http://dx.doi.org/10.1257/aer.p20161112
Besides the dreadful direct effects, enduring reduces the profitability of schooling, leading to
epidemics like HIV/AIDS induce much more a decline of human capital accumulation in the
subtle effects in the economy, principally via medium and long run. Boucekkine, de la Croix,
labor productivity and factor accumulation. and Licandro (2002) have proposed a theory of
Concerning the former, the impact of morbid- endogenous growth based on the Ben-Porath
ity on effective labor has been recognized and mechanism. Notice also that the same essential
accounted for since the early work of Over argument (that shorter time horizons discourage
(1992) and Cuddington and Hancock (1994). investment) can be applied to physical capital
The impact of mortality shocks on savings, accumulation; the latter may collapse not only
and therefore on factor accumulation, has also because of the massive increase in health expen-
been documented in this early literature, with an ditures undermining savings as in the early liter-
emphasis on the detrimental role of the induced ature but also via this indirect mechanism. Both
health expenditures. Recently, Asiedu, Jin, and channels are taken into account by Freire (2004)
Kanyama (2015) show that HIV/AIDS has a in her analysis of saving behavior in South Africa
negative but diminishing effect on foreign direct after the outbreak of the HIV/AIDS crisis.
investment (FDI), and this adverse effect occurs This paper proposes a reappraisal of the eco-
even when the prevalent HIV rate is low. In con- nomic growth cost of HIV/AIDS accounting,
trast, the consequences of HIV/AIDS on human among others, for Ben-Porath-type effects on
capital accumulation have been more rarely labor productivity and factor accumulation. We
treated. In particular, its impact on schooling use an economico-epidemiological Solow-Swan
time has been generally omitted in the macro- model. Mortality and morbidity effects on effec-
economic literature. Ferreira, Pessôa, and Dos tive labor are taken into account. The Ben-Porath
Santos (2011) is an exception. The inherent mechanism affects the dynamics of the saving
mechanism is known as the Ben-Porath mech- rate and reduces labor productivity (which is a
anism (Ben-Porath 1967): a downward trend useful shortcut to the decline of human capital
in life expectancy following mortality shocks accumulation). The main contribution of the
paper is the application to South Africa. Based
on optimal projections of the demographic
* Azomahou: UNU-MERIT, Maastricht University (including HIV prevalence rates) and economic
and CERDI, Boschstraat 24, 6211 AX Maastricht, (including saving rates, based on the work of
The Netherlands (e-mail: azomahou@merit.unu.edu);
Boucekkine: Aix-Marseille University (AMSE), CNRS Freire 2004) South African series for the period
and EHESS, 2 rue de la Charité, 13002 Marseille, France 2000–2050, two main and novel findings are
(e-mail: raouf.boucekkine@univ-amu.fr); Diene: CERDI, presented. First, we identify a delayed effect of
65 Boulevard F. Mitterrand, 63009 Clermont-Ferrand, HIV/AIDS on economic growth: the growth rate
France (e-mail: bity.diene@udamail.fr). We gratefully gap between the AIDS and no-AIDS scenarios is
acknowledge insightful discussions with Elizabeth Asiedu.
This work was supported by the FERDI (Fondation pour les rather stable between 2010 and 2020, but then it
Etudes et Recherches sur le Developpement International) increases sharply between 2020 and 2030, keeps
and the Agence Nationale de la Recherche of the French increasing at a much lower pace between 2030
government through the program “Investissements d’avenir and 2040, and finally stabilizes after 2040. Most
of the related papers (e.g., Ferreira, Pessôa, and
ANR-10-LABX-14-01.”
†
Go to http://dx.doi.org/10.1257/aer.p20161112 to visit
the article page for additional materials and author disclo- Dos Santos 2011; or Corrigan, Glomm, and
sure statement(s). Mendez 2005), based on comparative statics
472
VOL. 106 NO. 5 HIV/AIDS AND DEVELOPMENT 473
of steady state equilibria, cannot deliver such a where pans indicates the active population
picture. Second, we identify the impact of the without HIV/AIDS by age and sex, u is the
different demographic mechanisms; in partic- number of unemployed people by age and sex,
ular, we find that while the fall in active pop- and h is the number of hours worked in a year.2
ulation is the main factor behind the AIDS The basic economico-epidemiological pro-
impact on economic growth during the decade duction function above is further extended in
2020–2030, the Ben-Porath mechanism on labor two essential directions. First, the morbidity
productivity seems more relevant in the last term, xi, f (t) , is assumed to depend on health
decade, physical capital accumulation playing a expenditures devoted to fighting HIV/AIDS.
minor role. Accordingly, for given AIDS-specific health
expenditures mi, f (t) , xi, f (t) is determined by
I. The Economico-Epidemiological Model
xi, f (t) = Φ(mi, f (t)),
The framework is a Solow-Swan model in the
tradition of Cuddington and Hancock (1994). where function Φ ( · ) is decreasing. This is
The aggregate production function at time t is largely consistent with Corrigan, Glomm, and
given by Mendez (2005), who argued that health expen-
ditures improve the productivity of the infected
Y (t) = A (t) K(t) θ E(t) 1−θ, people. In contrast to the latter work, where
health expenditures are determined by individ-
where Y (t) denotes the total output (or GDP) of ual choices, we shall use real health expendi-
the economy, A (t) the total factor productivity tures data. In the implementation, we have faced
(TFP,) K (t) the physical capital, and E (t) the a data unavailability problem. Instead of the
effective labor, θ being the capital share. E (t) intended indicators Φ(mi, f (t)) , we have been
is age-structured and gender-structured as it forced to use Φ(m (t)) , where m (t) is given by
should be since epidemics do not affect all age
the relation: m (t) = txHIV _ t M
and gender classes in the same way:1 where txHIV
HIV +
represents the percentage of AIDS-related
(1) E (t) = ∑[1 − ϕi, f (t) xi, f (t)] %i, f (t) Li, f (t) ,
expenditures in the total health expenditures
i, f (Mt), and HIV + is the number of infected people.
As in Corrigan, Glomm, and Mendez (2005),
where i stands for the age index, f = 1, 2 the we pick the following analytical for Φ ( · ):
gender class, and Li, f (t) is labor services provided
by individuals of age class i and gender f. ζ
%i, f (t) captures labor-specific productivity for (2) Φ(m (t)) = 1 − ψ1 + (ψ1 − ψ0) _ ,
m (t) + ζ
age class i and gender f , and in this sense, it
could be interpreted as a human capital indicator. with: Φ (0) = 1 − ψ0 , limx→∞ Φ (x) = 1 − ψ1 ,
xi, f (t) reflects morbidity at age class i and 0 < ψ0 < ψ1 < 1 , Φ′(x) < 0, and Φ′′(x) > 0;
gender f ; a larger xi, f (t) amounts to a lower ψ1 , ψ0, and ζ are positive productivity
effective labor effort due to morbidity. Finally, parameters.3
the production function includes a term ϕi, f (t) , Second, we introduce the Ben-Porath effect
which stands for the AIDS prevalence rate at and model it as follows. We start with the tradi-
date t for age class i and gender f. In the no-AIDS tional modeling of age and gender specific pro-
case, the prevalence rates are set to zero and E (t) ductivity (see Cuddington and Hancock 1994):
= ∑ i, f μi, f (t) Li, f (t) with Li, f (t) = ( pans − u ) h ,
(3) %i, f (t) = ρ1 + ρ2(i − i ) − ρ3(i − i ) ,
– – 2
1
We insist on dealing with gender here since a crucial
recent feature of HIV/AIDS in sub-Saharan Africa (and
2
notably in South Africa) is the increasing percentage of We use 49 weeks and 40 hours by week, that is
females affected. In its 2004 report, UNAIDS mentioned h = 1960.
that close to 60 percent of HIV/AIDS infected people are 3
In our numerical experiments, we fix ψ1 = 1 , ψ0 = 0.5 ,
females in sub-Saharan Africa, the youngest being the most and we let ζ vary from 1 to 10 , with ζ = 5 in the reference
exposed to the infection risk for obvious reasons. simulation, as in Corrigan, Glomm, and Mendez (2005).
474 AEA PAPERS AND PROCEEDINGS MAY 2016
where ρk , k = 1, 2, 3 and i are given positive series constructed by Freire (2004) for South
–
constants, the last being, for example, the min- Africa under the AIDS and no-AIDS scenario
imal age to work. Productivity is an inverse as the basis of our calibration and forecasting in
U-shaped function of age, maximal productiv- this respect. Freire’s model closely follows Galí
ρ2
ity being reached at age i + 2 _
–
. With a declin- (1990), a Blanchard-Yaari type of model, which
ρ3
ing life expectancy, such a magnitude is likely is more likely to capture the effects of increas-
to go down, not only mechanically but also via ing mortality on saving decisions than the arbi-
the Ben-Porath mechanism, according to which trary computation rule adopted by Cuddington
a decreasing life expectancy lowers schooling and Hancock (1994). It is worth pointing out
time, and therefore human capital and produc- that since the saving rate is not constant in our
tivity at work. We capture this mechanism by setting, there is no long-run steady state (and
assuming the following functional relationship: the inherent standard comparative statics) as in
Ferreira, Pessôa, and Dos Santos (2011).5
where νi, f (t) is the life expectancy at birth of A. Data and Variables
workers of age i (thus born at t − i) and of gen-
der f. αi, f and β are calibrated on real data. For The variables used include: (i) Economically
the calculation of ρ2 , we set for the reference active population, by age and sex (1980–2020);
year t = 0 , which corresponds to the year 2000, (ii) HIV-seroprevalence (1980–2015); (iii)
in the data, ρ2(i, f, t0) = αi νi, f (t) β = 0.005 for a HIV + (1990–2015) indicator; (iv) Life expec-
given β , then we calibrate αi = _ 0.005
, taking the tancy at birth by sex (1920–2015); (v) Total
νi, f (t)
β
health expenditure (1960–2000); (vi) Percentage
life expectancy series from the data. The deter-
expenditure of AIDS in total health expenditure
mination of αi allows us to determine the ρ2.4 β
(1960–2000); (vii) Unemployed by age and
is given values ranging from 0 (no Ben-Porath
sex (2000–2003); (viii) GDP per capita (1960–
mechanism) to 1 (a “linear” mechanism). The
2000); and (ix) Saving rates with and without
setup is completed by a capital accumulation
AIDS. See the online Appendix for further
equation:
details of these indicators and related sources.
( γ E2000 ) .
ant issue in Azomahou et al. 2015). To deal TFP at rate γ , K0 is computed by K0 = K2000 = _
Y2000 θ
dynamic equation above with the saving rate The physical capital K (t) series is then obtained thanks to
the standard accumulation equation (5) with the saving rate
series constructed from Freire (2004). Moreover, to make
our results comparable with earlier findings, we use the same
4
We set the values of the other parameters as follows: parameter values as Corrigan, Glomm, and Mendez (2005):
ρ1 = 0.8 , ρ3 = − 0.0001. θ = 0.33 , γ = 0.012, and δ = 0.05.
VOL. 106 NO. 5 HIV/AIDS AND DEVELOPMENT 475
[1 + e ] [1 + e ]
α(t−τ) −β(t−τ) 54
(6) p (t) = e
_
α(t−τ)
ae
_
β(t−τ)
+ b , 52
50
48
46
where α is the initial growth rate, a denotes the 44
peak value, β is the rate of convergence, b is the
2000 2010 2020 2030 2040 2050 2060
Table 1—Growth Rates of GDP Per Capita: Gap the yearly growth rate of GDP per capita, which
between the AIDS and No-AIDS Scenarios for Varying
ζ and β
is considerable. Our distinctive contribution
is in the timing of the growth effect after the
β= 1 incorporation of the most likely medium and
ζ 1 2.5 5 10 long-run demographic trends induced by AIDS.
Incidentally we show that there might not be any
2010 0.86 0.86 0.87 0.89
2020 0.89 0.89 0.89 0.89 conflict between the results obtained by authors
2030 2.78 2.78 2.78 2.78 like Bloom and Mahal (1997), and others like
2040 3.18 3.18 3.18 3.18 Corrigan, Glomm, and Mendez (2005): a rel-
2050 2.84 2.84 2.84 2.84 atively short term assessment as in the former
β = 1/2
might not reveal any dramatic AIDS growth
effect, while a medium or long term perspective,
2010 0.85 0.85 0.86 0.89 relying on demographic dynamics and those
2020 0.88 0.88 0.88 0.88
2030 2.72 2.72 2.72 2.72
of human capital, might deliver the opposite
2040 3.02 3.02 3.02 3.02 message.
2050 2.74 2.74 2.74 2.74
C. Robustness Check
β= 0
2010 0.83 0.84 0.85 0.87 We conduct two kinds of sensitivity analysis.
2020 0.86 0.86 0.86 0.86 First, we allow the coefficient β to vary taking
1
2030 2.65 2.65 2.65 2.65 the values 1 , _
2
, and 0. See the online Appendix
2040 2.87 2.87 2.87 2.87
2050 2.65 2.65 2.65 2.65 for additional sensitivity parameters results. The
second sensitivity analysis concerns function
Φ(m (t)) (relation 2). We allow ζ to vary from 1
to 10 for each value of β. The results are given
in Table 1.
B. The Economic Growth of AIDS Quantified A first conclusion should be that the varia-
tion of parameter ζ of the health expenditures
To start with a benchmark case, let us set function does not generally change the first two
β = 1. Table 1 gives the results for different decimals of the obtained growth per capita gap
values of the health expenditures productivity figures. The same type of findings is reported by
parameter ζ. Corrigan, Glomm, and Mendez (2005). Much
Consider the case ζ = 5. One can see that the more importantly, and not surprisingly, the gap
gap in growth of GDP per capita does not move figures increase with the sensitivity parameter β .
significantly between 2010 and 2020. From A comparison between the polar cases β = 0
2020 to 2030, the gap rises sharply from 0.89 and β = 1 is useful. In the first case (and ζ = 5),
percent to 2.78 percent, which translated into the gap jumps from a 0.65 percentage point
annual rates means that AIDS causes GDP per in 2020 to 2.65 percentage points in 2030,
capita to decline by 0.19 percent in the period rises slightly to 2.87 in 2040, before coming
2020–2030, while such a decline is around 0.09 back to the 2030 level in 2050. When β = 1 ,
percent in 2000–2010 for example. Accordingly, the size of the jump between 2020 and 2030
the growth impact of AIDS in 2020–2030 more is naturally bigger, but the difference is even
than doubles compared to 2000–2010. The gap larger in 2040: the difference between the two
keeps increasing from 2030 to 2040, it reaches gaps is about 0.13 in 2030, it is more than the
3.18 percent in 2040 but at a lower pace. Finally, double (equal to 0.31) in 2040. Therefore,
the gap shrinks to 2.84 percent at 2050 mainly while the fall in active population is the main
due to HIV/AIDS prevalence rates’ values at factor behind the impact of AIDS on growth
this horizon. during 2020–2030, the Ben-Porath mecha-
Our figures for the growth gaps are a bit nism seems more relevant in the last decade.
lower than those put forward by Over (1992) Both demographic factors are consequently
for the ten most affected sub-Saharan African key to understanding the medium and long-run
countries for the period 1990–2025. In our case, economic growth effects of AIDS in our
the growth gap may be close to 0.2 percent for scenarios.
VOL. 106 NO. 5 HIV/AIDS AND DEVELOPMENT 477