You are on page 1of 11

Applied Economics

ISSN: 0003-6846 (Print) 1466-4283 (Online) Journal homepage: https://www.tandfonline.com/loi/raec20

Public capital stock and economic growth: an


analysis of the Spanish economy

Rafael Flores De Frutos , Mercedes Gracia-Díez & Teodosio Pérez

To cite this article: Rafael Flores De Frutos , Mercedes Gracia-Díez & Teodosio Pérez (1998)
Public capital stock and economic growth: an analysis of the Spanish economy, Applied
Economics, 30:8, 985-994, DOI: 10.1080/000368498325156

To link to this article: https://doi.org/10.1080/000368498325156

Published online: 01 Oct 2010.

Submit your article to this journal

Article views: 127

Citing articles: 36 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=raec20
Applied Economics, 1998, 30, 985± 994

Public capital stock and economic growth:


an analysis of the Spanish economy

R A F A E L F L O R E S D E F R U T O S , ME R C ED E S G R A C I A - D IÂ E Z and
T E O D O S I O P EÂ R E Z - A M AR A L
Departamento de Economõ a Cuantitativa, Facultad de Ciencias Econo micas,
Universidad Complutense de Madrid, 28223 Madrid, Spain

The objective of this paper is to evaluate the short- and long-term e€ ects of public
investment in infrastructure on aggregate output, labour and capital formation in the
private sector. The problem is analysed in a dynamic multivariate framework, which
allows for explicit consideration of feedback among all the variables. This approach
generalizes the current literature, which relies on a single-equation model to estimate
production functions and implicitly assumes the absence of feedback relationships.
For the Spanish economy, our results suggest positive long-term e€ ects of public
investment on the private sector variables.

I. INTRODUCTION tive has focused on the size of the elasticity of output with
respect to public capital.
The e€ ect of public investment in infrastructure on the In this paper we argue that the focus of the debate on the
growth of the private sector has been the issue of recent size of elasticities obtained from single-equation estimates
debate. The pioneering work of Aschauer (1989a, b) suggests might be misleading in providing an answer to the previous
that while current public expenditure decreases productivity question. This is because the single-equation approach con-
and economic growth, public investment in infrastructure siders explicitly only one of the four dynamic relationships
increases private productivity, ® nding a positive net e€ ect in that may exist among the four basic variables in the produc-
the USA for the period 1952± 86. Using these results, tion function and, therefore, excludes the likely presence of
Aschauer (1989b) explains the stagnation of the American feedback among these variables. This is important because if
economy during the 1970s by a crumbling infrastructure feedback exists, the size of elasticity of output with respect to
base. This explanation has raised a wide debate on the public capital as obtained from a single-equation produc-
subject and several studies have emerged showing evidence tion function estimate is only a part of the total e€ ect of
for and against the Aschauer e€ ect. Among others, Munnell public capital on output. Indeed, public capital can a€ ect
(1990a, b), Munnell and Cook (1990) and Garcõ a-Mila and output directly, i.e., as an additional input to the production
McGuire (1992) support the Aschauer e€ ect, while Aaron function, or indirectly through capital and labour. Then
(1990), Ebert, (1990), Tatom (1991) and Ford and Poret output can also a€ ect labour as well as private and public
(1991) do not. For the Spanish economy, Bajo and Sosvilla capital. Finally, labour and private capital can a€ ect public
(1993) and ArgimoÂn et al. (1994) ® nd a positive and signi® - capital formation. Therefore the total e€ ect of public capital
cant e€ ect of public investment on the productivity of the on output will be the result of a direct e€ ect and many
private sector. All these studies use a single-equation model indirect e€ ects. In addition, these e€ ects among variables
within the framework of a neoclassical theory of production. might not be instantaneous, so that the total e€ ect on output
They estimate a production function for the private sector, will be achieved after several periods. If this is so, the impulse
usually a Cobb± Douglas function, in which public capital response function of output to a shock in the stock of public
enters as an input in addition to private capital and labour. capital will be more revealing than the size of the elasticity of
Thus the debate on whether or not public capital is produc- output with respect to public capital in a production function.
0003± 6846 Ó 1998 Routledge 985
986 R. Flores de Frutos et al.
In this work we address the question of estimating the public sector determines pkt . Both the public and the private
total and dynamic partial e€ ects of public capital on output, sectors are assumed to know at the beginning of each period
employment and stock of private capital by making use of a the past values of all the variables determined in both sectors.
structural VARMA approach (see Hamilton, 1994, Chap. 11 The public sector is assumed to determine pkt at the
for a clear exposition). As is known, this approach relies on beginning of each period and to announce in advance what
a full diagonalization of the contemporaneous error public capital investment will be during that period. We also
covariance matrix of a VARMA model ® tted to data, such assume that the public sector implements the plan that has
that each particular diagonalization implies a particular been previously announced. Then at the beginning of period
structural form and therefore a particular set of impulse t the private sector knows the current values of pkt . On the
response functions. In the present analysis it is not necessary contrary the public sector does not know the values of
to have a full diagonalization of the contemporaneous error zt since they are determined by the private sector at the end
covariance matrix since we are only interested in the e€ ects of the period. Therefore the current values of zt are not
of a shock in public capital on the other variables. This is an included in the information set of the public sector and none
important issue because only identifying assumptions for of its components will a€ ect pkt .
isolating the shock of public capital have to be made. To Denoting by V zt and V pt respectively the information sets
estimate the public capital e€ ects we use a conceptual of the private and public sectors at t, these sets are de® ned
framework (adapted from Flores and Pereira, 1993) which as:
has a number of desirable characteristics. First, it requires V zt = {zt - j , pkt ± j ; pkt }, j = 1, 2 , ¼
minimal identi® cation assumptions in the sense that only
one structural shock has to be isolated. Second, it does not V pt = {zt - j , pkt ± j }, j = 1, 2, ¼
impose a priori restrictions on the dynamic relations among and the problem can be formalized as follows.
the variables, allowing all kinds of feedback e€ ects. Third, it
is based on the observed institutional behaviour of the
Spanish public sector, but it can also be extended to other Private sector
economies with similar characteristics. The present analysis
Each period the private sector determines the levels of yt ,
is performed for the Spanish economy using yearly data for
lt and kt using information on V zt . That is, in general zt will
the 1964± 92 period.
depend on the current and past values of pkt as well as the
Finally, we would mention that the approach of this
past values of zt . This can be speci® ed as:
paper is particularly appropriate for dealing with other
technical problems previously recognized in the literature zt = m z (B) pkt + e zt
(Tatom, 1991; Munnell, 1992), such as: (i) the endogeneity of (1)
labour and private capital, which may generate simulta-
p z (B) e zt =a zt

neous-equation biases and invalidate OLS estimates; and (ii) where:


the inadequate treatment of the statistical properties of the m z (B) is a (33 1) vector of stable transfer functions (Box
time series, i.e. non-stationarit y and the possible existence of and Jenkins, 1970): m z (B) = (n y (B) n l (B) n k (B))9 and each
cointegration relationships. transfer function is given by n j (B) = n j 0 + n j 1 B +
The rest of the paper is organized as follows. Section II m j 2 B2 + ¼ , for j = y, l, k, where B is the rational lag
presents the conceptual framework to identify public capital operator;
shocks in a dynamic multivariate framework. Section III
contains the empirical analysis for the Spanish economy. e zt = (e yt e lt e kt )9 is a (33 1) vector of random variables;
The main conclusions are summarized in Section IV. p z (B) is a (33 3) polynomial matrix: p z (B) =
I - p 1 B - p 2 B2 - ¼ where p i is the coe cient matrix
associated to lag i. The roots of the determinant of p z (B)
I I . I D E N T I F Y I N G P U B L I C C A P I TA L must lie on or outside the unit circle;
SHOCKS a zt = (a yt a lta kt )9 is a (33 1) white-noise vector with con-
temporaneous covariance matrix R z .
We use the same type of economic variables as in previous
literature: private output Y t , private employment L t , stock
of private capital K t and stock of public capital in infra- Public sector
structure PKt (hereafter lower-case letters will denote the
The public sector determines pkt using information on V pt .
logs of these variables). Then we consider the existence of
That is, pkt will depend on past values of both zt and pkt .
two sectors in the economy, the private sector and the Formally:
public sector, which are assumed to have full control over
pkt = m p (B) zt + e pt
di€ erent variables. The private sector controls yt , lt and kt , (2)
which in vector notation will be zt = (yt lt kt )9 , while the p p (B) e pt = a pt
Public capital stock and economic growth: the Spanish economy 987
where: where m z 0 = (n y0 n l 0 n k0 )9 is the vector of contemporaneous
m p (B) is a (13 3) vector of stable transfer functions: e€ ects of pkt on zt . However, the model can be normalized
m p (B) = (n py (B) n pl (B) n p k (B)); by premultiplying (3) by V ± 1 :
e pt is a scalar noise; P w* (B) wt = a wt
* (6)
p p (B) is a scalar polynomial in B of in® nite order with where:
roots on or outside the unit circle: p p (B) = I - p p 1 B - P w* (B) = V - 1 P w (B)
p p 2 B2 - ¼ ;
* = V -1 a
a wt wt
2
a pt is a white-noise scalar with variance s p and indepen-
dent of the elements of a zt . and with a contemporaneous covariance matrix of a wt * given
by

3 4
Equation 2 explicitly allows for feedback of the private S z + n z 0 n z9 0 s p n z 0 s p
2 2
R *=V ± 1 S V± 1 = (7)
 Â

sector variables to the public sector. If the public sector does n z9 0 s p


2
s p
2

not use information on previous values of the private sector


variables no feedback rules exist and public capital is truly Note that partition (1, 2) of matrix (7) helps us to explain
an exogenous variable. the empirical contemporaneous correlations between the
Note that (i) the restriction n py (0) = n pl (0) = n pk (0) = 0 in residuals of pkt and the residuals of the private sector vari-
Equation 2, due to the information set used by the public ables that are likely to exist when a VARMA model is ® tted
sector, and (ii) the independence of a p t from the elements of to data. Under this conceptual framework, these correla-
the vector a zt , jointly represent su cient conditions to ident- tions are a consequence of n z0 , i.e., the instantaneous e€ ects
ify the underlying public capital shocks from the residuals of of pkt on the other variables.
model (1)± (2). Both restrictions are a consequence of the
assumption that the investment plan announced by the Impulse response functions
public sector is implemented.1 This is the basic assumption
of the model which does not seem very restrictive, at least in From model (3) the vector zt can be written as
the case of the Spanish economy. In Spain the public budget zt = C p (B) a pt +C z (B) a zt (8)
is approved during the last quarter of the previous year and,
usually, unexpected budgetary cuts are made in current where:
public expenditure rather than in public investment. Note C p (B)= [ I - n z (B) n p (B) ] ± 1 n z (B) p p (B) ± 1

also that a sporadic intra-year change in the public invest- (9)


ment plan in infrastructure as a response to an exceptional
= F p 0 + F p 1 B + F p 2 B2 + ¼ ,
shock in any element of a zt does not imply a permanent C z (B) = [ I - n z (B) n p (B) ] ± 1 p z (B) ± 1
relationship between a zt and a pt . (10)
The model (1)± (2) can be written in matrix form as:
= I + F z 1 B + F z 2 B2 + ¼ ,
The sequence of coe cients associated with the lag poly-

3 4 3 4 3 4
p z (B) - p z (B) m z (B) zt a
= zt
nomial C p (B) of Equation 9 is to be interpreted as the
- p p (B) m p (B) p p (B) pkt a pt response function of zt to an impulse in a pt ; that is, ¶ zt / ¶ a pt ± j
(3) for j = 0, 1, 2, ¼ . This function measures the dynamic con-
sequences for the private sector variables of a shock in pkt .
or in the compact notation as P w (B)wt a wt , with the follow- Therefore, the estimation of this function is the key for
ing contemporaneous covariance matrix of a wt : describing the e€ ects of public capital upon the performance

3 4
R 0 of the private sector. The information of the impulse re-
R = z
2 (4) sponse function. The value of this function at moment s is
0 s p
complemented by the step response function. The value of
The stochastic multivariate model in (3) is not nor- this function at moment s is the sum of the cumulative
malized, since e€ ects from the initial moment t up to s of a transitory unit

3 4
I - m z0 shock in a pt . Therefore, it is obtained as the sum of the
P w (0) = V = (5)
0 1 cumulative impulse responses.

1
This assumption is incompatible with either (a) a relation like a pt = b 9 a zt + e pt or (b) the simultaneous determination of a pt and a zt .
Indeed, when the investment plan is implemented, the elements of a zt cannot a€ ect pkt since they are not known at the end of the previous
period when the public sector decides the level of pkt . On the contrary, if pkt is determined at the end of period t because changes in a zt a€ ect
a pt and hence pkt , then the announced investment plan is not likely to be implemented. This is also the case of the simultaneous
determination of a p t and a zt , which forces pkt to be determined at the end of the period. The implementation of the public investment plan
only allows for a relationship like a zt = a a pt + e zt , but its substitution in (1) leads to an identi® cation problem in the sense that n z (0) and
a could not be estimated separately.
988 R. Flores de Frutos et al.
Note that in order to study the e€ ects of changes in pkt PK: Public capital stock in infrastructure, measured in
on zt it is not necessary to specify a whole structural thousand million 1980 pesetas. This variable includes
model and the model given in (3)± (4) is all that is needed. public investment in infrastructure in transport and
Nevertheless, this model does not allow one to determine communications as de® ned in the National Ac-
the speci® c channels through which those e€ ects might counting. This series has been computed by ArgimoÂn
operate. For this structural analysis the matrix R z should and MartõÂ n (1993) using a method of permanent stock,
be fully diagonalized and this is beyond the purpose of in which a constant rate of depreciation is assumed.
this paper. Additionally one may be interested in consider- K: Private productive capital stock, also measured in
ing the e€ ects of shocks to kt and lt on yt , but again thousand million 1980 pesetas. This series has been
this requires isolating the shocks in kt and lt by the corres- constructed by Corrales and Taguas (1989), also using
ponding block diagonalization of matrix R z . Both this and a method of permanent stock.
the previous analysis are clear extensions of the present
study.
Order of integration analysis
Table 1(a) contains the values of the augmented
Estimation strategy
Dickey± Fuller (ADF) test for a unit root in the series y, l,
Note that (6)± (7) is an exactly identi® ed theoretical VARMA k and pk. The results suggest that all the variables are at
model written in its in® nite VAR representation. Using the least I(1). Table 1(b) shows the univariate and intervention
methodology developed by Tiao and Box (1981) or Jenkins models (Box and Jenkins, 1970) for the same variables. The
and Alavi (1981), a VARMA model can be ® tted to data; the intervention model of y shows a permanent decrease of 4.6
resulting model will be referred to as the empirical VARMA, percentage points in the rate growth of output in 1975: the
which can also be written in its in® nite VAR representation. annual average growth of output decreases from 6.4% in the
Then all the parameters in the theoretical VARMA model period before 1975 to 1.8% in the remaining period. The
can be estimated from the empirical VARMA model.
Matrix V can be estimated by equating (7) to the residual
covariance matrix of the empirical model. Note that n z 0 can Table 1(a). ADF test for I (1)
be obtained by multiplying the partition (1, 2) of this resid-
ual covariance matrix by the partition (2, 2) of the same p =1 p =2 p =3
matrix. Next we estimate P w (B) by premultiplying the em-
y - 2.30 - 1.84 - 2.15
pirical VARMA model by the estimate of V. Finally, after l - 1.98 - 1.39 - 1.48
estimating (3) and (4), the estimation of the impulse response k - 1.42 - 1.46 - 1.75
functions is immediate from (9). pk - 0.50 - 0.40 - 0.40
Note: H0 : r = 1 in the model zt = m + r zt ± 1 + + p
g j = z t ± j + ut .
j=1
The table shows the values of the statistics ( r W - 1)/s W d r W (95%
I I I . E M P I R I C A L A N A LY S I S critical value = - 3.00).

T he data Table 1(b). Univariate and intervention models

We use yearly data of the Spanish economy for the 1964± 92 Rates v 0 / m s a% Q(6)
period (the data set is included at the end of the paper). The
variables are de® ned as follows: = yt - 0.046 0.27 0.064 1.7 5.3
(0.009) (0.16) (0.007)
Y: Private gross domestic product (GDP), measured in = lt ± 0.75 ± 1.5 3.2
thousand million 1980 pesetas. This series has been (0.12)
= kt ± 0.93 ± 1.6 9.7
computed by Molinas et al. (1990) as the di€ erence
(0.02)
between total real GDP at factor cost and public = pkt ± 0.93 ± 1.7 4.9
GDP. (0.03)
L: Private employment, measured in thousand workers.
It has been computed by GarcõÂ a-Perea and GoÂmez Notes: The model speci® cation for all the variables is:
(1993) from the Spanish Survey of Labour (Encuesta = zt =v 0 j E7 5
t + h t
de PoblacioÂn Activa, EPA) as the di€ erence between
(1 - / B) [h t - m ]=a

{
t
total employment and employment in public adminis-
tration. It should be noted that the series of private 1.0 t > 1975
j E7 5
=
t < 1975
t
employment could understate the total occupation in 0.0
this sector, since people working in both sectors have Standard deviations are in parentheses, s a is the residual standard
only been counted as public workers. deviation and Q (6) is the Ljung± Box statistic for six lags.
Public capital stock and economic growth: the Spanish economy 989
Table 2. Johansen test equilibrium long-run relationship, in which the term ecmt
represents a measure of the disequilibrium at each period t.
Null hypothesis Statistic Critical
(95%)
Empirical model
At most 3 cointegration
relationships 0.49 8.08 Incorporating the above cointegration restriction in model
At most 2 cointegration (3) and using the methodology of Tiao and Box (1981) for
relationships 11.04 17.84
At most 1 cointegration constructing VARMA models, we obtain the following em-
relationship 27.14 31.26 pirical representation of (6)± (7) which has been estimated by
0 relationships versus 1 37.66 27.34 the exact maximum likelihood procedure developed by Hill-
mer and Tiao (1979) (standard deviations are shown in
Note: Speci® cation AR(3) in levels with a constant. parentheses):

(1 - 1.15B + 0.45B2 ) ecmt = (0.35 - 0.25B) = lt ± 1


ADF test for the presence of two unit roots in the variables (0.20) (0.17) (0.15) (0.12)
can be easily computed from models in Table 1(b) as
(/ = - 1) /s W d/ = (note that residuals are white noise and lags of - (0.32 - 0.31B) = kt ± 1 +a et
(12)
twice-di€ erenced variables should not be included). The (0.20) (0.19)
ADF tests suggest that once the e€ ect of the permanent (1 - 0.54B) = lt = 0.34 ecmt ± +a
1 lt
decrease in 1975 is removed from the y series, it does not (13)
(0.15) (0.18)
need a second di€ erence; the ADF value for this variable is
- 4.56 with a 95% critical value of - 3.00. The same (1 - 0.78B) = kt = (0.41 - 0.24B) ecmt ± 1
occurs for the l, k y pk series, with ADF values of - 2.08, (0.06) (0.14) (0.13) (14)
- 3.50 and - 2.33 respectively and a 95% critical value of
- 1.95. Therefore we cannot reject the hypothesis that all + 0.20 = pkt ± 1 +a kt
(0.07)
the series are I(1).
(1 - 0.97B) = pkt = 0.26 ecmt ± 1 +a pkt
(15)
(0.05) (0.15)
Cointegration
Table 2 contains the results of the Johansen (1988) test to Residual means, residual standard deviations and ad-
determine the number of existing cointegration relation- justed R2 coe cients for each equation are:
ships. The ® rst column shows the di€ erent null hypotheses
that have been considered, the second column shows the
a e = - 0.0003 a l = - 0.0027 a k = - 0.0002 a pk = - 0.0014
computed values of the statistics and the third column s ae 3 100 = 1.13 s al 3 100 = 1.43 s ak 3 100 = 1.06
contains the 95% critical values. From this table we do not
reject the existence of one cointegration relationship,2 which s apk 3 100 = 1.53
is given by the following OLS regression: 2
R̀ecm (R̀ 2 y ) = 0.72 (0.64) R̀ 2 l = 0.60 R̀=2 k = 0.91 R̀=2 pk = 0.77
= =
yt = 0.43 + 0.34 lt + 0.46 kt + 0.21 pkt + ecmt
(11) and the estimated matrix of contemporaneous correlations
(0.07) (0.08) (0.06) (0.07) for the error vector is the following:

3 4
1
This equation can be interpreted as a Cobb± Douglas
production function with constant returns to scale in all the 0.34 1
q (0) =
inputs, which implies decreasing returns to scale over pri- 0.55 0.41 1
vate inputs.3 Equation 11 can also be considered as an - 0.27 0.37 - 0.21 1
2
All regressions including a constant term and three of the four variables in (11) have been estimated and the ADF tests indicate that
residuals are I(1) in all cases. In fact, in the regression of y on l and k, of y on l and pk, of y on k and pk, and of l on k and pk, the ADF values
are respectively - 2.08, 0.87, - 2.19 and - 2.60, with a 95% critical value of - 3.27. Then the existing cointegration relationship must
include the four variables considered in the analysis.
3
Equation 11 can be interpreted as a production function without ignoring the e€ ect of technical progress if we assume that advances in
technical progress are embodied in capital and labour. In such a function, technical change is non-neutral and production factors are not
assumed to be homogeneous (Wallis, 1979, pp. 63± 4). However, it is important to note that Equation 11 is a cointegration relationship that
allows a more e cient estimation of model (3) and its implementation as a production function is not necessary to obtain the results of this
paper.
990 R. Flores de Frutos et al.
The cross-correlation matrices of orders, 1, 2 and 3 for the (12)± (15) in compact notation as:
error terms of the di€ erent equations are:
P x (B) xt = a xt (16)

3 4
- 0.05 0.25 - 0.07 0.19 where
0.26 0.25 - 0.07 0.25 D(B) Mwt = xt
0.06 0.35 0.03 0.32 with

3 4
0.09 - 0.14 0.01 0.07 1 - 0.34 - 0.46 - 0.21

3 4
0 1 0 0
0.02 0.12 - 0.13 0.07 M=
0 0 1 0
- 0.05 - 0.19 - 0.24 0.17 0 0 0 1
- 0.08 - 0.09 - 0.39 - 0.10
and

3 4
- 0.07 - 0.06 0.30 - 0.03
1 0 0 0

3 4
0 = 0 0
- 0.03 0.03 - 0.15 - 0.01 D (B ) =
0 0 = 0
0.08 - 0.08 - 0.01 - 0.19
0 0 0 =
0.10 - 0.14 - 0.12 - 0.32
0.10 0.18 0.17 0.02 Premultiplying (16) by M ± 1
we obtain
[M - 1 P x (B) D(B) M ] wt = M - 1 a xt
where each (i, j ) element is the cross-correlation coe cient
between each pair of error series when series j leads to series which is the estimated version of (6), where:
i. These matrices show that no cross-correlation is larger
P w* (B) = M - 1 P x (B) D (B) M
than two standard deviations (6 2/Ï n = 6 0.40), so they
a w*t = M -1 a
are not statistically di€ erent from zero at the 95% level. xt

Moreover, the likelihood ratio tests for autoregressive struc-


ture in the error series, presented in Table 3, indicate the Response functions
absence of additional structure. These results suggest that
model (12)± (15) adequately represents the existing dynamic As we have pointed out in the previous section, the estima-
correlation structure. tion of (3) and (4) permits the estimation of (9) and, therefore,
This model shows the existence of dynamic relationships the estimation of the impulse response functions.
among all the variables. Speci® cally Equation 15 estab- Figures 1 and 2 show, respectively, the impulse response
lishes the existence of lagged feedbacks from the private functions and the step response functions of the four vari-
sector variables to public capital stock, i.e. pk is not ables to a transitory unit shock in the rate of growth of
exogenous. In fact, this equation shows that the rate of public capital stock. Tables 4(a) and 4(b) present the values
growth of public capital formation responds slowly and of these functions for the period 1993± 2007. The Appendix
positively to a disequilibrium situation in the previous peri-
od. That is, a positive value of ecmt (which occurs when
output is above its equilibrium level) increases the rate of
growth of public capital in the next period. This suggests
that in the Spanish economy, the stock of public capital is
procyclical: public investment is higher in periods of high
productivity relative to periods of low productivity.
Finally, to relate the empirical model (12)± (15) to the
theoretical VARMA representation in (6)± (7), we write

Table 3. L ikelihood ratio test

Null hypothesis Statistic

AR(0) versus AR(1) 10.48


AR(1) versus AR(2) 14.54
AR(2) versus AR(3) 13.00

Note: 95% critical value: 26.3. Fig. 1. Impulse response functions, rates of growth: 1993± 2007
Public capital stock and economic growth: the Spanish economy 991
Table 4(a). Impulse response functions

Years PK K L Y

1993 1 0 0.321 0.321498


1994 0.966 0.195 0.174068 0.450983
1995 0.957905 0.378689 0.12741 0.348363
1996 0.931147 0.46627 0.076883 0.338896
1997 0.879383 0.505644 0.014957 0.352026
1998 0.810375 0.520271 - 0.04394 0.353374
1999 0.732212 0.517759 - 0.09119 0.344079
2000 0.651498 0.503107 - 0.12375 0.329568
2001 0.573578 0.480974 - 0.14134 0.313014
2002 0.502524 0.455257 - 0.14526 0.296253
2003 0.441077 0.428907 - 0.13781 0.28052
2004 0.390713 0.404002 - 0.12179 0.266607
2005 0.351801 0.381872 - 0.10013 0.254924
2006 0.323827 0.363224 - 0.07558 0.24557
2007 0.305627 0.348275 - 0.05052 0.238415
Fig. 2. Step response functions, levels: 1993 ± 2007
Table 4(b). Step response functions

contains a simple procedure to test the signi® cance of the Years PK K L Y


estimated impulse response values and the resulting 90%
con® dence intervals are shown in Table A1. 1993 1 0 0.321 0.321498
1994 1.966 0.195 0.495068 0.772481
Table 4(a) and Table A1 suggest that an increase of one 1995 2.923905 0.573689 0.622478 1.120844
percentage point in = pk has a signi® cant and positive re- 1996 3.855052 1.039959 0.699361 1.459741
sponse on both output and private capital stock up to 1997 4.734435 1.545603 0.714318 1.811766
approximatel y the year 2000. The response of private em- 1998 5.54481 2.065874 0.67038 2.16514
ployment is di€ erent since the only signi® cant e€ ect is the 1999 6.277022 2.583633 0.579189 2.509219
2000 6.92852 3.086739 0.455439 2.838787
one for 1993. Speci® cally, for the period 1993± 2000 the 2001 7.502098 3.567713 0.3141 3.151801
e€ ects of an increase of one percentage point in = pk are: 2002 8.004622 4.022971 0.168841 3.448054
2003 8.445699 4.451878 0.031035 3.728574
(i) An increase of 6.9 percentage points in the level of public 2004 8.836411 4.85588 - 0.09075 3.995181
capital stock. 2005 9.188212 5.237752 - 0.19088 4.250105
(ii) An increase of 3.1 percentage points in the level of 2006 9.512039 5.600976 - 0.26646 4.495676
private capital stock. 2007 9.817666 5.949251 - 0.31398 4.734091
(iii) An increase of 0.3 percentage points in the level of
private employment.
(iv) An increase of 2.8 percentage points in the level of
output. From these tables we can conclude that:

Note that when dynamic e€ ects among all variables are (i) In the absence of shocks (S0), the equilibrium levels
taken into account, the elasticity of public capital in of PK, K and Y are considerably higher than the
a Cobb± Douglas production function underestimates the (disequilibrium) values in 1992. However, the equilib-
long-run e€ ect of public capital on output. From Equa- rium level of L is lower than the one registered in
tion 11 this elasticity is 0.21, while the long-run total 1992. This re¯ ects an inertia of the system to reduce
e€ ect is 2.8. the number of jobs. In the absence of shocks, the system
These results suggest that the long-run equilibrium levels also tends to reduce public capital productivity and
of the system will be the values reached in the year 2000. private capital productivity, whereas private labour
These estimated values are presented in Table 5(a), where productivity increases. This is consistent with the
the second columns shows the initial levels in 1992 (refer- past evolution of these variables: since 1964 public
ence year), column 3 contains the equilibrium levels to capital productivity and private capital productivity
which the system would move in the absence of future has been decreasing, while there has been a continued
shocks (simulation S0) and column 4 presents the equilib- increase in the productivity of private labour. This
rium levels with a transitory unit shock in = pk in 1993 situation is compatible with a process of substitution
(simulation (S1). Table 5(b) includes the same information of capital (private and public) for labour which
for the inverse of factor productivities (PK/Y , K/Y , L /Y ) might be due to changes in technology and production
and the inverse of public investment productivity (PI/Y ). costs.
992 R. Flores de Frutos et al.
Table 5(a). L ong-run equilibrium in the year 2000 (variables in private sector have been analysed. Our main results are as
levels) follows:
Variables Initial Long-run equilibrum levels (i) We have used a conceptual model that allows us to
levels Simulation Simulation identify public capital shocks from the residuals of
(1992) S0 S1 a VARMA model ® tted to data. This conceptual frame-
work relies on the assumption that the public sector
PK 1486.1 1803.1 1932.4 implements the investment plan in infrastructure that
(129.3)
has been previously announced.
K 27073.4 27922.1 43229.7
(848.7) (ii) The empirical analysis shows the existence of
L 10590 8124.2 8161.3 dynamic relationships among all the variables in the
(37.1) model. In fact, not only does public capital formation
Y 16768.7 17609.7 18116.8 a€ ect the private sector variables, but also previous
(507.1)
values of the private sector variables a€ ect the current
PI 196.3 74.7 92.1
(17.4) level of public capital stock. The response of the public
capital stock is procyclical; that is, it increases in peri-
Note: The numbers in parentheses indicate the di€ erence between ods in which previous values of output have been above
the value to which they are referred (S1) and the equivalent value in the equilibrium level.
column 3 (S0).
(iii) There exists only one cointegration relationship among
the variables in the model. This relationship can be
Table 5(b). L ong-run equilibrium in the year 2000 (inverses of the interpreted as a long-run Cobb± Douglas production
producti vities) function with constant returns to scale in all the inputs.
Nevertheless, when dynamic e€ ects among all the vari-
Variables Initial Long-run equilibrum levels
ables exist, the elasticity associated with each input is
levels Simulation Simulation
(1992) S0 S1 no longer interpreted as the long-run e€ ect on output of
a permanent unit change in the level of that input. In
PK/Y 0.089 0.102 0.107 this case, the production function must be interpreted
K/Y 1.430 1.537 1.541 as a technical relationship that the levels of all the
L /Y 0.632 0.461 0.450
variables must ful® l in the long run to be in equilibrium.
PI/Y 0.012 0.004 0.005
(iv) The response of the private sector variables to an
increase in public capital stock is positive. A transitory
increase of one percentage point in the rate of growth
of public capital stock implies a permanent increase in
(ii) A transitory unit shock in the rate of growth of public the levels of equilibrium of output, employment, private
capital stock (S1) leads to higher long run equilibrium capital and public capital of 2.8, 0.3, 3.1 and 6.9 percent-
levels of PK, K and Y than those obtained under the age points respectively. However, the response of the
simulation S0. With respect to L , it should be noticed system is slow: with a shock in 1993 the equilibrium is
that although the equilibrium level of L is lower than the not reached until 2000. The only exception is employ-
value registered in 1992, it is higher than the level ment, with an instantaneous response. It is important to
reached in the absence of shocks. Public investment has note that the response of labour is positive but the
the e€ ect of reducing the number of jobs that might be equilibrium level of private employment is lower than
destroyed in the long run. the one registered in 1992. This re¯ ects the inertia of the
system to substitute capital (private and/or public) for
These empirical results must be interpreted with care, labour which might be a consequence of a labour-
given the limited data set. Quarterly data would have been saving technical change.
preferable but, unfortunately, neither public nor private
capital stock time series for the Spanish economy are avail-
able with that frequency. AC K N O W L E D G E M EN T S

We are grateful to I. ArgimoÂn, P. GarcõÂ a-Perea and


I V . C O N CL U S I O N S D. Taguas for providing us with data, to A. Novales
and an anonymous referee for their useful comments
Using data on the Spanish economy for the period and to DGICYT (PB95-0912) for its partial ® nancial
1964± 92, the empirical e€ ects of public investment in infras- aid. We assume responsibility for all remaining
tructure on output, employment and capital stock in the errors.
Public capital stock and economic growth: the Spanish economy 993
REFERENCES Munnell, A. (1992) Infrastructure investment and economic
growth, Journal of Economic Perspectives, 6, 189± 98.
Munnell, A. and Cook, L. (1990) How does public infrastructure
Aaron, H. (1960) Discussion of why is infrastructure important? in a€ ect regional performance?, New England Economic
A. Munnell (ed.) Is T here a Shortfall in Public Capital Invest- Review, Federal Reserve Bank of Boston, pp. 11± 32.
ment?, Conference Series no. 34, Federal Reserve Bank of Tatom, J. (1991) Public capital and private sector performance,
Boston, pp. 51± 63. Review, Federal Reserve Bank of St. L ouis, pp. 3± 15.
ArgimoÂn, I. and MartõÂ n, M. J. (1993) Series de Stock de Infraes- Tiao, G. C. and Box, G. E. (1981) Modeling multiple time series
tructuras del Estado y de las Administraciones PuÂblicas en with applications, Journal of the American Statistical Associ-
Espan4 a, Servicio de Estudios del Banco de Espan4 a, ation, 76, 802± 16.
documento de trabjo no. 9315. Wallis, K. F. (1979) Topics in Applied Econometrics, Basil Black-
ArgimoÂn, I., GonzaÂlez-PaÂramo, J. M., MartõÂ n, M. J. and RoldaÂn, well, Oxford.
J. M. (1994) Productividad e infraestructuras en la economõÂ a
espan4 ola, Moneda y Cre dito 198, 207± 52.
Aschauer, D. (1989a) Does public capital crowd out private capi-
tal?, Journal of Monetary Economics, 24, 171± 88. AP P EN D I X A
Aschuer, D. (1989b) Is public expenditure productive?, Journal of
Monetary Economics, 23, 177± 200. Multiplying Equation 8 by = , we have
Bajo, O. and Sosvilla, S. (1993) Does public capital a€ ect private
sector performance? an analysis of the Spanish case, = zt = h(B) a pt + g (B) a zt (A1)
1964± 1988, Economic Modelling, 10 (3), 179± 85.
where h(B) = = C p (B) and g (B) = = C z (B). Note that the
Baxter, M. and King, R. (1993) Fiscal policy in general equilibrium,
American Economic Review, 83 (3), 315± 34. coe cients in h(B) are the impulse response functions of = zt
Box, G. and Jenkins, G. (1970) T ime Series Analysis, Forecasting to a transitory shock in a pt .
and Control 2nd edn, Holden Day, San Francisco. Since in (A1) the elements of vector a zt are independent of
Corrales, A., and Taguas, D. (1989) Series macroeconoÂmicas parael a pt , the order k correlation between = ztj ( jth) element of = zt ,
perõÂ odo 1954± 88: un intento de homogeneizacioÂn, monografõÂ a
j = y, l, k ) and a pt will be given by:
no. 75, Instituto de Estudios Fiscales, Madrid.
Eberts, R. (1990) Public infrastructure and regional economic s
(k) = hj k
ap
r (A2)
Development: a simultaneous equation approach, Economic = zj , ap
s
Review, Federal Reserve Bank of Cleveland, 26, 15± 27. = zj
Flores, R. and Pereira, A. (1993) Public capital and economic From (A2) it is evident that the null hypothesis H0 : hj k = 0 is
growth in the United States. Is Public Capital Productive?, equivalent to H0 : r zj , ap (k) = 0. Then we will not reject
University of California, San Diego, Discussion paper 93± 31. =
H0 when (Box and Jenkins, 1976)
Ford, R. and Poret, P. (1991) Infrastructure and private sector
t t
productivity, OECD Economic Studies, 17, 63± 89. - < r= zj , ap (k) < (A3)
GarcõÂ a-MilaÂ, T. and McGuire, T. (1992) The contribution of pub- Ï n Ï n
licly provided inputs to States’ economies, Re gional Science
and Urban Economics, 22, 229± 41. where t is the critical value of the t-student distribution for
GarcõÂ a-Perea, P. and GoÂmez, R. (1993) ElaboracioÂn de series a speci® c critical level, n is the sample size and r zj , ap (k) is
histoÂricas de empelo a partir de la encuesta de poblacioÂn =
the method of moments estimator of r zj , ap (k). Then from
activa (1964± 1992), mimeo, servicio de estudios de banco de =
Espan4 a. (A2) and (A3) we will not reject H 0 : hj k = 0 when
Hamilton, J. D. (1994) T ime Series Analysis, Princeton University ts W zj ts W zj
Press, Princeton, NJ. - = < h= j k < = (A4)
Hillmer, S. C. and Tiao, G. C. (1979) Likelihood function of Ï n s W ap Ï n s W ap
stationary multiple autoregressive moving average models,
Journal of the American Statistical Association, 74 (367), where hà j k is the method of moments estimator of hj k . Since
652± 60. h= j k is less e cient than the maximum likelihood estimator of
Jenkins, G. M. and Alavi, A. S. (1981) Some aspects of modeling hj k , the bounds in expression (A4) are overestimated when
and forecasting multivariate time series, Journal of T ime the H0 : hj k = 0 is tested by using the maximum likelihood
Series Analysis, 2, 1± 47.
Johansen, S. (1988) Statistical analysis of cointegration vectors’,
estimator of hj k instead of hà j k .
Journal of Economics, Dynamics and Control, 12, 231± 54.
Molinas, C., Ballabriga, F. C., Canadell, E., Escribano, A., LoÂpez, Table A1. Con® dence intervals in (A.4 ) (90% con® dence level ) *
E., Manzanedo, L., Mestre, R., SebastiaÂn, M. and Taguas, D.
(1990) MOISEES. Un Modelo de Investigacio n y Simulacio n de Y L K
la Economõ Â a Espan4 ola, Antoni Bosch edn and Instituto de
Estudios Fiscales. 6 0.30 6 0.33 6 0.50
Munnell, A. (1990a) Why has productivity declined? Productivity
and Public Investment, New England Economic Review, Fed- Note that the con® dence intervals in Table A1 have been construc-
eral Reserve Bank of Boston, pp. 3± 22. ted for the impulse response estimates obtained by the method of
Munnell, A. (1990b) How does public infrastructure a€ ect regional moments, which are less e cient than the maximum likelihood
economic performance?, in A. H. Munnell (ed.) Is T here estimates. Since the test is performed with the maximum likelihood
a Shortfall in Public Capital Investment?, Federal Reserve estimates, it can be considered that the con® dence level of the test
Bank of Boston, Conference Series no. 34, pp. 69± 103. is higher than 90%.
994 R. Flores de Frutos et al.
APPENDIX B

Time series used in the paper: levels (1964± 92)

Year PK K L Y

1964 343.9375 4725.959 11015.8 6217.933


1965 375.0972 5288.623 11312.3 6613.399
1966 408.8559 5965.164 11432.2 7147.264
1967 445.6529 6528.834 11486.9 7516.708
1968 488.5923 7091.527 11523.2 8016.343
1969 525.1142 7874.398 11585.9 8758.965
1970 557.5542 8697.349 11568.2 9199.379
1971 607.9068 9327.783 11645 9684.519
1972 655.2556 10201.12 11829 10546.92
1973 694.0858 11283.67 12004.4 11393.02
1974 730.1529 12447.23 12132.5 11975.51
1975 771.3623 13380.59 11863 12042.4
1976 794.5914 14222.41 11580.3 12381.42
1977 822.8978 14913.12 11574.8 12701.69
1978 822.8956 15571.31 11288.5 12925.1
1979 820.1098 16139.78 11041.5 12833.96
1980 818.8977 16683.14 10608.3 12949.37
1981 832.1184 17032.49 10244.1 12767.2
1982 860.1015 17260.81 10079.4 12877.96
1983 891.6164 17467.24 9951 13005.45
1984 897.4133 17483.62 9591.2 13221.79
1985 914.5094 17505.66 9421.4 13372.28
1986 936.2418 17777.72 9576.5 13674.59
1987 970.8297 18480.07 9961.5 14472.84
1988 1035.115 19463.38 10286.6 15304.03
1989 1133.099 20734.82 10655 15983.32
1990 1240.304 21952.66 10891 16506.75
1991 1357.651 23083.25 10851 16803.86
1992 1486.102 23983.58 10590 16768.7

You might also like