National Saving
National Saving
National Saving
IGNAZIO VISCO
Research Department, Bank of Italy, Rome, Italy
Summary
1. Introduction
for net rates, for national and for private saving ratios, for both
“adjusted” and “unadjusted” magnitudes, major trends appear to
be beyond dispute and provide a somewhat unexpected description
of the behaviour of Italian consumers. As it turns out, from a peak
of about 24% at the beginning of the 1960s, the net national saving
rate has declined to about 9% in 1993.
In particular, in the early 1990s private saving rates (defined
as net of depreciation, adjusted for durables and inflation and
computed as ratios to net national disposable income) have been
fluctuating around 10%, nine percentage points down from the
average saving rate prevailing in the 1960s. In the same years,
government saving (adjusted for inflation) has plunged to an
unprecedented −1%, after more than 30 years of current account
surpluses averaging to 2%. As a result, today’s overall Italian
saving rate is about 15 percentage points below the one observed
during the so-called “Italian economic miracle”.
As can be easily gauged from Figure A where national, private
and government saving rates (corrected for durables and inflation)
are presented, the beginning of the decline in the Italian saving
rate can be placed just short of the middle of the 1970s, in
conjunction with the first oil shock, when the growth rate of the
Italian economy first started to fall. It is only in the 1980s,
however, that the reduction of the national saving rate became quite
substantial, fully reflecting the progressive decline of government
saving along with the negative trend of private saving.
Several alternative explanations for the decline of the saving
rate have been put forward by the most recent research.† In a
recent article (Rossi & Visco,1994), we have scrutinized the impact
of fiscal policy on private sector behaviour to suggest that a
thorough assessment of the interplay of private and public decisions
can shed some light on the evolution of the Italian saving rate in
the most recent decades. In particular, focusing on the 1951–1990
purchasing power that inflation induces on the stock of nominally denominated
debt. This is particularly relevant when national saving is split between its
private and government components. Ignoring the loss of principal due to inflation
implies an overestimation of private savings: this simply amounts to neglecting
a possibly important source of revenue for the government, the so-called “inflation
tax”. Finally, even if it is controversial, it is likely that saving should be considered
net of the depreciation of the capital stock. In this section, the movements of
national, private and government saving rates (all computed in relation to the
net national disposable income) are examined allowing for a proper treatment of
consumer durables and accounting for the capital stock depreciation and the
“inflation tax”. For detailed statistics on post war Italian saving trends see the
Statistical Appendix to Ando, Guiso and Visco (1994). The updated time series
used in this paper are contained in its working paper version, available on request
from the authors.
† Besides the papers by Fornero and Castellino (1990) and Jappelli (1991), the
interested reader is referred to Modigliani and Jappelli (1993) and to Ando, Guiso
and Visco (1994).
NATIONAL SAVING AND SOCIAL SECURITY IN ITALY 331
25
20
15
%
10
–5 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93
52 54 56 58 60 62 64 66 68 70 72 74 76 78 80 82 84 86 88 90 92
Year
0 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93
52 54 56 58 60 62 64 66 68 70 72 74 76 78 80 82 84 86 88 90 92
† It should be recalled that, if the retirement age is not fixed, such effect may
be partly offset by induced phenomena of early retirement.
‡ It is implicitly assumed here that the increase in transfer payments is
perceived as sustainable by the agents. This warning forcefully applies to the
Italian case where the inherent risk of unsustainability of the social security
system had been recognized since the late 1970s.
§ See the pioneering works of Feldstein (1974) and Barro (1974). Apart from
Rossi and Visco (1994), aggregate (but inconclusive) time-series evidence for Italy
is reported in Modigliani and Sterling (1983).
336 N. ROSSI AND I. VISCO
s
yd
=u 0+u 1 g+u2 r+u3
wr wsg
yd
+h
A B
yd
+u 4
ys
yd
, (1)
† In what follows, all consumption, income and wealth variables are in real
terms.
‡ The alternative specification, with disposable income defined as gross of
current social security contributions and pension wealth being net of their present
value counterpart, could also be tested. In the empirical analysis summarized in
Appendix 1, this has been done with such a specification always turning out to
be dominated by the one discussed in the text.
§ See Appendix A in the working paper version of this paper.
338 N. ROSSI AND I. VISCO
(yl), net income from capital (rwr) and net social transfers to
households (ys).†
A thorough statistical analysis, summarized in Appendix 1,
led to the error-correction representation of the data generating
process reported in Table 1,‡ where coefficient estimates, associated
standard errors and diagnostic statistics, when available, for both
least squares and instrumental variables estimates are also re-
ported. Comparison of LS and IV estimates suggests, however,
simultaneity bias to be negligible. In Table 1,§ ln(c/yd) is the
logarithm of the consumption to income ratio,¶ where c stands for
“economic” consumption and yd is net private disposable income,
adjusted for inflation;†† z stands for real government consumption;
and r is defined as ln[(1+R)/(1+p)], where R is the nominal
interest rate on long-term Government bonds and p is the expected
annual rate of change of consumer prices.‡‡
The error-correction estimates have a plausible economic in-
terpretation, and appear to possess acceptable statistical prop-
erties. Needless to say, to assess the relevance of Auerbach and
Kotlikoff’s (1983) assertion about the likely instability of aggregate
− 0·545[ln(c/yd)t−1+(0·288−0·033((wr/yd)t−1+0·658(wsg/yd)t−1)+0·321rt−1+(1–0·288)(ys/yd)t−1)]
(0·091) (0·041) (0·003) (0·091) (0·106) (0·041)
− 0·567[ln(c/yd)t−1+(0·298−0·033((wr/yd)t−1+0·650(wsg/yd)t−1)+0·336rt−1+(1–0·298)(ys/yd)t−1)]
(0·104) (0·041) (0·004) (0·090) (0·102) (0·041)
NATIONAL SAVING AND SOCIAL SECURITY IN ITALY
Note: Changes in the aggregate private saving rate are computed from period
averages of ratios of actual saving figures to the net disposable income of the
private sector adjusted for inflation. The contribution of g is arrived at setting the
short-run components of the error-correction representation at their equilibrium
values and solving for the growth equilibrium rate.
in the private saving rate. Finally, comparing the 1980s with the
1960s, slightly less than half of the entire fall of the private
equilibrium saving rate appears to be due to the increased gross
social security wealth to income ratio. The negative pension wealth
effect on saving in the period (some six percentage points) is
partially offset by the direct effect of pension expenditures on
disposable income (with a positive impact on the saving rate of
about four percentage points). To put it differently, it might be
said that, had the Italian households adhered to the standard life-
cycle framework (with h=0), the net private saving rate currently
observed would not be much different from the one prevailing in
the mid-1970s.
† See Brandolini and Cannari (1994) for a full length description of the Bank
of Italy SHIW.
NATIONAL SAVING AND SOCIAL SECURITY IN ITALY 343
si/yin=w0+w1(wri/yin)+w2(wisg/yin), (2)
legislation expectations
The central tenet of this paper is that the changes in social security
laws and regulations which took place in the late 1960s and
early 1970s severely weakened the link between contributions and
benefits permitting a time path of aggregate consumption in excess
of what would have occurred in the absence of such changes. It is
worth recalling that those changes entailed both a replacement
effect due to the introduction of an unfunded social security system
for specific segments of the population as well as, for a given-
institutional structure, a lifetime wealth increment.
The evidence emerging from aggregate time series as well as
from cross-section data provides a rather clear-cut outcome, in-
dicating that a large part of the decline of private saving (in terms
of net private disposable income) observed since the 1960s may be
attributed to the rise in the social security wealth to income ratio.
Given the specification adopted in this work, a major role is played
by the productivity slowdown.
Past trends have been, however, substantially reversed in the
last couple of years. A major reform was enacted in 1992 and a
further attempt to halt social security expenditure has taken place
in the spring of 1995. As Figure B shows, the 1992 reform may
have determined a substantial fall (as large as 20%) in the gross
social security wealth (a gigantic “capital levy” of about 2000
trillions lire), leading the gross pension wealth to net national
disposable income ratio to fall to 4·6 in 1993 from 5·9 in 1992. This
outcome appears to be confirmed by the information available at the
individual level: preliminary computation of expected household
social security wealth for 1991 and 1993 suggests an average fall
of about 10% (the difference being given by postponed retirement).
To the extent that the overall reform of the Italian pension
system just approved by Parliament will confirm the implications
of the 1992 measures, on the basis of the parameter estimates of
Table 1 it might be inferred that such dramatic changes in the
social security arrangements could bring about a rise of the private
saving ratio of up to 3 percentage points in the steady state, with
figures lower than 3% accounting for the fact that the saving
augmenting replacement effect may be counterbalanced by a saving
reducing retirement effect.
Acknowledgements
referee for their most helpful comments and criticisms, Luca Bel-
trametti for providing his social security wealth data, Massimo
Rostagno for helpful discussions on the same topic and Alessandro
Venturini for excellent research assistance. The usual disclaimer
applies. The views expressed herein are those of the authors and
do not necessarily reflect those of the Bank of Italy. Financial
support from the Italian National Research Council is gratefully
acknowledged.
References
Ando, A., Guiso, L. & Terlizzese, D. (1994). Dissaving by the elderly, transfer
motives and liquidity constraints. In A. Ando, L. Guiso & I. Visco, Eds.
Saving and the Accumulation of Wealth, pp. 188–213. Cambridge: Cambridge
University Press.
Ando, A., Guiso, L. & Visco, I. Eds. (1994). Saving and the Accumulation of
Wealth. Cambridge: Cambridge University Press.
Ando, A. & Modigliani, F. (1963). The “life cycle” hypothesis of saving: aggregate
implications and tests. American Economic Review, 53, 55–84.
Auerbach, A.J. & Kotlikoff, L.J. (1983). An examination of empirical tests of
social security and saving. In E. Helpman, A. Razin & E. Sadka, Eds. Social
Policy Evaluation: An Economic Perspective, pp. 161–179. New York: Academic
Press.
Barro, R.J. (1974). Are government bonds net wealth? Journal of Political Eco-
nomy, 82, 1095–1117.
Beltrametti, L. (1995). On pension liabilities in Italy. Ricerche Economiche, 49,
405–428.
Brandolini, A. & Cannari, L. (1994). Methodological appendix: the Bank of Italy’s
survey of household income and wealth. In A. Ando, L. Guiso & I. Visco, Eds.
Saving and the Accumulation of Wealth, pp. 369–386. Cambridge: Cambridge
University Press.
Brugiavini, A. (1991). Effetti delle pensioni sul risparmio nella teoria del ciclo
vitale. In T. Jappelli, Ed. Bilancio pubblico e risparmio privato, pp. 43–68.
Milano: F. Angeli.
Cannari, L., D’Alessio, G., Raimondi, G. & Rinaldi, A.I. (1990). Le attività
finanziarie delle famiglie italiane. Temi di Discussione No. 172. Roma: Banca
d’Italia.
Doornick, J.A. & Hendry, D.F. (1994). PcFiml 8.0. London: International Thomson
Publishing.
Engle, R.F. & Granger, C.W. (1987). Co-integration and error correction rep-
resentation, estimation and testing. Econometrica, 55, 255–276.
Feldstein, M. (1974). Social security, induced retirement, and aggregate ac-
cumulation. Journal of Political Economy, 82, 905–926.
Feldstein, M. (1976). Social security and saving: the extended life cycle theory.
American Economic Review, 66, 77–87.
Fornero, E. & Castellino, O. Eds. (1990). Formazione e impiego della ricchezza
delle famiglie. Torino: Fondazione Giorgio Rota.
Franco, D. (1992). Alcune note sull’espansione della spesa pubblica in Italia
(1960–1990). In Ente Einaudi, Il Disavanzo Pubblico in Italia: Natura Strut-
turale e Politiche di Rientro, pp. 211–355. Bologna: Il Mulino.
Franco, D. (1995). Pension liabilities. Their use and misuse in the assessment of
fiscal policies. Economic Papers, No. 110. Brussels: European Commission.
Franco, D. & Morcaldo, G. (1990). The Italian pension system: development and
NATIONAL SAVING AND SOCIAL SECURITY IN ITALY 347
−ln(c/yd)≈s/yd=u0+u1g+u2r+u3
A B
wr ws
yd
+h d
y
(1.1)
−ln(c/y)=u0+u1g+u2r+u3
A B
wr wsg
y
+h
y
(1.2)
and
−ln(c/yd)=u0+u1g+u2r+u3
A B AB
wr wsw
yd
+h
yd
+u4
ys
yd
(1.3)
† In this paper, as in Rossi and Visco (1994), the income flows are net of taxes
while the present values of current and future discounted flows are not.
‡ For a formal test of the alternative specification, see footnote † on p. 23.
§ Being aware of the double counting issue and to evaluate the robustness of
our results, in our 1994 paper we also considered a specification such as (1.1)
with ws replaced by wsw (but without allowing ys to enter the regression with a
different coefficient than y), to obtain results very close to those eventually
reported in that paper, with, if anything, a slightly stronger effect of social
security wealth on private consumption. Moreover, in conducting a simulation
experiment on microeconomic data we fixed h at 0·5 (as opposed to an estimate
of 0·3 in the time series model) to account, again, for double counting.
NATIONAL SAVING AND SOCIAL SECURITY IN ITALY 349
−ln(c/yd)=u0+u1g+u2r+u3
A B AB
wr wsg
yd +h d +u4
y
ys
yd
(1.4)
with u4=1−u0.
If the above constraint is satisfied, equations (1.2) and (1.4) may
be considered equivalent representations of consumer behaviour.
In what follows we shall first focus on the somewhat simpler
equation (1.2) and proceed, subsequently, to test the restriction
u4=1−u0.
Recent developments in the theory of cointegration have allowed
to explicitly link economic equilibrium relationships between a set
of variables such as those depicted by equation (1.2) with statistical
models of those variables (Engle & Granger, 1987). In testing for
the existence of cointegration, it is necessary to establish first that
the individual series are I(1) and then that there exists some non
trivial function of them which is I(0). The most common approach
is to start testing for a unit root in the univariate representations
of the individual series and then in the least squares linear
combination. Dickey–Fuller (DF) and augmented Dickey–Fuller
(ADF) statistics have been computed for the following variables:†
(i) the log of the consumption to income ratio, ln(c/y), (ii) the
private wealth to income ratio, wr/y, (iii) the gross social security
wealth to income ratio, wsg/y, (iv) the real interest rate, r.
In the 1951–1993 sample, the hypothesis of a unit root at the
5% level could never be rejected for the variables listed above.‡
In a multivariate context, though, there may be a number of
cointegrating regressions linking the saving rate, the real interest
rate and the wealth to income ratios. We therefore revert to the
† All estimates reported in this Appendix have been obtained using Microfit
3.0 (Version 386, see Pesaran and Pesaran, 1991) or Pcfiml (Version 8.0, see
Doornick and Hendry, 1994), where specific references are reported.
‡ The hypothesis of a unit root could, instead, be rejected for productivity and
population growth.
350 N. ROSSI AND I. VISCO
ln c ln y r wr/y wsg/y
−1·000 0·966∗ −0·155 0·022∗ 0·023∗
† In Table 5 White’s standard errors are reported under the coefficient es-
timates. The mis-specification tests statistics are distributed as follows: serial
correlation F(1, n−k−1), functional form F(1, n−k−1), normality v2(2), het-
eroskedasticity F(1, n−2), predictive failure F(n1, n−k), where n1 is the number
of observations in the shorter time period (1991–1993), additional regressors F(8,
n−k−8).
352 N. ROSSI AND I. VISCO
Short-run terms
D ln(yd)t 0·456 (0·044) 0·457 (0·045) 0·459 (0·042)
D ln(ys/yd)t−1 0·093 (0·041) 0·093 (0·044) 0·098 (0·043)
D ln zt −0·482 (0·126) −0·507 (0·142) −0·476 (0·130)
D ln zt−1 −0·332 (0·168) −0·358 (0·174) −0·340 (0·168)
D ln ct−1 0·271 (0·078) 0·253 (0·083) 0·275 (0·081)
Error-correction term
ln(c/y)t−1 −0·538 (0·088)
ln(c/yd)t−1 −0·516 (0·102) −0·545 (0·091)
Long-run terms
intercept −0·126 (0·031) −0·128 (0·057) −0·157 (0·039)
rt−1 −0·198 (0·066) −0·184 (0·064) −0·175 (0·063)
(wr/y)t−1 0·015 (0·003)
(wr/yd)t−1 0·016 (0·005) 0·018 (0·004)
(wsg/y)t−1 0·010 (0·002)
(wsg/yd)t−1 0·014 (0·004) 0·012 (0·002)
(ys/yd)t−1 −0·546 (0·209) −0·388 (0·063)
Misspecification tests
R̄2 0·866 0·864 0·866
dw 2·030 2·079 2·010
r̂100 0·836 0·845 0·838
Serial correlation 0·147 0·273 0·095
Functional form 1·387 1·135 1·237
Normality 1·017 1·117 0·952
Heteroskedasticity 0·843 1·449 0·537
Predictive failure 0·911 0·729 0·894
Additional regressors 0·485 0·489 0·337
No. of observations (n) 40 40 40
No. of regressors (k) 10 11 10
TABLE 6 Estimates of h
Note: in parentheses, White’s standard errors. Equations (1), (2) and (3) as in
Table 5. Equations (4), (5) and (6) as in Table 5 with additional regressors as
indicated. Equations (7), (8) and (9) as (1), (2) and (3), respectively, with wsg
t−1
s
replaced by wt−1 .
The evidence from the ECM estimates then confirms that stem-
ming from the analysis of cointegration. Table 6 presents a sum-
mary of the estimates of h in equations (1.2) and (1.4) as can
be obtained by the estimated coefficients presented in Table 5.
Estimates of the respective standard errors are also provided. It
is clear that this parameter appears to be sizeable and, as already
observed, not far from one. To further check the robustness of this
result a few attempts have been made, as shown in Table 6, to
check whether restricting to zero the coefficient of wsc would
dramatically change the picture. Again we obtain very similar
results, with, if anything, estimates of h even closer to one.†
Since model (1.4) can be derived from model (1.2) if the restriction
u4=1−u0 is satisfied and since such restriction turns out not to
be rejected by the data (F(1, 29)=0·555) it seemed safe to use in
the text the specification of column (3) of Table 5, which is written
in terms of the standard saving to disposable income ratio. The
corresponding (non-linear) restricted least-squares and in-
strumental variables estimates are presented in Table 1 of the
text.
Legislation Expectations
Notes: (i) in parentheses, White’s standard errors, (ii) the sample excludes
households with pension income only, (iii) the reported estimates and standard
errors refer to a household living in a small town in Northern Italy, formed by
two adults, both income recipients, and two children, and with the head of the
household in his late twenties or early thirties.