Sustainable Living Standards and Adaptive Economizing in Economic Growth
Sustainable Living Standards and Adaptive Economizing in Economic Growth
Sustainable Living Standards and Adaptive Economizing in Economic Growth
91
University of Bielefeld
Department of Economics
Center for Empirical Macroeconomics
P.O. Box 100 131
33501 Bielefeld, Germany
http://www.wiwi.uni-bielefeld.de/∼cem
Sustainable Standards of Living and Adaptive
Economizing in Economic Growth‡§
Herbert Dawid∗ and Richard H. Day∗∗
∗
Department of Business Administration and Economics and
Institute of Mathematical Economics, Bielefeld University
∗∗
Department of Economics, University of Southern California
Abstract
We consider economic growth in the presence of a minimal survival
consumption level when preferences are lexicographic giving highest
priority to the survival of the current and then all future generations.
Knowledge of production is incomplete so future output made possible
by current saving must be estimated. The resulting capital accumu-
lation path is compared with the inter–temporally optimal one with
full knowledge of the production function. Although previous results
with the standard utility function show that for sufficiently strong
discounting the adaptive strategy converges to the optimal path, in
the present case in several scenarios the effect of adaptive economiz-
ing under incomplete information is drastic leading to demise of the
economy although long-run survival would be possible.
JEL Classification: C61, O41, D83
Keywords: Economic Growth, Adaptive Economizing, Minimal Consump-
tion Level
‡
Corresponding author: Herbert Dawid, Department of Business Administration and
Economics, Bielefeld University, P.O. Box 100131, 33501 Bielefeld, Germany. Phone:
+49-521-1064843; fax: +49-521-10689005; email: hdawid@wiwi.uni-bielefeld.de
§
We are grateful to Laura Dawid-Hofer for preparing the figures in this manuscript.
1 Introduction
A minimal sustainable standard of living is a realistic and important feature
to take account of in the theory of growth, for if a given generation does
not have enough product, it will not survive. Moreover, the only way future
generations can survive is to make sure that the current generation survives
and can bring its children up to form the next generation of adults. The issue
is relevant for—in contrast to the high level of living achieved in the developed
countries—substantial numbers of people in developing countries are close to
or even below the subsistence threshold. Several authors have studied the
implications of the existence of a minimal consumption level for optimal
growth by incorporating Stone-Geary preferences into neo-classical growth
models (see e.g. Rebelo (1992), Easterly (1994), Chatterjee and Ravikumar).
Christiano (1989) uses this approach to demonstrate that consideration of
a subsistence level allows to capture the essential features of the evolution
of postwar Japanese saving rates much better than growth models assuming
standard preferences. Additional empirical motivation for the consideration
of subsistence consumption is provided in Steger (2000) where it is shown
that a linear growth model with subsistence consumption is able to reproduce
several stylized of economic growth.
In this paper we investigate the existence of a subsistence threshold in
the context of macroeconomic growth theory by representing household pref-
erences with a lexicographic ordering. The first priority in the lexicographic
order is to maximize consumption up to the subsistence threshold. The sec-
ond priority, given satisfaction of the first, is to insure a subsistence level
for all future generations. The third priority, given satisfaction of the first
two, is to optimize consumption for all generations. From the formal point of
view we think of a Swiss Family Robinson type private ownership economy.
The adults are manager-worker-owners to whom all proceeds of production
accrue. They determine current consumption and savings based on the trade-
off between their own consumption and the standard of living which their
descendants could enjoy in the future. The savings are invested, and the aug-
mented capital stock that results constitutes their children’s endowment. The
new generation of adults repeats the same economizing decision but on the
basis of the capital stock inherited from their parents. Such an economy was
investigated in some detail in the context of the one–sector growth model in
Day and Lin (1992) and in Day (1999), both for the standard inter–temporal
1
equilibrium formulation and for a boundedly rational, adaptive economizing
formulation, where agents have incomplete information about future produc-
tion technology and evaluate the capital stock left to the next generation
based on a simple heuristic rather than on infinite horizon optimization.1
In this paper we consider both cases in the same one–sector growth setting
but when each generation faces a minimal consumption level necessary for
survival.
Infinite horizon inter–temporal optimization models provide the appro-
priate analytical framework when designing controls for relatively simple
mechanical systems whose physical characteristics are fully understood and
when the objective is simple and stationary. Such is assuredly not the case
with human endeavor in general. Consequently, individuals do not form
detailed economic plans over the very long–run. To the extent tradeoffs are
considered, we generally give explicit attention to horizons much shorter than
the potential duration of the process as a whole and each generation consid-
ers the future on its own terms. Considerable support for this view has been
provided by Rust (1994) who found in a variety of settings that the optimal
strategies derived from dynamic optimization had weak explanatory power.
Further support, especially relevant in the present setting of the one–sector
growth model has been derived from laboratory experiments by Noussair and
Matheny (2000) who found that over–investment and non–monotonic capital
accumulation paths typically occurred. Such phenomena are ruled out for
inter-temporally optimal paths but the adaptive economizing model consid-
ered in this paper can generate such behavior. A theoretical argument for
the use of short horizon planning under incomplete information about the
1
The relationship between adaptive and optimal or Nash equilibrium strategies has
been studied in various contexts: a very early one is Kirman (1977) who showed that linear
least squares learning by duopolists who face nonlinear demand curves need not converge
to Cournot–Nash equilibria. Situations with various types of strategic interaction in game
theoretic settings have been explored by Fudenberg and Levine (1998). The representation
of learning by agents who use econometric methods has been explored in depth by Evans
and Honkapohja (2001) but for linear models. They emphasize conditions that guarantee
convergence to competitive equilibria. Here our concern is to explore what happens when
people do not use econometric or other sophisticated strategies, not in the belief that
people are irrational, but because such strategies depend on information and knowledge
that are seldom available or which can only be acquired at great cost in time and resource.
After all, it is possible that only trained mathematical economists use such rules and then
only in their research.
2
economic environment has been put forward in Dawid (2005), where it has
been shown in a standard one-sector growth model that in cases where indi-
viduals have incomplete information about the production function the use
of adaptive economizing might lead to higher total discounted utility than
infinite horizon optimization.
Given these anecdotal, experimental and theoretical arguments, consid-
erations about the impact of incomplete information and short horizon plan-
ning should be included in the economic theory of growth. However, the in-
finite horizon dynamic optimization formulation under complete information
is still an appropriate way to formalize the idea of inter–temporal equilib-
rium in a one–sector economy, and provides a benchmark against which to
evaluate more realistic formulations. For that reason we consider that for-
mulation first. Before taking it up, we describe in §2 the basic ingredients
of the analysis: the production function, the lexicographic preferences, and
the relationship of the survival consumption threshold to viable capital accu-
mulation. In §3 inter–temporal optimal trajectories are derived and charac-
terized. In §4 the behavior of adaptive economizing trajectories is described
when knowledge of production is incomplete and future income must be pro-
jected. Then the two approaches are compared. For very small initial capital
stocks (not surprisingly) both approaches give identical results: demise. For
larger capital stocks, however, their behaviors are different. In addition to
the convergence to or fluctuation about optimal growth paths, our analysis
shows that for some conditions adaptive economizing may lead to extinction
when in fact long–run survival is possible. In particular, extinction always
occurs for small discount factors.
Under the inter–temporally optimal policy an increase in the minimal
level of consumption has only marginal long–run effects if the initial capi-
tal stock is sufficiently large. With adaptive economizing, however such an
increase might trigger the long–run extinction of capital even if the initial
capital stock is way above the minimal capital stock which can sustain this
level of consumption. Hence the implications of a minimal standard of living
can indeed be quite dramatic. Nonetheless, if our economy is sufficiently
wealthy and productive, a steady state exists which is optimal in the usual
sense. Moreover, the adaptive economy can converge to it, given appropriate
conditions on its time preference and productivity parameters.
3
2 Production, Preferences and Viability
2.1 Production and Capital Accumulation
We assume the standard neoclassical production function with capital and
labor inputs which in per labor terms is expressed by a function, f : IR+ 7→
IR+ which is twice continuously differentiable on (0, ∞) with the properties
4
consumption has reached this level, they give future generations’ consump-
tion priority over more consumption for themselves. When survival at this
level is apparently attainable for themselves and their heirs, the current gen-
eration considers the trade–off between its own consumption and that of its
descendants, discounting the utility of future generations in the usual way.
We suppose that a generation’s satisfaction from consumption, given that
survival is assured, can be represented by a twice continuously differentiable
utility function u : IR+ +
0 7→ IR0 that satisfies
( Ã ! )
1 h i 1−δ
kt+1 = φ(kt ) := max (1 − δ)kt + f (kt ) − c̄ , kt . (6)
1+n 1+n
5
Every capital accumulation trajectory is under our lexicographic preferences
bounded above by this equation.
We note that, φ(k) is strictly concave and monotonically increasing for
all k > k̄. If c̄ is not too big2 , there will exist two stationary states of (6),
ˆ ˆ
say k̂ and k̂, such that k̄ < k̂ < k̂, as illustrated in figure 1. The first one is
unstable, the second one asymptotically stable if consumption is constant at
ct = c̄ for all t.3 If initial wealth is above the disaster level but insufficient to
allow survival of all future generations (k̄ ≤ k0 < k̂), consumption must follow
subsistence until the finite time, t∗ , is reached when wealth falls below the
disaster level and population must die out. The maximal survival period is t∗ .
Consumption greater than c̄ for any generation below t∗ would lower wealth
and cause an earlier demise. When k̂ < kt , income is large enough to allow
existence forever and for some generations to exceed the subsistence level. If
ˆ
kt > k̂, the subsistence level can be maintained indefinitely, but even under
minimal consumption capital must decumulate and converge from above to
ˆ
k̂. Accordingly, any steady state of the capital accumulation process with
ˆ
lexicographic preferences must lie in the interval [k̂, k̂].
2
If c̄ is so large that no fixed point of φ(k) exists, it is impossible to sustain minimal
consumption c̄ for all future periods regardless of the initial capital endowment. Hence,
long-run viability is impossible. We will not deal with such a scenario but only situations
where long run viability is possible for a sufficiently large initial capital stock.
3 ˆ
From figure 1 we see that both k̄ and k̂ increase while k̂ decreases if c̄ increases. These
preferences are a special case of the general class of L∗∗ utility functions discussed for
example in Day (1996).
6
3 Optimal Growth
3.1 Inviable Societies
Imagine a society that experiences a natural catastrophe or devastating war
that reduces its capital stock proportionally more than its population, or
one that arises from a new colony that started with inadequate capital for its
members, or for which its capital was quite unproductive. If initial wealth
is below the disaster level saving for the future is pointless. The best our
impoverished agents can do is simply consume their entire income. The pop-
ulation dies out and aggregate capital stock simply decays exponentially at
the depreciation rate. Or, if wealth is above the disaster level, the initial
survival is assured. According to the second priority, consumption is main-
tained at the survival threshold until the maximal survival period is reached,
when the population dies out. From that point on aggregate capital stock
decays exponentially.
Proposition 1. Inviable Societies
(i) Immediate Demise: 0 < k0 < k̄. The initial population, N0 , consumes
the entire production (c0 = f (k0 ) and immediately dies out; the aggre-
gate capital stock, Kt , decays according to
(ii) Finite Survival: k̄ < kt < k̂. Population subsists for t∗ − 1 periods and
dies out in period t∗ , where t∗ is the smallest t such that φt (k0 ) < k̄.
ct = c̄ , t = 0, . . . , t∗ − 1,
ct∗ = f (kt∗ ) , t = t∗ ,
kt+1 = φ(kt ) , t = 0, 1, . . . , t∗ − 1.
7
3.2 Viable Societies
When initial wealth is sufficient to assure the survival of future generations,
i.e. k0 ≥ k̂, the optimal consumption strategy takes into account the entire
future in the usual way and is characterized by the optimal growth problem4
∞
X
V (k0 ) := max αt u(c),
t=0
(8)
s.t. c̄ ≤ ct ≤ f (kt ),
h i
1
kt+1 = 1+n
(1 − δ)kt + f (kt ) − ct ,
where α is the time preference or discount parameter. With these definitions
the equation of L∗∗ optimal capital accumulation can be denoted by
1 h i
kt+1 = τ ` (kt ) = (1 − δ)kt + f (kt ) − h∗ (kt ) . (9)
1+n
Here h∗ (kt ) denotes the corresponding optimal consumption function. Ex-
istence and uniqueness of the optimal accumulation path follows from stan-
dard arguments (see Stokey and Lucas (1989)). When c̄ = 0, the optimal
growth problem reduces to the conventional one which is defined by (8) for
all k0 > 0. In this case, denote the capital accumulation map by τ (k).
The equation of optimal capital accumulation where c̄ = 0 has the following
properties (see Stokey and Lucas (1989)).
(i) For all k0 ∈ (0, k m ) there exists a unique steady state, k̃(α) ∈ (0, k m ),
that depends on the discount factor, α.
(ii) At the steady state, k̃ associated with α, the net rate of return on
capital satisfies
4
Note that in principle such a formulation is compatible with growth models with Stone-
σ
Geary preferences u(ct ) = (ct −c̄)
1−σ
−1
which have been used in the literature to model
subsistence consumption. However, in our setting we implicitly assume that marginal
utility of consumption stays finite for c → c̄ whereas under Stone-Geary preferences it
goes to infinity for c → c̄. In spite of this, the characterization of optimal growth paths
for different parameter ranges that we obtain in this subsection corresponds exactly to the
one derived under Stone-Geary preferences with a linear production function (see Rebelo
(1992)). Considering preferences of the form u(c − c̄) would make our analysis more
cumbersome without altering the qualitative insights concerning the impact of adaptive
economizing on the growth paths.
8
f 0 (k̃) − (n + δ) 1−α
ρ(k̃) := = . (10)
1+n α
(iii) All optimal trajectories converge monotonically to the steady state.
What happens when c̄ > 0? To answer this, we must find out when
optimal trajectories generated by τ (k) satisfy the minimal consumption con-
straints. That is, if {ct }∞
0 is an optimal trajectory when c̄ = 0, when does ct
exceed c̄ ?
For inviable societies, we know from Proposition 1 that the capital accu-
mulation map is given by
for all periods where the population can survive. Therefore τ ` (k) in gen-
eral differs from the optimal accumulation map τ (k). In particular, for
k < min[k̃, k̂] we have τ ` (k) = φ(k) < k < τ (k). Consequently, optimal
trajectories would be increasing if c̄ = 0 whereas for c̄ > 0 they are de-
creasing as long as the population survives. Consequently τ (k) does not
characterize the L∗∗ optimal trajectories for c̄ > 0 if the initial capital stock
is small.
Now consider scenarios where k0 > k̂. A first observation is that the
optimal capital accumulation map τ ` (·) has to satisfy
9
Insert figure 2 here!
1
α̂ = . (12)
1 + ρ(k̂)
ˆ ˆ
Note that ρ(k̂) < 0 and therefore k̃ < k̂ holds for all α ∈ [0, 1].
This implies the following relationships concerning k̃(α).
ˆ
Weak Discounting: k̂ ≤ k̃(α) ≤ k̂ for all α ≥ α̂ (a)
(13)
Strong Discounting k̃(α) < k̂ for all α < α̂ (b)
Figure 2 shows the situations that correspond to the two inequalities (13).
When k̃ > k̂, as in figure 2(a), the situation is relatively straight forward.
Basically, L∗∗ optimal trajectories are generated by the mapping
n o
kt+1 = min φ(kt ), τ (kt )
(i) Weak Discounting: α > α̂. There exists a unique wealth k A ∈ [k̂, k̃]
à ∗∗ optimal capital accumu-
such that τ (k A ) = φ(k A ). The equation of L
lation is
10
φ(kt ) , kt ∈ (k̂, k A ]
kt+1 = τ ` (kt ) :=
τ (kt ) , kt ∈ (k A , k m ]
à ∗∗
(ii) Strong Discounting: α < α̂. For all k0 ∈ (k̂, k m ], the equation of L
capital accumulation is
11
4.1 Adaptive Preferences
As before, our Swiss Family Robinson adults have an absolute preference for
survival; first for themselves and next, if their own survival is assured, for fu-
ture generations. They do not compare the discounted utility of all possible
consumption trajectories. Instead, we assume they compare the trade–off be-
tween their family’s current consumption, c, with a level of consumption, c1 ,
that they believe could be enjoyed forever, given the capital stock they leave
to their descendants. This means they compare present consumption with a
constant sequence, c1 , c1 , c1 , . . ., that could be sustained if future generations
maintain their wealth endowment. At the same time, that endowment can
be exploited as the next generation pleases. The lexicographic preference
ordering thus described is represented by the utility function,
u(c)
, if 0 ≤ c < c̄
α
u` (c, c1 ) = u(c̄) + 1−α
u(c1 ) , if c ≥ c̄, 0 ≤ c1 ≤ c̄ , (14)
α
u(c) + 1−α
u(c1 ) , if c ≥ c̄, c1 > c̄
Because the preference ordering does not explicitly take into account all fu-
ture generations, to distinguish this preference ordering from the L∗∗ optimal
case, we refer to feasible trajectories that sequentially maximize (14) as `∗∗
optimal.6
12
which is equivalent to a first order approximation of the production function.
Projected output with the planned capital stock k 1 becomes
f 0 (k) − (n + δ)
ρ(k) = .
1+n
7
While it is usually assumed that producers know their production functions, in reality
they typically understand their current operating conditions but can only estimate output
at points removed from current practice. Large companies have engineering departments
that estimate production relationships while farmers have to guess, sometimes with the
help of experimental studies and advisors. The use of linear approximations of unknown
non–linear relationships is common practice in many planning heuristics used by business
firms, see e.g. Nahmias (1993) for examples in the field of Operations Management. Well
trained economists can substitute econometric methods for the “naive” projection assumed
in (15). Our concern is not with sophisticated methods used by economists but with the
behavior that approximates ordinary individuals and firms. Obviously, to do that we must
know their environment exactly.
13
We denote the capital stock where ρ(k) = 0 by k r . Existence and unique-
ness of this point follows directly from the assumptions about production.
0 ≤ c ≤ f (k). (19)
The solution of this problem is the adaptive `∗∗ consumption strategy,
c = h` (k). (20)
Given the imperfect information about production and limited foresight, the
agents act rationally but boundedly so. The implied `∗∗ adaptive economizing
equation of capital accumulation is
1 h i
kt+1 = θ` (kt ) := (1 − δ)kt + f (kt ) − h` (kt ) . (21)
1+n
For convenience we define θ(kt ) ≡ θ` (kt ) when c̄ = 0.
Guaranteeing sustenance for all future generations can be achieved if the
initial stock is large enough and the subsistence level, c̄, small enough. But,
if the capital endowment is not big enough, all generations may not be able
to reach c̄. Moreover, given that our agents are boundedly rational, they may
not be able to provide consumption c̄ forever even if that level is technically
feasible.
(i) There exists a unique, positive steady state, k̃, which coincides with
the optimal steady state and satisfies (10).
14
(ii) When discounting is strong enough, capital converges to the optimal
steady state, k̃.
A formal proof is found in Day and Lin (1992) or Day (1999, Chapter 17).
Ironically, these results imply that the more the current generation values
the standard of living its heirs can enjoy, the greater the chance of fluctua-
tions and suboptimal savings; whereas, the less the current generation cares
about the future, the more likely consumption will converge to the inter–
temporally optimal path. The reason for this seemingly counter intuitive re-
sult is that the first–order approximation of the production function leads to
an overestimation of the marginal benefits of savings along increasing capital
accumulation paths (but to an underestimation along decreasing capital ac-
cumulation paths). Consequently, the present generation saves more than is
inter–temporally optimal. if the resulting capital stock overshoots the steady
state, the rate of return falls enough so that the next generation saves less
than is optimal; capital declines, the rate of return and investment recover,
and the fluctuations persist.
So what happens when c̄ > 0? To answer this question, the discrepancy
between the estimated production and the realized production for capital
stock k 1 (kt , ct ) must be taken into account. Given (17), the estimated output
over–estimates the realized output at k 1 (kt , ct ) if k 1 (kt , ct ) 6= kt . The future
standard of living will, therefore, be over–estimated as well. This may lead
to a level of savings that the current generation believes will enable future
generations to survive when in fact that is not possible.
15
n o (1 + ρ)(f (k) − (n + δ)k) − c̄
c̃(k) := max c|c1 (k, c) ≥ c̄ = .
ρ
Accordingly, the projected or subjective future survival endowment is
1
`(kt ) := ((1 − δ)k + f (k) − c̃(k)), k > k̄. (22)
1+n
Whenever k 1 (kt , ct ) ≥ `(kt ), the current generation believes that viability for
all future generations is assured. On the other hand, the current generation
can only survive if k 1 (kt , ct ) ≤ φ(kt ). So, all-together the current generation
believes that whenever
`(kt ) ≤ kt+1 ≤ φ(kt )
survival of the current and all future generations is guaranteed. Under which
circumstances is it possible to choose such a kt+1 ? To answer this question
we use the following characterization of the function `(·) which is illustrated
in figure 3.
(ii) `(k) is a continuous function such that for all k 6= k̂, `(k) < k̂.
(iii) There exists a capital stock k B ∈ (k̂, k r ) such that for all k ≥ k B ,
`(k) = 0.
16
A proof is contained in the Appendix.
Points (iv) and (v) of the lemma show that the range of current capital
stocks, kt , where the current generation believes in the survival of the current
and all future generations is [k̂, k m ] and therefore exactly coincides with the
range of current capital stocks where long term survival is indeed possible.
However, under preferences (14) the restrictions on kt+1 imposed by (subjec-
tive) viability considerations for kt ≥ k̂ (i.e. kt+1 ≥ `(kt )) are less stringent
than those needed to guarantee actual survival (i.e. kt+1 ≥ k̂). Conditions (i)
and (ii) imply that only for a single initial capital stock, namely k0 = k̂ is ac-
tual long–run viability guaranteed by subjective viability considerations. For
initial capital stocks above the threshold k̂ the optimal consumption strat-
egy under complete information would guarantee continuing survival. Given
the limited information of the adaptive economizer, however, it is not clear
a priori whether such an outcome will always be achieved. In particular,
if `(kt ) ≤ kt+1 < k̂t the current generation believes that long-run viability
is assured by their actions although the choice of consumption level of this
generation actually buries the prospects of long-run survival.
ct = c̄ , t = 0, . . . , t∗ − 1,
ct∗ = f (kt∗ ) , t = t∗ ,
kt+1 = φ(kt ) , t = 0, 1, . . . , t∗ − 1.
17
If the initial capital stock is sufficient to allow for a viable economy various
possibilities may occur which are illustrated in figure 4.
18
(a) Let k C be the largest capital stock k ≤ k̃ such that φ(k) = θ(k).
Then k̂ < k C < k̃ < k D ≤ k m and `(k) < θ(k) < φ(k) for all
k ∈ (k C , k D ).
(b) For kt ∈ [k̂, k m ] the capital accumulation equation under adaptive
economizing is given by
min[φ(kt ), θ(kt )] , kt ∈ [k̂, k C ]
θ(k )
t , kt ∈ (k C , min[k D , k r ])
kt+1 = θ` (kt ) :=
max[`(kt ), θ(kt )] , kt ∈ [min[k D , k r ], k r )
1−δ
θ(kt ) = 1+n kt , kt ∈ [k r , k m ].
(b) For almost all initial capital stocks k0 ∈ [k̂, k m ] the population dies
after a finite number of periods and paths of capital accumulation
generated by adaptive economizing agents converge towards zero.
The only initial capital stocks where long–run viability is achieved
are those where the path hits k̂ after a finite number of iterations.
19
5 Discussion
Comparing proposition 4 with proposition 2 we realize that the answer to our
question ’Can adaptive economizing guarantee survival when it is possible?’
is often negative. In particular, for strong discounting adaptive economizing
generically leads to eventual demise although long run survival would be
feasible. Even if discounting is sufficiently weak to allow for a steady state
above k̂, long run survival is not guaranteed. Discount factors close to one
imply instability of the steady state and the resulting fluctuations might push
the capital stock below k̂ inducing demise after a finite number of periods.
For initial capital stocks below k̂ this fate is unavoidable but for k0 > k̂ the
early depletion of the capital stock is due to the limited information and the
simplified model of the world used by the decision maker.
Since α̃ increases for increasing c̄ the set of discount factors leading to
unnecessary extinction becomes larger the higher the desired consumption
level c̄ is. Strictly speaking these observations apply only to cases where ini-
tial capital endowment is not too large. A particularly striking implication
of proposition 4 is that in cases of strong capital depreciation, (1−δ)
1+n
k m < k̂,
any sufficiently high initial capital endowment of k0 ≥ k r leads to demise un-
der adaptive economizing regardless of the discount factor and the stability
of the steady state k̃. The high initial endowment leads to strong overes-
timation of the productivity of capital for lower capital stocks and strong
overconsumption by the current generation. It should be noted that the
drastic implications of adaptive economizing discussed here occur although
in our setting the adaptive economizers primary concern is survival of all
generations and they are always able to tell correctly whether the current
capital stock is in principle sufficient for long run survival or not.
Our results are quite surprising for in the standard growth model adaptive
economizing generates paths that are qualitatively similar to the optimal ones
as long as the discount factor is small. Also, intuitively one could expect
that heavy discounting minimizes the effects of prediction errors of future
outputs and imperfect foresight. However, the smaller the discount factor
the larger is current consumption, the more severe is over–prediction of future
production possibilities and the stronger is the resulting over-consumption.
Although all generations are interested in allowing future generations the
minimal consumption level, this can lead to the demise of the population
and gradual decay of the capital stock.
20
The question now arises whether this unnecessary extinction is due to the
simplistic way that the future is considered or to the crude manner of esti-
mating future output. The answer seems to be the latter. Obviously, agents
with perfect knowledge about the production function who are endowed with
a capital stock above the crucial level k̂ would never allow the capital stock
to sink below this level, even if they follow the adaptive economizing strategy
assumed here. On the other hand, Dawid (2005) has shown in a one-sector
growth model framework with standard preferences that, as long as decision
makers re–estimate the production function in a linear way every period,
oscillations about the positive steady state can occur even if the planning
horizon is infinite. Hence, in the scenario with lexicographic preferences, it
is possible that unnecessary extinction could occur even if agents consider
prospective inter–temporal trade–offs over the entire future.
Under the inter–temporally optimal policy an increase in the minimal
level of consumption c̄ has only marginal long run effects if the initial capital
stock is sufficiently large. Our analysis shows that with adaptive economizing
policies such an increase might trigger the long run extinction of capital even
if the initial capital stock is way above the minimal capital stock k̂ which
can sustain this level of consumption. Hence the implications of a minimal
standard of living can indeed be quite dramatic.
Throughout the analysis the population growth rate has been assumed
to be constant. It influences the critical wealth level k̂ as well as the steady
state k̃ and the stability of the adaptive economizing process. If this rate were
endogenized, the results would be modified. Likewise, as duly noted above,
a change in c̄ would influence the possible outcomes if it were determined
culturally rather than biologically. Even in that event, however, there is
surely a lower bound to any reduction in the subsistence level.
As it is, the analysis is suggestive of various extreme situations in which
national or man–made disasters reduce capital stocks enough to precipitate
decline, or in which a society combined low regard for future generations with
an overestimation of labor productivity under reduced per capita capital en-
dowment such that it decumulated its per capita wealth so much that future
viability is endangered. To make this point we have relied on a standard
representative agent framework. Important issues like wealth distribution
call for a heterogenous agents model and have been ignored here. For ex-
ample societies could attempt to redistribute wealth so that some members
survive and be pushed above the critical wealth level k̂. Other related con-
21
siderations such as heterogeneous initial wealth and/or preferences would be
worth exploring.
22
Appendix
Proof of Proposition 2:
(i) First we show that a unique stock k A ∈ [k̂, k̃] exists. It follows from k̃ ≥ k̂
that τ (k̂) ≥ k̂ = φ(k̂). On the other hand, we have τ (k̃) = k̃ ≤ φ(k̃) since
ˆ
k̃ ∈ [k̂, k̂]. Denote by h∗0 (k) the optimal consumption function for c̄ = 0.
Because φ(k) − τ (k) = h∗0 (k) − c̄ ∀ k > k̄ the continuity and monotonicity
of, h∗0 (k), induces monotonicity and continuity of φ(k) − τ (k) on [k̂, k m ] and
this establishes the existence of a unique stock k A with τ (k) = φ(k).
The fact that both h∗0 (k) and τ (k) increase monotonically further implies
that for any path with initial stock larger than k A , which is generated by
τ (k), the inequalities ct ≥ c̄ and kt ≥ k A hold for all t. Thus for k0 > k A
the optimal path of the problem with and without consumption threshold
coincide and we have τ ` (k) = τ (k) on [k A , k m ].
This leaves us with showing that on the interval [k̂, k A ] the optimal con-
sumption is given by the minimal possible value c̄. It follows from the concav-
ity of u and the concavity of the value function of the problem that optimal
consumption cannot decrease with an increasing capital stock. Thus, the
fact that optimal consumption at k A equals c̄ implies that the same has to
hold true on [k̂, k A ].
Finally, it is a standard result in growth theory that τ (k) is monotonically
increasing. Accordingly, τ ` (k) is monotonically increasing on the interval
(k̂, k m ] and any path starting in this interval has to converge monotonically
to the unique steady state k̃ ∈ (k̂, k m ).
(ii) With strong discounting we have k̃ < k̂ and accordingly τ ` (k) has no
ˆ ˆ
fixed point in the interval (k̂, k̂). Furthermore, it is easy to see that k̂
ˆ
cannot be a fixed point of τ ` . Staying at k̂ yields consumption of c̄ ev-
ery period. Higher consumption can for example be generated by consuming
ˆ ˆ
f (k̂) + (1 − δ)k̂ − k̂ > c̄ in the current period, moving to capital stock to k̂,
ˆ ˆ
and consuming c̄ in all following periods. Hence, τ ` (k̂) < k̂ and together we
get τ ` (k) < k ∀ k ∈ (k̂, k m ), which implies that any L∗∗ optimal path with
k0 > k̂ is monotonically decreasing. To ensure long run viability τ l further
has to satisfy τ ` (k) ≥ k̂ for all k ∈ (k̂, k m ] and we immediately get the claims
of (ii).
23
(iii) Follows directly from φ(k̂) = k̂ and the condition k̂ ≤ τ ` (k) ≤ φ(k). 2
Proof of Lemma 1:
(i): For k = k̂ we have c1 (k̂, f (k̂) − (n + δ)k̂) = c̄, thus l(k̂) = k̂.
(ii): Continuity of ` follows from the continuity of c1 (k, c) with respect to k
and its monotonicity with respect to c. Note further that c1 (k, (1 − δ)k +
f (k)−(1+n)k 1 ) = f (k)−f 0 (k)k +(1+n)ρ(k)k 1 is increasing in k 1 for k < k 0 .
Concavity of f implies that agents always overestimate future growth and
hence c1 (k, (1 − δ)k + f (k) − (1 + n)k 1 ) > f (k 1 ) − (n + δ)k 1 for all k 1 6= k.
Since f (k̂) − (n + δ)k̂ = c̄ we have c1 (k, (1 − δ)k + f (k) − (1 + n)k̂) > c̄ and
thus `(k) < k̂ ∀k 6= k̂.
(iii): For current capital stock k = k r and planned capital stock k 1 = 0 we
get for the projected future output y 1 = f (k r ) + f 0 (k r )(0 − k r ) = f (k r ) − (n +
δ)k r > c̄, which implies that `(k r ) = 0 and by continuity and monotonicity
of f (k) − f 0 (k)k the existence and uniqueness of k B < k r with the given
properties follows. Since `(k̂) = k̂ > 0 we must have k B > k̂.
(iv): We first show that `(k) > k for k < k̂. To see this note that c1 (k, f (k) −
(n + δ)k) = f (k) − (n + δ)k increases with k on [0, k r ]. This implies that
c1 (k, f (k) − (n + δ)k) < c̄ for k < k̂ and using the monotonicity of c1 (k, (1 −
δ)k + f (k) − (1 + n)k 1 ) with respect to k 1 establishes `(k) > k. Furthermore,
it follows from k̂ < k r and φ(k̂) = k̂ that φ(k̂) < k for k < k̂. Hence our
result.
(v): Follows directly from φ(k) > k̂ and `(k) < k̂ for all k ∈ (k̂, k m ]. 2
Proof of Proposition 3:
For k0 < k̄ the claim of the proposition follows directly from the lexicographic
structure of the preferences u` . For k̄ ≤ k0 < k̂ we know from lemma 1 that
`(k0 ) > k0 > φ(k0 ). Accordingly, c1 (k0 , c̄) < c̄ and due to the lexicographic
structure of u` current consumption is reduced to the minimal level c̄. 2
24
Proof of Proposition 4:
(i):
(a): For α > α̂ we have k̃ > k̂ and therefore φ(k̃) > θ(k̃) = k̃ > k̂ > `(k̃).
Furthermore, θ(k̂) > k̂ = φ(k̂). Continuity of `(k), φ(k), θ(k) establishes that
k̂ < k C < k̃ < k D , where θ(k) > `(k) might hold on the entire interval [k̃, k m ]
in which case we set k D = k m .
(b): The optimization problem of the adaptive economizing agent for k > k̂
can be written as
α
maxk1 u((1 − δ)k + f (k) − (1 + n)k 1 ) + 1−α
u((1 − δ)k + f (k) − (1 + n)k 1 )
h i
1−δ
s.t. max 1+n
k, `(k) ≤ k 1 ≤ φ(k). (23)
1−δ
The optimal solution to this problem under the weaker constraint 1+n k≤
1
k 1 ≤ 1+n ((1 − δ)k + f (k)) is given by k 1 = θ(k). Hence, whenever θ(k)
satisfies (23) we have θ` = θ. Concavity of the objective function implies
further that whenever θ(k) lies outside the range given by (23) the optimal
solution lies on the corresponding boundary of the interval. For all k ∈ [k̂, k̃)
we have θ(k) > k ≥ k̂ ≥ `(k) and by definition we have θ(k) > `(k) for all
k ∈ [k̃, k D ). Therefore, θ(k) ≥ `(k) ∀k ∈ [k̂, k D ]. Comparing φ(k) and θ(k)
we first observe that by definition φ(k) ≥ θ(k) for all k ∈ [k C , k̃]. To show
that φ(k) ≥ θ(k) holds also for k ∈ [k̃, k m ] we show that the adaptive `∗∗
consumption strategy h` (k) for c̄ = 0 is non-decreasing in k on [k̃, k m ]. We
denote this consumption strategy by ha (k). First note that θ(k) < k and
therefore ha (k) > f (k) − (n + δ)k for all k > k̃. Furthermore, if ha (k) < f (k)
the first order condition
u0 (ha (k))
αρ(k)) 0
− u (f (k) − (n + δ)k + ρ(k)(f (k) − (n + δ)k − ha (k)))
1−α
= 0 (24)
has to hold. This implies that for all capital stocks where ha (k) < f (k) we
must have ρ(k) > 0. For any k where ρ(k) ≤ 0 we have ha (k) = f (k) and
1−δ
therefore θ(k) = 1+n k < φ(k). Since `(k) = 0 ∀k ≥ k B and k B < k r we get
1−δ
θ` (k) = θ(k) = 1+n k < φ(k) for all k ∈ [k r , k m ].
If ρ(k) > 0 we get by implicit differentiation of (24)
αρu00 ((1+ρ)ρ(1+n)+ρ0 (f −(n+δ)k−ha ))+αu0 ρ0
ha 0 (k) = (1−α)u00 +αρ2 u00
. (25)
25
To simplify notation we have omitted the arguments of all functions in this
expression. Both the numerator and the denominator are negative, where
the negativity of the numerator follows from (f − (n + δ)k − ha ) < 0 and
ρ(k) > 0. Therefore, we have ha 0 (k) > 0 for all values of k where θ(k) > 1+n1−δ
k.
1−δ
Accordingly, φ(k) − θ(k) increases with k as long as θ(k) > 1+n k and since
1−δ m
φ(k m ) > 1+n k we have shown that φ(k) > θ(k) for all k ∈ [k B , k m ]. The
claims of (b) now follow directly.
(c) Follows directly from (b) since θ` = θ in the neighborhood of k̃.
(d) Due to l(k̂) = φ(k̂) the adaptive economizing problem (23) has k 1 = k̂ as
the only admissible solution for k = k̂ and therefore θ` (k̂) = k̂ and k̂ is a fixed
point of θ` . It is however easy to see that k̂ is an unstable fixed point and
therefore only accumulation paths hitting k̂ after a finite number of periods
converge to k̂. Since θ` (k) = θ(k) in the neighborhood of k̃, instability of k̃
with respect to θ yields instability with respect to θ` . Since `(k) < k̂ for all
k > k̂ the resulting fluctuating capital accumulation paths might eventually
hit a capital stock below k̂. In this case there will be demise with finite
survival time as described in point (iv) of proposition 2.
(ii):
(a): Since k̃ < k̂ we have θ(k̂) < k̂ = l(k̂). On the other hand we have
θ(k B ) > 0 = `(k B ). Therefore k D has to lie in the interval k̂, k B and due to
continuity `(k) > θ(k) also holds for all capital stocks in the interval [k̂, k D ).
Thus, θ` (k) = `(k). For θ(k) > `(k) the consumption threshold is not bind-
ing and we have θ` (k) = θ(k). Note that ha (k̂) > c̄ and the arguments
given in the proof of point (b) in part (i) show that current consumption
under θ(k) is therefore larger than c̄ for k ∈ [k̂, k m ]. This establishes that
θ` = max(θ(k), `(k)) on [k D , k m ]. Finally, analogous arguments as in the
case of weak discounting show that for k ∈ [k r , k m ] we have `(k) = 0 and
1−δ
θ(k) = 1+n k.
(b): Because θ` (k) < k holds for all capital stocks except k = k̂ we con-
clude that any path with k0 > k̂ which does not hit k̂ after a finite number
of periods either has to converge to k̂ from above or enter the region [0, k̂].
We know from part (ii)(a) of this proposition and part (ii) of lemma 1 that
θ` (k) = `(k) < k̂ for k > k̂ in the neighborhood of k̂ which rules out conver-
gence of a path towards k̂ from above. Hence, every path that does not hit k̂
after a finite number of periods eventually has to enter [0, k̂) which according
to proposition 3 induces within a finite number of periods consumption below
26
the minimal level and therefore the collapse of the economy. 2
27
References
Chaterjee, S. and Ravikumar, B., 1999. Minimal Consumption Require-
ments: Theoretical and Quantitative Implications for Growth and Distribu-
tion. Macroeconomic Dynamics 3, 482-505.
Dawid, H., 2005. Long Horizon versus Short Horizon Planning in Dynamic
Optimization Problems with Incomplete Information. Economic Theory 25,
575-597.
Day, R., 1999. Complex Economic Dynamics, Volume II. MIT Press, Cam-
bridge, MA.
Easterly, W., 1994. Economic Stagnation, Fixed Factors, and Policy Thresh-
olds. Journal of Monetary Economics 33, 525-557.
28
Fudenberg, D. and Levine, D.K., 1998. The Theory of Learning in Games.
MIT Press, Cambridge, MA.
Kirman, A., 1975. Learning by Firms about Demand Conditions, in: R. Day
and T. Groves (Eds.), Adaptive Economizing Models. Academic Press, NY,
pp. 137-156.
Nahmias, S., 1993. Production and Operations Analysis. Irwin, Boston, MA.
29
Figure Captions
Figure 1: Maximal capital accumulation trajectories given minimum con-
sumption, c̄ > 0.
30
kt+1
[(1-δ)kt+f(kt)]/(1+n)
kt+1=kt
φ(kt)
_ kt
k ^k ^^
k
Figure 1
kt+1
[(1-δ)kt+f(kt)]/(1+n)
kt+1=kt
φ(kt)
τl(kt)
τ(kt)
_ kt
^ A ~
k k k k
Figure 2a
kt+1
[(1-δ)kt+f(kt)]/(1+n)
kt+1=kt
φ(kt)
τl(kt)
τ(kt)
_ kt
~ ^
k k kA k
Figure 2b
kt+1
[(1-δ)kt+f(kt)]/(1+n)
kt+1=kt
φ(kt)
θ(kt)
^
k
l(kt)
_ kt
k ^ k B
k
Figure 3
kt+1
[(1-δ)kt+f(kt)]/(1+n) kt+1=kt
φ(kt)
θ(kt)
θl(kt)
l(kt)
_ kt
^ B ~
kk k k
Figure 4a
kt+1
[(1-δ)kt+f(kt)]/(1+n) kt+1=kt
φ(kt)
θ(kt)
θl(kt)
l(kt)
_ kt
^ ~ B
k k kk
Figure 4b
kt+1
[(1-δ)kt+f(kt)]/(1+n)
kt+1=kt
φ(kt)
θ(kt)
θl(kt)
l(kt)
_
^ ~ kt
k k k kB
Figure 4c
kt+1
[(1-δ)kt+f(kt)]/(1+n) kt+1=kt
φ(kt)
θ(kt)
θl(kt)
l(kt)
_ kt
^ k B
k~ k
k
Figure 4d