DEMO)

.40

.35

.30

.25

.20

.15

.10

.05
1984 1986 1988 1990 1992 1994 1996 1998 2000

Recursive C(2) Estimates ± 2 S.E.

Figure 8. Recursive estimation of MNEGDP (including DEMO)

.6

.5

.4

.3

.2

.1

.0
1984 1986 1988 1990 1992 1994 1996 1998 2000

Recursive C(2) Estimates ± 2 S.E.

19
 These results are confirmed if we repeat all the analysis for the period 1960-1993.
The final year has been chosen because of both the complete trade integration of Spain
into the EU and the beginning of an intense period of budgetary consolidation, with a
view towards integration into the EMU. We repeat the analysis of cointegration
including the dummy in the VAR and maintaining the optimal length, although, in this
case, the information criteria choose model 3 for all the measures of openness and
protection. The institutional coherence of this analysis is also reflected in econometric
terms. Now all the measures of openness show cointegration with public expenditure (as
can be seen in Table 6).

Table 6. Cointegration test of Johansen:

Total public expenditure, trade openness and democracy, 1960-1993
Cointegration based on max eigenvalues:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 13.92 14.26 0.06
r=0 r≥1 24.33 14.26 0.00
MGDP
r≤1 r=2 2.63 3.84 0.11
r=0 r≥1 18.64 14.26 0.01
XMGDPCOM
r≤1 r=2 2.70 3.84 0.10
r=0 r≥1 18.19 14.26 0.01
XMGDPREAL
r≤1 r=2 1.37 3.84 0.24
r=0 r≥1 19.52 14.26 0.01
MGDPREAL
r≤1 r=2 1.66 3.84 0.19
r=0 r≥1 16.13 14.26 0.02
MNEGDP
r≤1 r=2 1.65 3.84 0.19
Cointegration based on trace of stochastic matrix:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
r=0 r≥1 15.60 15.49 0.04
XMGDP
r≤1 r=2 1.68 3.84 0.19
r=0 r≥1 26.96 15.49 0.00
MGDP
r≤1 r=2 2.62 3.84 0.10
r=0 r≥1 21.34 15.49 0.01
XMGDPCOM
r=2 2.70 3.84 0.10
r=0 r≥1 19.56 15.49 0.01
XMGDPREAL
r≤1 r=2 1.37 3.84 0.24
r=0 r≥1 21.18 15.49 0.01
MGDPREAL
r≤1 r=2 1.66 3.84 0.19
r=0 r≥1 17.78 15.49 0.02
MNEGDP
r≤1 r=2 1.65 3.84 0.19

20
 LTPEGDP = 0.96LXMGDP
(5.37)
LTPEGDP = 1.41LMGDP
(6.69)
LTPEGDP = 0.67LXMGDPCOM
(7.94)
LTPEGDP = 0.49LXMGDPREAL
(7.92)
LTPEGDP = 0.59LMGDPREAL
(6.65)
LTPEGDP = 1.06LMNEGDP
(5.58)

All the variables of openness have the predicted sign and are significant. In
addition, the sign of the error correction mechanism supports the cointegration relation
(Table 7).

Table 7. Coefficients of adjustment in equation ∆GPTPIB (α2)

XMGDP MGDP XMGDPCOM XMGDPREAL MGDPREAL MNEGDP

-0.16 -0.11 -0.27 -0.29 -0.17 -0.12
(3.62) (4.54) (4.50) (4.55) (4.04) (3.20)

In the case of measures of protection, only taxes on imports have a long-term

relation with expenditure.

LTPEGDP = -0.65LTMMNE
(5.61)

But this variable also has the same characteristics as the measures of openness in
terms of the expected sign of the coefficient and the error correction term.

We have, therefore, determined that there is a positive long-term relation between

trade openness and government expenditure. The following step will be to establish the
direction of causality between the two variables, since cointegration does not establish
the direction. The causality concept was initially proposed by Granger (1969),
considering that a variable x causes another variable y if the present and past behaviour
of x allows us to predict the behaviour of y. This analysis can be carried out with
autoregressive vectors (VAR) through Wald tests. Nevertheless, in cointegrated VAR,

21
the usual tests of causality can lead to erroneous results, since they were initially
considered for stationary series. Extended tests of causality (of long-term) of Granger
(1988) arose from the theory of cointegration and its relation to the Error Correction
Model:
k k
ΔOPt = μ1 + α 1 ECM + ∑ γ i ΔOPt −i + ∑ φi ΔPEt −i + ε 1t
i =1 i =1

k k
ΔGPt = μ 2 + α 2 ECM + ∑ γ i ΔPE t −i + ∑ φi ΔOPt −i + ε 2t
i =1 i =1

where ECM stands for the error correction mechanism, which is the lagged residual of
the cointegration relation. The coefficients of this term (α1 and α2) reflect short-term
adjustments towards long-term equilibrium. For there to be long-term causality between
the variables, the coefficients αi will have to be significantly different from zero 10 .
Otherwise, there will only be weak or short-term causality. In our case, the Error
Correction Model is the following 11 :

ΔOPt = α1[PEt −1 − ( ρ 0 βOPt −1 )] + ε 1t

ΔPEt = α 2 [PEt −1 − ( ρ 0 βOPt −1 )] + ε 2t

in which we test the null hypothesis αi = 0, that is to say, non causality. In particular, if
α1 = 0 and α2 ≠ 0, we will say that trade openness causes public expenditure in the long-
term in the Granger sense; if α1 ≠ 0 and α2 = 0, we will say that public expenditure
causes trade openness in the long-term in the Granger sense; and if α1 ≠ 0 and α2 ≠ 0, we
will say that both variables cause each other in the long-term in the Granger sense.

As we can see, long-term causality runs in both directions (Tables 8-13) although,
in general, it is stronger from openness to expenditure.

10
In econometric terms, the analysis of long-term causality is equivalent to the analysis of weak
exogeneity of Johansen (1992).
11
Since we start with a VAR of one lag, the Error Correction Model has zero lags.

22
 Table 8. Causality test between TPEGDP and XMGDP*

H0: αi = 0 χ2(1) Probability

α1 4.26 0.04
α2 10.55 0.00
* LR test with a rank of cointegration = 1

Table 9. Causality test between TPEGDP and MGDP

H0: αi = 0 χ2(1) Probability

α1 14.62 0.00
α2 15.57 0.00

Table 10. Causality test between TPEGDP and XMGDPCOM

H0: αi = 0 χ2(1) Probability

α1 2.93 0.09
α2 14.61 0.00

Table 11. Causality test between TPEGDP and XMGDPREAL

H0: αi = 0 χ2(1) Probability

α1 6.15 0.01
α2 16.02 0.00

Table 12. Causality test between TPEGDP and MGDPREAL

H0: αi = 0 χ2(1) Probability

α1 14.58 0.00
α2 13.19 0.00

Table 13. Causality test between TPEGDP and MNEGDP

H0: αi = 0 χ2(1) Probability

α1 14.17 0.00
α2 8.80 0.00

23
 Furthermore, when we use social public expenditure and the variable DEMO, we
also find cointegration with openness in 1960-1993 (Table 14) although, in this case,
only with total trade in real terms (XMGDPREAL) 12 . The magnitude of the openness
variable is very similar to that of total public expenditure. The sign of the openness
variable and of the error correction term are as expected and causality is also
bidirectional (Tables 15 and 16).

Table 14. Cointegration test of Johansen:

Social public expenditure, trade openness and democracy, 1960-1993
Cointegration based on max eigenvalues:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 11.46 14.26 0.13
r=0 r≥1 21.13 14.26 0.00
MGDP
r≤1 r=2 5.40 3.84 0.02
XMGDPCOM r=0 r≥1 11.26 14.26 0.14
r=0 r≥1 46.16 15.89 0.00
XMGDPREAL
r≤1 r=2 7.00 9.16 0.13
r=0 r≥1 48.67 15.89 0.01
MGDPREAL
r≤1 r=2 11.08 9.16 0.02
r=0 r≥1 47.88 15.89 0.00
MNEGDP
r≤1 r=2 9.23 9.16 0.05
Cointegration based on trace of stochastic matrix
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 13.02 15.49 0.11
r=0 r≥1 26.53 15.49 0.00
MGDP
r≤1 r=2 5.40 3.84 0.02
XMGDPCOM r=0 r≥1 12.41 15.49 0.14
r=0 r≥1 53.16 20.26 0.00
XMGDPREAL
r≤1 r=2 7.00 9.16 0.13
r=0 r≥1 59.74 20.26 0.01
MGDPREAL
r≤1 r=2 11.07 9.16 0.02
r=0 r≥1 57.10 20.26 0.00
MNEGDP
r≤1 r=2 9.23 9.16 0.05

LSPEGDP = 1.67 + 0.44LXMGDPREAL

(7.45) (5.18)

12
The same analysis, with and without including DEMO in 1960-2000, show very similar results when
using social public expenditure. These results appear in the Annex.

24
 Table 15. Causality test between SPEGDP and XMGDPREAL*

H0: αi = 0 χ2(1) Probability

α1 18.04 0.00
α2 17.83 0.00
* LR test with a rank of cointegration = 1

Table 16. Coefficients of adjustment in equation ∆GPTPIB (α2)

XMGDP MGDP XMGDPCOM XMGDPREAL MGDPREAL MNEGDP

-0.14
(5.06)

4.3 Analysis of robustness and stability

4.3.1 Robustness

Our analysis of the bivariant VAR has shown the existence of a relation of
cointegration between diverse measures of trade openness/protection and public sector
size measured as total expenditure over GDP. Nevertheless, many other variables have
been employed to explain public expenditure growth. So, it is important to take into
account that, in our case, other factors may have taken part in this growth.

In this section we aim, therefore, to analyze the robustness of the previous

results. In order to carry out this exercise, we start from the test of extreme bounds of
Leamer (1983, 1985), applied in the much-cited paper of Levine and Renelt (1992) to
discriminate between the explanatory factors of economic growth. The methodology
consists of estimating of the following equations:

γ =αj + βyj * y + βzj * z + βxj * xj +ε

where y is a vector of fixed variables that always appears in regressions, z is the interest
variable and xj is a vector of up to three variables taken from a set X of possible control
variables. This model is estimated for the M possible combinations of xj Є X. For each
model j, both βzj and the corresponding standard deviation, σzj, are estimated. The
inferior extreme bound is defined as the smallest value of βzj - 2 σzj and the superior

25
extreme bound as the highest value of βzj + 2 σzj. If the inferior bound is negative and
the superior bound is positive, then variable z is not robust.

Sala-i-Martin (1997) considers that this methodology is very strict, because, if in

one regression the sign of the coefficient βzj changes or is not significant, the variable is
not robust. So, he considers the complete distribution of the estimators of βz. In our
case, in line with Fölster and Henrekson (2001), we have calculated the percentage of
regressions in which the coefficient of openness/protection has the expected sign, is
significant and, in addition, displays cointegration with public expenditure.

Sala-i-Martin (1997) includes three fixed variables and a set of control

variables taken in groups of three, arguing that in the literature, the usual regression of
growth has seven variables. Here we consider that it is not necessary to include the fixed
variables, but we do introduce the groups of three. Gadea (1993) and Jaén and Palma
(2004) have demonstrated the existence of four types of explanatory factors of the
growth of the Spanish public expenditure: structural, demographic, institutional and
economic. Of these variables, with the purpose of carrying out an exercise of
robustness, we have chosen six control variables (Table 17). These six variables,
combined in groups of three -in order not to lose many degrees of freedom- involve the
estimation of 20 equations.

Table 17. VARIABLES OF CONTROL

Economic
GDPPC GDP per capita in real terms
Factors
Demographic
POP Total population
Factors
Demographic
P65 Percentage of population over 65
Factors
Percentage of workers employed in the Structural
EA
agricultural sector Factors
Deflator of public expenditure on goods and Structural
DPE
services over deflator of GDP Factors
Institutional
DIRTAX Direct taxes over total taxes
Factors

26
 Following the previous steps, we verified that the optimum length of the VAR is
still one lag and chose model 3. According to the results of the test (Table 18), the main
conclusion is that we can accept that the studied relation is robust, except in the case of
total trade in current terms. Results are especially satisfactory for XMGDPCOM,
XMGDPREAL and MNEGDP. If we choose criterion B, the relation for import taxes is
not robust. However, the criterion applied here is much stricter than the one applied in
studies of economic growth due to the econometric methodology of time series where,
in addition to the expected sign and significance of the variable of interest, cointegration
is also presumed. Therefore, it can be concluded that with measures of openness there is
a long-term relation, with the expected sign and robust.

Table 18. TEST OF ROBUSTNESS*:

Total public expenditure, openness/protection and democracy, 1960-1993
Percentages
Variables A B
XMGDP 30 30
MGDP 45 45
XMGDPCOM 85 85
XMGDPREAL 50 45
MGDPREAL 45 40
MNEGDP 55 55
TMMNE 45 25
*All equations include total public expenditure and demo.
A Percentage of total equations (20) in which there is cointegration and the sign of openness/protection is expected
and significant.
B Percentage of total equations (20) in which there is cointegration, the sign of openness/protection is expected and
significant and the coefficient of adjustment fulfils the cointegration requirements.

The best behaviour of the expected sign of considered factors corresponds to the
structural variables, but the demographic effect and direct taxes are also important.
GDPPC, on the other hand, has a negative coefficient. We should also highlight that we
do not find clear evidence that openness is catching the effect of the variable population
on expenditure.

We repeat the same analysis for the relation of cointegration found between
social public expenditure and total trade in real terms. In this case, the relation is no
robust (Table 19).

27
 Table 19. TEST OF ROBUSTNESS*:
Social public expenditure, openness/protection and democracy, 1960-1993
Percentages
Variables A B
XMGDPREAL 35 35

4.3.2 Stability

The question of structural changes is very usual in a time series context and has
been studied by many authors from the point of view of unit roots 13 . Nevertheless, tests
proposed in the econometrical literature to study the stability of the relation of
cointegration between variables are more recent. A simple way to analyze the stability
of parameters through the OLS model is to apply the classic test of Chow (1960), that is
to say, dividing the sample into two groups and test whether there are important
differences in estimated equations. Other more complex tests, also based on
uniequational models, are those developed by Hansen (1992) and Gregory and Hansen
(1996). The first is based on the FM estimator of residuals and the second is an
extension of the test of cointegration of Engle and Granger (1987). Both of them allow
us to find the point in time at which the structural change occurs.

In our case, the considered break point is the restoration of democracy, as it makes
possible the materialization of demands of public expenditure through the voting
mechanism. After including the dummy variable DEMO in the VAR and restricting the
end of the sample period to 1993, the results of the analysis are satisfactory in terms of
cointegration and robustness. We then continue testing the stability of the estimated
parameters 14 . Given the approach followed, namely, that of a cointegrated VAR, we
apply recursive estimation of Hansen and Johansen (1993, 1999). This method starts
from a base period, 1... T0, from which eigenvalues, parameters and statistics are
recursively estimated in increasing subsamples 1.... ,n with n = T0... ,T. Calculations are
carried out in two ways: allowing all parameters to vary (X-Model) and restimating only
the long-term parameters (R-Model). In general, the scale of the test statistics is the
critical value of 5%, so a value greater than one is significant and indicates non-
stability.
13
Perron (1989), Zivot and Andrews (1992) and Lee and Strazicich (2003), among others.
14
In 1960-1993 and for robust indicators.

28
 In Figure 9, we show the results of the test of fluctuation of both the transformed
eigenvalues and their sum. Dennis (2006) indicates that this test is considered to be
quite conservative, so its rejection is a strong indication of the non-constancy of
eigenvalues. As can be observed, according to the results of this test, the hypothesis of
stability is clearly accepted for the robust indicators of openness.

29
 Figure 9. Transformed eigenvalue fluctuation test

TPEGDP AND MGDP

Eigenvalue Fluctuation Test

1.00
Tau(Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index)
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

1.00
Tau(Ksi(1)+...+Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index)
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

Tau(Ksi) = C(T)||Ksi(t)-Ksi(T)||

TPEGDP AND XMGDPCOM

Eigenvalue Fluctuation Test

1.00
Tau(Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index)
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

1.00
Tau(Ksi(1)+...+Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index)
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

Tau(Ksi) = C(T)||Ksi(t)-Ksi(T)||

30
 TPEGDP AND XMGDPREAL

Eigenvalue Fluctuation Test

1.00
Tau(Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index)
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

1.00
Tau(Ksi(1)+...+Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index)
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

Tau(Ksi) = C(T)||Ksi(t)-Ksi(T)||

TPEGDP AND MGDPREAL

Eigenvalue Fluctuation Test

1.00
Tau(Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index)
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

1.00
Tau(Ksi(1)+...+Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index)
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

Tau(Ksi) = C(T)||Ksi(t)-Ksi(T)||

31
 TPEGDP AND MNEGDP

Eigenvalue Fluctuation Test

1.00
Tau(Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index )
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

1.00
Tau(Ksi(1)+...+Ksi(1)) X(t) = R1(t)
5% C.V. (1.36 = Index )
0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

Tau(Ksi) = C(T)||Ksi(t)-Ksi(T)||

In Figure 10, we show the results of the test of stability of the cointegration space
(β). It is based on the difference between β(n) and β(T)15 and, like the previous test, is
also considered to be conservative. In this case, we also observe a stable behaviour of
the model.

15
With n = T0, …, T.

32
 Figure 10. Test Max of constancy of β

TPEGDP AND MGDP

Test of Beta Constancy
1.00
Q(t) X = R1
5% C.V. (1.81 = Index)

0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

TPEGDP AND XMGDPCOM

Test of Beta Constancy
1.00
Q(t) X = R1
5% C.V. (1.81 = Index)

0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

TPEGDP AND XMGDPREAL

Test of Beta Constancy
1.00
Q(t) X = R1
5% C.V. (1.81 = Index)

0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

33
 TPEGDP AND MGDPREAL
Test of Beta Constancy
1.00
Q(t) X = R1
5% C.V. (1.81 = Index)

0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

TPEGDP AND MNEGDP

Test of Beta Constancy
1.00
Q(t) X = R1
5% C.V. (1.81 = Index)

0.75

0.50

0.25

0.00
1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992

In summary, after including the variable DEMO, the robust relation of

cointegration found between total public expenditure and several indicators of openness
in 1960-1993 is stable, according to most conservative tests of recursive estimation of
Hansen and Johansen (1993, 1999), so we can accept the conclusions derived from this
relation.

34
5. CONCLUSIONS

The time series analysis for the Spanish economy in the period 1960-2000, based
on the cointegration test of Johansen, reveals a long-term relationship between public
expenditure and both trade openness and several protection indicators. However, the
Error Correction Model (ECM) is not coherent with the cointegration theory because
the correction term is positive. For this reason, we incorporate a very important variable
for public expenditure growth in Spain, namely, a dummy that captures the effect of the
restoration of democracy in 1977. With this new variable, measures of trade openness
show contradictory results, so, after carrying out a simple recursive analysis, we end the
study period in 1993 -maintaining the democracy dummy in the model-. We have
chosen this year for two institutional reasons: the end of the intense trade openness
process after integration into the EU and the beginning of a budgetary restriction stage
in order to get access to the European Monetary Union. If we consider the period 1960-
1993, the cointegration analysis shows that there is a long-term relation between all the
trade openness measures and one of the protection indicators and the total public
expenditure. Now, openness has the expected sign, and the correction term of ECM
fulfils the cointegration requirements. We complete our analysis with several
econometric techniques which reveal that the aforementioned relation is both robust and
stable.

6. BIBLIOGRAPHY

ABIZADEH, S. (2005): “An analysis of government expenditure and trade

liberalization”, Applied Economics, 37, pp. 1881-1884.

ALESINA, A. and WACZIARG, R. (1998): “Openness, country size and government”,

Journal of Public Economics, 69, pp. 305-321.

ÁLVAREZ, S., PASCUAL, M. and ROMERO, D. (2003): “Protección social,

globalización y crecimiento económico”, Hacienda Pública española,
Monografía, pp. 63-77.

35
BORCHERDING, T.E., FERRIS, J.S. and GARZONI, A. (2004): “Growth in the real
size of government since 1970”, en Backhaus, J. y Wagner, R.E. (eds.), Handbook
of Public Finance, Hingham, Ma: Kluwer Academic Press.

CHOW, G.C. (1960): “Tests of equality between sets of coefficients in two linear
regressions”, Econometrica, 28, pp. 591-605.

DENNIS, J.G. (2006): CATS in RATS. Cointegration Analysis of Time Series, version 2.

DREHER, A. (2006): “The influence of globalization on taxes and social policy: An

empirical analysis for OECD countries”, European Journal of Political Economy,
22, pp. 179-201.

DREHER, A. (2008): “The impact of globalization on the composition of government

expenditures: evidence from panel data”, Public Choice, 134, pp. 263-292.

ENGLE, R.F. and GRANGER, C. (1987) ‘Cointegration and error correction:

representation, estimation and testing’, Econometrica, 55 (2), pp. 251-276.

FERRIS, J.S. (2003): “Do alternative measures of government result in alternative

explanations for government size?”, Economics Bulletin, 8 (4), pp. 1-11.

FERRIS, J.S. and WEST, E.G. (1996): “Testing theories of real government size: U.S.
experience, 1959-1989”, Southern Economic Journal, 62 (3), pp. 537-553.

FÖLSTER, S. and HENREKSON, M. (2001): “Growth effects of government

expenditure and taxation in rich countries”, European Economic Review, 45, pp.
1501-1520.

GADEA, M.D. (1993): “Un modelo sobre los determinantes del crecimiento del gasto
público en España (1964-1991), Revista de Economía Aplicada, 1 (2), pp. 31-65.

GARRETT, G. (2001a): “Globalization and government spending around the world”,

Studies in comparative international development, 35 (4), pp. 3-29.

GARRETT, G. and MITCHELL, D. (2001): “Globalization, government spending and

taxation in the OECD”, European Journal of Political Research, 39 (2), pp. 145-
177.

GRANGER, C. (1969): “Investigating Causal Relations by Econometric Methods and

Cross-Spectral Methods," Econometrica, 34, pp. 424-438.

36
GREGORY, A.W. y HANSEN, B.E. (1996): “Residual-based tests for cointegration in
models with regime shift”, Journal of Econometrics, 70, pp. 99-126.

HANSEN, B.E. (1992): “Tests for parameter instability in regressions with I(1)
processes”, Journal of business and economics statistics, 10 (3), pp. 321-335.

HANSEN, H. and JOHANSEN, S. (1993): “Recursive estimation in cointegrated VAR-

models”, Manuscript, Institute of Mathematical Statistics, University of
Copenhagen.

HANSEN, H. and JOHANSEN, S. (1999): “Some tests for parameter constancy in

cointegrated VAR-models”, Econometrics Journal, 2, pp. 306-333.

IVERSEN, T. and CUSACK, T.R. (2000): “The causes of welfare state expansion:
deindustrialization or globalization”, World Politics, 52 (3), pp. 313-349.

JAÉN, M. y PALMA, L. (2004): “Public expenditure dynamics in Spain: a simplified

model of its determinants”, Papeles de Trabajo, Nº 9, Instituto de Estudios
Fiscales, pp. 3-48.

JOHANSEN, S. (1992): “Testing weak exogeneity and the order of cointegration in UK

money demand data,” Journal of Policy Modelling, 14 (3), pp. 313-334.

JOHANSEN, S. (1995): Likelihood-based inference in cointegrated vector

autoregressive models, Oxford University Press.

JUSELIUS, K. (2006): The cointegrated VAR model. Methodology and applications,

Oxford University Press.

KITTEL, B. and WINNER, H. (2005): “How reliable is pooled analysis in political

economy? The globalization-welfare state nexus revisited”, European Journal of
Political Research, 44, pp. 269-293.

LEAMER, E. (1983): “Let’s take the con out of econometrics”, American Economic
Review, 73 (1), pp. 31-43.

LEAMER, E. (1985): “Sensitivity analysis would help”, American Economic Review,

75 (3), pp. 308-313.

LEE, J. and STRAZICICH, M. (2003): “Minimum LM unit root tests with two
structural breaks”, Review of Economics and Statistics, 84 (4), pp. 1082-1089.

37
LEVINE, R. and RENELT, D. (1992): “A sensitivity analysis of cross-country growth
regressions”, American Economic Review, 82 (4), pp. 942-963.

MARTÍNEZ-MONGAY, C. (2002): “Fiscal policy and the size of governments”, in

Bute, M., Von Hagen, J. and Martínez-Mongay, C. (eds.), the behaviour of fiscal
authorities: Stabilisation, growth and institutions, Basingstoke: Pal grave
Macmillan.

MOLANA, H., MONTAGNA, C. and VIOLATO, M. (2004): “On the causal

relationship between trade openness and government size: evidence from 23
OECD countries”, Research Paper Series: Internationalisation of Economic
Policy, Nº 14, University of Nottingham.

PERRON, P. (1989): “The great crash, the oil price shock and the unit root hypothesis”,
Econometrica, 57 (6), pp. 1361-1401.

QUINTOS, C. (1998a): “Stability Tests in Error–Correction Models”, Journal of

Econometrics, 82, pp. 289–315.

QUINTOS, C. (1998b): “Fully Modified Vector Autoregressive Inference in Partially

Nonstationary Models”, Journal of the American Statistical Association, 93, pp.
783–795.

RAM, R. (2009): “Openness, country size and government size: additional evidence
from a large cross-country panel”, Journal of Public Economics, 93 (1-2), pp.213-
218.

RODRIK, D. (1996): “Why do more open economies have bigger governments?”,

National Bureau of Economic Research Working Paper n.º 5537.

RODRIK, D. (1997a): “International trade and big government”, in Cohen, B.J. (ed.),
International trade and finance: new frontiers for research: essays in honour of
Peter B. Kenen, New York: Cambridge University Press, pp. 89-121.

RODRIK, D. (1997b): “Trade, social insurance and the limits to globalization”,

National Bureau of Economic Research Working Paper Nº 5905.

RODRIK, D. (1998): “Why do more open economies have bigger governments?”,

Journal of Political Economy, 106 (5), pp. 997-1032.

38
SALA-I-MARTIN, X. (1997): “I just run four million regressions”, The American
Economic Review, 87 (2), Papers and Proceedings of the Hundred and Fourth
Annual Meeting of the American Economic Association, pp. 178-183.

SAUNDERS, P. and KLAU, F. (1985): “The role of the public sector: causes and
consequences of the growth of government”, OCDE Economic Studies, 4, pp. 5-
239.

SHELTON, C.A. (2007): “The size and composition of government expenditure”,

Journal of Publics Economics, 91 (11-12), pp. 2230-2260.

ZIVOT, E. and ANDREWS, D. (1992): “Further evidence on the great crash, the oil
price shock and the unit root hypothesis”, Journal of business and economics
statistics, 10 (3), pp. 251-270.

39
ANNEX

We repeat the same analysis using the variable social public expenditure.

• Period 1960-2000

Results of the Johansen test are shown in Table 1.

Table 1. Cointegration test of Johansen:

Social public expenditure and trade openness, 1960-2000
Cointegration based on max eigenvalues:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 8.83 14.26 0.30
MGDP r=0 r≥1 14.13 14.26 0.05
XMGDPCOM r=0 r≥1 8.23 14.26 0.36
r=0 r≥1 62.53 15.89 0.00
XMGDPREAL
r≤1 r=2 2.38 9.16 0.70
r=0 r≥1 56.87 15.89 0.00
MDPREAL
r≤1 r=2 0.92 9.16 0.96
r=0 r≥1 59.79 15.67 0.00
MNEGDP
r≤1 r=2 1.28 9.24 0.91
Cointegration based on trace of stochastic matrix
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 10.56 15.49 0.24
r=0 r≥1 21.07 15.49 0.01
MGDP
r≤1 r=2 6.94 3.84 0.01
XMGDPCOM r=0 r≥1 8.72 15.49 0.39
r=0 r≥1 64.91 20.26 0.00
XMGDPREAL
r≤1 r=2 2.38 9.16 0.70
r=0 r≥1 57.80 20.26 0.00
MDPREAL
r≤1 r=2 0.92 9.16 0.96
r=0 r≥1 61.07 20.26 0.00
MNEGDP
r≤1 r=2 1.28 9.16 0.91

LSPEGDP = 4.99 - 0.15LXMGDPREAL

(6.30) (0.59)
LSPEGDP = 4.43 - 0.21LMGDPREAL
(8.28) (1.06)
LSPEGDP = 6.05 - 0.80LMNEGDP
(6.07) (2.05)

As in total expenditure, there is cointegration with XMGDPREAL, MGDPREAL

and MNEGDP. However, their coefficients do not have the expected sign and the
variables in real terms lose their significance.

40
 Table 2 shows the results of Johansen’s test for the measures of trade policy. It can
be seen that, with both statistics there is a cointegration relation between public
expenditure and all the measures the trade policy.

Table 2. Cointegration test of Johansen:

Social public expenditure and measures of trade policy, 1960-2000
Cointegration based on max eigenvalues:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
r=0 r≥1 25.80 15.89 0.00
TMMNE
r≤1 r=2 7.80 9.16 0.09
r=0 r≥1 26.69 15.89 0.00
TMCPMNE
r≤1 r=2 10.31 9.16 0.03
r=0 r≥1 26.88 15.89 0.00
REGTRADE
r≤1 r=2 3.45 9.16 0.50
r=0 r≥1 24.98 15.89 0.00
XTD
r≤1 r=2 7.86 9.16 0.09
Cointegration based on trace of stochastic matrix
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
r=0 r≥1 33.60 20.26 0.00
TMMNE
r≤1 r=2 7.80 9.16 0.09
r=0 r≥1 37.00 20.26 0.00
TMCPMNE
r≤1 r=2 10.31 9.16 0.03
r=0 r≥1 30.33 20.26 0.00
REGTRADE
r≤1 r=2 3.45 9.16 0.50
r=0 r≥1 32.84 20.26 0.00
XTD
r≤1 r=2 7.86 9.16 0.09

LSPEGDP = 4.76 + 1.82LTMMNE

(2,37) (1.80)
LSPEGDP = 3.08 - 0.81LREGTRADE
(6,49) (4,34)
LSPEGDP = 3.18 + 1.38LXTD
(3.92) (3.21)

In this case, only REGTRADE has the expected negative sign and, although the t-
ratio decreases, the variable remains significant. As can be deduced from Table 3,
causality is also unidirectional, running from protection to expenditure. Furthermore, as
in total expenditure, the error correction term is positive (Table 4).

41
 Table 3. Causality test between SPEGDP and REGTRADE

H0: αi = 0 χ2(1) Probability

α1 3.50 0.06
α2 16.57 0.00

Table 4. Coefficients of adjustment in equation ∆GPTPIB (α2)

TMMNE TMCPMNE REGTRADE XTD

0,02
(4,83)

• Period 1960-2000, including DEMO

Results of Johansen’s test are shown in Table 5.

Table 5. Cointegration test of Johansen:

Social public expenditure, trade openness and democracy, 1960-2000
Cointegration based on max eigenvalues:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 3.90 14.26 0.87
MGDP r=0 r≥1 9.33 14.26 0.26
XMGDPCOM r=0 r≥1 5.11 14.26 0.73
r=0 r≥1 50.45 15.89 0.00
XMGDPREAL
r≤1 r=2 0.45 9.16 0.99
r=0 r≥1 54.63 15.89 0.00
MDPREAL
r≤1 r=2 0.52 9.16 0.99
r=0 r≥1 56.52 15.67 0.00
MNEGDP
r≤1 r=2 0.87 9.24 0.97
Cointegration based on trace of stochastic matrix
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
XMGDP r=0 r≥1 5.65 15.49 0.74
r=0 r≥1 15.75 15.49 0.04
MGDP
6.42 3.84 0.01
XMGDPCOM r=0 r≥1 5.25 15.49 0.78
r=0 r≥1 50.89 20.26 0.00
XMGDPREAL
r≤1 r=2 0.45 9.16 0.99
r=0 r≥1 55.16 20.26 0.00
MDPREAL
r≤1 r=2 0.52 9.16 0.99
r=0 r≥1 57.39 20.26 0.00
MNEGDP
r≤1 r=2 0.87 9.16 0.97

42
 LSPEGDP = 3.32 + 0.02LXMGDPREAL
(7.29) (0.10)
LSPEGDP = 2.87 + 0.05LMGDPREAL
(13.54) (0.57)
LSPEGDP = 3.39 - 0.13LMNEGDP
(9.02) (0.85)

The variables XMGDPREAL and MGDPREAL have the expected sign, but
they are not significant. The error correction term is negative and significant and long-
term causality is bidirectional (Tables 6, 7 and 8).

Table 6. Causality test between SPEGDP and XMGDPREAL

H0: αi = 0 χ2(1) Probability

α1 23.36 0.00
α2 16.66 0.00

Table 7. Causality test between SPEGDP and MGDPREAL

H0: αi = 0 χ2(1) Probability

α1 15.38 0.00
α2 16.88 0.00

Table 8. Coefficients of adjustment in equation ∆GPTPIB (α2)

XMGDP MGDP XMGDPCOM XMGDPREAL MGDPREAL MNEGDP

-0.05 -0.08
(4.45) (4.48)

43
 Trade policy measures (Table 9).

Table 9. Cointegration test of Johansen:

Social public expenditure, measures of trade policy and democracy, 1960-2000
Cointegration based on max eigenvalues:
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
r=0 r≥1 20.34 15.89 0.03
TMMNE
r≤1 r=2 1.01 9.16 0.95
r=0 r≥1 20.73 15.89 0.01
TMCPMNE
r≤1 r=2 1.30 9.16 0.91
r=0 r≥1 23.53 15.89 0.00
REGTRADE
r≤1 r=2 3.15 9.16 0.55
r=0 r≥1 22.97 15.89 0.00
XTD
r≤1 r=2 6.38 9.16 0.16
Cointegration based on trace of stochastic matrix
Endogenous Null Alternative Critical Value
Statistic Probability
Variable Hypothesis Hypothesis 5%
r=0 r≥1 21.34 20.26 0.03
TMMNE
r≤1 r=2 1.01 9.16 0.95
r=0 r≥1 22.04 20.26 0.03
TMCPMNE
r≤1 r=2 1,30 9.16 0.91
r=0 r≥1 26.68 20.26 0.01
REGTRADE
r≤1 r=2 3,15 9.16 0.55
r=0 r≥1 29.35 20.26 0.00
XTD
r≤1 r=2 6.38 9.16 0.16

LSPEGDP = 2.95 + 0.57LTMMNE

(3.63) (1.78)
LSPEGDP = 2.78 + 0.59LTMCPMNE
(3.40) (2.00)
LSPEGDP = 1.56 + 1.28LREGTRADE
(0.93) (2.82)
LSPEGDP = 2.64 + 0.30LXTD
(9.41) (2.86)

In this case, none of the measures of protection has the expected negative sign.

44