You are on page 1of 12

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/24095455

Growth Rates Of Per-Capita Income And Aggregate Welfare: An International


Comparison

Article  in  Review of Economics and Statistics · February 1997


DOI: 10.1162/003465397556782 · Source: RePEc

CITATIONS READS

21 122

1 author:

Nanak Kakwani
UNSW Sydney
193 PUBLICATIONS   6,842 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Essay on Economic Growth and Poverty Reduction View project

Social welfare functions and Development: Measurement and Policy Applications View project

All content following this page was uploaded by Nanak Kakwani on 15 May 2015.

The user has requested enhancement of the downloaded file.


GROWTH RATES OF PER-CAPITA INCOME AND AGGREGATE WELFARE:
AN INTERNATIONAL COMPARISON

Nanak Kakwani*

Abstract —This paper is concerned with the measurement of aggregate have a time series on per-capita growth domestic product
growth rates, where the aggregation is over time. The paper demon-
strates that any mechanical procedures for computing aggregate growth (GDP) (income) or some other indicator of welfare. It is
rate has welfare implications, and value judgments implicit in various common to compute an average growth rate, and to interpret
commonly used procedures are not appealing. A new procedure sug- a higher average growth to mean higher welfare. This
gested in the paper captures all the essential properties of a welfare
function. The methodology of the paper is applied to an analysis of growth interpretation of growth rates suggests that there must exist a
rates of per capita GNP of 83 developing countries during the 1970 –1987 relationship between growth rates and changes in welfare
period. levels. This paper explores this relationship using alternative
growth procedures.
I. Introduction The most common method used to compute an average
growth rate is to regress the natural logarithm of income on
T HE gross national product (GNP) per head and the
related income measures are widely used to appraise the
economic well-being of people living in different countries.
time. The average growth rate is the antilog of the least-
squares coefficient estimate on time minus 1. This method is
used widely by the United Nations and other international
These measures have been subject to much criticism for
organizations and individual countries.
their failure to give any indication of how the total output of
The issue to be discussed in this paper is whether the
a country is distributed among its people. 1 Recently many
widely used least-squares procedure is the appropriate one if
economists, the most notable of them being Sen (1985),
our objective is to see how the economic welfare of people
have been concerned with whether these aggregated income
measures re ect the well-being of people. Despite these has changed over a period of time. The paper develops a
criticisms, aggregate income statistics continue to be widely conceptual framework that may be used to derive aggregate
used to distinguish rich and poor countries, perhaps because growth rates from a welfare function deŽ ned in terms of
of readily available national accounts for most countries of levels of per-capita incomes in different years. Using this
the world. framework, welfare implications of alternative growth rate
The present paper is concerned with the measurement of estimation procedures are derived. A new procedure sug-
growth rates of various broad economic indicators. It is gested here is shown to be better because it captures the
clearly important to determine whether and to what extent essential properties of a welfare function.
people are becoming better or worse off over time. Not Ideally the aggregate growth rate should take into account
surprisingly, therefore, magnitudes of growth rates are the changes in income inequality among individuals. This is
always the focus of attention when economists discuss seldom done in practice—because the computation of growth
alternative strategies of economic development. However, rates is never related to welfare. The methodology presented
how these growth rates should be computed is seldom in this paper can effectively handle the distributional issue
discussed in the economic literature.2 by deŽ ning the welfare function over a period of time in
There may be several purposes for measuring growth terms of equally distributed equivalent levels of income for
rates. One purpose may be to see how the structure of an each year (Atkinson, 1970).
economy has been changing over time by looking at sectoral In this paper we also address the issue of structural change
growth rates; another is whether the economy has moved in the economy. The question asked is whether the economy
from a lower (higher) to a higher (lower) growth path. has moved from a lower (higher) to a higher (lower) growth
Although we deal with the latter issue in this paper, our main path. It has been demonstrated that one can obtain subperiod
focus is on the welfare aspect of growth rates. Suppose we growth rates based on a welfare function deŽ ned over the
entire period. Boyce (1986) has observed that the least-
squares growth rate may give rise to an anomaly that the
Received for publication April 12, 1994. Revision accepted for publica- whole-period growth rate may lie outside the range of
tion July 18, 1995. subperiod growth rates or, in other words, the total growth
* University of New South Wales.
I am grateful to the two referees of the REVIEW whose comments proved rate may be negative (or positive) when subperiod growth
very helpful.
1 See Sen (1973, 1974) and Kakwani (1981), who have attempted to
rates are positive (or negative). Our proposed procedure of
derive alternative welfare measures, which take into account both the size
computing subperiod growth rates does not give rise to such
and the distribution of income. anomalies.
2 Although we recognize that the reliability and comparability of national
The methodology developed in the paper is applied to an
accounts data are subject to serious limitations, this issue has not been
considered in this paper. However, the World Bank puts enormous effort analysis of growth rates of per-capita GNP of 83 developing
into improving these data on a continual basis. countries during the 1970 –1987 period.

r 1997 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology [ 201 ]
202 THE REVIEW OF ECONOMICS AND STATISTICS

II. Least-Squares Growth Rate such that S nt5 2 wt 5 1.0. Equation (5) implies that the
least-squares growth rate for the whole period is approxi-
Let x 5 (x1, x2, . . . , xn ) be the vector of values of an
mately equal to a weighted average of the annual growth
economic indicator (such as real GDP or real GNP per head)
rates. It can be seen that the weight wt increases with t
given for n periods. We ask: what is the most appropriate
until t 5 n/2 1 1, and then it decreases.
single growth rate of x for these n years? The most
Thus the least-squares growth rate (LSGR) procedure
commonly used procedure for estimating this growth rate is
gives maximum weight to the growth rates in the middle of
the least-squares method. The least-squares growth rate R is
the time period (over which it is computed). The lower
estimated by Ž tting a trend line
weights are given to the growth rates at the beginning and at
the end of the time period.
log xt 5 a 1 b t1 e t (1)
We have demonstrated that the LSGR procedure, which
appears to be a mechanical one of Ž tting a time trend, has an
where t varies from 1 to n, a and b are the parameters to be
intuitive interpretation. A natural question is whether the
estimated, and e t is assumed to be the white noise error
weighting scheme implied by it is generally acceptable. Are
term.3 This model is motivated by the logarithmic transfor-
there alternative procedures which are intuitively more
mation of the compound growth rate equation
appealing? We discuss alternative procedures in the next
section.
xt 5 x1 (1 1 R ) t2 1. (2)

a 5 log x1 2 log (1 1 R), x1 being the value of x in the Ž rst III. Alternative Procedures
period and b 5 log (1 1 R).
If b ˆ is the least-squares estimate of b , the estimated In the previous section we demonstrated that the LSGR
growth rate R̂ is obtained as antilog (b ˆ ) 2 1. It can be shown procedure gives different weights to the annual growth rates
that within the period. One obvious alternative is to give equal
weights to the annual growth rates in equation (5), i.e., wt 5
n 1/(n 2 1). Denoting the estimated growth rate by this
o t5 1
log xt (t 2 t) procedure as R̂ 1, equation (5) then gives4

bˆ 5 log (1 1 R̂ ) 5 (3) n

o
n 1
o t5 1
(t 2 t) 2 log (1 1 R̂1 ) 5 wt 5
n2 1 t5 2
log (1 1 rt ) (7)

where t5 (n 1 1)/2. Let which can further be simpliŽ ed to

rt 5
xt 2
xt2
xt2
1
1
(4) R̂ 1 5 ()
xn 1/(n2
x1
1)
2 1. (8)

be the growth rate for the tth year, where t varies from 2 to n. R̂ 1 uses only the Ž rst and last observations of the period.
The question then arises as to how the total growth rate R̂ Clearly, this procedure, which will be referred to as geomet-
(for the entire period) is related to the growth rates in each ric mean growth rate (GMGR), is inefficient because the
year. To answer this question, substitute equation (4) into ‘‘average’’ growth rate obtained from it is completely
equation (3), which after algebraic manipulations leads to insensitive to the values of x between the end points of the
period. If we are interested in using R̂ 1 as long-term growth
n rate in order to draw welfare conclusions, this procedure
bˆ 5 log (1 1 R̂ ) 5 o t5 2
wt log (1 1 rt ) (5) may yield particularly misleading results.
The time trend equation (1) was obtained from the
compound growth rate equation (2), with a and b estimated
where by the least squares. Since the initial value x1 is known, an
alternative procedure for estimating R would be to impose
6(t 2 1)(n 1 12 t) the restriction that a 5 log x1 2 log (1 1 R). This restricts
wt 5 (6)
n (n 1 1)(n 2 1)
4 Note that log (1 1 R̂1 ) in equation (7) can also be obtained by applying
the OLS to equation (1) after taking the Ž rst differences. This estimator
It means that E(e t ) 5 0, E (e 2t ) 5 s 2 (constant), and E(e te t8) 5 0 for all
3 will have the minimum variance only if it is assumed that D e t 5 e t 2 e t2 1 is
tÞ t8. Under these assumptions, the ordinary least-squares (OLS) method the white noise error term, which means e t is generated from a Ž rst-order
applied to equation (1) will result in estimates of a and b which have the autoregressive process AR(1) with autocorrelation equal to 1. Such a
minimum variance. process is not stationary.
COMPARISON OF INCOME GROWTH RATES AND AGGREGATE WELFARE 203

the time trend equation (1) to pass through x1. Therefore TABLE 1.—CHARACTERISTICS OF COUNTRIES WITH NEGATIVE PER-CAPITA
GROWTH RATES , 1970–1987
equation (1) becomes
Total Countries with Negative
Number Growth Rates
(log xt 2 log x1 ) 5 b (t 2 1) 1 e t (9) of
Region Countries LSGR GMGR RSLGR IWGR AMGR
where t 5 2, 3, . . . , n. Applying the least squares to equation Africa 36 19 20 16 23 19
(8) yields Middle East and
eastern Europe 13 0 0 0 1 0
n Asia 12 0 0 1 0 0

o
Latin America 22 9 7 7 11 6
(log xt 2 log x1 )(t 2 1) All developing
t5 2 countries 83 28 27 24 35 25
bˆ 5 log (1 1 R̂ 2 ) 5 (10)
n

o t5 2
(t 2 1) 2

which is nothing but the simple average of yearly growth


where R̂ 2 is the estimated ‘‘average’’ growth rate by this rates. This will be called the arithmetic-mean growth rate
procedure. After algebraic manipulations, equation (10) (AMGR).
yields We may now compare average growth rates computed by
the alternative procedures. The calculations were performed
n on the World Bank data from 83 developing countries
log (1 1 R̂ 2 ) 5 o t5 2
wt log (1 1 rt ) (which were taken from the World Bank data Ž les by means
of the ANDREX system). Real GNP per capita was used to
compare the economic performance of countries over the
where
period of 1970 –1987. All growth rates were computed in
constant prices using the local currencies. Details of the
3[n (n 2 1) 2 (t 2 1)(t2 2)]
wt 5 (11) numerical results are not presented here, but are available in
n (n 2 1)(2n 2 1) Kakwani (1991).
Our results show that growth rates computed by the
such that S nt5 2 wt 5 1.0. This procedure will be referred to as alternative procedures vary substantially for a large number
the restricted least-squares growth rate (RLSGR) procedure. of countries. For example, most Latin American countries
It can be seen that wt in equation (11) is a decreasing performed well in the early seventies and extremely badly in
function of t, which means that RLSGR gives maximum the eighties. Any aggregation procedure giving higher
weights to the growth rates in the beginning of the period. weight to growth rates in the beginning of the period would
An alternative weighting scheme would be the one in which show higher values of the average growth rate for these
wt is a monotonically increasing function of t. A simple countries. This was quite evident in the numerical results. In
weight function satisfying this property is 18 out of 22 Central and South American countries, RLSGR
(for which the weight given to growth rates decreases
2t monotonically with time) showed higher values of the total
wt 5 (12) growth rate than the other alternative procedures. According
n2 1 n2 2
to RLSGR, Brazil’s per capita GNP grew at an annual rate of
such that S tn5 2 wt 5 1.0. Denoting the estimated ‘‘average’’ 4.33% during the 1970 –1987 period, but LSGR and IWGR
growth rate by R̂ 3, equation (12) gives procedures showed a growth performance of only 2.84 and
2.58%, respectively, during the same period.
n India’s growth performance was considerably better in the
2 o t5 2
t log (1 1 rt ) 1980s than in the 1970s. Accordingly IWGR, which gives
higher weights to growth rates in the most recent period,
log (1 1 R̂ 3 ) 5 . (13) showed a higher growth performance. Bangladesh’s growth
n2 1 n2 2
rate varied from 2 0.4% to 1.07%, depending on which
This procedure will be referred to as an increasing-weight procedure is used.
growth rate (IWGR) procedure. Table 1 reports the number of countries that had negative
Finally, we present the simplest way of computing the per capita GNP growth rates during the period of 1970 –1987
average growth rate, according to the various procedures. It can be seen that the
number of countries that experienced negative growth rates
n vary from 24 to 35, depending on which procedure is used to
o
1
R̂ 4 5 rt (14) compute growth rates. RLSGR yields the smallest number
n2 1 t5 2 of such countries, IWGR largest.
204 THE REVIEW OF ECONOMICS AND STATISTICS

The above analysis clearly demonstrates that inferences AXIOM 1: W(x) should be strictly increasing in each argu-
concerning the economic performance of countries can vary ment, i.e., ­ W/­ xi . 0 for all
substantially with respect to the procedure employed. It is,
therefore, of considerable importance to choose the most i5 1, 2, . . . , n.
appropriate computational procedure. The next two sections
provide an evaluation of the alternative procedures from a AXIOM 2: W(x) 5 W(p (x)), where p is any permutation of
welfare point of view. x. Axiom 1 is a natural requirement of a welfare function
whose arguments are annual per-capita income of a country.
Welfare of a country over a period should always increase
IV. Aggregate Growth Rate and Welfare when per-capita income in any year is increased.
Axiom 2 implies a symmetric welfare function, which
Let r2, r3, . . . , rn be the (n 2 1) annual growth rates means that if we interchange incomes in 2 years, the welfare
calculated over the period from 1 to n years. We deŽ ne the function remains unchanged. For example, two income
aggregate growth rate R as a function of the annual growth streams (100, 200, 300) and (100, 300, 200) should be iden-
rates, tical with respect to their welfare levels. This requirement
implies that exactly the same weight is given to income
R5 R (r2, r3, . . . , rn ) (15) levels irrespective of the year in which they are earned.
Intuitively, an income of $100 should make the same
where rt is given by xt 5 xt2 1 (1 1 rt ). contribution to total welfare irrespective of whether $100 is
We noted in the previous two sections that there exist earned (or enjoyed) in 1970 or in 1975. This requirement of
several mechanical procedures to compute R and inferences a symmetric welfare function seems natural, particularly
concerning the economic performance of countries vary when a reasonable justiŽ cation for a nonsymmetric welfare
substantially with respect to the procedure employed. Which function does not exist.6
procedure should one then employ to compute the aggregate We may now present Axiom 3, which relates to the
growth rate? aggregate growth rate with the welfare function.
Since the aggregate growth rate is computed to determine
whether and to what extent people are becoming better off or AXIOM 3: If R and R* are the aggregate growth rates
worse off over time, it is natural to relate the growth rate computed from the income streams x and x*, respectively,
with the level of welfare people enjoy. In this section we then the following statements are equivalent:
present some welfare criteria (in the form of axioms) which (a) R $ R*.
can be employed to evaluate the available procedures for (b) W(x) $ W(x*).
computing aggregate growth rates.
The social welfare function as introduced by Bergson in This axiom is the main focus of our paper. It implies that
1938 and subsequently developed by Samuelson in 1947 is the higher growth rate is always welfare superior, i.e., if
generally deŽ ned over a set of individuals. Following the there is an improvement (deterioration) in the growth rate
during a period (for instance, between 1970 and 1980), then
same idea, one can deŽ ne a welfare function for an
it should imply a higher (lower) level of welfare for that
individual or a group of individuals over a period of time.
period.
For instance, let x 5 (x1, x2, . . . , xn ) be a vector of values of
We propose the following procedure to obtain the aggre-
a country’s real per-capita income for n years. We then
gate growth R from a welfare function. Let rt be the growth
deŽ ne a welfare function for that country over n years as5 rate of a country in the year from (t 2 1) to t. Then, by
deŽ nition, xt 5 xt2 1 (1 1 rt ) must hold. Substituting sequen-
W (x) 5 W (x1, x2, . . . , xn ). (16) tially xt in terms of x1 gives
We may now propose the following two axioms, which must xt 5 x1 (1 1 r2 )(1 1 r3 ) · · · (1 1 rt ) (17)
be satisŽ ed by this welfare function.
6 In many economic policy formulations we discount the future utility,

which means that the greater weight is given to the current utility
5 To make the aggregate growth rate sensitive to changes in the
compared to that in the future. The main motivation for discounting the
distribution of income within a country, the welfare function in equation future utility is that the individuals prefer to consume in the current period,
(16) should be deŽ ned as but if they forgo current consumption, they must be compensated so that
they can have a higher consumption in the future. This issue is different
W (x * ) 5 W (x *1, x *2, . . . , x *n ) from the one we are dealing with here. We are not concerned with
maximization of an individual’s utility. We are dealing with the measure-
where x*t, t 5 1, 2, . . . , n, is the equally distributed equivalent level of ment of welfare over time given a stream of consumption or income. How
income for year t (Atkinson, 1970). This can be computed if we have the individuals should allocate consumption in different years is not our
income distribution data for each year. Thus the methodology presented concern. Moreover, many writers are against the discounting the future
here is easily extendable so that the country’s aggregate growth rate takes utility (see, for instance, Sen (1961), Weizsäcker (1965), and Ramsey
account of changes in income inequality within the country. (1928)).
COMPARISON OF INCOME GROWTH RATES AND AGGREGATE WELFARE 205

which on substituting into equation (16) yields substituting in equation (20) gives

W (x) 5 W [x1, x1 (1 1 r2 ), x1 (1 1 r2 ) n

(1 1 r3 ), . . . , (1 1 rn )].
(18) W (x) 5 log x1 1 log (1 1 R) o t5 1
(t 2 1) vt. (21)

Similar to the idea of an equivalent per-capita GNP, we Equating (20) to (21) yields
may introduce a new concept, to be called an equivalent
uniform growth rate. This is the constant growth rate that n
would result in the same level of welfare as the observed
incomes in n years. Letting R be the equivalent uniform
o t5 2
vt (log xt 2 log x1 )
growth rate, then the welfare level obtained by this growth log (1 1 R) 5 n
(22)
rate R must be equal to the welfare level obtained from the
observed incomes in n years.
o t5 2
(t 2 1) vt
Thus if W is homothetic, the condition
which expresses R as a function of x1, x2, . . . , xn. To express
W [x1, x1 (1 1 r2 ), x1(1 1 r2 )(1 1 r3 ), R in terms of the growth rates r2, r3, . . . , rn, we write
x1 (1 1 r2 )(1 1 r3 ) · · · (1 1 rn)]
(19) t
5 W {x1, x1 (1 1 R ), x1 (1 1 R ) 2, . . . , log (xt ) 2 log (x1 ) 5 o log (1 1 rj ) (23)
x1 (1 1 R ) n2 1
} j5 2

will give R as a function of r2, r3, . . . , rn.7 which is derived from log (1 1 rt ) 5 log xt 2 log xt2 1.
Thus using equation (19) R can be derived directly from a Substituting equation (23) into equation (24) gives
welfare function. It is clearly in variant with respect to any
n

o
positive linear transformation of the welfare function W. If
W is homothetic, R will be scale independent, implying that log (1 1 R) 5 wt log (1 1 rt ) (24)
t5 2
x can be measured in any unit. Further it can be seen that if W
satisŽ es Axioms 1 and 2, then R derived by the above where
procedure will always satisfy Axiom 3.
The above formulation suggests that all procedures for n
estimating growth rates rely on an implicit welfare function.
In the next section we evaluate alternative procedures by o j5 t
vj
examining these implicit welfare functions. wt 5 n
(25)

o t5 2
(t 2 1) vt
V. An Evaluation of Alternative Procedures
Let us consider a general class of welfare functions that such that S nt5 2 wt 5 1. Thus we have proved that the class of
are homothetic in incomes; welfare functions (20) results in the aggregate growth rate of
the whole period being approximately equal to the weighted
n average of the growth rates in each year. All the growth rate
W (x) 5 o t5 1
vt log xt (20) estimation procedures discussed in sections II and III (with
the exception of AMGR) belong to the class of growth rates
in equation (24), which are generated from the welfare
where log xt may be interpreted as the welfare or utility
function (20).
enjoyed by a country in year t and vt is the weight attached to
Using equation (25), we can derive implications for
tth year utility, such that
welfare weights vt given wt. Thus we can evaluate alternative
n growth procedures on the basis of welfare weights implied
o t5 1
vt 5 1.0. by them.
Let us Ž rst consider the GMGR procedure for which wt 5
1/(n 2 1) for all t $ 2. Equation (25) yields the welfare
If R is the uniform growth rate in period 1 to n years, then function
xt 5 x1 (1 1 R) t2 1 must hold for all t 5 1, 2, . . . , n, which on
n2 2 1
W (x) 5 log x1 1 log xn (26)
7 Note that x1 will cancel out when W is homothetic. n2 1 n2 1
206 THE REVIEW OF ECONOMICS AND STATISTICS

which shows that the welfare weights are zero for the years This implies that only the utility in the most recent year
from 2 to (n 2 1). Thus the welfare function is completely receives a positive welfare weight, all other utilities receive
insensitive to the utilities enjoyed by the country between negative weight. This procedure violates all three axioms.
the end points of the whole period. Such a welfare function Finally we consider the AMGR procedure. It can be seen
is not acceptable because it violates Axiom 1, a minimum that this procedure does not belong to the class of growth
requirement of a welfare function. rate procedures in equation (24). However, it is easy to show
Next we consider the most commonly used LSGR proce- that the welfare function
dure. Using equation (6) we obtained the welfare weights as

vt 5
6k(2t 2
n (n 1
n2
1)(n 2
1)
1)
W (x) 5
n2
x1
(
1 x1
x2
1
x3
x2
1 ··· 1
xn
xn2 1
)
will lead to the AMGR procedure. This welfare function is
for t $ 2, k being the constant of proportionality. This also not intuitively appealing because an increase in per-
equation implies that vt 5 0 for t 5 (n 1 1)/2, and positive capita GDP in a particular year (holding other incomes
(negative) for t being greater (less) than (n 1 1)/2. This is a constant) may lead to a reduction in welfare. This can be
peculiar welfare function. It assigns negative weights to the seen from the two country examples considered above.
utilities enjoyed by the countries in the Ž rst half of the total According to AMGR, country 1 has an average growth rate
period. To highlight this peculiarity, let us consider a simple of 52.08%, whereas country 2 has an average growth rate of
example of income streams of two countries in Ž ve periods: 64.58%. This implies an inverse relationship between the
in country 1—100, 200, 300, 400, and 500—and in country average aggregate growth rate and aggregate welfare, which
2—100, 100, 300, 400, and 500. Country 1 clearly enjoys a violates our main Axiom 3.
greater level of welfare than country 2, because its income in The above analysis suggests that none of the mechanical
period 2 is twice as large (incomes in the other periods being procedures for computing average growth rates are appropri-
the same). However, the LSGR for countries 1 and 2 are ate from the welfare point of view. In the next section we
48% and 58%, respectively. This demonstrates the possibil- propose a new procedure which satisŽ es all the three axioms
ity of an inverse relationship between aggregate growth rate proposed in this paper.
and aggregate welfare. This feature of the LSGR procedure
is unacceptable if we want to interpret a higher growth rate VI. New Procedure
as indicating a higher level of welfare, as is commonly done.
The LSGR procedure is used widely by the United Nations Using the framework given in section IV, we derive a new
and other international organizations, but it violates all three procedure for estimating growth rates. We begin with the
axioms proposed here. simple symmetric welfare function
Does the RLSGR have the desirable welfare properties?
n

o
To answer this question we utilized equation (10) and (11) to 1
obtain the welfare weights, W (x) 5 u (xt ) (28)
n t5 1

k(t 2 1)
vt 5 , for t . 1. (27) where u(xt ) is the utility enjoyed by a country in year t. If
n (n 2 1)(2n 2 1) u8(xt ) . 0, W(x) will be a strictly increasing function of
incomes in each year. This welfare function satisŽ es Axioms
These weights are all positive, implying that every year’s 1 and 2.
utility has a positive impact on the total welfare. This To obtain a scale-independent aggregate growth rate R as
welfare function satisŽ es Axiom 1 and is clearly more deŽ ned in equation (15), it is necessary that u(xt ) be
acceptable. However, the weights given to each year’s utility homothetic. The class of homothetic utility functions is
are not uniform. It gives least weight to utility enjoyed in the given by
beginning of the period and the largest weight to utility in
the most recent year. Thus it violates Axiom 2. e
Bx 1t 2
The IWGR procedure yields (on using equations (25) and u (xt ) 5 A1 , eÞ 1
(26)) the following welfare weights: 12 e (29)

2k 5 A1 B log xt, e5 1
vt 5 2 , for 1 , t, n2 1
n2 1 n2 2
where e . 0 is the measure of relative risk aversion, which is
constant for this utility function (Atkinson, 1970).
2k
5 , for t5 n. Applying equation (19) on equation (28) and using
n2 1 n2 2 equation (29), we can solve R in terms of r2, r3, . . . , rn. It is
COMPARISON OF INCOME GROWTH RATES AND AGGREGATE WELFARE 207

obvious that for practical purposes, R should be a unique where


function of r2, r3, . . . , rn. When e Þ 1, this unique function is
not obtainable. Thus we must assume that e 5 1; or in other 2 2(t 2 1)
words, the utility function must be of the logarithmic type wt (s) 5 , 2# t# s
(which satisŽ es the well-known Bernoulli hypothesis). For n (n 1 12 2s)
this utility function the aggregate growth rate is given by
2(n 2 t1 1)
n
t.
o
5 , s.
log (1 1 R) 5 wt log (1 1 rt ) (30) n (n 1 12 2s)
t5 2

It can be seen that when s 5 1, wt (s) 5 wt, as given by


where
equation (31). If xn is chosen as the focal point, then wt (s) is
given by
2(n 2 t1 1)
wt 5 (31)
n (n 2 1)
2(t 2 1)
wt (n) 5 .
such that S n
wt 5 1.0.
t5 2
n (n 2 1)
It can be seen that wt in equation (31) is a decreasing
function of t, which means that the proposed procedure gives Thus our methodology allows us to choose any focal point,
maximum weight to the growth rates in the beginning of the but x1 seems to be the most natural choice.
period. The least weight is given to the most recent growth Summarizing the above discussion we can say that of all
rates. Is this weighting scheme appropriate from the welfare the aggregate growth procedures, the proposed method is the
point of view? The answer would seem to be yes. Consider, most desirable one. First of all, we have derived it from a
for instance, two situations: in the Ž rst there is a growth rate welfare function, and therefore, it provides a positive
of 10% in the Ž rst year and a growth rate of 20% in the relationship between the aggregate growth rate and the
second year; and in the second situation there is a growth aggregate welfare. If a higher growth rate is preferred to the
rate of 20% in the Ž rst year and 10% in the second year. The lower growth rate, then an increase in growth rate should
proposed procedure would dictate that the second situation imply a higher level of welfare. Second, it is simple to
be preferred to the Ž rst. Obviously income levels in three compute. It is equal to the weighted average of the logarith-
years (starting with 100) for the Ž rst and second situations mic function of yearly growth rates.
would be 100, 110, 132 and 100, 120, 132, respectively. The
second situation is obviously welfare superior to the Ž rst
situation. Therefore higher growth rates in the beginning of VII. Subperiod Growth Rates
the period are preferable to higher growth rates in later This section addresses the issue of structural change in the
periods. Thus the weights given in equation (31) are economy. The question we ask is whether the economy has
meaningful from the welfare point of view. moved from a lower (higher) to a higher (lower) growth
It should be pointed out that the aggregate growth rate R path. Therefore we divide the total period into two subperi-
in equation (30) is not invariant to the choice of a focal point. ods. Let R1 be the aggregate growth rate of the Ž rst
Since R measures the aggregate change in the per-capita subperiod (1 to n1 ) and R2 that of the second subperiod (n1 to
income, it is essential to specify a focal point (or the initial n) so that n1 # n. The economy has moved into a higher
point) from which the change should be measured. We have (lower) growth path if R2 . R1 (R 2 , R 1 ). Therefore the
chosen the focal point to be x1, which seems to be a natural compound growth rate, equation (2), can be rewritten as
choice when we calculate the aggregate growth rate over
periods 1 to n. But we do not have to be restricted to a single
focal point. 8 An alternative focal point may be the other end xt 5 x1 (1 1 R1 ) t2 1, if t # n1
point xn, though any intermediate points may also be used. (33)
Suppose we choose a focal point xs, where s lies between 1 5 x1 (1 1 R1 ) n12 1 (1 1 R 2 ) t2 n1
, if t $ n1
and n, 1 # s # n. Then with some algebraic manipulations
we can demonstrate that the aggregate growth rate R(s) is If the aggregate growth rates in the two subperiods are equal,
given by i.e., R 1 5 R2 5 R, then equation (33) reduces to
n

log [1 1 R (s)] 5 o t5 2
wt (s) log (1 1 rt ) (32) xt 5 x1 (1 1 R ) t2 1
(34)

8 This point was made by one of the referees of the REVIEW . where R is the aggregate growth rate for the entire period.
208 THE REVIEW OF ECONOMICS AND STATISTICS

Let x* be the average welfare level (equivalent level of or, in other words, R1 # R # R 2 may not hold. It means that
per-capita GNP) in the entire period. Then x* is given by the total growth rate may be negative (or positive) when the
subperiod growth rates are both positive (or negative). Our
n

o
1 empirical investigations suggested that in 18 out of 77
log x* 5 log xt. (35) countries, the total growth rate computed by the LSGR
n t5 1
procedure was outside the range of the subperiod growth
rates. For instance, Bangladesh achieved an average growth
Substituting equation (34) into equation (35) gives
rate of 1.3% over the period of 1970 –1987 while the growth
rates for the subperiods of 1970 –1979 and 1979 –1987 were
n2 1
log x* 5 log x1 1 log (1 1 R) (36) found to be less than unity. Similarly, Chad achieved a
2 growth rate of 2.3% during the 1970 –1987 period while the
growth rates for the subperiods were 2 1.88% and 1 1.48%,
which provides the relationship between the aggregate respectively. Thus absurdities arising from the conventional
welfare level measured by x* and the aggregate growth rate LSGR procedure can occur frequently.
R. (Axiom 3 is satisŽ ed.) The main cause of anomalies in the LSGR procedure as
Let x *1 and x *2 be the welfare levels in the two subperiods pointed out by Boyce is that the exponential trend lines used
(1 to n1 ) and (n1 1 1 to n), respectively. Then are likely to be discontinuous. He proposed a restricted
dummy variable procedure to eliminate such discontinuities.
n1

o
1 Our procedure of computing subperiod growth rates implies
log x *1 5 log xt (37) a continuous trend line (the continuity is achieved by means
n t5 1
of equation (33)), and therefore it does not give rise to any
n
anomalies.
o
1
log x*2 5 log xt. (38)
n2 n 1 t5 n11 1
VIII. Growth Performance of 83 Developing Countries
Substituting equation (33) into equations (37) and (38) gives Table 2 presents the growth rates for 83 countries
computed using the proposed method. Globally growth
n1 2 1 slowed down in the 1980s. Per-capita incomes declined in
log x *1 5 log x1 1 log (1 1 R1 ) (39) many developing countries, particularly in Africa and Latin
2
America. Therefore it seems appropriate to analyze the
log x *2 5 log x1 1 (n1 2 1) log (1 1 R1 ) growth performance of countries in the two subperiods of
1970 –1979 and 1980 –1987. The countries were ranked
n2 n1 1 1 (40) according to their growth performance in each period; the
1 log (1 1 R2 ).
2 higher (lower) the rank, the better (worse) the country’s
growth performance. These ranks are also presented in the
Since n log x* 5 n1 log x *1 1 (n 2 n1 ) log x *2, equations table.
(36), (39), and (40) give Table 3 provides a summary of table 2. The conclusions
emerging from tables 2 and 3 are summarized below. It will
(2n 2 n1 )(n1 2 1) be observed that among 36 African countries, per-capita
log (1 1 R) 5 log (1 1 R1 ) growth rates deteriorated considerably in the period of
n (n 2 1) 1980 –1987 compared to the 1970 –1979 period. In the
1970 –1979 period there were 14 countries whose growth in
(n 2 n1 )(n 2 n1 1 1) (41) GNP was less than the growth in population. This number
1
n (n 2 1) increased to 26 in 1980 –1987. Aggregate growth deterio-
rated in 27 countries. The countries which suffered the most
3 log (1 1 R2 ) deterioration are Gabon, Gambia, Kenya, Lesotho, Malawi,
and Nigeria. In the 1970 –1979 period, Gabon enjoyed a
which shows that log (1 1 R) is a weighted average of per-capita annual growth rate of 7.6%, but in the 1980 –1987
log (1 1 R 1 ) and log (1 1 R2 ). This implies that R lies period, per-capita GNP declined at an annual rate of 11.88%.
between R1 and R2. Also, it can be shown that R is an It is interesting to note that many African countries suffered
increasing function of R 1 and R 2. a substantial deterioration in their incomes even in the
R and R1 can be computed from equations (36) and (39), 1970s. Some examples of such countries are Ghana, Guinea-
respectively. Therefore given R 1 and R, we compute R2 from Bissau, Madagascar, Mauritania, Niger, Uganda, and Zam-
equation (41). bia. At the same time it may be noted that among all the
Boyce (1986) has argued that the LSGR for the whole developing countries, Botswana enjoyed the highest growth
period may lie outside the range of subperiod growth rates rate, 11.38%. Although its growth dropped to 5.1% in the
COMPARISON OF INCOME GROWTH RATES AND AGGREGATE WELFARE 209

TABLE 2.—GROWTH RATES OF GNP PER CAPITA TABLE 2.—(CONTINUED)


1970–1979 1980–1987 1970–1987 1970–1979 1980–1987 1970–1987
Country Value Rank Value Rank Value Rank Country Value Rank Value Rank Value Rank

Africa Costa Rica 3.42 57 2 3.45 19 1.76 43


Burundi 2 0.13 18 2.19 62 0.41 28 Dominican Rep. 4.16 61 2 1.38 35 2.83 57
Benin 2 0.63 11 2.25 63 0.04 21 Ecuador 7.05 76 2 3.34 20 4.51 74
Botswana 11.38 83 5.1 75 9.87 83 El Salvador 2.51 46 2 6.24 5 0.38 26
Burkina Faso 1.25 33 3.44 67 1.76 44 Guatemala 3.31 55 2 3.24 22 1.73 42
Cameroon 1.51 37 10.2 82 3.49 64 Haiti 1.88 42 2 0.65 44 1.28 39
Central African Rep. 0.6 24 2 3.21 23 2 0.31 16 Honduras 1.86 40 2 1.66 33 1.02 36
Congo 2.89 54 5.03 73 3.39 63 Jamaica 2 0.25 16 2 5.66 7 2 1.55 10
Cote d’Ivoire 1.86 41 2 2.73 26 0.76 34 Mexico 2.89 53 0.87 54 2.41 50
Ethiopia 0.26 20 2 0.8 42 0.01 20 Nicaragua 0.37 21 2 8.77 2 2 1.86 8
Gabon 7.6 79 2 11.88 1 2.66 55 Panama 2.25 44 1.87 60 2.16 48
Gambia 4.65 65 2 4.07 13 2.53 53 Papua New Guinea 1.06 29 2 2.83 25 0.13 23
Ghana 2 1.67 5 2 3.78 17 2 2.17 4 Paraguay 4.56 64 3 65 4.19 70
Guinea 1.74 39 2 1.49 34 0.97 35 Peru 1.14 31 2 2.05 29 0.38 27
Guinea-Bissau 2 1.76 4 2 1.67 32 2 1.74 9 Trinidad & Tobago 2.44 45 1.3 57 2.17 49
Kenya 6.03 73 2 3.92 15 3.6 65 Uruguay 1.46 36 0.49 53 1.23 37
Lesotho 8.99 81 2 1.84 31 6.34 79 Venezuela 0.41 22 2 4.76 11 2 0.83 13
Liberia 2 0.5 13 2 6.52 4 2 1.95 7
Madagascar 2 1.39 8 2 4.75 12 2 2.19 3
Malawi 4.85 68 2 4.98 9 2.45 51
Mali 1.62 38 1.28 56 1.54 41
Mauritania 2 1.5 7 0.21 50 2 1.1 11 1980s, this was found to be one of the highest in the
Mauritius 5.84 72 2 1 40 4.19 71 developing world.
Niger 2 3.66 1 1.1 55 2 2.56 2
Nigeria 2.88 52 2 6.11 6 0.69 32 Among 37 low-income countries, 23 suffered a deteriora-
Rwanda 2 0.56 12 5.08 74 0.74 33 tion in their growth rates in the 1980s, but this Ž gure among
Senegal 2 0.42 14 2 1.18 36 2 0.6 14 46 middle-income countries was 38. In 1970 –1979 only two
Sierra Leone 2 0.19 17 2 0.27 47 2 0.21 18
Somalia 0.93 26 2 0.93 41 0.49 29 middle-income countries had a negative growth rate (Chile
South Africa 0.97 27 2 0.72 43 0.57 31 and Jamaica). This number increased to 25 in the 1980 –
Sudan 1.43 35 2 3.78 16 0.18 25 1987 period. From these observations we might conclude
Tanzania 0.74 25 2 3.28 21 2 0.22 17
Togo 1.07 30 2 2.95 24 0.11 22 that the middle-income countries suffered a greater decline
Uganda 2 2.67 3 2 4.86 10 2 3.19 1 in their standard of living than did the low-income countries
Zaire 2 1.09 9 2 5.27 8 2 2.09 6 during the 1980s.
Zambia 2 1.67 6 2 3.69 18 2 2.15 5
Zimbabwe 3.44 58 2 2.71 27 1.96 45 It is quite evident that most Asian countries improved
Middle East and Eastern Europe their growth performance considerably during the recession-
Algeria 5.09 69 14.17 83 7.16 81 ary period of the 1980s. The two most populated countries in
Egypt 2.52 47 5.22 76 3.15 60
Greece 4.76 66 2 1.14 37 3.34 62 the world, India and China, increased their growth rates
Hungary 8.05 80 0 48 6.1 78 signiŽ cantly.
Jordan 2.75 49 6.76 80 3.68 67 In the 1970s a large number of countries in the central and
Morocco 2.55 48 0.27 51 2.01 46
Oman 1.32 34 6.74 79 2.57 54 south American region enjoyed high growth rates. This
Portugal 4.37 63 2 0.35 46 3.24 61 performance deteriorated considerably in the 1980s, when
Syrian Arab Rep. 7.05 75 2 2.03 30 4.84 76 more than 80% of them registered negative growth rates in
Tunisia 6.04 74 2 0.35 45 4.5 73
Turkey 4.32 62 2 1.05 39 3.03 59 per-capita GNP. Aggregate growth deteriorated in 21 out of
Yemen Arab Rep. 9.9 82 2.13 61 8.02 82 22 countries (Chile was the only country where this deterio-
Yugoslavia 5.36 71 0.11 49 4.1 68 ration did not occur). These countries were severely affected
Asia
Bangladesh 2 2.77 2 5.25 77 2 0.94 12 by the debt crisis precipitated by high interest rates and
Burma 1.21 32 4.74 72 2.03 47 declining demand for their exports. Many countries fol-
China 2.76 50 8.71 81 4.13 69 lowed restrictive domestic Ž scal and monetary policies in
India 0.56 23 3.66 68 1.28 38
Indonesia 4.83 67 3.27 66 4.46 72 order to cope with external payment constraints and in a-
Korea 7.27 78 4.16 69 6.53 80 tion.
Malaysia 5.16 70 2.97 64 4.64 75 An interesting question that arises is whether there has
Nepal 2 0.27 15 1.53 59 0.15 24
Pakistan 0.02 19 5.84 78 1.36 40 been a signiŽ cant change in relative growth performances of
Philippines 3.59 59 2 1.13 38 2.46 52 countries between the 1970 –1979 and 1980 –1987 periods.
Sri Lanka 2.08 43 4.65 70 2.68 56 This requires testing the hypothesis that the ranking of
Thailand 3.32 56 4.69 71 3.64 66
Latin America countries according to their growth performances in 1980 –
Argentina 1.04 28 2 3.92 14 2 0.15 19 1987 was not signiŽ cantly different from that in 1970 –1979.
Bolivia 2.86 51 2 6.8 3 0.5 30 If the relative performance of all countries was exactly the
Brazil 7.17 77 2 2.19 28 4.89 77
Chile 2 0.94 10 1.38 58 2 0.4 15 same in the two periods, the rank correlation of growth rates
Colombia 3.86 60 0.34 52 3.02 58 r in the two periods would be equal to unity. So we need to
210 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 3.—SUMMARY OF TABLE 2


Number of Countries whose
Countries with Negative Growth Rates Per-Capita GNP Growth Rate
Number of Deteriorated between
Country Groupings Countries 1970–1979 1980–1987 1970–1987 1970–1979 and 1980–1987

Low income 37 16 22 14 23
Middle income 46 2 25 5 38
Africa 36 14 26 13 27
Middle East and Eastern Europe 13 0 5 0 9
Asia 12 2 1 1 4
Latin America 22 2 15 5 21
Highly indebted 17 2 13 4 17
Others 66 16 34 15 45
Oil exporters 13 0 5 1 9
Primary producers 41 14 32 15 32
Manufacturing exporters 29 4 10 3 21
All developing countries 83 18 47 19 61

test whether the computed value of r is signiŽ cantly less than TABLE 4.—RELATIVE GROWTH PERFORMANCE OF COUNTRIES CLASSIFIED BY
I NCOME, REGIONS, AND OTHER CHARACTERISTICS
unity. The test statistic
Performance

Î
in Welfare
(1 2 r) n 2 2 1970–1979 to
z5 Country Groupings 1970–1979 1980–1987 1970–1987 1980–1987
Î 12 r2 Low income 27.9 40.56 29.4 34.0
Middle income 53.3 43.15 52.2 48.4
under the null hypothesis of no signiŽ cant difference of r Africa 32.9 34.80 31.5 31.2
from unity is approximately a student’s t distribution with Middle East and
Eastern Europe 63.0 155.69 66.7 65.2
(n 2 2) degrees of freedom. This approximation suggested Asia 42.9 67.75 52.6 63.7
by Pitman (1937) has been shown to perform better than the Latin America 43.9 31.63 38.8 34.1
usual normal approximation (Iman and Canover, 1978). Highly indebted 46.4 28.2 40.4 34.5
The value of r was computed to be 0.52 with a standard Others 40.9 45.6 42.2 43.9
error of 0.11, which gave a value of t 5 4.21, which is Oil exporters 54.6 49.6 57.2 54.1
signiŽ cant at the 5% level. Therefore the null hypothesis is Primary producers 31.9 32.1 29.8 30.2
rejected. This leads to the conclusion that the overall relative Manufacturing
exporters 50.6 52.6 52.4 53.2
growth performance of countries has changed signiŽ cantly
All developing
between the 1970 –1979 and 1980 –1987 periods. countries 42.0 42.0 42.0 42.0
Having established that the overall relative growth perfor-
mance of countries has changed signiŽ cantly between the
1970 –1979 and 1980 –1987 periods, the next step is to see
increased to 40.56, indicating a substantial increase in
for which country groups the relative growth performance
relative growth performance. Accordingly the relative growth
has improved (deteriorated) in the two periods. To accom-
performance of the middle-income countries declined sub-
plish this task, we computed the average rank of country
stantially from an average rank of 53.3 in 1970 –1979 to
groups when the countries are ranked according to their
43.15 in 1980 –1987. The highly indebted countries suffered
growth rates. The results are presented in table 4. The table
the most decrease in their relative growth performances
also presents the average rank values when the countries are
between the two periods.
ranked according to the percent change in welfare between
the 1970 –1979 and 1980 –1987 periods. The average rank of IX. Some Concluding Remarks
all 83 developing countries is always equal to 42.0. The rank
average of each country group must be compared with a In comparing the performances of two countries over a
value of 42.0. If the average is greater (smaller) than 42.0, certain period it is common practice to compute some
the relative growth performance of that country group is measure of the average growth rate. Some people then argue
better (worse) than the overall average. that the country with the higher growth rate has done
It can be seen from the table that in 1970 –1979 the ‘‘better.’’ This suggests that there must exist a relationship
relative growth performance of the low-income countries between growth rates and changes in welfare. In practice,
was extremely poor, with an average rank value of 27.9. growth rates are estimated by purely mechanical procedures,
However, in the 1980 –1987 period the average rank value with no regard to their welfare implications. In this paper we
COMPARISON OF INCOME GROWTH RATES AND AGGREGATE WELFARE 211

analyze what different mechanical procedures imply about growth rates. These issues are being researched in our
the welfare weights attached to growth in different years, ongoing work on the standard of living.
when the welfare function is deŽ ned over a time series on
per-capita income. All the methods commonly used to REFERENCES
calculate the average growth rate (least-squares, geometric Atkinson, A. B., ‘‘On the Measurement of Inequality,’’ Journal of
mean, arithmetic mean, etc.) are shown to imply unreason- Economic Theory 2 (1970), 244–263.
Bergson, A., ‘‘A Reformulation of Certain Aspects of Welfare Econom-
able or bizarre welfare weights. The welfare weights turn out ics,’’ Quarterly Journal of Economics 52 (1938), 310–334.
to be sensible only if the average growth rate procedure is Boyce, James, ‘‘Kinked Exponential Models for Growth Rates Estima-
derived directly from a symmetric welfare function. The tion,’’ Oxford Bulletin of Economics and Statistics 48:4 (1986),
385–391.
‘‘equivalent uniform growth rate’’ is thus calculated so as to Iman, R. L., and W. J. Canover, ‘‘Approximation of the Critical Region for
take proper account of the importance of the time path of Spearman’s Rho with and without Ties Present,’’ Communication in
income. Statistics, ser. B, 7 (1978), 269–282.
Kakwani, Nanak, ‘‘Welfare Measures: An International Comparison,’’
Since the average growth rate is derived directly from a Journal of Development Economics 8 (1989).
welfare function, it should always be possible to take into ——— ‘‘Growth Rates and Aggregate Welfare: An International Compari-
account the changes in inequality over time (if the welfare son,’’ Policy, Research and External Affairs, World Bank, Washing-
ton, DC, Working Paper 647 (1991).
function is appropriately deŽ ned). This was not attempted in Pitman, E. J. G., ‘‘SigniŽ cance Tests Which may be Applied to Samples
the paper, mainly because of nonavailability of the yearly from Any Population: II. The Correlation Coefficient Test,’’ Journal
income distribution data. of Royal Statistical Society (supplement 4) (1937), 225–232.
Ramsey, F. P., ‘‘A Mathematical Theory of Savings,’’ Economic Journal
The international organizations such as the United Na- (Dec 1928), 543–559.
tions and the World Bank routinely employ mechanical Samuelson, P. A., Foundation of Economic Analysis (Cambridge, MA:
procedures to compute aggregate growth rates of a wide Harvard University Press, 1947).
Sen, Amartya, ‘‘On Optimising the Rate of Saving,’’ Economic Journal
range of socioeconomic variables. In this paper we have (Sept. 1961), 479–496.
argued that these procedures can give rise to unreasonable ——— On Economic Inequality (Oxford: Clarendon Press, 1973).
economic implications. Our methodology, although focused ——— ‘‘International Bases of Alternative Welfare Approaches: Aggrega-
tion and Income Distribution,’’ Journal of Public Economics 4
on measuring growth rates of per-capita GNP, can be (1974), 387–403.
modiŽ ed to measure a country’s performance in other ——— ‘‘The Standard of Living,’’ Tanner Lectures given at Clare Hall,
indicators of individual well-being such as life expectancy at Cambridge (1985).
Weizsäcker, C. C. van, ‘‘Existence of Optimal Programs of Accumulation
birth or infant mortality rate. These indicators pose special for an InŽ nite Time Horizon,’’ Review of Economic Studies 32
problems, which must be considered in the computation of (1965), 85–104.

View publication stats

You might also like